Transportation Technologies for Sustainability

This volume collects selected topical entries from the Encyclopedia of Sustainability Science and Technology
(ESST). ESST addresses the grand challenges for science and engineering today. It provides unprecedented,
peer-reviewed coverage of sustainability science and technology with contributions from nearly 1,000 of the
world’s leading scientists and engineers, who write on more than 600 separate topics in 38 sections.
ESST establishes a foundation for the research, engineering, and economics supporting the many
sustainability and policy evaluations being performed in institutions worldwide.

Editor-in-Chief
ROBERT A. MEYERS, RAMTECH LIMITED, Larkspur, CA, USA

Editorial Board
RITA R. COLWELL, Distinguished University Professor, Center for Bioinformatics and Computational
Biology, University of Maryland, College Park, MD, USA
ANDREAS FISCHLIN, Terrestrial Systems Ecology, ETH-Zentrum, Zürich, Switzerland
DONALD A. GLASER, Glaser Lab, University of California, Berkeley, Department of Molecular & Cell
Biology, Berkeley, CA, USA
TIMOTHY L. KILLEEN, National Science Foundation, Arlington, VA, USA
HAROLD W. KROTO, Francis Eppes Professor of Chemistry, Department of Chemistry and Biochemistry,
The Florida State University, Tallahassee, FL, USA
AMORY B. LOVINS, Chairman & Chief Scientist, Rocky Mountain Institute, Snowmass, USA
LORD ROBERT MAY, Department of Zoology, University of Oxford, Oxford, OX1 3PS, UK
DANIEL L. MCFADDEN, Director of Econometrics Laboratory, University of California, Berkeley, CA, USA
THOMAS C. SCHELLING, 3105 Tydings Hall, Department of Economics, University of Maryland, College
Park, MD, USA
CHARLES H. TOWNES, 557 Birge, University of California, Berkeley, CA, USA

EMILIO AMBASZ, Emilio Ambasz & Associates, Inc., New York, NY, USA
CLARE BRADSHAW, Department of Systems Ecology, Stockholm University, Stockholm, Sweden
TERRY COFFELT, Research Geneticist, Arid Land Agricultural Research Center, Maricopa, AZ, USA
MEHRDAD EHSANI, Department of Electrical & Computer Engineering, Texas A&M University, College
Station, TX, USA
ALI EMADI, Electrical and Computer Engineering Department, Illinois Institute of Technology, Chicago,
IL, USA
CHARLES A. S. HALL, College of Environmental Science & Forestry, State University of New York,
Syracuse, NY, USA
RIK LEEMANS, Environmental Systems Analysis Group, Wageningen University, Wageningen,
The Netherlands
KEITH LOVEGROVE, Department of Engineering (Bldg 32), The Australian National University,
Canberra, Australia
TIMOTHY D. SEARCHINGER, Woodrow Wilson School, Princeton University, Princeton, NJ, USA
Mehrdad Ehsani, Fei-Yue Wang and Gary L. Brosch (Eds.)

Transportation Technologies
for Sustainability

With 826 Figures and 172 Tables

Editors
Mehrdad Ehsani
Department of Electrical & Computer Engineering
Texas A&M University
College Station, TX, USA

Fei-Yue Wang
Institute of Automation
Chinese Academy of Sciences
Beijing, China

Gary L. Brosch
Center for Urban Transportation Research
University of South Florida
Tampa, FL, USA

ISBN 978-1-4614-5843-2 ISBN 978-1-4614-5844-9 (eBook)

ISBN 978-1-4614-5845-6 (print and electronic bundle)
DOI 10.1007/978-1-4614-5844-9
Springer New York Dordrecht Heidelberg London
Library of Congress Control Number: 2012953371

© Springer Science+Business Media New York 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the
rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known
or hereafter developed. Exempted from this legal reservation are brief excerpts in connection with reviews or scholarly analysis or material
supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Duplication of this publication or parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location, in its
current version, and permission for use must always be obtained from Springer. Permissions for use may be obtained through RightsLink at the
Copyright Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of
a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors
nor the publisher can accept any legal responsibility for any errors or omissions that may be made. The publisher makes no warranty, express or
implied, with respect to the material contained herein.

This book consists of selection from the Encyclopedia of Sustainability Science and Technology edited by Robert A. Meyers, published by
Springer New York in 2012.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Editors

Electric, Hybrid, and Fuel Cell Vehicles Mass Transit Science and Technology

MEHRDAD EHSANI (also one of the board members) GARY L. BROSCH

Department of Electrical & Computer Engineering Center for Urban Transportation Research
Texas A&M University University of South Florida
College Station, TX Tampa, FL
USA USA

Intelligent Vehicles Technology

FEI-YUE WANG
Institute of Automation
Chinese Academy of Sciences
Beijing
China
Sections

Electric, Hybrid and Fuel Cell Vehicles

Editor: Mehrdad Ehsani
AC Machines: Permanent Magnet Synchronous and Induction Machines
Batteries, Battery Management, and Battery Charging Technology
Battery Technologies
Electric, Hybrid, and Fuel Cell Vehicles, Introduction
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
Energy Storage: Ultracapacitor
Fuel-Cell-Powered HEV Design and Control
Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
Hybrid Energy Storage Systems for Vehicle Applications
Internal Combustion Engines, Alternative Fuels for
PHEVs and BEVs in Coupled Power and Transportation Systems
Plug-in Hybrid Electric Vehicles
Regenerative Braking
Sustainable Transportation
Sustainable Vehicle Fuels, Well-to-Wheel Analysis
Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV
Vehicle Biofuels
Vehicle Dynamics and Performance
Vehicle Energy Storage: Batteries
Vehicle Traction Motors

Intelligent Vehicle Technology

Editor: Fei-Yue Wang
3D Pose Estimation of Vehicles Using Stereo Camera
Active Multifocal Vision System, Adaptive Control of
Active Pedestrian Protection System, Scenario-Driven Search Method for
Cooperative Group of Vehicles and Dangerous Situations, Recognition of
Driver Assistance System, Biologically Inspired
Driver Assistance Systems, Automatic Detection and Site Mapping
Driver Behavior at Intersections
Driver Characteristics Based on Driver Behavior
Driver Inattention Monitoring System for Intelligent Vehicles
Driving Under Reduced Visibility Conditions for Older Adults
Dynamic Environment Sensing Using an Intelligent Vehicle
Intelligent Vehicles Technology, Introduction
Night Vision Pedestrian Warning in Intelligent Vehicles
True Color Night Vision Video Systems in Intelligent Vehicles
viii Sections

Unscented Kalman Filter in Intelligent Vehicles

Vehicle Detection, Tightly Coupled LIDAR and Computer Vision Integration for
Vehicular Ad Hoc Networks, Enhanced GPSR and Beacon-Assist Geographic Forwarding in

Mass Transit Science and Technology

Editor: Gary L. Brosch
Advanced Public Transport Systems, Simulation-Based Evaluation
Bicycle Integration with Public Transport
Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
Bus Rapid Transit, Institutional Issues Related to Implementation
Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental, and Planning Aspects
Bus Versus Rail Implications for Transit-Oriented Development
High Speed Rail, Technology Development of
High-Occupancy Vehicle and Toll Lanes
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit, Shared Infrastructural Issues
Light Rail Transit, Systemic Viability
MAGLEV Technology Development
Mass Transit Science and Technology, Introduction
Personal Rapid Transit and Its Development
Transit-Oriented Development and Land Use
Table of Contents

3D Pose Estimation of Vehicles Using Stereo Camera

Björn Barrois, Christian Wöhler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
AC Machines: Permanent Magnet Synchronous and Induction Machines10.1007/978-1-4419-0851-3_870
A. H. Ranjbar, Babak Fahimi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Active Multifocal Vision System, Adaptive Control of10.1007/978-1-4419-0851-3_472
Alois Unterholzner, Hans-Joachim Wuensche . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Active Pedestrian Protection System, Scenario-Driven Search Method for10.1007/978-1-4419-0851-3_486
Alberto Broggi, Pietro Cerri, Stefano Ghidoni, Paolo Grisleri, Ho Gi Jung . . . . . . . . . . . . . . . . . . . . . . . 74
Advanced Public Transport Systems, Simulation-Based Evaluation
Haris N. Koutsopoulos, Moshe Ben-Akiva . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Batteries, Battery Management, and Battery Charging Technology
Robert S. Balog, Ali Davoudi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Battery Technologies
Ramakrishnan Mahadevan, Mohit Chhabra, Puneet Pasrich, Frank Barnes . . . . . . . . . . . . . . . . . . . . 158
Bicycle Integration with Public Transport10.1007/978-1-4419-0851-3_490
John Pucher, Ralph Buehler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
Cliff Henke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Bus Rapid Transit, Institutional Issues Related to Implementation
Mark A. Miller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
Dario Hidalgo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and
Planning Aspects
Vukan R. Vuchic, Richard M. Stanger, Eric C. Bruun . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
Bus Versus Rail Implications for Transit-Oriented Development
Ming Zhang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Cooperative Group of Vehicles and Dangerous Situations, Recognition of10.1007/978-1-4419-0851-3_792
Kym Watson, Christian Frese, Thomas Batz, Jürgen Beyerer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
Driver Assistance System, Biologically Inspired
Moustapha Doumiati, Daniel Lechner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
Driver Assistance Systems, Automatic Detection and Site Mapping
Andreas Wimmer, Klaus C. J. Dietmayer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
Driver Behavior at Intersections
Toshihisa Sato, Motoyuki Akamatsu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
Driver Characteristics Based on Driver Behavior
Jianqiang Wang, Lei Zhang, Xiaojia Lu, Keqiang Li . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
x Table of Contents

Driver Inattention Monitoring System for Intelligent Vehicles

Yanchao Dong, Zhencheng Hu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
Driving Under Reduced Visibility Conditions for Older Adults
Rui Ni . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
Dynamic Environment Sensing Using an Intelligent Vehicle
Huijing Zhao, Long Xiong, Yiming Liu, Xiaolong Zhu, Yipu Zhao, Hongbin Zha . . . . . . . . . . . . . . . . . 438
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
Yimin Gao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
Electric, Hybrid, and Fuel Cell Vehicles, Introduction
Mehrdad Ehsani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492
Energy Storage: Ultracapacitor
Andrew F. Burke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494
Fuel-Cell-Powered HEV Design and Control
Etim U. Ubong, Jim Gover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
High Speed Rail, Technology Development of
Eda Beyazit, Moshe Givoni . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537
High-Occupancy Vehicle and Toll Lanes
Katherine F. Turnbull . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
Douglass B. Lee, Jr., Kent M. Hymel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
M. Abul Masrur, Vijay K. Garg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593
Hybrid Energy Storage Systems for Vehicle Applications
Mehrdad Ehsani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 614
Intelligent Vehicles Technology, Introduction
Fei-Yue Wang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627
Internal Combustion Engines, Alternative Fuels for
Thomas Wallner, Scott A. Miers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
Light Rail Transit in the US and Abroad, Examination of History and Innovations
José I. Farrán . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 667
Light Rail Transit, Shared Infrastructural Issues
Rongfang Rachel Liu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683
Light Rail Transit, Systemic Viability
Gregory L. Thompson, Jeffrey R. Brown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 704
MAGLEV Technology Development
James R. Powell, Gordon Danby . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 753
Mass Transit Science and Technology, Introduction
Gary L. Brosch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 804
Night Vision Pedestrian Warning in Intelligent Vehicles
Matthias Serfling, Otto Löhlein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 806
Personal Rapid Transit and Its Development
Shannon S. McDonald . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 831
 Table of Contents xi

PHEVs and BEVs in Coupled Power and Transportation Systems10.1007/978-1-4419-0851-3_824

Mladen Kezunovic, S. Travis Waller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 851
Plug-in Hybrid Electric Vehicles10.1007/978-1-4419-0851-3_818
Mehrdad Ehsani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 870
Regenerative Braking
Yimin Gao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 882
Sustainable Transportation
Mehrdad Ehsani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 890
Sustainable Vehicle Fuels, Well-to-Wheel Analysis
Amgad Elgowainy, Michael Wang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 911
Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV
Mehrdad Ehsani . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 939
Transit-Oriented Development and Land Use
Robert Cervero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947
True Color Night Vision Video Systems in Intelligent Vehicles
Dirk Hertel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 959
Unscented Kalman Filter in Intelligent Vehicles
Moustapha Doumiati, Daniel Lechner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 991
Vehicle Biofuels
Mark Holtzapple . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1006
Vehicle Detection, Tightly Coupled LIDAR and Computer Vision Integration for
Lili Huang, Matthew Barth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1033
Vehicle Dynamics and Performance
Yimin Gao . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1060
Vehicle Energy Storage: Batteries
Y. S. Wong, C. C. Chan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1082
Vehicle Traction Motors
C. C. Chan, Ming Cheng . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1103
Vehicular Ad Hoc Networks, Enhanced GPSR and Beacon-Assist Geographic Forwarding in
Ping Wang, Fuqiang Liu, Wenjie Shu, Wenqiang Xu . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1133
List of Contributors

MOTOYUKI AKAMATSU MOSHE BEN-AKIVA

Human Technology Research Institute, National Department of Civil and Environmental Engineering
Institute of Advanced Industrial Science and MIT
Technology (AIST) Cambridge, MA
Tsukuba-shi, Ibaraki USA
Japan

EDA BEYAZIT
ROBERT S. BALOG Transport Studies Unit, School of Geography and the
Department of Electrical and Computer Engineering Environment
Texas A & M University University of Oxford
College Station, TX Oxford
USA UK

FRANK BARNES JÜRGEN BEYERER

Electrical, Computer, and Energy Engineering Institute for Anthropomatics, Vision and Fusion
Department Laboratory
University of Colorado Karlsruhe Institute of Technology
Boulder, CO
Karlsruhe
USA
Germany
and
BJÖRN BARROIS Fraunhofer Institute of Optronics, System
EvoBus GmbH Technologies and Image Exploitation IOSB
Neu-Ulm Karlsruhe
Germany Germany

MATTHEW BARTH ALBERTO BROGGI

Department of Electrical Engineering VisLab – Dipartimento di Ingegneria
University of California, Riverside dell’Informazione
CA Università degli Studi di Parma
USA Parma
Italy

THOMAS BATZ
Department Information Management GARY L. BROSCH
Fraunhofer Institute of Optronics, System Center for Urban Transportation Research
Technologies and Image Exploitation IOSB University of South Florida
Karlsruhe Tampa, FL
Germany USA
xiv List of Contributors

JEFFREY R. BROWN MING CHENG

Department of Urban and Regional Planning School of Electrical Engineering
Florida State University Southeast University
Tallahassee, FL Nanjing
USA China

ERIC C. BRUUN MOHIT CHHABRA

Department of Electrical & Systems Engineering Electrical Engineering
University of Pennsylvania University of Colorado
Philadelphia, PA Boulder, CO
USA USA

GORDON DANBY
RALPH BUEHLER Maglev-2000 Corporation
School of Public and International Affairs Wading River, NY
College of Architecture and Urban Studies Virginia USA
Tech
Alexandria, VA
ALI DAVOUDI
USA
University of Texas
Arlington, TX
ANDREW F. BURKE USA
Institute of Transportation Studies, University of
California-Davis KLAUS C. J. DIETMAYER
Davis, CA Institute of Measurement, Control, and
USA Microtechnology
University of Ulm
Ulm
PIETRO CERRI
Germany
VisLab – Dipartimento di Ingegneria
dell’Informazione
Università degli Studi di Parma YANCHAO DONG
Parma Graduate School of Science and Technology
Italy Kumamoto University
Kumamoto
Japan
ROBERT CERVERO
Department of City and Regional Planning MOUSTAPHA DOUMIATI
University of California B2i Automotive Engineering Company
Berkeley, CA Massy
USA France

C. C. CHAN MEHRDAD EHSANI

Department of Electrical and Electronic Engineering Department of Electrical & Computer Engineering
University of Hong Kong Texas A&M University
Hong Kong College Station, TX
China USA
 List of Contributors xv

AMGAD ELGOWAINY JIM GOVER

Center for Transportation Research Department of Electrical Engineering
Argonne National Laboratory Kettering University
Argonne, IL Flint, MI
USA USA

BABAK FAHIMI PAOLO GRISLERI

University of Texas VisLab – Dipartimento di Ingegneria
Arlington, TX dell’Informazione
USA Università degli Studi di Parma
Parma
JOSÉ I. FARRÁN Italy
Adavant Consulting
San Francisco, CA CLIFF HENKE
USA Parsons Brinckerhoff
Newark, NJ
CHRISTIAN FRESE USA
Institute for Anthropomatics, Vision and Fusion
Laboratory DIRK HERTEL
Karlsruhe Institute of Technology Quincy, MA
Karlsruhe USA
Germany
DARIO HIDALGO
YIMIN GAO EMBARQ, The World Resources Institute Center for
Department of Electrical and Computer Engineering Sustainable Transport
Texas A&M University Washington, DC
College Station, TX USA
USA
MARK HOLTZAPPLE
VIJAY K. GARG Department of Chemical Engineering
Dublin, OH Texas A&M University
USA College Station, TX
USA
STEFANO GHIDONI
Intelligent Autonomous System Laboratory LILI HUANG
(IAS-Lab), Department of Information Engineering Department of Electrical Engineering
University of Padua University of California
Padova Riverside, CA
Italy USA

MOSHE GIVONI ZHENCHENG HU

Transport Studies Unit, School of Geography and the Graduate School of Science and Technology
Environment Computer Science Department
University of Oxford Kumamoto University
Oxford Kumamoto
UK Japan
xvi List of Contributors

KENT M. HYMEL KEQIANG LI

US Department of Transportation State Key Laboratory of Automotive Safety and
Volpe National Transportation Systems Center Energy
Cambridge, MA Tsinghua University
USA Beijing
China

HO GI JUNG FUQIANG LIU

The School of Mechanical Engineering Department of Information and Communication
Hanyang University Engineering
Seondong-gu, Seoul Tongji University
South Korea Shanghai
China

MLADEN KEZUNOVIC RONGFANG RACHEL LIU

Department of Electrical and Computer Engineering Department of Civil and Environmental Engineering
Texas A&M University New Jersey Institute of Technology
College Station, TX Newark, NJ
USA USA

YIMING LIU
HARIS N. KOUTSOPOULOS Key Lab of Machine Perception (MOE)
Department of Transport Sciences Peking University
The Royal Institute of Technology, KTH Beijing
Stockholm China
Sweden
XIAOJIA LU
China Agricultural University
DANIEL LECHNER Beijing
Department of Accident Mechanism Analysis China
IFSTTAR-MA Laboratory
Salon de Provence RAMAKRISHNAN MAHADEVAN
France Telecommunications
University of Colorado
Boulder, CO
DOUGLASS B. LEE, JR. USA
US Department of Transportation
Volpe National Transportation Systems Center M. ABUL MASRUR
Cambridge, MA College of Engineering & Science, University of
USA Detroit Mercy
Detroit, MI
USA
OTTO LÖHLEIN
Machine Perception Systems and Reliability SHANNON S. MCDONALD
Environment Perception Daimler AG Southern Illinois University
Ulm Carbondale, IL
Germany USA
 List of Contributors xvii

SCOTT A. MIERS A. H. RANJBAR

Mechanical Engineering Department of Electrical Engineering
Michigan Technological University Amirkabir University of Technology
Houghton, MI Tehran
USA Iran

MARK A. MILLER TOSHIHISA SATO

California Partners for Advanced Transit and Human Technology Research Institute, National
Highways (PATH) Program, Institute of Institute of Advanced Industrial Science and
Transportation Studies Technology (AIST)
University of California Tsukuba-shi, Ibaraki
Berkeley, CA Japan
USA
and
MATTHIAS SERFLING
Institute of Transportation Studies, School of
Machine Perception Systems and Reliability
Public Affairs
Environment Perception Daimler AG
University of California
Ulm
Los Angeles, CA
Germany
USA

WENJIE SHU
RUI NI
Department of Information and Communication
Department of Psychology
Engineering
Wichita State University
Tongji University
Wichita, KS
Shanghai
USA
China

PUNEET PASRICH RICHARD M. STANGER

Electrical, Computer, and Energy Engineering Los Angeles, CA
Department USA
University of Colorado
Boulder, CO
USA GREGORY L. THOMPSON
Department of Urban and Regional Planning
Florida State University
JAMES R. POWELL Tallahassee, FL
Maglev-2000 Corporation USA
Shoreham, NY
USA
KATHERINE F. TURNBULL
Department of Landscape Architecture & Urban
JOHN PUCHER Planning, Texas Transportation Institute
Bloustein School of Planning and Public Policy College of Architecture – The Texas A&M University
Rutgers University System
New Brunswick, NJ College Station, TX
USA USA
xviii List of Contributors

ETIM U. UBONG MICHAEL WANG

Center for Fuel Cell Research and Powertrain Center for Transportation Research
Integration, Electrical & Computer Eng Argonne National Laboratory
Kettering University Argonne, IL
Flint, MI USA
USA
PING WANG
ALOIS UNTERHOLZNER Department of Information and Communication
Institute for Autonomous Systems Technology Engineering
Department of Aerospace Engineering Tongji University
University of the Bundeswehr Munich Shanghai
Neubiberg China
Germany
KYM WATSON
VUKAN R. VUCHIC Department Information Management
Department of Electrical & Systems Engineering Fraunhofer Institute of Optronics, System
University of Pennsylvania Technologies and Image Exploitation IOSB
Philadelphia, PA Karlsruhe
USA Germany

ANDREAS WIMMER
S. TRAVIS WALLER
Institute of Measurement, Control, and
Department of Civil Engineering
Microtechnology
The University of Texas at Austin
University of Ulm
Austin
Ulm
USA
Germany

THOMAS WALLNER CHRISTIAN WÖHLER

Energy Systems Division Image Analysis Group
Argonne National Laboratory Dortmund University of Technology
Argonne, IL Dortmund
USA Germany

FEI-YUE WANG Y. S. WONG

Institute of Automation State Grid Energy Research Institute
Chinese Academy of Sciences State Grid Corporation of China
Beijing China
China

JIANQIANG WANG HANS-JOACHIM WUENSCHE

State Key Laboratory of Automotive Safety and Institute for Autonomous Systems Technology
Energy Department of Aerospace Engineering
Tsinghua University University of the Bundeswehr Munich
Beijing Neubiberg
China Germany
 List of Contributors xix

LONG XIONG MING ZHANG

Key Lab of Machine Perception (MOE) Graduate Program in Community and Regional
Peking University Planning
Beijing School of Architecture, The University of Texas
China at Austin
Austin, TX
WENQIANG XU USA
Department of Information and Communication
Engineering HUIJING ZHAO
Tongji University Key Lab of Machine Perception (MOE)
Shanghai Peking University
China Beijing
China

HONGBIN ZHA
Key Lab of Machine Perception (MOE) YIPU ZHAO
Peking University Key Lab of Machine Perception (MOE)
Beijing Peking University
China Beijing
China
LEI ZHANG
State Key Laboratory of Automotive Safety and XIAOLONG ZHU
Energy Key Lab of Machine Perception (MOE)
Tsinghua University Peking University
Beijing Beijing
China China
 3D Pose Estimation of Vehicles Using Stereo Camera 1

3D Pose Estimation of Vehicles space. Projecting the scene flow into the image
plane yields the optical flow.
Using Stereo Camera Stereo vision 3D reconstruction of a scene based on
BJÖRN BARROIS1, CHRISTIAN WÖHLER2 a pair of images.
1
EvoBus GmbH, Neu-Ulm, Germany Yaw angle Angle measuring the rotation of a vehicle
2
Image Analysis Group, Dortmund University of around its vertical axis.
Technology, Dortmund, Germany Yaw rate Temporal derivative of the yaw angle.

Definition of the Subject

Article Outline
Glossary
Definition of the Subject
Introduction and State of the Art
Discussion
A System for Stereo-Based 3D Pose Estimation of
Vehicles
Experimental Evaluation
Summary and Conclusion
Future Directions
Bibliography
Glossary
3D Pose estimation Estimation of the translational
and rotational degrees of freedom of an object,
usually corresponding to its position in 3D space
and its orientation angles. For articulated objects
additional, internal degrees of freedom are
estimated.
Driver assistance system A system which helps the
driver of a car in specific traffic situations in order
to increase safety.
Iterative closest point (ICP) algorithm An algo-
rithm which finds a set of translation and rotation
parameters fitting a measured point cloud to
a model.
Optical flow The optical flow vector denotes the
motion in image space of the projection of a scene
point between two subsequent images of
a sequence.
Quaternion Number system which extends the system
of complex numbers. In computer vision and
computer graphics, quaternions are used for the
efficient representation of rotations.
Scene flow Vector field describing the 3D scene points
visible in the image and their current motion in 3D
For advanced driver assistance systems, the 3D poses and
motion states of oncoming and intersecting vehicles
represent important information. This work describes
methods for 3D vehicle pose estimation based on
a motion-attributed 3D point cloud generated. First,
stereo and optical flow information is computed for
the investigated scene. A four-dimensional clustering
approach separates the static from the moving objects
in the scene. The iterative closest point algorithm (ICP)
estimates the vehicle pose using a cuboid as a weak
vehicle model. Classical ICP optimization is based on
the Euclidean distance metric. Its computational effi-
ciency can be significantly increased by applying
a quaternion-based optimization scheme. In vehicle-
based small-baseline stereo systems, it is favorable to
use a polar distance metric which especially takes into
account the error distribution of the stereo measure-
ment process. To derive the pose parameters and simul-
taneously their temporal derivatives, i.e., the motion
state of the vehicle, from two subsequent stereo image
pairs in order to avoid a temporal filtering stage,
a model-based scene flow method is applied. Here,
3D points on the object surface are reprojected into
the stereo images and similarities between image win-
dows extracted around the reprojected points are max-
imized. An experimental evaluation is performed on
seven different real-world sequences, where different
stereo algorithms, baseline distances, distance metrics,
and optimization algorithms are examined. The results
show that the proposed polar distance metric yields
a higher accuracy especially for the estimation of the
yaw angle of a vehicle than the common Euclidean
distance metric, especially when using pixel-accurate
stereo points. The model-based scene flow approach
yields a refined 3D pose estimation along with the
instantaneous motion state of the vehicle.

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
2 3D Pose Estimation of Vehicles Using Stereo Camera
Introduction and State of the Art
Some approaches for detection of obstacles around the
ego-vehicle can be found in the literature, but the
existence of an obstacle is not enough information to
evaluate the situation entirely. The knowledge about
pose and motion of oncoming and intersecting vehicles
is important to avoid accidents. Especially at intersec-
tions, analysis of the behavior of other road users is
important to help the driver to pass through the
intersection in a safe and comfortable manner. Further-
more, pose information is important for the recogni-
tion and prediction of driving maneuvers and the
analysis of complex traffic situations involving several
traffic participants. This section provides an overview
of the state of the art in the field of vehicle pose
estimation, where approaches relying on different
sensor concepts are discussed.
Systems Based on Monocular Images
A system for segmentation and detection of vehicles
using a monocular camera system is described in [1],
where a vehicle model defined by seven parameters is
used. The position of the vehicle is determined by
taking into account the symmetry of the edges in the
image, the shape of the model, and the shadow of the
vehicle. An evaluation of this approach is not available.
In [2] a Kalman filter–based system for tracking
vehicles in front of the ego-vehicle is presented.
The contours of other vehicle candidates are extracted
using a Hough transform and circumscribed by a
bounding box. Furthermore, a classifier determines if
the bounding box corresponds to a vehicle or a back-
ground pattern. A 3D model is used for tracking the
vehicles in 3D space, where the measurement is
interpreted as the projection of the 3D model into the
2D image plane, which yields the 3D position of the
vehicle. System evaluation is performed based on radar
data, where an accuracy of the distance estimation
of the image-based approach of better than 1 m at
a distance of 60 m is achieved.
Early methods for vehicle detection in the field
of traffic surveillance are proposed in [3–5]. These
approaches are based on observing a road intersection
with a fixed camera. The vehicles are represented by
polyhedral models which are adapted to the edges
extracted from the images. A 3D scene model provides
information about free space and possible occlusions.
An illumination model is used to suppress the influ-
ence of shadows on the pose estimation. To obtain
a prediction of the motion behavior, a description of
the possible motion patterns based on a physical model
is used. The vehicle is tracked across the sequence using
optical flow analysis combined with an extended
Kalman filter.
Systems Based on Sensor Fusion
A system consisting of a camera and a radar sensor is
used in [6] to detect traffic participants. The advan-
tages of the two sensors (high lateral resolution of the
camera, high accuracy of the distances and radial
velocities measured by the radar sensor) are used to
estimate the position of vehicles driving ahead of the
ego-vehicle. The distance is measured based on radar
information while the lateral extension of the object is
inferred from the camera image. The evaluation of the
system is restricted to the accuracy of the estimated
bounding box in the image.
The combination of a monocular color camera and
a radar sensor is proposed in [7]. Shadows, rear lights,
line segments, and symmetric features are extracted
from the image and combined with the radar informa-
tion in a low-level fusion step. A subsequent post-
processing step aims for removing false detections
and determining the vehicle positions. An evaluation
of the metric accuracy of the system is not available.
A laser scanner and a video camera are used in [8]
to estimate the pose, size, and velocity of vehicles. The
detection and 3D pose estimation of vehicles is
obtained based on the laser scanner data. Relying on
the pose estimation result, rectified side views of the
detected vehicles are constructed, which are in turn
used by a cascade classifier to distinguish vehicles in
front of the ego-vehicle from background patterns. The
evaluation in [8] is limited to the classification perfor-
mance of the system, while the geometric detection
accuracy is not examined.
Similarly, a combination of a laser scanner and
a camera is used in [9] for the recognition of objects
in traffic scenes. This approach is based on sensor
fusion relying on vehicle-specific features extracted
from both sensors. Besides the application of

autonomously following the vehicle driving ahead,
traffic jam and road intersection scenarios are exam-
ined. It is shown that the proposed system is able to
track vehicles even in the presence of high
accelerations.
Stereo Camera Systems
The early approach to stereo-based vehicle detection
described in [10] relies on a stereo camera system with
a relatively long baseline approximately corresponding
to the width of the ego-vehicle. Vehicles driving behind
the ego-vehicle are extracted from the 3D point cloud,
and a Kalman filter is applied, where the highway
scenario is addressed specifically. Due to the long base-
line, an estimation of the vehicle position is possible for
distances of 100 m and more.
Stereo information and optical flow data are com-
bined in [11] to determine the position and the velocity
of vehicles. The 3D points and the extracted optical
flow field are processed simultaneously by an extended
Kalman filter framework. Objects are inferred from the
scene along with their velocities using a clustering tech-
nique. A considerable settling time of the temporal
filtering stage is observed. The accuracy of the proposed
method is not evaluated.
A fairly complex system relying on 3D scene recon-
struction based on the fusion of structure from motion
data and stereo vision is described in [12]. A detection
step yields the objects present in the scene along with
their approximate orientation in discrete steps and an
assignment to different object classes using the
approach introduced in [13]. The results are reliable
in a depth range between 15 and 30 m. The pose of the
objects is not compared to ground truth data.
A stereo camera is used in [14] to detect vehicles
driving ahead and estimate their distance to the ego-
vehicle. In a first step, a stereo-based 3D reconstruction
of the scene is performed. A symmetry analysis extracts
rear views of vehicles from one of the camera images,
and a bounding box is computed. The vehicle is then
tracked in the 2D image space and associated with the
3D data in a separate step.
Stereo analysis based on affine warping is used in
[15] to detect obstacles ahead of the ego-vehicle. The
image of the left camera is warped and mapped onto
the image of the right camera using affine warping and
3D Pose Estimation of Vehicles Using Stereo Camera 3
edge comparison. Dynamic programming estimates
a boundary between the road surface and obstacles on
the road. However, the orientations of the obstacles are
not estimated.
In [16], a stereo camera system is used for several
applications, e.g., the detection and tracking of vehicles
and pedestrians. An occupancy grid distinguishing
between free and occupied regions in front of the
vehicle is created using the 3D points of an initial stereo
analysis to detect objects in front of the car. Once an
object is found, its size, aspect ratio, and orientation are
used to find a corresponding model in an object model
database. The most suitable 3D model from the data-
base is projected into the image and fitted to the image
of the object using chamfer matching. The distance up
to which this approach yields reliable results is
about 35 m. No evaluation of the pose estimation
accuracy is given.
The approach introduced in [17] uses a sparse scene
flow field which is established by a point-wise Kalman
filter–based integration of stereo and optical flow
information. An extended Kalman filter tracks the
point cluster representing the vehicle. The resulting
motion-attributed 3D point cloud is segmented
according to their position and motion state, where
the independently moving parts of the scene are
regarded as objects. The objects are tracked using
a second, object-specific Kalman filter stage. An explicit
vehicle model is not utilized, but the size of the vehicle
is estimated based on the size of the point cloud. The
proposed system is evaluated on synthetically rendered
image sequences. The settling time of the Kalman filter
framework typically amounts to about 25 frames
(cf. also [11]). An extension of this approach is pro-
posed in [18], where the observed object trajectories are
analyzed by a particle filter stage in order to obtain
a prediction of the vehicle motion for time intervals
of several seconds. Example images and pose estima-
tion results are shown in Fig. 1.
Three approaches to the adaptation of a temporal
filtering stage are discussed in [19] for vehicle tracking
based on stereo information. Specifically, an
interacting multiple model (IMM), a maximum likeli-
hood estimation of the filter parameters based on the
yaw rate, and an extended model taking into account
the yaw acceleration are discussed. The proposed
tracking approaches are evaluated using synthetically
4 3D Pose Estimation of Vehicles Using Stereo Camera
3D Pose Estimation of Vehicles Using Stereo Camera. Figure 1
Tracking results obtained with the stereo-based method described in [17, 18] for an oncoming vehicle first detected at
48 m distance. The predicted driving path indicating curve radius and velocity for the next second and the tracked 3D
points are shown as well as a bounding box indicating the object pose (from [18])
generated images as well as 3D point clouds syntheti-
cally generated based on real trajectories of the vehicle
center. On these data sets, the positional accuracy of
the system amounts to 0.1–0.2 m and the accuracy
of the filtered yaw rate to about 2–6
per second.
Discussion
In many state-of-the-art systems, the orientation angle
of the vehicle remains undetermined or needs to be
inferred from the specific application scenario, e.g.,
when vehicles driving ahead of the ego-vehicle are
detected. Other systems estimate the orientation angle
by a temporal filtering stage, which has the disadvan-
tage of delays occurring due to the finite settling time.
For an accurate prediction of the motion behavior the
direct measurement of the vehicle orientation is essen-
tial. Furthermore, metric evaluations of the estimated
poses in comparison to independently determined real-
world reference data can hardly be found in the existing
literature, such that it is difficult or impossible to
compare different existing approaches with respect to
measurement accuracy.
A general disadvantage of systems using monocular
cameras is the low accuracy of the resulting distance
measurements, while systems relying on sensor fusion
require a considerable instrumental and calibration
effort. In principle, a laser scanner is a sensor which is
well suited for measuring the 3D pose of vehicles due to
its high lateral resolution and distance accuracy, but
laser scanners still tend to be too expensive for com-
mercial automotive systems. Hence, stereo cameras are
a promising cost-efficient alternative as they provide
both a high lateral resolution and a reasonable accuracy
of the measured distances in ranges relevant for typical
traffic scenarios.
A System for Stereo-Based 3D Pose Estimation
of Vehicles
This section describes approaches for 3D pose estima-
tion of vehicles at high accuracy without relying on
temporal filtering. A tracking-by-detection approach
is performed where the pose estimation is performed
individually for each frame. First, the input images
from a calibrated camera system are rectified to stan-
dard geometry with epipolar lines parallel to the image
rows. A sparse scene flow field is computed to obtain
reliable depth information combined with motion
information about the investigated scene, where three
different algorithm combinations are used to compute
a sparse incomplete scene flow field, i.e., a motion-
attributed 3D point cloud. Separate objects are extracted
by clustering, where each point is defined by three
spatial coordinates and one or two velocity coordi-
nates. The 3D points associated with the object are
used by an iterative closest point (ICP) method to
determine the object pose, where the classical
 3D Pose Estimation of Vehicles Using Stereo Camera 5

1 1

0.9 0.9
Intensity

Intensity
0.8 0.8

0.7 0.7
0.6 0.6

0.5 0.5
20 20
15 15
10 20 10 20
v,t [pixels] 15 v,t [pixels] 15
a 5 10 b 5 10
0 0 5 0 5
u [pixels] 0 u [pixels]

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 2

(a) Spatiotemporal intensity profile of a moving edge, measured over a time interval of three time steps. The size of the
spatiotemporal matching window is 21  7  3 pixels. For visualization, the (v, t) axis is divided such that each time step
comprises an interval of seven pixels. (b) Modeling result according to Eq. 1 (from [37])

Euclidean distance or a polar distance metric can be
used. In a last step, a model-based scene flow stage
yields a refined estimation of the pose parameters and
their instantaneous temporal derivatives.
Extraction of Depth and Motion
Spacetime Stereo (adopted from [20] and [36]) The
spacetime stereo algorithm is based on local intensity
modeling and yields 3D points with the associated
motion component parallel to the epipolar lines [20].
Image regions corresponding to a sufficiently high ver-
tical intensity gradient are extracted in the left and right
camera image, and their local spatiotemporal neigh-
borhood is modeled by the model function hð~ P; u; v; tÞ
where u and v denote the pixel coordinates, t the time
coordinate in a spatiotemporal region of interest, and ~
the vector of function parameters:
hð~
P; u; v; tÞ ¼ p1 ðv; tÞ tanh ½p2 ðv; tÞu þ p3 ðv; tÞ
þ p4 ðv; tÞ
The tanh function approximates the shape of an
ideal edge blurred by the optical system. The functions
pi(v, t) are polynomials in v and t of (in principle)
arbitrary order. In the experiments described in section
“Experimental Evaluation,” linear polynomials were
used. Correspondence analysis is then based on
a comparison of the modeled edges in the left and the
right image in terms of the sum of squared differences
(SSD). The SSD values on each epipolar line are
analyzed, where different constraints can be taken
into account: uniqueness, ordering, or the minimum
weighted matching constraint [21]. Analysis of the
model function parameters yields the velocity compo-
nent along the epipolar line and sub-pixel-accurate
disparity values. An example of the intensity profile
of an observed moving edge versus a moving edge
modeled according to Eq. 1 is shown in Fig. 2.
Feature-Based Stereo and Optical Flow (adopted
from [36]) The utilized feature-based method for
computing stereo and optical flow is based on local
P
features to obtain a representation of the investigated
scene [22]. The features are established in both images
(left and right camera or current and previous time
step), then the comparison is performed using a hash
table technique. This approach provides a fast imple-
ð1Þ
mentation of the algorithm. The correspondences have
pixel accuracy, which is a disadvantage in comparison
with the other regarded stereo algorithms. The combi-
nation of the stereo and optical flow results yields
3D points associated with the motion parallel to the
image plane.
6 3D Pose Estimation of Vehicles Using Stereo Camera

Correlation Stereo and Feature-Based Optical Flow Clustering

(adopted from [36]) The combination of correla-
This section provides an extended description of the
tion-based stereo with the feature-based optical flow
clustering technique outlined in [36]. An initial seg-
technique yields optical flow and more precise depth
mentation of the attributed 3D point cloud extracted
information in video real time. The correlation-based
with any of the sparse scene flow techniques (cf. Fig. 3a)
stereo method performs a SSD comparison of left and
is obtained by means of a graph-based unsupervised
right image patches [23]. An interest operator (Prewitt
clustering technique [24] in a space of 4, 5, or 6
filter) is used to separate reliable from unreliable depth
dimensions spanned by the three spatial coordinates
information. The combination yields 3D points with
and the extracted velocity components of the
two-dimensional motion vectors.

30 30

25 25

20 20
z[m]
z[m]

15 15

10 10

5 5

0 0
–10 –5 0 5 10 –10 –5 0 5 10
b x [m] c x [m]

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 3

Example of an incomplete sparse scene flow field. (a) Original image with projected 3D points (green) and associated
motion vectors (red). (b) Bird’s-eye view of the same scene. (c) Clustered points belonging to the vehicle (blue) together
with the adapted model (black) (adapted from [36])
 3D Pose Estimation of Vehicles Using Stereo Camera 7

3D points [25]. This clustering stage generates a scene- traversed, indices are assigned to the clusters based on
dependent number of clusters, essentially separating a recursive algorithm.
the moving object from the (stationary or differently Important advantages of this clustering algorithm
moving) background. in the context of vehicle detection are its high compu-
To avoid the need for a parameter that weights tational efficiency, its simple parameterization, and
spatial against velocity coordinates, the graph-based the fact that the overall number of clusters does not
method in [24] is used which takes into account the need to be known in advance. The nontrivial prob-
spatial and velocity coordinates separately. Accord- lem of appropriately weighting spatial against veloc-
ingly, for two scene flow points ity coordinates does not occur. Since only velocity
differences between neighboring points are regarded,
0 1 0 1
x1 x2 the algorithm avoids a decomposition of rotating
B y1 C B y2 C objects such as turning vehicles into different
B C B C
~ B z1 C B z2 C
P1 ¼ B x_ 1 C and ~
P2 ¼ B x_ 2 C; ð2Þ clusters – as a result of the velocity components
B C B C
@ y_ 1 A @ y_ 2 A varying continuously along the object axis, such
z_ 1 z_ 2 a decomposition would occur for many classical
clustering approaches which rely on single scalar
the distance measure ~
d ¼ ðde ; dv ÞT of the cluster- distance metrics.
ing algorithm is a vector with two components
given by Pose Estimation Based on the ICP Algorithm
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi The outline in this section of the ICP algorithm is
de ¼ ðx1  x2 Þ2 þ ðy1  y2 Þ2 þ ðz1  z2 Þ2 adopted from [25] and [36]; it is extended by the
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi description of a computationally efficient quater-
dv ¼ ðx_ 1  x_ 2 Þ2 þ ðy_ 1  y_ 2 Þ2 þ ð_z 1  z_ 2 Þ2 :
nion-based optimization approach. A classical
ð3Þ method to register a point cloud to a geometric
object model is the iterative closest point (ICP)
The clustering approach is based on two thresholds algorithm introduced in [27]. Given an initial esti-
ye (for the spatial distance) and yv (for differences in mate of the object pose, the pose parameters are
velocity). Two scene flow points can be assigned to updated by minimizing the mean squared distance
the same cluster if de < ye and dv < yv. Hence, it is between the scene points and the model, and the
not necessary to combine the different features into point assignment is repeated accordingly. This
a single scalar value as it would be necessary for procedure is applied in an iterative manner. As
classical clustering approaches such as the k-means a result, the algorithm yields the three-dimensional
algorithm [26]. object pose. In [27], this method is applied to the
The method in [24] relies on a graph-based technique registration of point sets, curves, and surfaces. How-
to establish clusters, where the scene flow points are ever, the approach can only be used in situations
regarded as the nodes and the distances between them as where all scene points belong to the object as it is
the edges of the graph. The points are sorted according to a pose estimation rather than a scene segmentation
the norms of their associatedvelocities, and the resulting technique.
sorted list S ¼ ~ P 1 ; :::; ~
P n is traversed completely. In the ICP algorithm proposed in [28], the scene
For each point ~ Pi ¼ ðxi ; yi ; zi ; x_ i ; y_ i ; z_ i ; ÞT with points as well as the object model are represented as sets
i = 1, . . ., n its distances according to Eq. 3 to the of chained points. During each iteration step, the pose
subsequent points in the sorted list are computed. If parameters are updated, and at the same time some
the velocity distance dv to one of the subsequent points scene points are assigned to the object model while
falls below the threshold yv , the search is terminated. If others are rejected, based on the distance to the object
de < ye and dv < yv , the two points are assigned to the and the similarity of the tangent directions of the scene
same cluster. Once the complete sorted list has been and model curves. Thus, outliers in the point cloud
8 3D Pose Estimation of Vehicles Using Stereo Camera

z z

pmi
rmi

pi ri

ji – jm
i

x (0,0) x
a b

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 4

Illustration of the Euclidean and the polar distance metric

data are automatically rejected, and the algorithm is is defined by four real numbers qa, qb, qc, and qd
robust with respect to disappearing and reappearing according to
object parts as well as partial occlusions. As a result, the q_ ¼ qa þ iqb þ jqc þ kqd : ð5Þ
algorithm yields the subset of scene points belonging to
The last three summands in Eq. 5 can be seen as
the object, i.e., a scene segmentation, along with the
three “imaginary parts” analogous to the imaginary
object pose.
part of a complex number [31]. Similar to complex
numbers, it is i2 = j2 = k2 = 1. Unit quaternions are
Euclidean Distance Metric (adopted from [36]) For
quaternions of norm 1, i.e., k q_ k¼ 1. Quaternions
pose estimation of vehicles, a cuboid is utilized as
allow a much more efficient representation of rota-
a weak geometric model representing several vehicle
tions in 3D space than classical rotation matrices,
types. An initial pose is estimated based on the centroid
especially when several rotations are performed sub-
and the first principal component of the vehicle point
sequently. Calculation rules for quaternions are
cloud. Afterward, the ICP algorithm fits the geometric
described in [31].  
model to the point cloud. Thus, the translational pose
The measured 3D point cloud is denoted by ~ w pi ,
parameters tx and tz and the yaw angle y are updated by
the set of corresponding model points by fw ~mi g, where
minimizing the mean squared distance between the
i = 1, . . ., n with n as the number of measured
scene points and the model. The distance di is the
3D points. Subtraction of the centroids yields the
perpendicular between the visible model side and the n o
3D point in Euclidean space (cf. Fig. 4a), where ~ pi zero-mean transformed 3D point clouds ~ w 0pi and
n o
~
denotes the 3D point and pmi the corresponding point
w 0mi . It is shown in [30] that the determination of
~
on the model plane:
the transformation (translation and rotation) between
di ðtx ; tz ; yÞ ¼ k ~
pi  ~
pmi k : ð4Þ two point clouds requires the determination of the unit
To obtain the vehicle pose, a nonlinear minimiza- quaternion q_ which maximizes the expression
X n  
tion of the error function with respect to the pose
q_ w_ 0pi q
_ w_ 0mi ð6Þ
parameters tx, tz, and y is performed using the i¼1
Levenberg–Marquardt algorithm or the Downhill Sim-
Applying the calculation rules for quaternions to
plex algorithm [29].
expression (6) yields

Quaternion-Based Optimization A closed-form q_ T N q_ ð7Þ

solution for the problem of relative orientation _ where
to be maximized with respect to q,
between two point clouds with known correspon- X
n
dences is proposed in [30], where quaternions are N¼ RpTi Rmi : ð8Þ
used for direct parameter estimation. A quaternion q_ i¼1
 3D Pose Estimation of Vehicles Using Stereo Camera 9

In Eq. 8, the matrices Rpi and Rmi are given by of corresponding points on the model surface by min-
0 1 imizing the mean-squared Euclidean distance.
0 x 0 pi y 0 pi z 0 pi
To “imitate” the behavior of an M-estimator in the
B x0p 0 z 0 pi y 0 pi C
B i C quaternion-based optimization scheme, each point  is 
Rpi ¼ B 0 C
@ y pi z 0 pi 0 x 0 pi A weighted with the fair weighting function wf di
ð0Þ
ð0Þ
z 0 pi y 0 pi x 0 pi 0 prior to the optimization procedure, where di is the
0 0 0 0 1 ð9Þ initial Euclidean distance of the i-th 3D point from the
0 x mi y mi z mi
B x0m 0 z 0 mi y 0 mi C model surface. The fair weighting function is given by
B i C
Rmi ¼ B 0 C:
@ y mi z 0 mi 0 x 0 mi A 1
wf ðxÞ ¼ ð11Þ
z 0 mi y 0 mi x 0 mi 0 1 þ jxj
c

The unit quaternion q_ that maximizes expression with c as a user-defined parameter. The corresponding
emax corresponding
(8) is then given by the eigenvector~ error function Equat then amounts to
to the largest eigenvalue of N. The characteristic poly- N h  i2
X ð0Þ
nomial is of degree 4 and can therefore be solved Equat ¼ k~
pi  ~
p m i k w f di : ð12Þ
analytically. Based on q_ ¼ ~ emax ¼ ðq0 ; qx ; qy ; qz ÞT , i¼1

the rotation matrix between the 3D point clouds
n o n o
w 0pi and w
~ ~0mi is given by
0 1
q02 þ qx2 þ qy2  qz2 2ðqx qy  q0 qz Þ
B 2ðqy qz þ q0 qx Þ C
R ¼ @ 2ðqy qx þ q0 qz Þ q02 qx2 þ qy2  qz2
2ðqz qx þ q0 qy Þ 2ðqz qy þ q0 qx Þ
This analytical approach is computationally highly
efficient. For noisy data, however, it suffers from
a low robustness as it inherently yields a solution defined
by the least-mean-squared Euclidean distance between
the points, which is strongly influenced by outliers.
Furthermore, when an object model defined by planes
is used, the point-to-plane matching can only be
performed once before determining the transformation.
As a consequence, it is favorable to combine the
analytical optimization technique according to [30]
with the ICP approach in order to obtain an algorithm
which is faster than the nonlinear optimization scheme
but also robust with respect to outliers. The initial pose
is again determined based on the centroid and the first
principal component of the point cluster representing
the moving object. In a first step, point-to-plane
matching is performed to obtain the closest point ~ pmi
on the model surface for each 3D point ~ pi , where the
Euclidean distance between ~ pi and ~ pmi is denoted by
di ¼ k ~pi  ~
pmi k . The previously described quater-
nion-based technique according to [30] yields an ana-
lytical solution for the rotation and translation
parameters between the 3D point cloud and the cloud
Once the pose refinement step has been performed,
remaining outliers are eliminated from the point cloud.
The quaternion-based optimization procedure is iter-
2ðqx qz þ q0 qy Þ ated once or twice, where the points are reweighted
A: according to Eq. 12 for each iteration. The behavior of
q02  qx2  qy2 þ qz2
the quaternion-based optimization is similar to that of
ð10Þ a direct nonlinear minimization of the Euclidean dis-
tances with an M-estimator error function as long as
the pose refinement determined by that operation is
small.
For the real-world data regarded in section “Exper-
imental Evaluation” a general observation is that if the
original, unweighted 3D points are used in the quater-
nion-based optimization scheme, the resulting pose
accuracy strongly degrades especially for the yaw
angle. The computation time of the quaternion-based
optimization scheme is about 40 times lower than that
of the nonlinear ICP methods (cf. section “Computa-
tional Efficiency”).
Polar Distance Metric (adopted from [36]) The ICP
algorithm with the Euclidean metric does not take into
account the properties of the stereo measurement pro-
cess. When reconstructing 3D points from two cam-
eras, the relation between the disparity and the depth is
nonlinear. This results in a low accuracy for points
which are far away from the camera, while points
close to the camera have a higher accuracy. The noise
of the point positions is neither Gaussian nor
symmetric. The Euclidean metric used in the standard
10 3D Pose Estimation of Vehicles Using Stereo Camera

ICP assumes at least a symmetric error distribution of
the measurements. To overcome this problem one
might think of transforming the geometric model
into disparity space (pixel coordinates and disparities),
which may be very complex depending on the model. It
is thus preferable to use a distance metric which takes
into account the nonlinearity of the measurement
error.
The 3D points are represented in polar coordinates
(cf. Fig. 4b)pon the xz plane,
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi where the distance is
rðx; y; zÞ ¼ x 2 þ y 2 þ z 2 and the polar angle
amounts to ’ = arctan(x/z). Assuming an ideal pinhole
camera model, the pixel coordinates (u, v) can be
transformed into the 3D coordinates according to
u/f = x/z and v/f = y/z, where f denotes the camera
constant measured in pixel units. Assuming a standard
epipolar geometry, the depth coordinate z can be
calculated by z = bf/d, where b is the baseline distance
in meters and d the disparity in pixels. Using these
equations, the distance to the camera r(u, v, d) is
given by
s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  2  2ffi
bu 2 bv bf
rðu; v; dÞ ¼ þ þ
d d d zi2
measurement error dr largely depends on the disparity
error dd. Accordingly, dr is approximately proportional
to z2.
Hence, this error analysis yields the result that the
normalized measurement error dr/z2 has an approxi-
mately Gaussian distribution. The error distribution of
the polar angle ’ is also approximately of Gaussian
shape. This results in an error function E for the
model fit which consists of an error term regarding
the distance of a 3D point to the model and a second
error term regarding the polar angle difference, com-
bined by the weight factor l:
N h i
X
E ðfri g; fzi g; f’i gÞ ¼ Er2 ðri ; zi Þ þ lE’2 ð’i Þ :
i¼1
ð16Þ
The first error term Er only depends on distances
from the camera, where ri is the distance of the object
point i to the camera and rmi the distance to the camera
of the intersection point of the same line of sight with
the model:
ri  rmi
: ð13Þ Er ðri ; zi Þ ¼ : ð17Þ

An error calculation based on the total differential The second error term E’ only depends on the
dr is used to analyze the relations of the depth mea- polar angles:
surement process: ’i  ’M
E’ ð’i Þ ¼ ð1 þ tanh jaðj’i  ’M j  bÞjÞ:
2
@r @r @r
dr ¼ du þ dv þ dd ð14Þ ð18Þ
@u @v @d
The hyperbolic tangent enforces continuity of the
Converting these terms into three-dimensional
angular error function and is required for the conver-
world coordinate expressions results in the following
gence behavior of the optimization. The value ’M
relations:
denotes the polar angle of the center point of the
@r x z model. The parameter b corresponds to about half
¼
@u f r the angular width of the model projected into the
@r y z image plane, whereas a is a user-defined parameter
¼ ð15Þ
@v f r which influences the optimization behavior. Error
@r z3 function (16) can be minimized in the same manner
¼ : as the error function involving the Euclidean metric
@d bfr
using a nonlinear optimization method.
The term z/r is smaller than or equal to 1. The terms
x/f, y/f, and z/f all have the same order of magnitude. The Computational Efficiency For the ICP approaches
measurement errors du and dv of the pixel coordinates u, described in sections “Euclidean Distance Metric” and
v, and the disparity measurement error dd are of the “Polar Distance Metric” involving standard nonlinear
order 0.1–1 pixels and independent of z and r. In our minimization techniques, the repeatedly performed
scenario, one can assume that z/b  1, whereby the point-to-plane matching and distance evaluation may
 3D Pose Estimation of Vehicles Using Stereo Camera 11

lead to a significant computational burden of the ICP

algorithm with polar coordinates. The C++ implemen-
tation requires about 1 ms computation time on a sin-
gle 2 GHz core for 100 points, where the computation
time increases linearly with the number of points. In
the application scenario of section “Experimental Eval-
uation,” the object is typically represented by 200–400
points. The computational efficiency of the quater-
nion-based optimization scheme (cf. section “Quater-
nion-Based Optimization”) is about 40 times higher
than that of the nonlinear ICP methods.

Model-Based Scene Flow

In this section, a temporal extension of the concept of
model-based stereo is described, which is termed

i=1

model-based scene flow. This method aims for
a computation of the 3D pose of the object and an
instantaneous estimation of its temporal derivative
using two subsequent stereo image pairs, thus avoiding
a temporal filtering stage. To initialize the lateral velocity
t_x of the vehicle, information provided by the optical
flow field of the right camera is used. The velocity t_z
along the depth axis and the yaw rate y_ are initially
unknown, such that their values are initialized by zero.
The basic idea of model-based scene flow estima-
tion is to use the model surface together with the pose
parameters and their temporal derivatives to obtain
the homographies that determine image points
corresponding to each other in all four images. Pose
and motion parameters are used to connect images
taken at the same time by different cameras and images
acquired at different times, respectively (cf. Fig. 5).
Image windows of predefined size around the K
reprojected points are extracted from the four images
(cf. Fig. 6). Each of the four images Iit is identified by
two indices i and t, where the right image is denoted
by the index i = 0 and the left image by i = 1, while the
previous time step is denoted by the index t = 0 and the
current time step by t = 1.
The 3D model is assumed to be composed of
planes, which is always the case for tessellated models
like the cuboid used in section “Experimental Evalua-
ðkÞ
tion.” The projective vector ~qi0 t 0 of the pixel coordinates
of window k in image i0 at time step t0 is obtained from
ðkÞ
the corresponding projective vector ~qit in image i at
i 0 t 0 ðkÞ
time step t by the homography it H induced by the
i=0
3D Pose Estimation of Vehicles Using Stereo Camera.
Figure 5
Principle of model-based scene flow. Circles: 3D points,
triangles: corresponding points on the model surface
model plane on which the center of window k is
located. It is thus
ðkÞ 0 0 ðkÞ
~qi0 t 0 ¼ iitt H ðkÞ ~qit ð19Þ
0 0
with k = 1, . . ., K. The homography iitt H ðkÞ depends on
the known extrinsic and intrinsic camera parameters
and on the pose at time steps t and t0 of the model plane
on which patch k is located, which follows from the 3D
pose parameters of the object and their temporal
derivatives.
A detailed description of the concept of the
homography induced by a plane is given in [32]. Effec-
0 0
tively, the determination of the homography iitt H ðkÞ
between the views of camera i at time step t and camera
i0 at time step t0 relies on the following procedure
(cf. Fig. 5): The first step is the computation of the
ðkÞ
intersection point ~s t , at time step t between the ray
ðkÞ
that belongs to the image point ~qi0 t 0 and the model
plane at time step t on which the center of window
ðkÞ
k is located. If t 6¼ t0 , the point ~s t , which is a 3D point
in the scene, has to be adjusted according to the object
motion between the time steps t and t0 , resulting in the
12 3D Pose Estimation of Vehicles Using Stereo Camera

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 6

Determination of optical flow (F), disparity (D), and scene flow (S)

ðkÞ
point ~s t 0 . This adjustment consists of a rotation and
a translation readily inferred from the temporal
ðkÞ ðkÞ
derivatives of the pose parameters. If t = t0 , ~s t 0 ¼ ~s t .
ðkÞ 0
The point ~s t 0 , is then projected into camera i , resulting
ðkÞ
in the corresponding image point ~qi0 t 0 .
The error function Emsf minimized by the model-
based scene flow algorithm consists of the sum over the
six pairwise similarity measures between the projected
image patches according to
X     
ðkÞ ðkÞ
Emsf ¼ S Iit ~qit ; Ii0 t 0 ~qi0 t 0
k;i;t;i 0 6¼i;t 0 6¼t
 
ðkÞ
where S denotes the similarity measure, Iit ~qit the
ðkÞ
window around the point ~qit in image i at time step t,
 
ðkÞ
and Ii0 t 0 ~qi0 t 0 the corresponding window in image i0 at
0 0
time step t 0 , warped by the homography iitt H ðkÞ.
Warping is required to account for the difference in
perspective between the views.
The zero-mean sum of squared differences (SSD)
and the normalized cross-correlation coefficient
(NCC) are used as the similarity measure S. A Sobel
edge detector is used as an interest operator to select
windows which display a significant amount of con-
trast. The parameters describing the 3D pose and its
temporal derivatives are obtained by minimizing
(SSD) or maximizing (NCC) Eq. (20) using the
Gauss–Newton algorithm. If only one stereo image
pair is regarded, optimization of Eq. 20 yields a pose
estimation according to the concept of model-based
stereo.
The computational efficiency of the model-based
scene flow is typically comparable to that of the ICP
approach with classical optimization (i.e., without qua-
ternions) as described in sections “Euclidean Distance
ð20Þ Metric” and “Polar Distance Metric.”
Experimental Evaluation
This section presents an evaluation of the ICP-based
approach and the model-based scene flow stage. It
partially draws upon the description in [36]. Three
color cameras with a resolution of 1,034  776 pixels
were used for image acquisition. The cameras were
mounted side by side at different displacements,
resulting in three different baseline distances b of
0.102, 0.228, and 0.380 m. The frame rate of the
sequences is about 14 fps. The color information is
only used for ground truth calculation, whereas for
the stereo algorithms the images were converted to

grayscale. Several colored markers have been attached
to the vehicle, and a color classifier was used to extract
the markers in the images. Afterward, the corner posi-
tions in each image were estimated by a highly accurate
corner detector [33] at sub-pixel accuracy. The corner
positions were used for bundle adjustment to compute
the three-dimensional world coordinates of each
corner point. Four markers were attached to each side
of the vehicle, such that the corresponding plane is
overdetermined. The colored markers hardly display
a contrast in the grayscale images and therefore do
not generate additional 3D points or optical flow
vectors.
The system performs a frame-by-frame 3D pose
estimation without any temporal filtering (tracking by
detection). The 3D point cloud is clustered to obtain
the scene part which represents the vehicle. The first
principal component of this point cloud is used to
initialize the angular pose parameter.
A set of seven different scenes, each of which was
recorded with three different stereo baseline distances, is
used. These sequences display a moving vehicle on a road
crossing in typical intersection situations, either passing
straight through the intersection or turning left or right
at different velocities, where the distance of the object
from the camera amounts to 15–30 m. (The images and
reference data are publicly available at http://aiweb.
techfak.uni-bielefeld.de/content/vehicle-pose-data-set.)
The evaluation is based on two geometric indicators:
the yaw angle difference between the true yaw angle
and the pose estimation result and the mean distance
of the ground truth points to the corresponding
model planes. The differences are visualized using
error bars denoting the median of the full sequence
and the 25% and the 75% quantiles, respectively.
For the images rectified to standard stereo geome-
try, the effective principal distance amounts to f = 1,350
pixels, such that bf = 307.8 m pixel for the medium
baseline distance of b = 0.228 m. According to [33], the
difference between two checkerboard corner positions
(i.e., a disparity value) can be measured with an accu-
racy (here corresponding to one-half of the difference
between the 25% and the 75% quantiles) of Dd = 0.032
pixels. At a typical object distance from the camera of
z0 = 20 m, the disparity thus amounts to d0 = bf/z0 =
15.39
pffiffiffi pixels, resulting in a depth difference accuracy of
2½bf =d0  bf =ðd0 þ DdÞ ¼ 0:059 m (cf. law of error
3D Pose Estimation of Vehicles Using Stereo Camera 13
propagation). The distance in the scene between the
pairs of markers attached to the frontal and the rear
part of the vehicle amounts to about 3 m, such that the
lowest accuracy of the yaw angle, which occurs when
the vehicle is observed laterally, corresponds to 1.1
.
The absolute depth and yaw angle errors are propor-
tional to z02 .
Evaluation of the ICP-Based 3D Pose Estimation
This section provides an evaluation of the ICP-based
approach with respect to the influence of the baseline
distance of the stereo camera system, the employed
stereo algorithm, the distance metric, and the optimi-
zation algorithm used for minimizing the error func-
tion with respect to the pose parameters. In Figs. 7–12,
the median values and 25% and 75% quantiles
of the deviations of the inferred pose parameters
from their ground truth values are determined
(1) based on all images of each sequence (upper
diagram) and (2) based on all images in which the
extension of the 3D point cloud in x direction is at
least 20% larger than the vehicle width (lower dia-
gram). The second evaluation has been added because
the pose estimation result is expected to be rather
unreliable for frontal and rear views of the vehicle
due to the resulting small number of 3D points on its
longer side.
Baseline Distance (adopted from [36]) The first part
of the evaluation analyzes the different baseline dis-
tances of the sequences. Figure 7 shows the yaw angle
error and Fig. 8 the distance error for the three different
baseline distances for all seven sequences using the
three different stereo approaches described. The com-
parison between the different baseline distances shows
that the smallest baseline is not sufficient for our appli-
cation. Especially for vehicles turning left or right in
front of the camera (sequences 3 and 4) the errors are
large. The largest baseline also produces larger errors,
since due to the large camera displacement and the
resulting parallax effects the stereo algorithms produce
a smaller number of correct correspondences.
Stereo Algorithm (adopted from [36]) To evaluate
the different stereo algorithms, tracking results are
computed for all sequences with the intermediate
14 3D Pose Estimation of Vehicles Using Stereo Camera

Angle error, optimization: Levenberg−Marquardt, polar distance metric

40 STS, b = 10,2cm
STS, b = 22,8cm
20 STS, b = 33,0cm
FBS, b = 10,2cm
Δθ [deg]

FBS, b = 22,8cm
0 FBS, b = 33,0cm
CBS, b = 10,2cm
–20 CBS, b = 22,8cm
CBS, b = 33,0cm

–40
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

Angle error, optimization: Levenberg−Marquardt, polar distance metric

40 STS, b = 10,2cm
STS, b = 22,8cm
20 STS, b = 33,0cm
FBS, b = 10,2cm
Δθ [deg]

FBS, b = 22,8cm
0 FBS, b = 33,0cm
CBS, b = 10,2cm
–20 CBS, b = 22,8cm
CBS, b = 33,0cm

–40
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 7

Dependence of the yaw angle error (denoted by Dy) on the stereo baseline; Levenberg–Marquardt optimization, polar
distance metric. Upper diagram: all images; lower diagram: rear views excluded. Blue color denotes the spacetime stereo
technique (STS), red color the feature-based stereo (FBS), and green color the correlation-based stereo (CBS). The small
baseline is marked by squares, the intermediate baseline by crosses, and the large baseline by circles. The intermediate
baseline distance is most favorable with respect to the yaw angle error

baseline distance and the Levenberg–Marquardt algo-
rithm. Figures 9 and 10 show the results of the different
stereo algorithms, where both distance metrics (Euclid-
ean and polar) are evaluated. The results for the yaw
angle error show that the difference between the pixel
accurate feature-based stereo and the other two algo-
rithms with sub-pixel accuracy is negligible. The results
are similar for all three stereo algorithms. However,
regarding the vehicle position, a difference between
pixel and sub-pixel accuracy is noticeable. The dis-
tances of the feature-based stereo results show an up
to three times higher error for sequences 1, 2, and 3. No
significant difference between the spacetime stereo and
the correlation-based stereo can be found.
Distance Metric (adopted from [36]) Figures 9 and
10 show the pose estimation errors for the two distance
metrics. For the position error both metrics have
a similar performance, whereas the polar distance met-
ric produces a smaller yaw angle error for most
sequences. The computation time for the polar metric
is the same as for the Euclidean metric.
Optimization Algorithm (adopted from [36]) The
results of the Downhill Simplex optimization are
shown in Fig. 11 for the yaw angle error and in
Fig. 12 for the error of the average distance between
the ground truth points and the corresponding model
planes. The yaw angle errors are all higher for the
Downhill Simplex algorithm. Considering that the
number of function calls of the Levenberg–Marquardt
algorithm is of the same order of magnitude (about 100
function calls) as for the Downhill Simplex algorithm
with a better pose estimation accuracy, the Levenberg–
Marquardt algorithm appears to be preferable. The
distance errors show a similar behavior.
 3D Pose Estimation of Vehicles Using Stereo Camera 15

Distance error, optimization: Levenberg−Marquardt, polar distance metric

80
STS, b = 10,2cm
STS, b = 22,8cm
60 STS, b = 33,0cm
FBS, b = 10,2cm
dist [cm]

FBS, b = 22,8cm
40 FBS, b = 33,0cm
CBS, b = 10,2cm
CBS, b = 22,8cm
20
CBS, b = 33,0cm

0
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

Distance error, optimization: Levenberg−Marquardt, polar distance metric

80
STS, b = 10,2cm
STS, b = 22,8cm
60 STS, b = 33,0cm
FBS, b = 10,2cm
dist [cm]

FBS, b = 22,8cm
40 FBS, b = 33,0cm
CBS, b = 10,2cm
CBS, b = 22,8cm
20
CBS, b = 33,0cm

0
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 8

Dependence of the distance error on the stereo baseline; Levenberg–Marquardt optimization, polar distance
metric. Upper diagram: all images; lower diagram: rear views excluded. Colors and markers as in Fig. 7

Angle error, baseline distance b = 22.8 cm, optimization: Levenberg−Marquardt

15
STS, polar
10 STS, eucli
FBS, polar
5 FBS, eucli
CBS, polar
Δθ [°]

0 CBS, eucli
−5

−10

−15
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

Angle error, baseline distance b = 22.8 cm, optimization: Levenberg−Marquardt

15
STS, polar
10 STS, eucli
FBS, polar
5 FBS, eucli
CBS, polar
Δθ [°]

0 CBS, eucli
−5

−10

−15
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 9

Dependence of the yaw angle error (denoted by Dy) on the stereo algorithm; Levenberg–Marquardt optimization,
intermediate stereo baseline. Upper diagram: all images; lower diagram: rear views excluded. The colors (blue, red, and
green) denote the stereo algorithm, while diamonds represent the polar metric and stars the Euclidean metric
16 3D Pose Estimation of Vehicles Using Stereo Camera

Distance error, baseline distance b = 22.8 cm, optimization: Levenberg−Marquardt

50
STS, polar
40 STS, eucli
FBS, polar
FBS, eucli
dist [cm]

30 CBS, polar
CBS, eucli
20

10

0
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

Distance error, baseline distance b = 22.8 cm, optimization: Levenberg−Marquardt

50
STS, polar
40 STS, eucli
FBS, polar
FBS, eucli
dist [cm]

30 CBS, polar
CBS, eucli
20

10

0
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 10

Dependence of the distance error on the stereo algorithm; Levenberg–Marquardt optimization, intermediate stereo
baseline. Upper diagram: all images; lower diagram: rear views excluded. Colors and markers as in Fig. 9

Angle error, baseline distance b = 22.8 cm, optimization: Downhill−Simplex

15
STS, polar
10 STS, eucli
FBS, polar
5 FBS, eucli
CBS, polar
Δθ [°]

0 CBS, eucli
−5

−10

−15
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

Angle error, baseline distance b = 22.8 cm, optimization: Downhill−Simplex

15
STS, polar
10 STS, eucli
FBS, polar
5 FBS, eucli
CBS, polar
Δθ[°]

0 CBS, eucli
−5

−10

−15
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 11

Yaw angle error for Downhill Simplex optimization, intermediate stereo baseline. Upper diagram: all images; lower
diagram: rear views excluded. Colors and markers as in Fig. 9
 3D Pose Estimation of Vehicles Using Stereo Camera 17

Distance error, baseline distance b = 22.8 cm, optimization: Downhill−Simplex

50
STS, polar
40 STS, eucli
FBS, polar
FBS, eucli
dist [cm]

30 CBS, polar
CBS, eucli
20

10

0
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

Distance error, baseline distance b = 22.8 cm, optimization: Downhill−Simplex

50
STS, polar
40 STS, eucli
FBS, polar
FBS, eucli
dist [cm]

30 CBS, polar
CBS, eucli
20

10

0
Seq 1 Seq 2 Seq 3 Seq 4 Seq 5 Seq 6 Seq 7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 12

Distance error for Downhill Simplex optimization, intermediate stereo baseline. Upper diagram: all images; lower diagram:
rear views excluded. Colors and markers as in Fig. 9

Evaluation of the Model-Based Scene Flow Stage

For the evaluation of the model-based scene flow stage,
an initial pose is computed using the quaternion-based
ICP algorithm according to section “Quaternion-Based
Optimization.” The stereo baseline is set to b = 0.228 m,
corresponding to the configuration which turned out
to yield the most accurate pose estimation results
according to section “Evaluation of the ICP-Based 3D
Pose Estimation.” Figure 13 depicts an example trajec-
tory and the estimated yaw angle for sequence S5. The
reference trajectory is obtained using the average
position of the four colored markers. The estimated
trajectory is generated based on the center of the
corresponding model surface.
The reference data for the temporal pose deriv-
atives are obtained by numerical differentiation of
the pose reference data with respect to time. Since
the resulting yaw rate is too noisy to provide suffi-
ciently reliable reference values, a fourth-order
polynomial fitted to the differentiated yaw angle
reference values is used as reference data for the
yaw rate. In Figs. 14 and 15, the deviations between
the estimated pose parameters and their temporal
derivatives and the corresponding reference values are
visualized by error bars denoting the median over the
sequence and the 25% and the 75% quantiles, respec-
tively. The model-based scene flow method is always
initialized with the result of the quaternion-based ICP
approach.
The average yaw angle accuracy of the nonlinear
ICP approach with polar coordinates is about three
times higher than that of the quaternion-based solu-
tion, i.e., 2.4
 2.1
versus 4.3
 6.3
(cf. Fig. 14,
top). The average yaw angle error of the model-based
scene flow method amounts to 1.8
 3.4
. The fairly
large error interval is mainly caused by the high uncer-
tainty observed for sequence S6 while the errors are
comparable to those of the nonlinear ICP approach for
all other sequences. The average positional error of the
quaternion-based ICP is lower by one third than that of
the nonlinear ICP (0.21  0.07 m vs. 0.31  0.08 m);
the model-based scene flow method yields a
18 3D Pose Estimation of Vehicles Using Stereo Camera

15
Reference trajectory
Estimated trajectory
14

13

12
70
Reference orientation
z [m]

11 65
Estimated orientation
60

Yaw angle [°]

10
55

9 50

45
8
40
7 35
–4 –2 0 2 4 0 5 10 15 20 25
x [m] Time step

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 13

Left: Trajectory for sequence S5 with the vehicle model drawn for one time step, depicting the position of the center of the
lateral model surface. Right: Behavior of the yaw angle for the same sequence

1
ICP (polar coord.) ICP (polar coord.)
40 ICP (quaternions) ICP (quaternions)
Refinement 0.8 Refinement
Distance error [m]
Yaw angle error [°]

20
0.6

0
0.4

−20 0.2

−40 0
S1 S2 S3 S4 S5 S6 S7 S1 S2 S3 S4 S5 S6 S7

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 14

Yaw angle error (left) and distance error (right). Symbols denote the median values and error bars the 25% and 75%
quantiles. “Refinement” denotes the result of model-based scene flow

comparable average error value of 0.19  0.11 m (cf.
median values in Fig. 14, bottom).
The accuracy of the yaw rate estimated with the
model-based scene flow approach, averaged over all
test sequences, corresponds to 1.2
 1.3
per time
step (cf. Fig. 15). The average horizontal velocity error
amounts to 0.06  0.07 m per time step, while the
average error of the velocity component along the
depth axis is somewhat higher and corresponds to
0.11  0.09 m per time step. For the model-based
scene flow method, the evaluation results do not
show significant differences between the zero-mean
SSD and the NCC similarity measures. Comparably
large errors mainly occur for sequences S4, S5, and
 3D Pose Estimation of Vehicles Using Stereo Camera 19

4 performing a turning maneuver (cf. Fig. 16). The

NCC
mSSD image size amounts to 640  480 pixels, the frame
Yaw rate error [° / time step]

2 rate is 12 fps, and the baseline distance corresponds to

0.3 m. In the regarded traffic setting, no photogram-
metric markers could be attached to the vehicle and
0 thus no reference data are available. The position (tx, tz)
and the yaw angle y determined with the model-based
−2 scene flow method illustrate that the vehicle moves
laterally from the right to the left, approaches the
camera, and performs a rotation by about 25
,
−4
S1 S2 S3 S4 S5 S6 S7 corresponding to an average yaw rate of about 1.4

per time step (cf. Fig. 17). The measurements of the
0.4 instantaneous yaw rate y_ are somewhat noisy but con-
Lateral velocity error [m / time step]

NCC
mSSD sistently indicate a value slightly decreasing from about
0.2 1.5 to 1.0
per time step. The measured instantaneous
lateral velocity t_x of 0.14–0.18 m per time step is con-
0 sistent with the behavior of the lateral position tx over
time. The measurements of the depth velocity t_z are
−0.2
rather noisy but illustrate that the model-based scene
flow method is able to instantaneously detect small
−0.4
velocity components along the depth axis of only
a few centimeters per time step, corresponding to less
−0.6
S1 S2 S3 S4 S5 S6 S7 than a percent of the depth itself. At the end of the
sequence, the depth velocities are close to zero since the
0.8
Depth velocity error [m / time step]

NCC vehicle moves more laterally with respect to the camera

mSSD than at the beginning.
0.6

0.4 Summary and Conclusion

0.2 This study has described and evaluated approaches for

0 a stereo camera. After computation of stereo and opti-
−0.2
S1
3D Pose Estimation of Vehicles Using Stereo Camera.
Figure 15
Error of the yaw rate (top), horizontal velocity (middle), and
the velocity along the depth axis (bottom). Symbols denote
the median values and error bars the 25% and 75%
quantiles
S6, where in most images the object is far away from the
camera (z 30 m) and only its rear side is visible.
As a further example, a short image sequence of
1.5 s length is regarded, showing a passenger car
three-dimensional pose estimation of vehicles using
cal flow information, a graph-based clustering stage in
the space spanned by the spatial and velocity coordi-
S2 S3 S4 S5 S6 S7
nates separates the static from the moving objects in
the scene. An iterative closest point algorithm (ICP)
estimates the vehicle pose using a cuboid as a weak
vehicle model. A classical nonlinear optimization and
a computationally efficient quaternion-based analytical
approach relying on the Euclidean distance metric have
been compared to a distance metric based on polar
coordinates which takes into account the properties
of the stereo measurement process. The tracking-by-
detection approach has been employed, such that no
temporal filtering (e.g., Kalman filtering) has been
applied. The algorithm has been evaluated on seven
different real-world sequences with respect to different
20 3D Pose Estimation of Vehicles Using Stereo Camera
3D Pose Estimation of Vehicles Using Stereo Camera. Figure 16
Example sequence showing a passenger car turning right
stereo algorithms, baselines, distance metrics, and opti-
mization algorithms. The first part of the evaluation
has regarded all images of all sequences, while in
a second part the images in which the vehicle is ori-
ented longitudinally with respect to the camera have
been excluded. For most of the sequences, the pose
estimation yields reliable results for about 90% of the
time steps. Only in sequence S6 the vehicle drives
nearly parallel to the z axis, leading to a small projected
extension of the 3D point cloud in x direction, resulting
in unreliable pose estimation results for more than 90%
of the frames of that sequence.
Furthermore, a method for model-based estima-
tion of 3D pose parameters and their instantaneous
temporal derivatives has been examined in the context
of moving vehicle detection. The quaternion-based
ICP algorithm has been used for initialization. Pose
refinement and estimation of the temporal pose
derivatives are again performed without relying on
temporal filtering, using a model-based scene flow
method where 3D points on the object surface are
reprojected into pairs of stereo images acquired at
subsequent time steps and similarities between image
windows extracted around the reprojected points are
maximized.
The evaluation shows that for obstacles in a range
between 10 and 30 m ahead of the ego-vehicle, the
intermediate baseline distance of 0.228 m is preferable.
In that scenario a sub-pixel accurate stereo algorithm
yields up to three times higher distance accuracies
when compared to a pixel accurate algorithm. To min-
imize the point distances in the ICP algorithm, the
Levenberg–Marquardt algorithm yields better
accuracies than the Downhill Simplex optimization
while the number of function calls is comparable. The
proposed polar distance metric is preferable to the
Euclidean distance metric as for most sequences it
yields a higher accuracy of the estimated yaw angle,
especially for the pixel accurate stereo algorithm. If this
most favorable configuration is employed, the pose
estimation yields yaw angle errors typically smaller
than 3
and absolute distance errors of about 0.1 m,
corresponding to relative distance errors around 0.5%.
For the model-based scene flow method, the yaw angle
and positional errors are similar. The accuracy of the
estimated instantaneous yaw rate is about 1
per time
step, while the accuracies of the lateral velocity and the
velocity along the depth axis are both of the order of
0.1 m per time step.
These evaluation results indicate that the described
methods for image-based 3D pose estimation of vehi-
cles yield useful information for driver assistance sys-
tems in which instantaneous knowledge about the
position and motion direction of vehicles is of essential
importance.
Future Directions
In driver assistance systems for complex traffic scenar-
ios such as road intersections or roundabouts in the
urban environment, highly dynamical motion patterns
need to be predicted several seconds into the future.
Such predictions may serve for the recognition of driv-
ing maneuvers or the analysis of situations involving
several vehicles with respect to the possible occurrence
of collisions.
 3D Pose Estimation of Vehicles Using Stereo Camera 21

−1.5 2.5

Yaw rate [° / time step]

−2 2

−2.5
x [m]

1.5
−3
1
−3.5

−4 0.5
0 5 10 15 20 0 5 10 15 20
Time step Time step

20.5 −0.1

Lateral velocity [m / time step]

−0.12
20

−0.14
z [m]

19.5
−0.16

19
−0.18

18.5 −0.2
0 5 10 15 20 0 5 10 15 20
Time step Time step

−20 0.05
Depth velocity [m / timestep]

−25
−30
Yaw angle [°]

−35
0
−40
−45
−50
−55 −0.05
0 5 10 15 20 0 5 10 15 20
Time step Time step

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 17

Pose parameters and their temporal derivatives estimated with the model-based scene flow method for the image
sequence shown in Fig. 16

As an example of ongoing work in this field, the
object position is predicted in [34] based on a Bayesian
framework in which the probability distribution of the
motion hypotheses is represented by a set of measured
example trajectories (particles) rather than motion
models. These particles are propagated over time
using a particle filter [18, 34]. The measurement noise
is inherent in the example trajectories and therefore
does not need to be modeled. An example result is
shown in Fig. 18. Effectively, this method avoids using
simple motion models by cognitive prediction of
motion patterns based on a large data base of known
motion behaviors.
A further step is the recognition and prediction of
situations involving two or more vehicles based on
their 3D poses simultaneously acquired over time.
22 3D Pose Estimation of Vehicles Using Stereo Camera

40 40 40

35 35 35

30 30 30

25 25 25
z [m]

z [m]

z [m]
20 20 20

15 15 15

10 10 10

5 5 5

−15 −10 −5 0 5 −15 −10 −5 0 5 −15 −10 −5 0 5

x [m] x [m] x [m]

3D Pose Estimation of Vehicles Using Stereo Camera. Figure 18

Particles (blue) for a vehicle driving across a roundabout (time t = 0.0 s, 0.1 s, 0.3 s, left to right) as obtained with the particle
filter method described in [34]. The estimated vehicle position (red) is compared with the true position (green) (from [34])

355 Recent work [35] focuses on the recognition of situa-

tions involving two vehicles at road intersections. For
350 each vehicle, a set of possible future motion trajectories
is predicted several seconds ahead by the trajectory
345
particle filter using a motion database (cf. Fig. 19). An
interaction model based on the mutual visibility of the
340
vehicles and the assumption that the drivers will
y [m]

335 attempt to avoid a collision is used to assign probabil-

ities to possible future situations (cf. Fig. 19), allowing
330 a classification of the situation by a probabilistic
framework.
325 These short descriptions of research activities in the
fields of cognitive motion prediction, maneuver recog-
320 nition, and situation classification illustrate possible
future directions of the evolution of driver assistance
315
515 520 525 530 535 540 545 550 systems. Fundamentally relying on 3D pose estima-
x [m] tion results, such systems will anticipate the behavior
3D Pose Estimation of Vehicles Using Stereo Camera. of the traffic participants and predict their positions,
Figure 19 motion patterns, individual intentions, and mutual
Example of an early stage of a situation in which both interactions.
vehicles are turning left (cf. [35]). The future positions
(ground truth) of the vehicles are represented by the Bibliography
dashed lines, while the solid lines represent the predictions 1. Collado JM, Hilario C, de la Escalera A, Armingol JM
(from [35]) (2004) Model based vehicle detection for intelligent vehicles.

In: Proceedings of the IEEE intelligent vehicles symposium,
Parma, pp 572–577
2. Dellaert F, Thorpe CE (1997) Robust car tracking using
kalman filtering and Bayesian templates. In: Proceedings
of the conference on intelligent transportation systems,
Boston
3. Koller D, Danilidis K, Nagel H-H (1993) Model-based object
tracking in monocular image sequences of road traffic scenes.
Int J Comput Vision 10(3):257–281
4. Kollnig H, Nagel H-H (1997) 3D pose estimation by directly
matching polyhedral models to gray value gradients. Int
J Comput Vision 23(3):283–302
5. Haag M, Nagel H-H (1999) Combination of edge element
and optical flow estimates for 3D model-based vehicle
tracking in traffic image sequences. Int J Comput Vis
35(3):295–319
6. Kato T, Ninomiya Y, Masaki I (2002) An obstacle detection
method by fusion of radar and motion stereo. IEEE Trans Intell
Transp Syst 3(3):182–188
7. Steux B, Laurgeau C, Salesse L, Wautier D (2002) Fade: a vehicle
detection and tracking system featuring monocular color
vision and radar data fusion. Proceedings of the IEEE Intell
Vehicle Symp 2:632–639
8. Wender S, Dietmayer K (2007) 3D vehicle detection using
a laser scanner and a video camera. In: Proceedings of
the sixth European congress on ITS in Europe, Aalborg
9. Kaempchen N (2007) Feature-level fusion of laser scanner
and video data for advanced driver assistance systems.
PhD thesis, Faculty of Engineering and Computer Science,
Ulm University
10. Knöppel C, Regensburger U, Schanz A, Michaelis B
(2000) Robuste Erkennung von Straßenfahrzeugen im
Rückraumbereich eines Straßenfahrzeuges. In: Proceedings
of the 22nd DAGM symposium. Springer, Heidelberg,
pp 35–42
11. Dang T, Hoffmann C, Stiller C (2002) Fusing optical flow and
stereo disparity for object tracking. In: Proceedings of the
IEEE international conference on intelligent transportation
systems, Singapore
12. Leibe B, Cornelis N, Cornelis K, van Gool L (2006) Integrating
recognition and reconstruction for cognitive traffic scene anal-
ysis from a moving vehicle. In: Proceedings of the 28th DAGM
annual pattern recognition symposium. Springer, Heidelberg,
pp 192–201
13. Leibe B, Schiele B (2004) Scale-invariant object categorization
using a scale-adaptive mean-shift search. In: Proceedings of
the 26th DAGM annual pattern recognition symposium.
Springer, Heidelberg, pp 145–153
14. Toulminet G, Bertozzi M, Mousset S, Bensrhair A, Broggi A
(2006) Vehicle detection by means of stereo vision-based
obstacles features extraction and monocular pattern analysis.
IEEE Trans Image Process 15(8):2364–2375
15. Kubota S, Nakano T, Okamoto Y (2007) A global optimization
algorithm for realtime on-board stereo obstacle detection
3D Pose Estimation of Vehicles Using Stereo Camera 23
systems. In: Proceedings of the IEEE intelligent vehicles
symposium, Istanbul, pp 7–12
16. Nedevschi S, Danescu R, Marita T, Oniga F, Pocol C, Sobol S,
Tomiuc C, Vancea C, Meinecke MM, Graf T, To TB, Obojski MA
(2007) A sensor for urban driving assistance systems based on
dense stereovision. In: Proceedings of the IEEE intelligent
vehicles symposium, Istanbul, pp 276–283
17. Barth A, Franke U (2008) Where will the oncoming vehicle be
the next second? In: Proceedings of the IEEE intelligent vehi-
cles symposium, Eindhoven
18. Hermes C, Barth A, Wöhler C, Kummert F (2009) Object motion
analysis and prediction in stereo image sequences. In: Pro-
ceedings of the Oldenburger 3D-Tage, Oldenburg
19. Barth A, Siegemund J, Franke U, Förstner W (2009) Simulta-
neous estimation of pose and motion at highly dynamic turn
maneuvers. In: Proceedings of the 31st DAGM symposium,
Springer, Heidelberg, pp 262–271
20. Schmidt J, Wöhler C, Krüger L, Gövert T, Hermes C (2007) 3D
scene segmentation and object tracking in multiocular image
sequences. In: Proceedings of the fifth international confer-
ence on computer vision systems, Bielefeld. http://biecoll.ub.
uni-bielefeld.de/volltexte/2007/29/
21. Fielding G, Kam M (1997) Applying the Hungarian
method to stereo matching. In: Proceedings of the IEEE
conference on decision and control, San Diego,
pp 549–558
22. Stein F (2004) Efficient computation of optical flow using
the census transform. In: Proceedings of DAGM annual
pattern recognition symposium, Tuebingen, pp 79–86
23. Franke U, Joos A (2000) Real-time stereo vision for urban traffic
scene understanding. In: Proceedings of the IEEE intelligent
vehicles symposium, Dearborn, pp 273–278
24. Bock HH (1974) Automatische Klassifikation. Vandenhoeck &
Ruprecht, Göttingen
25. Barrois B, Wöhler C (2008) Spatio-temporal 3D pose estimation
of objects in stereo images. In: Proceedings of the sixth inter-
national conference on computer vision systems, vol 5008,
Lecture notes in computer science. Springer, Heidelberg,
pp 507–516
26. Duda RO, Hart PE, Stork DG (2000) Pattern classification,
2nd edn. Wiley, New York
27. Besl PJ, McKay ND (1992) A method for registration of 3-d
shapes. IEEE Trans Pattern Anal Mach Intell 14(2):239–256
28. Zhang Z (1992) Iterative point matching for registration of
free-form curves. IN-RIA, Technical Report. ftp://ftp.inria.fr/
INRIA/publication/publi-ps-gz/RR/RR-1658.ps.gz
29. Press WH, Teukolsky SA, Vetterling WT, Flannery BP
(1992) Numerical recipes in C: the art of scientific computing.
Cambridge University Press, New Delhi
30. Horn BKP (1987) Closed-form solution of absolute
orientation using unit quaternions. J Opt Soc Am
A 4(4):629–642
31. Hamilton WR (1866) Elements of quaternions. Longmans,
Green, London
24 3D Pose Estimation of Vehicles Using Stereo Camera
32. Hartley R, Zisserman A (2003) Multiple view geometry
in computer vision, 2nd edn. Cambridge University Press,
Cambridge
33. Krüger L, Wöhler C (2011) Accurate chequerboard corner
localisation for camera calibration and scene reconstruction.
Pattern Recognit Lett (to appear)
34. Hermes C, Einhaus J, Hahn M, Wöhler C, Kummert F (2010)
vehicle tracking and motion prediction in complex urban
scenarios. In: Proceedings of the IEEE intelligent vehicles sym-
posium, San Diego
35. Käfer E, Hermes C, Wöhler C, Ritter H, Kummert F (2010)
Recognition of situation classes at road intersections. In:
Proceedings of the IEEE international conference on robotics
and automation, Anchorage
36. Barrois B, Hristova S, Wöhler C, Kummert F, Hermes C
(2009) 3D pose estimation of vehicles using a stereo
camera. In: Proceedings of the IEEE intelligent vehicles
symposium, Xi‘an
37. Wöhler C (2009) 3D computer vision. Springer-Verlag, Berlin/
Heidelberg
 AC Machines: Permanent Magnet Synchronous and Induction Machines
AC Machines: Permanent Magnet
Synchronous and Induction
Machines
A. H. RANJBAR1, BABAK FAHIMI2
1 a multiphase AC winding. The latter rotor form
Department of Electrical Engineering, Amirkabir
University of Technology, Tehran, Iran
2 this family of electric machines, speed of the rotor is
University of Texas, Arlington, TX, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Permanent Magnet Motor Drives
AC Induction Machines
Fundamental Equations of Induction Machine
Power Electronic Inverter
IM Drive in Synchronous Reference Frame
Fundamental Equations of the IM Drive Under
Field-Oriented Control
Field-Oriented Control Implementations
Future Directions
Bibliography
Glossary
Permanent magnet synchronous machine
(PMSM) PMSM is an electromechanical energy
conversion device that forms as an important cate-
gory of electric machines. This machine comprises
an AC stator and a brushless rotor that houses
permanent magnet. Since the frequency of the exci-
tation in the stator winding is proportional to the
mechanical speed of the rotor, this family of
machines is called “synchronous.” PMSM is being
considered for variety of applications including
electric propulsion of modern vehicles.
Surface-mounted PMSM (SM-PMSM) A type of
PMSM where permanent magnets are placed on
the surface of the rotor, making it easy to build,
and skewed poles are easily magnetized on this
surface-mounted type to minimize cogging torque.
Interior permanent magnet (IPM) A type of PMSM
where, unlike SM-PMSM, permanent magnets
are mounted such that a magnetic saliency is
established.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
25
Induction machine (IM) A type of electrical machines
that operates based on the principle of induction.
While stator is formed by a multiphase AC winding,
the rotor houses is (a) a series of short-circuited
bars in the form of a squirrel cage or (b)
will allow for external injection of the current. In
not an integer multiple of the stator frequency, and
as such, they are named asynchronous machines.
Permanent magnet material Permanent magnet used
in PMSM is a piece of magnetic material which,
once magnetized or “charged” by an external mag-
netic field, retains a usefully large magnetic
moment after the magnetizing force is removed.
Thus, a permanent magnet becomes a source of
magnetic field in the rotor.
Field-oriented control Field-oriented control was
introduced in the beginning of 1970s. It demon-
strated that an induction motor or synchronous
motor could be controlled similar to that of
a separately excited DC motor by the orientation
of the stator mmf or current vector in relation to the
rotor flux to achieve a fast dynamic response in the
electromagnetic torque.
Definition of the Subject
Electrical machines can be viewed as electromechanical
energy converters where at least two magnetic
fields interact with each other to produce torque.
There are different types of electrical machines
including: brushed DC machines, brushless DC
machines, permanent magnet synchronous machines
(PMSM), AC induction machines (IM), switched
reluctance machines (SRM), and etc.
In 1886, the first practical DC motor was invented by
Frank Julian Sprague. It was a motor capable of constant
speed under variable loads. Brushes were used inevitably
for this family of electric machinery which played
a dominant role for a good part of the nineteenth
century and early decades of the twentieth century.
In 1882, Nikola Tesla discovered the rotating
magnetic field and pioneered the use of a rotary field
of force to operate machines. Later on, he suggested
that the commutators could be removed from
# 2012, DOI 10.1007/978-1-4419-0851-3
26 AC Machines: Permanent Magnet Synchronous and Induction Machines
a machine, and the device could operate on a rotary
field of force. In 1888, Nikola Tesla invented the first AC
motor, and, consequently, the polyphase power trans-
mission system was developed. With the advent of AC
machine and the effective use of the rotating magnetic
field, development of brushless electric machinery
attracted significant attention.
This chapter is dedicated to analysis of AC machines
including induction machines (IM) and permanent mag-
net synchronous machines (PMSM). Both IM and
PMSM take advantage of the rotating magnetic field
that is produced by their multiphase stator windings.
Permanent magnet synchronous machines (PMSM)
also known as brushless DC machines (BLDC) form an
important group of electric machinery that are being
considered for variety of applications including electric
propulsion of modern vehicles. Due to their high
efficiency, compact structure, high torque-to-current
and torque-to-volume ratios, fast dynamic response, sim-
ple mechanical construction, absence of moving contact,
and easy maintenance, PMSMs are widely used in many
industrial applications. However, induction motors are
now the preferred choice for industrial motors due to
their rugged construction, absence of brushes compared
to DC machines, absence of permanent magnets (PM)
compared to PMSM, and the simplicity of speed control.
Induction motor is considered most suitable for almost
every application because of its robust and rugged con-
struction and reliability. Moreover, its maintenance is
easy, and costs are relatively lesser. It has widespread use
in industrial, commercial, and domestic applications.
AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 1
Rotor and stator of a two-pole DC machine
Introduction
Electromechanical energy conversion can be viewed as
an interaction between at least two magnetic fields.
These fields can be constituted by (1) electric current,
(2) induced magnetization in ferromagnetic materials,
and (3) permanent magnets. To sustain a productive
electromechanical energy conversion, the following
requirements should be met:
1. At least one of the magnetic fields must be caused
by a controllable current.
2. The two interacting magnetic fields should be perpen-
dicular to each other. They can be stationary in space
(i.e., DC-brushed machines) or rotating together at
the same angular speed (i.e., AC machines).
In order to understand the operation of AC machines,
a quick review of DC-brushed machines is helpful. Rotor
and stator of a typical two-pole DC machine are shown
in Fig. 1. To compensate for the motion of the rotor, the
resulting field of the rotor should be continuously kept in
a stationary location. This is accomplished by the virtue
of the mechanical commutator. Field of the stator can be
produced by permanent magnets or fixed coils which are
fed with DC current. Injection of the current to a moving
member requires a continuous contact between the
armature and the external supply of the direct current.
This makes the use of brushes inevitable for this family of
electric machinery which played a dominant role for
a good part of the nineteenth century and early decades
of the twentieth century.
 AC Machines: Permanent Magnet Synchronous and Induction Machines 27

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 2

Stator and rotor of a squirrel-cage induction machine

With the advent of induction machine and the

effective use of the rotating magnetic field, develop-
ment of brushless electric machinery attracted signifi-
cant attention. However, ease of control has made DC
machines more popular in applications that require
a wide range of motor speeds. DC-brushed machines
are still among the most popular electromechanical
energy converters especially for low powers where the
use of permanent magnets is prominent.
This chapter is dedicated to the analysis of AC
machines including induction machine (IM) and per-
manent magnet synchronous machine (PMSM) that
are shown in Figs. 2 and 3, respectively. As shown in
Figs. 2 and 3, both IM and PMSM take advantage of the
rotating magnetic field that is produced by their three-
phase stator windings.
AC Machines: Permanent Magnet Synchronous and
Since the magnetic field caused by the multiphase
Induction Machines. Figure 3
stator windings rotates with a frequency equal to sup-
Stator and rotor of a three-phase four-pole surface-
ply frequency, the corresponding field of the rotor has
mounted permanent magnet synchronous machine
to maintain an adequate speed and orientation to
establish a relative stationary displacement with respect expect that existence of a DC field on the rotor neces-
to the field of the stator. This can be established by sitates a mechanical speed which is proportional to that
rotating the rotor at a speed whose addition to the of the stator frequency. In a two-pole machine, if two
supply frequency of the rotor field (at stand still) sources of magnetic field (i.e., electric currents) with
equates that of the stator supply. While this mechanism frequencies of f1 and f2 rotate at angular speeds of
automatically happens in a squirrel-cage induction om1 and om2 respectively, then the following relation-
machine, special tuning of the rotor currents in doubly ship must be held at all time to maintain synchronism
fed electric machinery turns into a necessity which is between the two fields (i.e., f1 þ o2pm1 ¼ f2 þ o2pm2 ).
properly handled by using power electronic converters. Employment of permanent magnets as the source
Following this design philosophy, one can intuitively of DC magnetic field on the rotor elevates the use of slip
28 AC Machines: Permanent Magnet Synchronous and Induction Machines
rings and brushes which are otherwise used for injection of
current in a rotating circuit, hence, resulting in a brushless
configuration. Low conductivity of the permanent mag-
nets will prohibit the appearance of appreciable induced
currents at low speeds and effectively eliminates the need
for providing reactive power for magnetization of the
magnetic circuit. These attributes have turned perma-
nent magnet synchronous machines into attractive
alternatives for a wide range of applications with
power ranging from a few watts into several megawatts.
Yet cooling, mechanical integrity, and operation
under harsh ambient conditions have been reportedly
impeding a universal takeover of the PMSM drives.
Besides, the issue of supply chain and lower efficiency
at higher speeds when field weakening has to be done are
among other reasons which have raised concern over the
use of PMSM in even larger scales. Yet, PMSM drives
have been considered as main contenders for many
applications among which electric propulsion of mod-
ern electrified automobiles, advanced renewable energy
harvesting apparatus, and high-performance precision
equipments stand out. The compact, relatively low-cost,
and optimally sized power electronic converters pave the
way for the massive integration of PMSM drives for
many applications. The invention of affordable temper-
ature-resilient high-energy permanent magnet materials
(rare-earth permanent magnet) has undoubtedly accel-
erated the integration of PMSM into many industries.
The continued growth of automotive market for
PMSM, in large part, depends on efficient operation
in the field-weakening region. The existence of perma-
nent magnets as a steady source of magnetic field
can potentially lead to supply of fault currents even
when the source is disconnected leading to unsafe
operating conditions. For the foreseeable future,
PMSM drives continue to enjoy the affordable cost
of high-energy permanent magnets and research and
development will continue on development of novel
magnetic configurations which can increase the
efficiency of the electromechanical energy conversion
and improve the reliability of this family of actuators.
Rotating Magnetic Field Versus Stationary Magnetic
Field
As mentioned earlier, both IM and PMSM contain
multiphase stator windings. These multiphase stator
windings produce a rotating magnetic field. In this
section, a three-phase arrangement for the stator wind-
ing is considered although the same conclusion can be
derived for higher number of stator phases. The rotat-
ing magnetic field produced by the three-phase AC
windings in IM and PMSM can also be produced by
a set of two-phase AC windings that is stationary and
90 apart with respect to each other. This rotating field
can be symbolically represented as:
MMFs ¼ kCosðos t  fs Þ ð1Þ
In which os and fs denote the frequency of current
and displacement of an arbitrary point on the stator. If
the frame of reference is rotating at a speed of om1 then
for an observer located at another frame of reference
which has a mechanical speed of om2 (in the same
direction), the rotating field will be seen as:
MMFs ¼ kCosððos þ om1  om2 Þt  fs Þ ð2Þ
This suggests that if the stator is excited with
a frequency of os , an observer in the air gap will sense
the rotating field at a speed of os ; while for an observer on
the rotor (rotor rotates with a speed of om in counter-
clockwise direction), the sensed frequency of the rotat-
ing field due to stator excitation will be os  om .
Notably, the two-phase AC windings mentioned
above can be replaced by two-phase DC windings that
are rotating (at an angular speed equal to the AC
supply) and are 90 apart with respect to each other.
Therefore, instead of analyzing the magnetic field pro-
duced by the stationary three-phase AC windings, one
can imagine the rotating two-phase DC windings that
are 90 apart with respect to each other. This can be
a physical interpretation of Park’s transformation that
is a mathematical transformation used to simplify the
analysis of three-phase systems.
In the case of balanced three-phase systems, the appli-
cation of Park’s transformation reduces the three AC
quantities to two DC quantities. Simplified calculations
can then be carried out on these transformed quantities
before performing the inverse transform to recover the
actual three-phase AC quantities. This is often used in
order to simplify the analysis of three-phase synchronous
machines and to simplify calculations for the control of
three-phase inverters. The unique characteristic of Park’s
transformation is that it eliminates all time-varying
inductances from the voltage equations of the machine.
 AC Machines: Permanent Magnet Synchronous and Induction Machines 29

These time-varying inductances can originate from squirrel-cage IM, as shown in Fig. 2, the magnetic
motion or varying magnetic reluctance. Park’s field of the rotor originates from the voltage that is
transformation is expressed by Eq. 3 where f can be induced on the rotor bars. Since these bars are short
voltage, current, or flux linkage of a synchronous circuited through the end ring, the voltage induced on
machine. Park’s transformation is a special case of the their terminals translates into significant currents that
reference-frame theory (also referred to as direct- originate the magnetic field of the rotor. Consequently,
quadrature (dq0) transformation). In the case of the magnetic field of the stator interacts with the
balanced three-phase circuits, application of the dq0 magnetic field produced by rotor currents to produce
transform reduces the three AC quantities in the abc torque in IM. It must be kept in mind that the resulting
frame of references to two DC quantities in dq frame of field of the rotor rotates at the synchronous speed in
references. the air gap.
Unlike magnetic field of the rotor in IM that is
fqd0S ¼ KS ðyÞ:fabcS ð3Þ dependent on the voltage induced by stator windings
on the rotor, magnetic field of PMSM is generated by
where
permanent magnets that are placed on the rotor. As

T  
fdqoS ¼ fqS fdS foS ð4Þ a result, the electromagnetic torque in the PMSM
results from interaction between the magnetic field of
ðfabcS ÞT ¼ ½ faS fbS fcS  ð5Þ the stator (i.e., an electromagnet) and the magnetic
2 3 field of the permanent magnets on the rotor. The elec-
3 Þ cosðy þ 3 Þ
cos y cosðy  2p 2p
24 2p 5 tromagnetic torque that is the cause of rotation in
KS ðyÞ ¼ sin y sinðy  3 Þ sinðy þ 3 Þ
2p
3 1 1 1 electrical machines can also be defined in terms of
2 2 2
electromagnetic co-energy. The electromagnetic co-
ð6Þ energy that is stored in the coupling field of unsatu-
dy rated electrical machines can be expressed as Eq. 10.
o¼ ð7Þ
dt
W c ¼ Lt ðyr Þ:I ð10Þ
In Eq. 7, o and y are supply frequency and rotor where
position, respectively. Thus, the definite integral form  
IS
Eq. 7 can be written as Eq. 8, where yð0Þ is the initial I¼ ð11Þ
IR
rotor position.
Zt    
y¼ oðxÞ dx þ yð0Þ ð8Þ LS ðyr Þ LSS ðyr Þ þ LSR ðyr Þ
L¼ ¼ ð12Þ
0
LR ðyr Þ LRR ðyr Þ þ LRS ðyr Þ

It can be shown that for the inverse transformation, The electromagnetic torque is defined as the deriv-
the following form can be used: ative of electromagnetic energy with respect to rotor
position and is expressed by Eq. 13.
2 3
cos y sin y 1
2 @W c @Lt SS @Lt SR
ðKS Þ1 ¼ 4 cosðy  2p
3 Þ sinðy  3 Þ 1
2p 5 ð9Þ T¼ ¼ :IS þ :IS
3 cosðy þ 2p Þ sinðy þ 2p
Þ 1 @yr @yr @yr
3 3 ð13Þ
@Lt RR @Lt RS
þ :IR þ :IR
@yr @yr
Torque Production in Electrical Machines
Based on Eq. 13, in order to have nonzero torque,
The magnetic field of stator in both IM and PMSM is the electromagnetic energy in the coupling field of the
produced by the virtue of three-phase stator windings electrical machine should vary with respect to rotor
which are distributed 120 apart from each other [1]. position. Considering Eq. 10, since rotor and stator
However, the source of magnetic field for the rotor in currents are not dependent on the rotor position, the
PMSM is different from that of IM. Assuming the electromagnetic torque can only be produced whenever
30 AC Machines: Permanent Magnet Synchronous and Induction Machines

the self-/mutual inductance of the windings changes reluctance is not uniform as it was for SM-PMSM.
with respect to rotor position. Therefore, the electro- So, the self-inductance value of the stator will depend
magnetic torque can be expressed as Eq. 14. on the rotor position and, consequently, there will be
  nonzero reluctance torque in addition to the reaction
@wf 1 @ LSS LSR
T¼ ¼ IT : :I ð14Þ torque. Therefore, an IPM has high efficiency and
@yr 2 @yr LRS LRR
torque density because the motor can utilize both
reaction and reluctance torque due to the magnetic
saliency [2]. The interior permanent magnet motors
1 ∂Lss(qr) ∂Lrs(qr) ∂Lsr(qr) have demonstrated attractive performance in the field-
Ist Is + Irt t
T = Is + Is Ir
2 ∂qr ∂qr ∂qr weakening region as well. This is accomplished by
the virtue of the reluctance torque which allows for
an extended speed range in the constant power region.
IM Table 1 briefly compares PMSM and IM in terms
IPM
of important characteristics including torque,
SM-PMSM
cooling, fault tolerance, supply chain, speed ratio,
t
∂Lrr(qr) and noise.
+ Ir Ir
∂qr Equation 13 also suggests that the electromagnetic
torque can be expressed symbolically as:
ð15Þ
X
T¼ fi ðIs ; Ir Þgi ðyr Þ ð16Þ
In Eq. 15, the torque that is produced by change of i
self-inductance with respect to rotor position is called
“reluctance torque,” and the torque that is produced by The above equation illustrates the impact of power
change of mutual inductance with respect to rotor electronic-based control (i.e., f ð:; :Þ) and magnetic
position is called “reaction torque.” Therefore, design (i.e., gð:Þ). Power electronic-based control
depending on the specific structure of the machine, allows for injection of a desired waveform for electric
any of these two torque components or both of them current and, as such, many researchers have tried to
may exist. Considering the rotor and stator structures obtain optimal waveforms which results in maximum
in the IM and PMSM, as shown in Figs. 2 and 3, it can torque, minimum torque pulsation, maximum torque
be seen that the self-inductances do not change with per ampere, etc. For instance, in the case of PMSM,
respect to rotor position since the air gap between rotor a maximum torque production has pointed toward the
and stator is fixed. So, the rotation in the squirrel-cage following relationship:
IM and surface-mounted (SM) PMSM is due to the @lpm
is / ð17Þ
reaction torque. @yr
When the rotor magnets are placed inside the
rotor iron in PMSM, the machine is called interior where lpm represents the flux linkage in the stator phase
permanent magnet (IPM). In this case, the magnetic due to permanent magnets. Similarly, in switched

AC Machines: Permanent Magnet Synchronous and Induction Machines. Table 1 Comparison of characteristics in
PMSM and IPM
[N-m/kg] [W/kg] Cooling Fault tolerance Supply chain Speed ratio Noise
IPM ++ ++   –– + +
SM-PMSM ++ +   ––  +
IM + + ––  ++ + +
 AC Machines: Permanent Magnet Synchronous and Induction Machines
reluctance drives, the stator current is found to be
optimal when the following condition is met:
@Ls
is2 / ð18Þ
@yr
where Ls represents the self-inductance of the
stator phase. In general, the problem can be
mathematically formulated as the following opti-
mization task:
0 0 p 11
Z X
P which undermines an efficient use of permanent mag-
ðIs ;Ir Þ ¼ Arg @Max @ fi ðIs ;Ir Þ:gi ð yr Þdyr AA
i
2 nets. These motors are considered to have small
0
ð19Þ
where P represents the number of magnetic poles. The
boundary conditions imposed by the available DC-link
voltage, time constants of the circuit, and power elec-
tronic circuits have to be taken account in this optimi-
zation process. Similar expressions can be found for
other objective functions. It is clear that magnetic con-
figuration of the electromechanical energy conversion
device plays a central role in this process. In particular,
it influences the room for optimization using power
electronic converter.
Permanent Magnet Motor Drives
Classification
Permanent magnet synchronous machines (PMSM)
also known as brushless DC machines (BLDC) form
an important group of electric machinery that are
being considered for variety of applications including
electric propulsion of modern vehicles. PMSM is an
electromechanical energy conversion device formed by
a conventional AC stator (i.e., sinusoidally distributed
multiphase winding fed with a balanced set of cur-
rents). In this class of machines, permanent magnets
are mounted on the rotor which then removes the need
for injection of the current to a moving component.
Special magnetic arrangement for the rotor provides
the advantages of (a) brushless operation, (b) control-
lability of a DC machine, (c) high power density in
constant torque region and also (d) compatibility with
the commonly used multiphase inverters. The mag-
nets can be placed in number of ways on the rotor.
Depending on the placement, they are called either
a surface-mounted PMSM (SM-PMSM) or an
31
interior PMSM (IPM). In SM-PMSM, permanent
magnets are placed on the rotor surface, making it
easy to build, and skewed poles are easily magne-
tized on this surface-mounted type to minimize
cogging torque.
This configuration is used for low-speed applica-
tions because of the limitation that the magnets may fly
apart if they are used in high-speed applications. Intro-
duction of container shields to protect the permanent
magnets will effectively increase the air gap length
saliency, thus having practically equal inductances in
both direct and quadrature axes. Besides, the relatively
small self-inductance of SM-PMSM will result in
a small electrical time constant which in turn requires
fast switching for direct control of the current. Notably,
in SM-PMSM, reluctance torque is almost nonexistent
and reaction torque is the main source of motion. The
permeability of the permanent magnet is almost that of
the air, thus for calculation of self-inductances, the
magnetic material should be treated as an extension
of the air gap. In SM-PMSM, the rotor has an iron core
that may be made of punched laminations for simplic-
ity in manufacturing. Thin permanent magnets are
mounted on the surface of this core using adhesives.
Alternating magnets of the opposite magnetization
direction produce radially directed flux density across
the air gap. The magnetic field produced by these PMs
then interacts with magnetic field produced by the
stator current to produce torque. Figure 4 shows how
magnets are placed in SM-PMSM [3].
Interior PMSMs have embedded permanent mag-
nets which are mounted inside rotor as shown in Fig. 5
[3]. Unlike SM-PMSM, each PM is mounted inside
the rotor in IPM. This will make IPM a good can-
didate for high-speed applications. There is induc-
tance variation for this type of rotor because the
permanent magnet is equivalent to air in the mag-
netic circuit calculation. The use of axial lamination
will further intensify the magnetic saliency for this
family of machines. These machines are considered to
have magnetic saliency in the sense that q-axis induc-
tance is greater than the d-axis inductance. The q and d
axes represent the quadrature and direct axes values,
respectively. This frame of references rotate with the
mechanical speed of the motor and as such a DC
32 AC Machines: Permanent Magnet Synchronous and Induction Machines
AC Machines: Permanent Magnet Synchronous and
Induction Machines. Figure 4
Cross-sectional area of an SM-PMSM
AC Machines: Permanent Magnet Synchronous and
Induction Machines. Figure 5
Cross-sectional area of an IPM
current in q and d axes can produce the same field as in
the stationary AC windings of the machine. Since the
PMs are not uniformly distributed on the rotor surface,
the effective air gap between rotor and stator is not
constant. As a result, the stator self-inductance varies
with respect to rotor position. Consequently, in IPM,
both reaction torque and reluctance torque contribute
to the motion.
PMSM machines have recently attracted consider-
able attention from various sectors of industry due to
their compact, efficient, and maintenance-free opera-
tion. Recent advancements in manufacturing of mag-
nets have enabled the use of such machines in a wide
range of applications. This in turn, has initiated many
new opportunities for this family of electric machines.
These machines can further be classified into either
sinusoidal type or trapezoidal type according to the
type of flux produced in the air gap by the magnets.
The difference is created by the way the magnets are
magnetized initially before mounting on the rotor.
Machines with sinusoidal back-emf can be controlled
to achieve ripple-free torque whereas machines with
nonsinusoidal back-emf offer lower inverter size and
a reduction in losses.
Permanent Magnet Material
Permanent magnet used in PMSM is a piece of mag-
netic material which, once magnetized or “charged” by
an external magnetic field, retains a usefully large mag-
netic moment after the magnetizing force is removed.
Thus, a permanent magnet becomes a source of mag-
netic field in the rotor which can interact with the
magnetic field generated by stator currents to produce
torque. For a magnet to be technologically useful, this
magnetization must persist in the presence of fairly
high-opposing fields (the permanent magnet must
have a high “coercive force”) and often at elevated
and sometimes at very cold temperatures. Its magnetic
properties must be reasonably stable for long periods of
time in adverse environments.
NdFeB magnets set a vivid example of this trend
by offering an impressive energy density of about
200–440 mkJ3 . In addition, highly efficient performances,
more compact geometry, lower weight, and shorter
electrical time constants are among other attractive
attributes of these technologies.
Permanent magnets are described by their demag-
netization characteristic. This is a portion of the hys-
teresis cycle located in the second quadrant of the flux
density vs. field intensity (B-H) plane. Figure 6 shows
demagnetization characteristics of various magnets at
rated temperature (20 C) [5]. As can be observed, the
 AC Machines: Permanent Magnet Synchronous and Induction Machines 33

Demagnetization Curves for Permanent Magnets

1.4
AlNiCo

1.2

1
NdFeB

SmCo
Flux Density (T)

0.8

0.6

0.4

Ferrite
0.2

0
–900 –800 –700 –600 –500 –400 –300 –200 –100 0
Field Intensity (kA/m)

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 6

Demagnetization characteristics for various permanent magnets

demagnetization curve of a permanent magnet can be
fully characterized by the following parameters:
1. Remanent magnetism (Br), where characteristic
crosses the B-axis. Br is the magnetic flux density
remaining in a material, especially a ferromagnetic
material, after removal of the magnetizing field.
2. Coercive field intensity (Hc), where characteristic
crosses the H-axis. Hc shows the material’s resis-
tance to becoming demagnetized.
3. Curvature connecting these two points, mostly one
or two linear segments.
In fact, the second linear region that is close to the
H-axis is considered as an unstable region.
Therefore a proper operating point will be located
in the first linear part, which is expressed as:
B ¼ Br þ m0 mr H
where mr and m0 stand for relative and air permeability,
respectively. It must be noted that in permanent
magnets, mr is very close to 1 (relative permeability
of the air). Accordingly, the energy density of a perma-
nent magnet (assuming a linear characteristic) can be
computed as:
 
Br 2 kJ
E ¼ B:H ) EMax ¼ : ð21Þ
4m0 mr m3
Figure 7 illustrates the constant energy loci along
with demagnetization characteristics of various perma-
nent magnets [5]. AlNiCo alloys, the hexaferrites, and
the rare earth permanent magnets are the three types of
magnets that are used in electrical machines among
which rare earth PMs have attracted more attention
in PMSM applications.
Chemically, the AlNiCos are iron–cobalt–nickel–
based alloys with minor additions of aluminum and
copper, and in some grades titanium. The ferrites have
ð20Þ some peculiar properties setting them apart from other
magnets, for better and worse. As oxides, they are
electrical insulators, which eliminate eddy currents.
(Most permanent magnet materials, being metals, are
fairly good conductors.) This is advantageous under
34 AC Machines: Permanent Magnet Synchronous and Induction Machines
Demagnetization Curves for Permanent Magnets
1000 (kJ / m3)
1.2
1
Flux Density (T)
0.8
0.6
0.4
0.2
0
–1000 –900 –800 –700
AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 7
Constant energy loci and demagnetization curves of various PM
dynamic operating conditions as in electric motors and
also facilitates pulsed field charging. They are chemi-
cally inert, so the production steps of milling, pressing,
and surface grinding can be done in air and with water
lubrication; they also have no corrosion problems in
use even at high temperatures, but the magnetization in
ferrites drops more rapidly in the presence of heating.
Developed in the 1970s and 1980s, rare earth mag-
nets are the strongest type of permanent magnets
made, substantially stronger than ferrite or AlNiCo
magnets. The magnetic field typically produced by
rare-earth magnets can be in excess of 1.4 T, whereas
ferrite or ceramic magnets typically exhibit fields of
0.5–1 T. There are two types of rare-earth PMs: neo-
dymium magnets (Nd2Fe14B) and samarium–cobalt
(SmCo magnets). The newest family of hard magnetic
alloys, rare-earth PM, includes not only the
well-known samarium–cobalt and neodymium–
iron–boron magnets but also many other related mate-
rials. All are based on intermetallic compounds formed
when 3d-transition metals (TM), notably the strongly
250 (kJ / m3) AlNiCo
750 (kJ / m3) 500 (kJ / m3)
BH(Max) = Const. NdFeB
SmCo
Ferrite
–600 –500 –400 –300 –200 –100 0
Field Intensity (kA/m)
magnetic Co and Fe, are alloyed with rare-earth ele-
ments (RE) of the 4f-transition series, (i.e., the “lan-
thanides” Ce, Pr, Nd, Sm, Gd, Dy, and others,) or their
close chemical relatives La and Y. Addition of still
another element, such as B, C, Si, or Ti, often helps
form ternary compounds [4]. Addressing different
types of permanent magnet in the above graphs, it
must be mentioned that ferrite magnets represent
a low-cost solution while offering a limited flux density.
AlNiCo magnets, which are more expensive as com-
pared to ferrite magnets, demonstrate a very high rem-
anent magnetism. This, however, is undermined by
very limited coercive field intensity. The most expen-
sive SmCo magnets represent quality in every aspect
including high remanent magnetism, large coercive
field intensity, and a fully linear demagnetization
characteristic. They also are known as highly stable in
the presence of high-temperature variation. Finally, the
NdFeB magnets demonstrate very high-energy density
at room temperature. Furthermore, the cost associated
with NdFeB is much less than SmCo magnets.
 AC Machines: Permanent Magnet Synchronous and Induction Machines 35

However, coercive field intensity in NdFeB magnets is
highly sensitive to temperature changes. This, in turn,
results in an inadequate performance at high
temperatures.
In general, the main objectives in selecting
a permanent magnet for motor drive application can
be summarized as:
● High-energy density
● A linear demagnetization characteristic in the entire
vicinity of the second quadrant (B-H plane)
● High stability with respect to temperature
● High specific resistance to mitigate eddy currents
● Durability against corrosion and demagnetization
● Low cost
Although achieving all these attributes in a single
magnet is not possible, proper design can help in opti-
mizing the performance of the drive in the context of
the application.
Table 2 summarizes the magnetic characteristics of
four families of the permanent magnets.
Comparing various properties of permanent mag-
nets, the following observation can be made:
● A rise in temperature will reduce the remanent
magnetism in all magnet types. On a percentage
basis, ferrite magnets seem to be the most sensitive
magnets, while SmCo demonstrates the least
sensitivity.
● Coercive field intensity portrays different behavior
for various materials. While an increase in temper-
ature results in significant decrease of coercive field
in NdFeB, an opposite response is seen in ferrite
magnets. Overall, SmCo offers the least sensitivity
to temperature.
● SmCo is the heaviest and the most expensive alter-
native among all candidates. It also presents one of
the lowest specific resistances, which translates to
high eddy-current losses.
● AlNiCo offers the highest remanent magnetism.
This, however, is mainly undermined by a very lim-
ited coercive field intensity and extremely high
conductivity.
● The reversible permeability in most cases is close
to 1.
Impact of Temperature on Performance of PMSM
Temperature variations are an inevitable attribute of
many industrial applications such as automotive
products. In fact, in some automotive applications,
a temperature range of 40 – 150 C is reported. This,
in turn, can alter the properties of permanent magnets
used in PMSM. Figures 8 and 9 show demagnetization
curves for NdFeB and ferrite magnets at various
temperatures. As can be seen, a rise in temperature
reduces the remanent magnetism in both cases. How-
ever, a severe reduction in coercive field in NdFeB poses

AC Machines: Permanent Magnet Synchronous and Induction Machines. Table 2 Summary of attributes for various
permanent magnets
Parameters at 20 C Ferrite AlNiCo SmCo5 NdFeB
Remanent magnetism (Br [T]) 0.38. . .0.42 0.61. . .1.35 0.85. . .1.0 1.0. . .1.23
Coercive field intensity (Hc [KA/m]) 390. . .280 59. . .50 1000. . .1200 1600. . .960
3
Max. energy density (BHmax [kJ/m ]) 28. . .34 13. . .62 140. . .200 195. . .280
Temperature coefficient (KB [%/C]) 0.2. . .  0.23 0.02 0.04. . .  0.05 0.11. . .  0.13
Temperature coefficient (KH [%/C]) +0.4. . . + 0.22 +0.03. . .  0.07 0.25 0.6. . .  0.8
Reversible permeability (mrev ) 1.05 <5 1.05 < 1.2
Density ½r=ðkg=dm3 Þ 4.6. . .4.9 6.7. . .7.3 8.1. . .8.3 7.3. . .7.4
Specific electric resistance ½r=mOcm 10 . . .10
12 16
40. . .70 50. . .60 140
Cost (%) 10. . .15 40. . .60 600. . .800 200. . .300
36 AC Machines: Permanent Magnet Synchronous and Induction Machines

Effects of Temperature on Demagnetization Curve for NdFeB(Neomax 35)

1.2

1
Flux Density (T)

0.8

0.6

0.4

–20C 20C 100C 150C

0.2

0
–1600 –1400 –1200 –1000 –800 –600 –400 –200 0
Field Intensity (kA/m)

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 8

Thermal behavior of NdFeB magnets

a real limitation for armature currents at high temper- Br  Bro ð1 þ kB T =100Þ ð23Þ
atures. This will clearly undermine the practicality of
HK  Hk0 ð1 þ kH T =100Þ ð24Þ
using NdFeB magnets at very high temperatures. It
must be noted that a modest range of temperature, The no-load operating points are as follows:
below 40 C, can be appropriate for employing the
Br ðB ð1 þ kB T =100ÞÞ
NdFeB. The ferrite magnets, on the other hand, depict BM0 ffi  ¼ r0  ð25Þ
a reverse behavior at high temperatures; they tend to 1 þ hMs
1 þ hsM
have larger coercive fields. This, however, at very low
temperatures may cause a problem. According to in which s and hM represent air gap length and radial
thickness of permanent magnets, respectively. A change
Table 2, some rare-earth permanent magnets demon-
strate a superior performance in terms of temperature in the temperature, therefore, directly shifts the
location of operating point. This will affect the allowed
stability. However, this comes with a significantly
armature current as the maximum coercive field has
higher cost. Therefore, if a modest range of tempera-
also been changed. This is expressed as:
tures can be accommodated, employment of NdFeB
magnets is recommended.
Br
In order to quantify the effects of temperature DHM  HM0  Hk ¼  Hk
m 0 ð 1 þ hM =s Þ
change, the linear (stable) part of demagnetization
ðBr0 ð1 þ kB T =100ÞÞ
characteristics is expressed as follows: ¼  ðHk0 ð1 þ kH T =100ÞÞ
m0 ð1 þ hM =sÞ
B ¼ f ðHÞ ffi Br þ m0 mr H ð22Þ ð26Þ
 AC Machines: Permanent Magnet Synchronous and Induction Machines 37

Effects of Temperature on Demagnetization Curve for Ferril

Magnets (remanent)
0.45

0.4
–20C

0.35
20C

0.3
100C
Flux Density (T)

0.25
150C

0.2

0.15

0.1

0.05

0
–300 –250 –200 –150 –100 –50 0
Field Intensity (kA/m)

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 9

Thermal behavior of ferrite magnets

One can say that an increase in temperature drive. Table 2 provides a range of kH values for various
will reduce remanent magnetism and, hence, will permanent magnets. In order to illustrate the effects of
reduce the torque productivity of the machine by temperature, the following example is given. Table 3
a factor of: shows three various designs (two-pole design is

 assumed) using ferrite, SmCo, and NdFeB permanent
@B ðkB Þ Br0 magnets.
¼ ð27Þ
@T 1 þ ðs=hM Þ 100 Once the air gap flux density and armature mmf are
known, the following equation can be used to estimate
where T stands for temperature in degree Celsius.
electromagnetic torque:
Table 2 provides a range of kB values for various per-
manent magnets. At the same time, the permitted p
T / HBD 2 l ð29Þ
armature field to maintain a stable operation is limited 3
by: in which D and l represent outer diameter of the rotor

 and effective stack length of the machine, respectively.
@B kB Br0 1 Using Eq. 29, one can select the stack length such that
¼  kH H K 0 ð28Þ
@T m0 ð1 þ hM =sÞ 100 a total torque of 5 N - m is achieved. Table 4 shows the
results of torque calculation.
It must be noted that a more strict limit on the In the next step, using the above method, operating
armature excitation will further impact the torque gen- points and the electromagnetic torque at 20, 20, 100,
eration of the permanent magnet synchronous motor and 150 C are computed and listed in Table 5.
38 AC Machines: Permanent Magnet Synchronous and Induction Machines

AC Machines: Permanent Magnet Synchronous and 6

Induction Machines. Table 3 Example design for

SmCo5
illustration of temperature effects 5

NdFeB
Type s (mm) hM (mm) D (mm) 4

Torque (N-m)
Ferrite 0.5 4.5 50 Ferrite
SmCo 0.5 1 50 3

NdFeB 0.5 2 50
2

0
0 50 100 150
AC Machines: Permanent Magnet Synchronous and Temperature (C)
Induction Machines. Table 4 Operational condition at
20 C AC Machines: Permanent Magnet Synchronous and
Induction Machines. Figure 10
Br (T) @ HA (kA/m) @ Stack Torque Effects of temperature on electromagnetic torque in
Type 20 C 20 C length (N-m)
brushed DC motor drives
Ferrite 0.42 30 165 5
SmCo 1 30 96 5
NdFeB 1.23 30 64 5
Please note that the stack length of machines is kept
constant in this exercise.
The results of this study are graphically shown in
Fig. 10 [5]. As can be seen, reduction in remanent
magnetism, due to temperature variations, is the
AC Machines: Permanent Magnet Synchronous and main cause for reduction in torque. At high tempera-
Induction Machines. Table 5 Effects of temperature on tures, despite the relatively low armature mmf, low
operational characteristics of PMSM coercive/knee point field intensity also contributes to
this reduction. It is therefore expected that a higher
Hk Bair gap HM Torque
armature mmf would result in yet higher reduction
Type Br (T) (kA/m) (T) (kA/m) (N-m)
in torque under this design. It must also be noted that
Ferrite 20 0.46 246 0.414 36 5.36 SmCo depicts the most stable operation; however,
20 0.42 280 0.378 33 5 with a reasonable range of temperatures, employment
100 0.34 246 0.306 27 3.57 of NdFeB magnets seems to be the most appropriate
choice. NdFeB magnets not only are better
150 0.29 280 0.261 23 2.6
economic choices, they also offer higher specific
SmCo 20 1.02 1,320 0.68 270 5.12
resistance at a lower weight compared to SmCo mag-
20 1 1,200 0.667 265 5 nets. This in turn minimizes the effects of high-
100 0.96 960 0.64 254 4.83 frequency eddy-current losses caused by slot
150 0.94 810 0.627 250 4.73 harmonics.

NdFeB 20 1.29 980 1.032 205 5.18

Voltage and Torque Equations for SM-PMSM
20 1.23 900 0.984 196 5
100 1.1 220 0.88 175 4.42 Since SM-PMSM is more popular than IPM, this part is
dedicated to describing dynamic equations of a
150 1.02 190 0.816 162 3.82
typical two-pole three-phase SM-PMSM depicted
 AC Machines: Permanent Magnet Synchronous and Induction Machines 39

Sensor

bs axis 1

as′

wr

S q axis
cs × × bs

qr

as axis

cs ′
bs ′ N
3
2
Sensor
Sensor ×
as

cs axis

d axis

ibs

+ rs
+
vcs
rs Ns
vbs
ics Ns
n

Ns
vas

ias rs
+

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 11

A two-pole three-phase SM-PMSM

in Fig. 11 [6]. In this figure, stator includes three torque equations that are used to describe the dynamic
sinusoidally distributed windings placed 120 apart behavior of the PMSM will be established.
from each other. Each winding has Ns equivalent Voltage equations in machine variables are
turns and resistance rs. In this section, the voltage and expressed through Eqs. 30–32
40 AC Machines: Permanent Magnet Synchronous and Induction Machines


Vas ¼ rs ias þ plas ð30Þ 2p
Lcc ¼ Lls þ LA  LB cos2 yr þ
3
Vbs ¼ rs ibs þ plbs ð31Þ
2 3
Vcs ¼ rs ics þ plcs : ð32Þ sin yr
l0m ¼ l0m 4 sinðyr  2p
3Þ
5 ð38Þ
Writing stator voltage equations in matrix format sinðyr þ 2p Þ
3
results in Eq. 33.
Vabcs ¼ rs iabcs þ plabcs ð33Þ So, pl0m would be the open-circuit voltage induced
in each stator phase winding. In Eq. 38, it is assumed
where
that the voltages induced by PMs on the stator wind-
ðfabcs ÞT ¼ ½ fas fbs fcs  ð34Þ ings are sinusoidal.
The electromagnetic torque in machine variables
rs ¼ diag ½ rs rs rs  ð35Þ can be expressed by Eq. 39.
The flux linkage can be written as:

(
labcs ¼ Ls iabcs þ l0m ð36Þ P ðLmd  Lmq Þ
Te ¼
2 3
where Ls is the stator inductance matrix and "

is expressed by Eq. 37. Also, l0m is the flux linkage 
1 2 1 2
ias  ibs  ics  ias ibs  ias ics þ 2ibs ics sin 2yr
2

established by the PMs on the rotor as viewed 2 2

pffiffiffi #
from stator windings and is expressed by 3 2 2
þ ði i  2ias ibs þ 2ias ics Þ cos 2yr
Eq. 38. Where l0m is the flux linkage amplitude pro- 2 bs cs
duced by the PMs as viewed from stator phase "

1 1
winding. þ l0 m ias  ibs  ics cos yr
2 2
2 3 pffiffiffi #)
Laa Lab Lac 3
þ ðibs  ics Þ sin yr
Ls ¼ 4 Lba Lbb Lab 5 ð37Þ 2
Lca Lcb Lcc
ð39Þ

Where;
The dynamic dq0 modeling is used to study the
Lab ¼ Lba motor during transient and steady state. This can be
Lac ¼ Lca done by the virtue of Park’s transformation. As
discussed earlier, Park’s transformation converts the
Lbc ¼ Lcb
three-phase voltage, current, and flux variables into
And; dq0 variables. Park’s transformation for the three-
phase stator voltages can be expressed as Eq. 40.
Laa ¼ Lls þ LA  LB cos2yr Consequently, voltage equations in dq0 frame are
1  p expressed as Eq. 41.
Lab ¼  LA  LB cos2 yr 
2 3
2 3 2 32 3
1  p Vq cos yr cosðyr  120Þ cosðyr þ 120Þ Va
2
4 Vd 5 ¼ 4 sin yr sinðyr  120Þ sinðyr þ 120Þ 54 Vb 5
Lac ¼  LA  LB cos2 yr þ
2 3 V0
3 1 1 1
Vc

 2 2 2
2p ð40Þ
Lbb ¼ Lls þ LA  LB cos2 yr 
3
1
Lbc ¼  LA  LB cos2ðyr þ pÞ Vdq0s ¼ rs idq0s
r
þ or lrdqs þ plrdq0s ð41Þ
2
 AC Machines: Permanent Magnet Synchronous and Induction Machines 41

where 3 P
Te ¼ ð Þð Þl0m iqs
r
ð55Þ
 T   2 2
lrdqs ¼ lds lqs 0
r r
ð42Þ
2 32 3 2 3 In which P; l0m ; and iqs
r
represent number of mag-
Lls þ Lmq 0 0 r
iqs 0 netic poles, flux linkage due to permanent magnet as
lrdq0s ¼ 4 0 Lls þ Lmd 0 54 ids
r 5
þ l0rm 4 1 5 observed by the coils on the new frame of reference
0 0 Lls i0s 0 (which is mounted on the rotor), and the quadrature
ð43Þ component of the new two-phase system mounted on
the rotor, respectively. This equation resembles that of
Expanding these equations,
a separately excited DC machine in which the armature
r
vqs ¼ rs iqs
r
þ or lrds þ plrqs ð44Þ current is replaced with. Although there is a difference
between the separately excited brushed machine and
r
vds ¼ rs ids
r
 or lrqs þ plrds ð45Þ SM-PMSM in the new frame of references and that is
v0s ¼ rs i0s þ pl0s ð46Þ the fact illustrated in Eq. 50 and Eq. 51. As can be seen,
the current in d-axis will influence the voltage equation
where in q-axis and visa verse. In separately excited brushed
lrqs ¼ Lq iqs
r
ð47Þ machine, the flux generated by the armature current
does not link the field windings and as such does not
lrds ¼ Ld ids
r
þ l0rm ð48Þ appear in the phase voltage equation. From Eq. 55, it is
evident that increasing the d-axis current does not
l0s ¼ Lls i0s ð49Þ
contribute to the torque generation, and therefore, it
where Lq = Lls + Lmq and Ld = Lls + Lmd. is usually regulated around zero. However, to extend
Substituting Eq. 47 to Eq. 49 into Eq. 43 to Eq. 46, the constant power region of the machine, a negative
results in: value of the direct axis current can counteract the field
of the permanent magnets leading to a smaller
r
vqs ¼ ðrs þ pLq Þiqs
r
þ or Ld ids
r
þ ol0rm ð50Þ motional back-emf. This will allow for regulation of
r
vds ¼ ðrs þ pLd Þids
r
 or Ld iqs
r
ð51Þ the current in stator phases with a penalty on efficiency.
The second term in Eq. 54 represents the reluctance
v0s ¼ ðrs þ pLls Þi0s ð52Þ torque and only exists in interior permanent magnet
machines. In this case, direct and quadrature axis cur-
The expression for electromagnetic torque in dq0
rent will both contribute to the formation of electro-
reference frame is then written as Eq. 53).
magnetic torque, and as such, they have to be optimally


 controlled to ensure a maximum torque per ampere
3 P
Te ¼ ðlrds iqs
r
 lrqs ids
r
Þ ð53Þ operation. Due to the presence of the reluctance
2 2
torque component, an extended operation in the con-
stant power region is feasible. The latter makes IPM
Substituting Eq. 47 and Eq. 48 into Eq. 53, Eq. 54 is an attractive candidate for electric propulsion of elec-
achieved as the final expression for the electromagnetic trified power trains. Figure 12 illustrates a graphical
torque in dq0 reference frame. interpretation of change in frame of references in


h SM-PMSM.
3 P i
Te ¼ l0rm iqs
r
þ ðLd  Lq Þiqs
r r
ids ð54Þ
2 2
Field-Oriented Control in PMSM
The second term in Eq. 54 denotes the reluctance Field-oriented control was introduced in the beginning
torque which is not significant in SM-PMSM, and of 1970s. It demonstrated that an induction motor or
therefore, the equation for the electromagnetic torque synchronous motor could be controlled similar to that
in SM-PMSM can be simplified to: of a separately excited DC motor by the orientation of
42 AC Machines: Permanent Magnet Synchronous and Induction Machines
qss
as
N N
wr
bs cs
S S
ac − 3 coils ac − 2 coils
AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 12
Consecutive change in frame of references to reach a field-oriented arrangement with DC currents in quadrature and
direct axes
the stator mmf or current vector in relation to the rotor
flux to achieve a desired objective. The stator windings
of the PMSM are fed by an inverter that generates
a variable frequency and variable voltage magnitude.
In field-oriented control of PMSM, instead of control-
ling the inverter frequency independently, the fre-
quency and phase of the output wave are controlled
using position sensors that are shown in Fig. 11 [4].
In order for the motor to behave like DC motor, the
controller needs to know the position of the instanta-
neous rotor flux or rotor position of PMSM. This can
be accomplished using an optical encoder. Having con-
tinuous access to the rotor position, the three-phase
currents can be calculated.
There are generally two control modes for control-
ling PMSM. These control modes correspond to con-
stant torque and flux weakening. These modes are
determined by the physical limitation of the motor
and the inverter. The limit is established by the rated
speed of the motor, at which the constant torque oper-
ation finishes (i.e., back-emf partially exceeds the avail-
able inverter-phase voltage) and the flux weakening
starts as shown in Fig. 13.
The field-oriented control of PMSM is also called
vector control due to the fact that the direct and quad-
rature components of the current are controlled simul-
taneously. As shown in Fig. 14, accurate measurement
of three-phase currents and rotor position is an integral
part of this control strategy. Once the transformation
to the synchronous frame of references (tied to the
rotor) is completed, the desired values of the direct
and quadrature current will be compared to the
N
dsr
dss
wr qsr wr
S
dc − 2 coils
Te Constant Torque
Im
Flux-Weakening
ωrated ωm
AC Machines: Permanent Magnet Synchronous and
Induction Machines. Figure 13
Torque versus speed characteristics of a PMSM
computed values of these DC quantities. The existing
error will be then processed by a proportional + inte-
gral (i.e., PI) controller, and the controlled variable is
then transformed back to the physical quantities
(AC quantities) which are fed to the inverter control.
According to Equation 55, the quadrature axis current
of the stator will be controlled directly to achieve the
desired torque in constant torque region. The direct axis
current will be kept at zero to maximize the torque per
ampere and to eliminate any need for reactive power [7].
In order to increase the speed range in PMSM, the
flux in the d-axis current can be used to reduce the net
flux generated by the permanent magnet. This action is
called flux weakening.
For any given DC-link voltage, there exist a unique
angular speed beyond which, at least for a nonzero
portion of one electrical cycle, the induced back-emf
 AC Machines: Permanent Magnet Synchronous and Induction Machines 43

P
q
2 r
iqeds
+ vabcs
PI (K eS(qe))–1 INVERTER PMSM
v*abcs
–

iabcs

iqeds

(K eS (qe))
P
q
2 r

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 14

Block diagram of the field-oriented (vector-controlled) control of SM-PMSM drive

(at rated constant torque) in each phase will exceed the
available phase voltage. This speed is defined as base
speed beyond, which regulation of the current wave-
form will necessitate a net reduction in the back-emf.
This can be accomplish by application of a negative
direct axis current which will effectively reduce the flux
originated from permanent magnet, thereby reducing
the motional back-emf. This method, however, intro-
duces additional loss in the stator and cannot be prac-
ticed indefinitely, especially at high stator currents
which may lead to partial demagnetization of the per-
manent magnets. The appearance of direct axis current
will introduce the need for reactive power and a reduc-
tion in torque per ampere. This can be interpreted from
the following equations:
 as they are less expensive and more robust.
1 iq
a ¼ tan ð56Þ
id
It is also known that:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Im ¼ id 2 þ iq 2 ð57Þ
AC Induction Machines
The induction machine is widely used in the industry
for a variety of applications for converting electrical
energy into mechanical work. Until the early 1970s,
they had been mostly operated in the constant-voltage,
uncontrolled mode which even today is the most com-
mon approach in practice.
The advent of thyristors and, consequentially, the
development of variable-frequency converters allowed
the wide-range speed control of induction motors. The
scalar control method (e.g., Volts/Hertz control
method) requires the simultaneous adjustments of the
frequency and the magnitude of the sinusoidal voltage
or current supplied to the motor. This allows making
steady-state operating characteristics of an induction
motor similar to those of a DC motor. Adjustable-
speed AC drive systems, employing scalar control prin-
ciples have been replacing DC drives in numerous
industrial applications, such as pumps, fans, compres-
sors, etc. Induction machines are extensively preferred
The scalar control of an induction motor does not
make it dynamically equivalent to a DC motor. For
instance, in an induction machine accurate position,
control is not possible since precise control of the
instantaneous torque is needed to obtain the required
speed trajectory. This, in turn, means that it is the
instantaneous stator current which needs to be con-
trolled. Scalar control applies to steady-state operation
of the motor only, while no particular control strategies
are involved for the transient operating conditions.
The separately excited DC motor has physically
separated armature and field windings whose mmfs
44 AC Machines: Permanent Magnet Synchronous and Induction Machines

are orthogonally displaced from each other in space. Ia If

The armature current can be controlled without affect- Φa

ing the field current. Therefore, a decoupled control of Ia
torque and flux is possible, which results in fast torque
and speed response. This is not the case in induction DC Motor
motors. To obtain the optimal torque from an induc-
Φf
tion motor, an algorithmic equivalent of the DC motor
has to be established to obtain an appropriate If
decoupling between the stator and rotor flux. The
control technique used to obtain such a stator–rotor
physical link is named as field-oriented control or AC Machines: Permanent Magnet Synchronous and
vector control. Induction Machines. Figure 15
The Field Orientation Principle was first formu- DC machine
lated by Haase in 1968 and Blaschke in 1970. At that
time, due to inadequate means to implement this
mode, the concept of vector control seemed impracti- to control the torque-producing current separately
cal. The advent of microprocessors made vector control from the magnetic flux-producing current in order to
of AC drives increasingly acceptable in the 1980s. achieve the responsiveness of a DC machine.
Vector-controlled AC drive systems have proved to
provide better dynamic performance than DC drive Fundamental Equations of Induction Machine
systems because of higher allowable speeds and shorter
time constants of AC motors. A two-pole, three-phase wye-connected symmetri-
cal induction machine is considered for the pur-
The typical induction motor is supplied by three
pose, as shown in Fig. 16 [6]. The stator windings are
alternating currents displaced by 120 which are then
identical, sinusoidally distributed displaced by 120
applied to a set of stationary coils. The resulting flux
from each other with Ns equivalent turns and resistance
from the stator induces alternating currents in the
rs. The rotor windings are also considered to be
“squirrel-cage” conductors of the rotor to create its
three identical sinusoidally distributed displaced by
own field. These fields interact to create torque. Thus
the rotor currents in an AC induction motor cannot be 120 from each other with Nr equivalent turns and
resistance rr.
directly controlled from an external source. As a result,
the torque, which is dependent on the interaction of The voltage equations for the stator and rotor
circuits can be written as:
the stator field and the consequential rotor current,
cannot be inherently controlled to obtain a desirable Vas ¼ rs ias þ plas ð58Þ
torque.
Vbs ¼ rs ibs þ plbs ð59Þ
Vector control or field-oriented control of an AC
induction motor is analogous to the control of Vcs ¼ rs ics þ plcs ð60Þ
a separately excited DC motor. In a DC motor, as
Vas ; Vbs ; and Vcs are the three-phase balanced
shown in Fig. 15, the field flux Фf is produced by the
voltages which rotate at supply frequency.
field current If. Ia is the armature current which pro-
duces the armature flux Фa. The field flux Фf and the las ; lbs ; and lcs are the stator flux linkages.
armature flux Фa are perpendicular to each other. Thus Var ¼ rr iar þ plar ð61Þ
the fields are decoupled and stationary with respect to
each other. The armature current is controlled to con- Vbr ¼ rr ibr þ plbr ð62Þ
trol torque. The field flux is unaffected thus allowing
Vcr ¼ rr icr þ plcr ð63Þ
a fast-transient response.
The purpose of field-oriented control is to recreate Var ; Vbr ; and Vcr are the rotor voltages and the
these orthogonal components in the AC machine so as rotor flux linkages which rotate at rotor speed, or .
 AC Machines: Permanent Magnet Synchronous and Induction Machines 45

bs axis

br axis
cs
×

× ar axis
br bs
×

qr
× cr

as axis

ar
×

×
as

cs axis

cr axis

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 16

Two-pole, three-phase symmetrical induction machine

The above equations can be written in short as: The inductance matrix for the stator can be written as
following:
Vabcs ¼ rs iabcs þ plabcs ð64Þ 2 3
Lls þ Lms  1=2Lms  1=2Lms
Vabcr ¼ rs iabcr þ plabcr ð65Þ 6 7
Ls ¼ 4 1=2Lms Lls þ Lms  1=2Lms 5 ð69Þ
where 1=2Lms  1=2Lms Lls þ Lms
ðVabcs ÞT ¼ ½Vas Vbs Vcs  ð66Þ Lls and Lms are the leakage and magnetizing induc-
tances of the stator winding, respectively.
ðVabcr ÞT ¼ ½Var Vbr Vcr  ð67Þ
Similarly, the inductance matrix of the rotor
The subscripts “s” and “r” denote that the variables winding is written as:
and parameters are associated with the stator and the 2 3
rotor, respectively. The flux linkages may be written as: Llr þ Lmr  1=2Lmr  1=2Lmr
6 7
" #  #" # Lr ¼ 4 1=2Lmr Llr þ Lmr  1=2Lmr 5 ð70Þ
labcs Ls Lsr iabcs 1=2Lmr  1=2Lmr Llr þ Lmr
¼ ð68Þ
labcr ðLsr ÞT Lr iabcr
Llr and Lmr are the leakage and magnetizing induc-
Ls and Lr are the winding inductance matrices of tances of the rotor winding, respectively.
the stator and the rotor, respectively. Lsr is the The mutual inductance matrix can be written
mutual inductance of the stator and rotor windings. as:
46 AC Machines: Permanent Magnet Synchronous and Induction Machines

2p 2p 3 0 Ns 2
Cosðyr Þ Cosðyr þ Þ Cosðyr  Þ rr ¼ rr ð80Þ
6 3 3 7 Nr
6 7
6 2p 2p 7
Lsr ¼ Lsr 6 Cosðyr  Þ Cosðyr Þ Cosðyr þ Þ 7
2
6 3 3 7 Ns
4 5 Lmr ¼ Lms ð81Þ
2p 2p
Cosðyr þ Þ Cosðyr  Þ Cosðyr Þ Nr
3 3
ð71Þ
2
0 Ns
Lr ¼ Lr ð82Þ
Lsr is the amplitude of the mutual inductance Nr
between the stator and rotor windings. yr is the rotor One may therefore write:
position. For a machine with P number of poles, 2 3
0
Llr þ Lms  1=2Lms  1=2Lms
yr ¼
P
ð72Þ 6 7
Lr ¼ 6 7
0 0
yrm ð83Þ
2 4 1=2Lms Llr þ Lms  1=2Lms 5
0
where, yrm is the mechanical rotor position. The 1=2Lms  1=2Lms Llr þ Lms
speed or in terms of the rotor mechanical speed orm where
for a P-pole induction machine may therefore be
2
0 Ns
expressed as: Llr ¼ Llr ð84Þ
Nr
P
or ¼ orm ð73Þ Thus the flux linkages may be written as:
2
" #  #" #
It is convenient to express the physical variables of labcs Lsr
0
iabcs
Ls0
the rotor referred to the stator variables by appropriate ¼ ð85Þ
0
labcr ðLsr ÞT Lr 0
iabcr
turns-ratios:
0 Ns
V abcr ¼ Vabcr ð74Þ
Ns Power Electronic Inverter
0 Ns
iabcr ¼ iabcr ð75Þ The inverter consists of six switching devices which
Nr
produce three-phase AC power from a DC source.
0 Ns Figure 17 displays the inverter topology.
labcr ¼ labcr ð76Þ
Nr The DC voltage is assumed to be constant, and the
transistors are assumed to be ideal switches. From
The resistive and inductive elements of the rotor
Fig. 17 and Kirchhoff ’s Voltage Law.
coil as referred to the stator windings may be written as:
Ns Vag ¼ Vas þ Vng ð86Þ
Lms ¼ Lsr ð77Þ
Nr
0 Ns Vbg ¼ Vbs þ Vng ð87Þ
Lsr ¼ Lsr ð78Þ
Nr
Vcg ¼ Vcs þ Vng ð88Þ
Thus, one can conclude that:
2 2p 2p 3 Adding equations Eq. 86 to Eq. 88 yields,
Cosðyr Þ Cosðyr þ Þ Cosðyr  Þ
6 3 3 7
6 7
6 2p 2p 7
Lsr ¼ Lms 6 7
0

6 Cosðyr  3 Þ Cosðyr Þ Cosðyr þ Þ 7 Vag þ Vbg þ Vcg ¼ Vas þ Vbs þ Vcs þ 3Vng ð89Þ
6 3 7
4 5
2p 2p
Cosðyr þ Þ Cosðyr  Þ Cosðyr Þ
3 3 By Kirchhoff ’s Current Law,

ð79Þ ias þ ibs þ ics ¼ 0 ð90Þ

 AC Machines: Permanent Magnet Synchronous and Induction Machines 47

+ ias
+
Vas
–

a
Vdc
ibs + Vbs –
b n
c

–
Vcs
–
+
ics
g

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 17

Inverter drive

which implies, a synchronous reference frame, the reference frame

rotates with the same frequency as the supply frequency
dias dias dias oe (see Fig. 18).
þ þ ¼0 ð91Þ
dt dt dt The transformation of variables to the synchronous
Thus, adding Eqs. 58 to 60 and substituting Eqs. 90 reference frame can be obtained by using
and 91 into this sum results in: a transformation matrix Ks ðye Þ such that:

Vas þ Vbs þ Vcs ¼ 0 ð92Þ f qd0 ¼ Ks ðye Þf abc ð97Þ

Thus Eq. 89 reduces to: where f can be any of the variables V, i, or l and Ks ðye Þ
can be expressed as:
Vag þ Vbg þ Vcg ¼ 3Vng ð93Þ
2 3
Solving Eq. 93 for Vcg and substituting this result Cosðye Þ Cosðye  2p
3 Þ Cosðye þ 3 Þ
2p
24 2p 5
into Eqs. 86–88 yields: Ks ðye Þ ¼ Sinðye Þ Sinðye  2p
3 Þ Sinðye þ 3 Þ
3 1 1 1
2 2 2
2 1 1 ð98Þ
Vas ¼ Vag  Vbg  Vcg ð94Þ
3 3 3
2 1 1 Rt
Vbs ¼ Vbg  Vag  Vcg ð95Þ where ye ¼ oe ðxÞ dx þ yeo and yeo is the initial angle
3 3 3 0
2 1 1 of the reference frame.
Vcs ¼ Vcg  Vag  Vbg ð96Þ
3 3 3 Applying this transformation to the stator voltage
equations, one gets a set of equations transformed into
Thus, the given inverter line-to-ground voltages,
the synchronous frame of reference which can be
the machine line-to-neutral voltages may be calculated
expressed as follows:
using Eqs. 94–96.

Vqs ¼ rs iqs þ plqs þ oe lds ð99Þ

IM Drive in Synchronous Reference Frame
Vds ¼ rs ids þ plds  oe lqs ð100Þ
To reduce the complexity involved in analysis with
these differential equations, one can apply the refer- Vos ¼ rs ios þ plos ð101Þ
ence-frame theory to transform the variables and equa-
tions associated with the stator and the rotor to the Similarly, applying the transform to the rotor volt-
synchronous reference frame. In the case of age equations, one obtains a set of equations
48 AC Machines: Permanent Magnet Synchronous and Induction Machines

as qss qse we

cr drr dre
ar
wm qrr qre
wm we
bs cs
dss dse
br

ws = we ws = we
ws = 0
wr = we – wm wr = we – wm
wr = 0

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 18

Change in frame of references applied to three-phase induction machine

transformed into the synchronous frame of reference 3

which can be expressed as follows: LM ¼ Lms ð107Þ
2
0 0 0 0 0 2 0 3
Vqr ¼ rr iqr þ plqr þ ðoe  or Þldr ð102Þ Llr þ LM 0 0
Ks Lr Ks 1 ¼4 0 5
0
0 Llr þ LM
0 0 0 0 0 0
Vdr ¼ rr idr þ pldr þ ðoe  or Þlqr ð103Þ 0 0 Llr
ð108Þ
0 0 0
Vor ¼ rr ior þ plor
0
ð104Þ 2 3
LM 0 0
6 7
Ks Lsr Ks 1 ¼ 4 0 LM 05 ð109Þ
In Eq. 99 to Eq. 101, Vqs , iqs , and lqs are q-axis stator 0 0 0
variables. Vds , ids , and lds are the d-axis stator variables
and Vos , ios , and los are the zero-axis stator variables. Thus the flux linkage equations in the synchronous
Under balanced normal operating conditions, the zero- frame are expressed as:
axis variables have zero values. Similarly, the variables
0
with subscript “r” are the corresponding rotor lqs ¼ Lls iqs þ LM ðiqs þ iqr Þ ð110Þ
variables. 0
The flux linkage equation after transformation may lds ¼ Lls ids þ LM ðids þ idr Þ ð111Þ
be expressed in the qdo-axis as: los ¼ Lls ios ð112Þ
" # " #" # 0 0 0 0
lqdos Ks Ls Ks 1 Ks Lsr Ks 1 iqdos lqr ¼ Llr iqr þ LM ðiqs þ iqr Þ ð113Þ
¼ :
labcr Ks ðLsr ÞT Ks 1 Ks Lr Ks 1 iqdor 0 0 0 0
ldr ¼ Llr idr þ LM ðids þ idr Þ ð114Þ
ð105Þ 0 0 0
lor ¼ Llr ior ð115Þ
The terms of the matrix in Eq. 105 can be evaluated
to obtain the following forms: The expression for the electromagnetic torque in
2 3 terms of the reference frame variables can be written as:
Lls þ LM 0 0


6 7 3 P
Ks Ls Ks 1
0 0
¼4 0 Lls þ LM 05 ð106Þ Te ¼ LM ðiqs idr  ids iqr Þ ð116Þ
2 2
0 0 Lls
The electromagnetic torque in terms of flux link-
where ages can be written as:
 AC Machines: Permanent Magnet Synchronous and Induction Machines 49



3 P 0 0 0 0
Te ¼ ðl qr idr  l dr iqr Þ ð117Þ Current vector
2 2
r


 q
3 P
Te ¼ ðlds iqs  lqs ids Þ ð118Þ B
2 2
r
Fundamental Equations of the IM Drive Under
×
Field-Oriented Control
To understand the concept of vector control in an AC Machines: Permanent Magnet Synchronous and
induction machine, a current loop consisting of a coil Induction Machines. Figure 19
of length L, radius r, and with N number of turns Current loop
carrying a current i in the presence of a flux density
B, as shown in Fig. 19, is considered.
According to the Lorenz force equation, the torque iqsdr
s
acting on the current loop is given by: lqdr

q
Te ¼ 2BiNLr sin y ð119Þ ×

Thus, the magnitude of the torque is maximum when

the current vector is perpendicular to the direction of q axis
flux density, B. The same concept maybe extended to an
induction machine. Consider Fig. 20 representing
an induction machine torque production phenome-
non. The rotor current and flux linkage vectors respec-
tively are:
0  0 0 T
i qdr ¼ i qr i dr ð120Þ d axis

and AC Machines: Permanent Magnet Synchronous and

Induction Machines. Figure 20
0
h 0 0
iT
l qdr ¼ l qr l dr ð121Þ Induction machine torque production phenomenon

It is also known from Eq. 117

0


 lqre ¼ 0 ð124Þ
3 P 0 0 0 0
Te ¼ lqr idr  ldr iqr ð122Þ
2 2 This condition can be satisfied by choosing
this may be expressed as: a suitable ye on an instantaneous basis such that all


 the rotor flux is in the d-axis. The second condition is
3 P 0 0 that:
Te ¼ l qdr i qdr sin y ð123Þ
2 2
0
idre ¼ 0 ð125Þ
This is analogous to Eq. 117. Thus, one can con-
clude that for a given magnitude of flux linkage, torque This condition can be achieved if the d-axis stator
is maximum in the case when the rotor current vectors current is held constant.
are perpendicular to the flux linkage vectors. In vector Combining the d-axis stator and rotor flux linkage
0
control, the method to achieve the condition that the equations and the condition idre ¼ 0 results in:
rotor flux and rotor current vectors are always perpen-
dicular is twofold. The first condition is that: leds ¼ Lss ids
e
ð126Þ
50 AC Machines: Permanent Magnet Synchronous and Induction Machines

where Direct Field-Oriented Control

Lss ¼ Lls þ LM ð127Þ In direct field-oriented control, the magnitude and
angular position of the reference flux vector are either
and
measured or estimated from the stator voltage and
0 currents using flux observers. Hall sensors can be used
ldre ¼ LM ids
e
: ð128Þ
to measure magnetic fields. If the position of the sta-
Combining Eqs. 107 and 114, the torque equation tionary reference frame is set to zero then the rotor flux
may now be expressed as: linkage in the q-d-axes of the synchronous reference
frame may be written as:



3 P 0 0
" # " #" #
Te ¼  ldre iqre ð129Þ 0 0
2 2 lqre Cos ye Sin ye lqrs
0
¼ 0
ð133Þ
0
ldre Sin ye Cos ye ldrs
Since lqre ¼ 0, from the q-axis rotor flux linkage
equation, it is known that: 0
In order to achieve lqre ¼ 0 , the position of the
0 LM e synchronous reference frame can be defined as:
iqre ¼  i ð130Þ
Lrr qs 0 0 p
ye ¼ angleðlqrs  jldrs Þ þ ð134Þ
where 2

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Lrr ¼ Llr þ LM ð131Þ 0 0 2 0 2
ldre ¼ ðlqrs Þ þ ðldrs Þ ð135Þ
Combining Eqs. 129 and 130 results in Eq. 132. 0 0
s s
The mutual flux linkages lqm and ldm may be




3 P LM 0 regarded as measurable. The mutual flux linkage in
Te ¼  ldre iqs
e
ð132Þ
2 2 Lrr the q-axis may be expressed as:

Equations 126–132 are the equations of the IM 0 0

s
lqm ¼ LM ðiqs
s
þ iqrs Þ ð136Þ
drive under rotor flux-oriented control.

Field-Oriented Control Implementations Also, the q-axis rotor flux linkage may be written as:

The basis of the field orientation algorithm is to use the 0 0 0

lqrs ¼ Llr iqrs þ LM ðiqs
s
þ iqrs Þ ð137Þ
flux angle, usually the rotor flux angle, to decouple the
torque and flux components of the stator current.
The most challenging feature of field orientation is the From the Eqs. 134 and 135, the expression Eq. 138 is
estimation or measurement of the flux angle. The direct obtained.
field-oriented method, in which direct measurement of 0
0 L 0
flux is performed using flux sensing coils and Hall-effect lqrs ¼ rr lsqm  Llr iqs
s
ð138Þ
LM
devices, proved to be impractical for general use. The
indirect field-oriented control (IFOC) method has Similarly, for d-axis, the following equation is
become more common. In this case, the flux angle achieved:
is not measured directly but is estimated from the 0
0 Lrr s 0
equivalent circuit model and from measurements of ldrs ¼ ldm  Llr ids
s
ð139Þ
LM
rotor speed, stator current, and the voltage. Given
sufficiently correct current and position sensors, The block diagram of a vector control system for an
this method performs reasonably well over the entire induction motor employing direct flux-oriented
speed range. control is shown in Fig. 21.
 AC Machines: Permanent Magnet Synchronous and Induction Machines 51

s
l̂qdm
DC
Inverter IM
Source

iabcs
Current
Ks Rotor Flux
Control
Calculator
Te* 4LrrTe* e*
iqs qe
ˆ ¢e
3PLm l dr

¢e
¢e * e * l̂dr
ldr 1 ids
Lm

AC Machines: Permanent Magnet Synchronous and Induction Machines. Figure 21

Direct field-oriented control

Indirect Field-Oriented Control

DC
Inverter wr
In the direct field orientation scheme, the angular posi- Source IM
tion yr of the rotor flux vector is measured directly,
using the air gap flux sensors. The indirect approach on
iabcs
the other hand is based on the calculation of slip speed,
Current
o r required for correct field orientation, and the impo- Control
sition of this speed on the motor.
From the q-axis rotor voltage equation of the iqe*s
induction machine, Eq.140 is obtained. ∫
qe we

0 ¼ rr iqr þ plqr þ ðoe  or Þldr ð140Þ ide*s iqe*s 1 Σ

0 tr
e ide*s
Applying the condition of vector control lqr ¼ 0
Eq. 138 becomes:
AC Machines: Permanent Magnet Synchronous and
0
e
0 iqr
Induction Machines. Figure 22
oe ¼ or  rr 0 e ð141Þ Indirect field-oriented control
ldr

Using Eqs. 117 and 114 will result in Eq. 142.

Thus, field-oriented control can be established by
0 e
rr iqs imposing the speed on the machine as calculated by the
oe ¼ or þ 0 e ð142Þ
Lrr ids Eq. 143. A vector control system for an induction
motor based on the indirect vector control technique
which can also be written in terms of rotor time is shown in Fig. 22.
constant tr , as: The rotor flux and the torque are controlled in
e
a feed-forward manner. As a result, the performance
1 iqs of the system strongly depends on the accurate knowl-
oe ¼ or þ e ð143Þ
tr ids edge of the motor parameters, that is, the rotor time
52 AC Machines: Permanent Magnet Synchronous and Induction Machines
constant. The advantage of this method is that
a standard motor can be used whose rotor position
can be easily measured by an external sensor.
Future Directions
Induction motors are subject to many kinds of different
internal and external faults during their operation [8].
Often the motor can run with these faults but ulti-
mately it results in performance degradation of the
machine, production shutdowns of the industrial pro-
cess and sometimes permanent burnouts of the motor.
Such shutdowns are costly in terms of lost production
time, maintenance costs, and wasted raw materials.
Therefore, schemes and methods for the diagnosis
and detection of the external faults of IMs are major
areas of research. Stator faults are one of the most
commonly reported faults. Almost 30–40% of all
reported faults are related to stator [9]. Stator faults
are essentially insulation failures, which ultimately lead
to commonly known line-to-ground or line-to-line
faults. The primary causes for stator faults are high
stator core or winding temperatures; slack core lami-
nation; slot wedges and joints; loose bracing for end
winding; contamination due to oil, moisture, and dirt;
short-circuit or starting stresses; electrical discharges;
and leakage in cooling systems, to name a few. Still,
quite a few techniques are available to detect stator
faults. So, detection of stator faults in induction
machines has attracted much attention in recent years.
Fault-tolerant control for induction machine drives
is another interesting issue since it enhances the reli-
ability of the drive without inducing high-cost invest-
ment. Improving reliability of electric motor drives can
be achieved by improving the fault tolerance of con-
verter switch modules and the fault tolerance of the
motor control algorithm. Field-oriented fault-tolerant
control algorithms and PWM strategies under fault
conditions have been recently introduced, and their
performances are studied [10].
Accurate assessment of the electro-magneto-
mechanical energy conversion process is another sig-
nificant field of research which usually requires
a Multiphysics Finite Element (FEA) analysis [11].
Since FE analysis is a time-consuming numerical pro-
cess, lumped parameter models, simplified equivalent
magnetic circuits, and separated field analysis are
usually used for design and control of electric motor
and generator drives. This computational burden is
particularly visible in the design of generators for
wind energy harvest and motors for propulsion of
electric and plug-in hybrid electric vehicles where sig-
nificant field statistics are often used in the design
process. Field reconstruction method (FRM) is
a technique in which a minimal set of field analysis is
used to establish basis functions for the magnetic flux
densities in the machine. Once the basis functions are
established, the performance of the machine is
predicted under arbitrary speed and excitation. It has
been shown that the FRM can reduce the computa-
tional time by two or three orders of magnitude while
maintaining the same level of precision that is usually
accomplished in a two-dimensional FE analysis. Due to
its high computational efficiency, FRM has created new
opportunity for health monitoring, fault-tolerant
operation, and multiphysics design of electric
machines. Over the past 6 years, this technique has
been successfully applied to different types of machines
[11]. This method still needs more researches.
The main disadvantage of PMSMs is nonuniformity
in the developed torque (torque ripple) [12]. These
torque pulsations vary periodically with rotor position
and are reflected as speed ripple. This ripple is parasitic,
and can lead to mechanical vibration, acoustic noise,
and drive system problems. Therefore, minimizing the
torque ripple is of great importance for PMSM design.
Cogging torque is the consequence of interaction
(magnetic attraction) between rotor-mounted perma-
nent magnets field and the stator teeth, which produces
reluctant variations depending on the rotor position; it
is stator current independent.
In the literature, numerous methods for reducing
cogging torque, such as employing a fractional number
of slots per pole, skewing of magnets or stator lamina-
tion stack (slots), displacing and shaping the magnets,
optimizing magnet pole-arc-to-pole-pitch ratio, intro-
ducing auxiliary slots or teeth, etc., have been proposed.
Reducing acoustic noise in an interior permanent
magnet motor during the design stage is another major
issue related to IPMs [13]. Noise sources in electric
machines are broadly categorized as magnetic,
mechanical, electronic, and aerodynamic sources. Gen-
erally, the mechanical and magnetic sources are con-
sidered as main noise sources of electric motors
 AC Machines: Permanent Magnet Synchronous and Induction Machines
because the interaction between the harmonic of mag-
netic forces and the resonant frequency of mechanical
structure is the major cause of vibration which pro-
duces the noise. So, when designing high-speed perma-
nent magnet (PM) machine, an effort to obtain eddy-
current losses reduction in the magnets is necessary in
order to improve their performances and diminish the
risk of demagnetization. The iron losses in machines
however, are strongly affected by the frequency of the
excitation, which is proportional to the speed of the
machine. In order to reduce iron losses, it is important
to consider using alternative magnetic materials in the
core of the machines [14]. So, optimal design of
PMSMs which still is an open area of research can be
translated into minimization of the loss while reducing
the flux and torque ripples.
Bibliography
Primary Literature
1. Fahimi B (2007) Qualitative approach to electromechanical
energy conversion: reinventing the art of design in adjustable
speed drives. In: ICEMS international conference on electrical
machines and systems, Beijing, pp 432–439
2. Jiang W, Moallem M, Fahimi B, Pekarek S (2006) Qualitative
investigation of force density components in electromechan-
ical energy conversion process. In: IEEE industrial electroniocs,
IECON 2006 – 32nd annual conference, Paris, pp 1113–1118
3. Arroyo ELC (2006) Modeling and simulation of permanent
magnet synchronous motor drive system. Master thesis in
electrical engineering, University of Puerto Rico
4. Strnat KJ (1990) Modern permanent magnets for applications
in electro-technology. Proc IEEE 78(6):923
5. Fahimi B (2002) Fault tolerant motor drives. In: Jurgen RK (ed)
Handbook of automotive electronics. CRC Press, Boca Raton
6. Krause PC, Wasynczuk O, Sudhoff SD (2002) Analysis of electric
machinery and drive systems, 2nd edn. Wiley Inter-Science,
New York
7. Rahman MF, Haque ME, Tang L, Zhong L (2004) Problems
associated with the direct torque control of an interior perma-
nent magnet synchronous motor drive and their remedies.
IEEE Trans Ind Electron 51(4):799–809
8. Yu H, Fahimi B (2008) A novel detection technique of non-
uniform airgap in frictionless linear induction transportation
systems. In: IEEE VEHICLE POWER AND PROPULSION CONFER-
ENCE (VPPC), Harbin, 3–5 Sept 2008
9. Nandi S (2006) Detection of stator faults in induction machines
using residual saturation harmonics. IEEE Trans Ind Appl
42(5):1201–1208
10. Bennani ABA, Ghodbane Cherif M, Belkhodja IS (2008) New fault
tolerant DTC control for induction machine drives. In: 13th
53
power electronics and motion control conference (EPE-PEMC),
Poznan
11. Fahimi B, Pekarek S (2010) Design and control of electric
machines utilizing a field reconstruction method (FRM). In:
36th annual conference on IEEE industrial electronics society,
IECON, Glendale
12. Guemes JA, Iraolagoitia AM, Del Hoyo JI, Fernandez P (2011)
Torque analysis in permanent-magnet synchronous
motors: a comparative study. IEEE Trans Energy Convers
26(1):55–63
13. Lee S, Hong J, Hwang SM, Lee WT, Lee JY, Kim YK (2009)
Optimal design for noise reduction in interior
permanent-magnet motor. IEEE Trans Ind Appl 45(6):1954–
1960
14. Liew GS, Ertugrul N, Soong WL, Gayler J (2006) An investiga-
tion of advanced magnetic materials for axial field brushless
permanent magnet motor drives for automotive applications.
In: 37th IEEE power electronics specialists conference
(PESC ‘06), Jeju
Books and Reviews
Belahcen A (1999) Overview of the calculation methods for forces
in magnetized iron cores of electrical machines. Presented at
the Seminar on modeling and simulation of multi-
technological machine systems, vol 29, pp 41–47
Desai PC, Krishnamurthy M, Schofield N, Emadi A (2010) Novel
switched reluctance machine configuration with higher num-
ber of rotor poles than stator poles: concept to implementa-
tion. IEEE Trans Ind Electron 57(2):469–476
Edrington CS, Kaluvagunta DC, Joddar J, Fahimi B (2005) Investi-
gation of electromagnetic force components in SRM under
single and multiphase excitation. IEEE Trans Ind Appl
41(4):978–988
Hayashi H, Nakamura K, Chiba A, Fukao T, Tungpimolrut K,
Dorrell DG (2009) Efficiency improvements of switched
reluctance motors with high-quality iron steel and
enhanced conductor slot fill. IEEE Trans Energy Convers
24(4):819–825
Holtz J (1996) Methods for speed sensorless control of AC drives.
In: Rajashekara K, Kawamura A, Matsuse K (eds) Sensorless
control of AC motor drives. IEEE Press, Piscataway
Hwu KI, Liaw CM (2003) Intelligent tuning of commutation for
maximum torque capability of a switched reluctance motor.
IEEE Trans Energy Convers 18(1):113–120
Khoobroo A, Fahimi B, Pekarek S (2008) A new field reconstruction
method for permanent magnet synchronous machines. In:
IECON international conference on industrial electronics,
Orlando, pp 2009–2013
Kioskeridis I, Mademlis C (2006) Optimal efficiency control of
switched reluctance generators. IEEE Trans Energy Convers
21(4):1062–1071
Krishnan R (2001) Switched reluctance motors drives: modeling
simulation, analysis, design, and applications. CRC Press, Boca
Raton
54 AC Machines: Permanent Magnet Synchronous and Induction Machines
Lin F, Yang S (2007) Instantaneous shaft radial force control with
sinusoidal excitations for switched reluctance motors. IEEE
Trans Energy Convers 22(3):629–636
Mao SH, Tsai MC (2005) A novel switched reluctance motor with
C-core stators. IEEE Trans Magn 41(12):3334–3336
Mecrow C, El-Kharashi EA, Finch JW, Jack AG (2004) Preliminary
performance evaluation of switched reluctance motors
with segmental rotors. IEEE Trans Energy Convers 19(4):
679–686
Zhu W, Fahimi B, Pekarek S (2006) A field reconstruction
method for optimal excitation of permanent magnet
synchronous machines. IEEE Trans Energy Convers 21(2):
305–313
 Active Multifocal Vision System, Adaptive Control of 55

Active Multifocal Vision System,
Adaptive Control of
ALOIS UNTERHOLZNER, HANS-JOACHIM WUENSCHE
Institute for Autonomous Systems Technology,
Department of Aerospace Engineering, University of
the Bundeswehr Munich, Neubiberg, Germany
Article Outline
Glossary
Definition of the Subject
Introduction: The Human Eye as a Template
Related Work
Design of the Visual Sensor
Realizing Visual Motor Routines
Saccades and Smooth-Pursuit Movements
Vestibulo-Ocular Reflex
Future Directions: The Importance of Visual
Perception
Bibliography
Glossary
Adaptive control The adaptation of the control mech-
anism to unknown or changing plant characteristics.
Multifocal vision system A set of cameras with differ-
ent focal length. Potentially the orientation of the
cameras can be actively changed. Equivalent to
camera platform.
Saccade Rapid, abrupt sensor movement to bring an
interesting object into the sensor focus.
Selective attention Selecting the most informative
parts within the sensor’s field of view for further
exploration.
Smooth-pursuit movement Slow sensor movement
to keep an object centered in the field of view.
Vestibulo-ocular reflex Sensor movement to compen-
sate disturbing motions of the sensor carrier.
Visual odometry Determining the position and ori-
entation of a moving camera by analyzing the
camera images.
Definition of the Subject
The perception of the environment is crucial for auton-
omous systems and visual perception in particular is
a substantial source of information for intelligent vehi-
cles. For a visual perception system of an intelligent
vehicle high measurement accuracy combined with
a large field of view is essential. Since these are contra-
dictory requirements for a single camera, multifocal
vision systems – equipped with tele- and wide-angle
cameras – are frequently used. However, due to their
small aperture angles high resolution telecameras have
to be actively directed toward scene regions of interest.
Moreover active inertial gaze stabilization is required
since telecameras are highly sensitive to rotational vehi-
cle motions. Gaze stabilization as well as adjusting the
orientation of the telecamera requires control of the
vision system. The use of adaptive control allows for
varying camera configurations and thereby further
enhances the applicability of multifocal vision systems.
Introduction: The Human Eye as a Template
Visual perception is one of the most important senses
for biological as well as for technical systems. However
for technical systems it is still challenging to extract
meaningful information from visual sensor input.
Due to its impressive performance in interpreting
visual scenes, the human sense of vision may serve as
a template for an artificial visual sense.
Similar to a camera in technical systems the eye is
the sensor of the human visual system. The human eye
provides high visual acuity in the central region of the
retina – the fovea – close to the optical axis. Simulta-
neously the eye offers a wide field of view due to a large
but low-resolution peripheral area surrounding the
fovea. Therefore peripheral vision delivers highly com-
pressed but less detailed data compared to the high-
resolution foveal vision. Since data compression already
takes place in the sensor, there is the need for a selec-
tion stage – often referred to as selective attention –
that decides which parts of the environment are most
informative and therefore should be discovered in
more detail.
Selective attention combined with high-resolution
foveal vision in turn requires directing the fovea
actively toward selected regions of interest (ROI).
Moreover during the measurement process the ROI
shall be kept foveated despite relative motions of the
ROI with respect to the sensor. Shifting the fovea

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
56 Active Multifocal Vision System, Adaptive Control of
toward a ROI as fast as possible is called a saccade.
Smooth-pursuit movements keep the selected ROI
within the fovea based on visual feedback. However
visual feedback is subject to large latencies and
therefore smooth-pursuit movements are inappropri-
ate to compensate for, e.g., fast head motions. The
vestibulo-ocular reflex (VOR) inertially stabilizes
the image on the retina based on measurements of the
vestibulo organ located in the inner ear. The combina-
tion of the vestibulo-ocular reflex and smooth-pursuit
movements ensures stable tracking of selected ROIs.
Binocular vision additionally requires vergence
movements – simultaneous movements of both eyes
in opposite directions – to keep a selected ROI foveated
in both eyes.
In summary the visual properties of the human eye
in combination with selective attention lead to
decreased data rates with increased information den-
sity compared to a sensor with high resolution in the
whole field of view.
An artificial visual sensor for intelligent vehicles
mimicking the desirable visual properties of the
human eye implies the need for a multifocal vision
system comprising at least one tele- and one wide-angle
camera. Such a sensor should be active and provide the
visual motor routines: saccade, smooth-pursuit
movement, and vestibulo-ocular reflex. The need for
vergence movements decreases with the increasing dis-
tance of the observed features. Therefore the capability
of vergence movements is not essential for intelligent
vehicles and can be omitted in favor of a considerable
reduction of the mechanical complexity of the resulting
multifocal vision system.
Before going into the details of the design and
control of active multifocal vision systems, this entry
gives a brief overview on related work. The design of
visual sensors for intelligent vehicles is discussed in the
subsequent section. In the following sections the real-
ization of the visual motor routines is presented. The
entry finally concludes with a discussion of possible
future research directions.
Related Work
This entry is mainly on the design and control of
multifocal vision systems. Yet to provide a more general
overview on the subject, literature on the human visual
sense is presented first.
A profound discussion of the human visual sense
including eye movements and visual attention is given
in [1]. A more general treatment of human perception
is presented in the textbooks [2] and [3]. Recent find-
ings regarding human gaze control and visual attention
can be found, e.g., in the reviews [4–9]. Human visual
perception during driving is investigated in [10–16],
for example.
Several approaches exist to the design and control
of active vision systems. Most of them are more or less
inspired by the human eye, mimicking to some extent
its sensing properties and visual motor routines. Vision
systems for robots – including humanoids – are
described, e.g., in [17–27]. Simulating human binocu-
lar vision is the purpose of the vision system presented
in [28]. The design and control of vision systems for
autonomous ground vehicles is discussed in [29–38].
The research on computational realizations of
visual attention may be divided into two – partially
overlapping – groups:
The first group deals mainly with computational
models of visual attention with the purpose of
identifying interesting scene regions and possibly
directing a camera toward this region. These
models mainly result in the computation of saliency
maps based on bottom-up feature extraction and
top-down weighting of these features, e.g., to find
a certain object of interest. However up to now
these approaches are not or rarely used for tracking
applications. The publications [39–58] provide by
no means a complete overview. Michalke et al. [59]
present an application of this approach in the con-
text of intelligent driver assistance systems.
In the second group the possibility of camera motions
is utilized to enhance the capabilities of a robotic
system. Some of the pioneering work on what is
commonly referred to as “active perception” or
“active vision” is presented in [60–64]. Dickmanns
[65] presents an early approach to vision-based
vehicle guidance. A lot of research is concerned
with active vision or gaze control respectively to
improve self-localization or obstacle avoidance of
(humanoid) robots: [66–76]. There are also some
works on simultaneous localization and mapping

(SLAM): [77–79]. Gaze control for autonomous
vehicles is presented, e.g., in [80–82] and [29].
Most of these works are based on simulations or
were conducted in comparably simple structured
and almost static environments.
Design of the Visual Sensor
Active multifocal vision systems – or equivalently cam-
era platforms – are able to provide simultaneously
a wide field of view, high resolution, and stereo vision.
A wide field of view is desirable to scan simulta-
neously as much of the vehicle’s environment as
possible. A high resolution provides detailed images
of points of interest in the vehicle’s surrounding.
Moreover telecameras are required for early detection
of distant objects, since the maximum range of object
detection limits the maximum speed of the vehicle.
Stereo vision is used to gather 3D information of the
environment.
Active vision systems that are pivotable in pitch and
yaw axis enable the exploitation of the high measure-
ment accuracy offered by telecameras within the entire
field of view of the vision system. Moreover the visual
field of the intelligent vehicle is enlarged this way.
However telecameras are more sensitive to distur-
bances than wide-angle cameras and therefore active
gaze stabilization is required to obtain stable images
and to keep objects of interest within camera scope.
For a sound discussion of these requirements and
the resulting design of multifocal vision systems refer
to [29] or [30]. According to [30], platforms pivotable
in pitch and yaw axis and equipped with several cam-
eras, that either provide a wide field of view or a high
resolution, offer a good compromise to meet the needs
of visual perception in autonomous ground vehicles.
Figure 1 depicts one possible multifocal camera
arrangement following the recommendations given
in [29].
Although multifocal vision systems offer a good
compromise covering more or less the needs of each
visual perception module of an intelligent vehicle, the
optimal camera configuration for a single perception
module can differ from the currently utilized camera
arrangement. To detect, e.g., lane markings black/white
cameras might be best; to detect drivable areas in off-
road terrain color cameras might be better. Divergent
Active Multifocal Vision System, Adaptive Control of 57
Stereo vision
High resolution
Wide
field of view
Active Multifocal Vision System, Adaptive Control of.
Figure 1
A multifocal vision system offering a wide field of view
combined with a high-resolution central area. Stereo vision
is possible within the intersection of the field of view of the
two wide-angle cameras
cameras offer a wide field of view. Parallel cameras, on
the contrary, increase the range where stereo vision is
possible. Depending on their range of interest different
perception modules require different focal lengths of
camera lenses. Therefore software designers prefer to
use the camera configuration that suits best their appli-
cation’s demands. For this reason the configuration of
the multifocal vision system of an intelligent vehicle
may be frequently changed during the development
process.
With every change of the cameras the extrinsic cam-
era parameters are lost and recalibration is required.
Moreover exchanging cameras may alter camera plat-
form dynamics due to different camera masses, which in
turn alters platform inertia leading to reduced motion
control performance of the multifocal vision system.
To overcome these problems the presented
multifocal vision system not only follows the design
recommendations given in [29], but additionally offers
the possibility to exchange cameras while retaining
camera calibration for later reuse of this configuration
as suggested in [83]. Two possible system configura-
tions are shown in Fig. 2. For easy exchange of the
cameras they are mounted on adapter plates (Fig. 2b).
These plates are attached to the yaw axis plate. For high
repeat accuracy the plates are positioned with adjust
pins relative to the yaw axis plate. Thus camera calibra-
tion is preserved as long as the cameras remain fixed to
the adapter plates. Pitch axis drive and position sensor
are mounted below the yaw axis plate. The pitch axis
camera is placed in the hollow shaft of the yaw axis
58 Active Multifocal Vision System, Adaptive Control of
a b
Active Multifocal Vision System, Adaptive Control of. Figure 2
Two possible configurations of the proposed multifocal vision system
drive. A mirror is used to reflect the horizontal view
toward the camera in the hollow shaft. Inertia of the
pitch axis is kept low this way since only the mirror has
to be rotated. Thus the weight of the pitch axis drive
can also be kept to a minimum, which in turn reduces
the inertia of the yaw axis. Placing the pitch axis camera
in the center of rotation of the yaw axis further reduces
yaw axis inertia.
Realizing Visual Motor Routines
An active multifocal vision system implies the need for
visual motor routines like saccades, smooth-pursuit
movement, and the vestibulo-ocular reflex.
Saccades are fast movements to a given reference
position. Smooth-pursuit movements are used to track
objects of interest or equivalently to keep selected
regions of interest foveated. Saccades and smooth-
pursuit movements require reference values either
from a single visual perception module, e.g., a vehicle
tracker or – based on the current situational context –
from an attention module. Both saccades (SAC) and
smooth-pursuit movements (SP) are based on feedback
control that directs the plant P – which is the multifocal
vision system – to a given reference position (Fig. 3).
The vestibulo-ocular reflex (VOR) provides active
gaze stabilization based on inertial measurements of
disturbing platform motions. A MEMS (micro-
electro-mechanical system) sensor S is placed on the
yaw axis plate of the multifocal vision system to mea-
sure the angular velocity Y _ p of the pitch axis induced
by vehicle rotations. Visual feedback from the camera C
is used to calibrate this open-loop gaze stabilization.
Due to the different control objectives, the realiza-
tion of saccades, smooth-pursuit movements, and the
Yaw axis drive
Yaw axis plate
Adapter
plate
Pitch axis
encoder Pitch axis
drive
Tele-lense inside
Pitch axis mirror hollow shaft
VOR S Θ̇p
SP/SAC P C
Active Multifocal Vision System, Adaptive Control of.
Figure 3
Control of the multifocal vision system comprising three
distinct controllers for saccades, smooth-pursuit
movements, and the vestibulo-ocular reflex
vestibulo-ocular reflex requires three distinct control-
lers (Fig. 3). The controllers for saccades and smooth-
pursuit movements are implemented for both the yaw
and the pitch axis of the multifocal vision system. The
vestibulo-ocular reflex is only realized for the platform
pitch axis, since disturbances like bumps, braking, or
accelerating mainly induce vehicle pitch rotations.
Saccades and Smooth-Pursuit Movements
Saccades can be considered to be a time-optimal con-
trol problem. Therefore a near time-optimal sliding-
mode control law (CSLC) is proposed for saccades
(Fig. 4). The smooth-pursuit movements are governed
by a state-space controller (CSSC) that is designed to
provide good tracking behavior. Both controllers share
a Kalman filter (KF) to derive full state information and
a recursive least squares (RLS) parameter estimator for
online identification of the parameters of the dynamic
model of plant P. The RLS estimator is necessary since
altering the camera configuration of the multifocal
 Active Multifocal Vision System, Adaptive Control of 59

of rotation of the platform yaw axis, g being the grav-

CSLC / SSC P itational acceleration, and yp being the adjustable pitch
angle of the platform yaw axis. The unbalanced mass
results from the pitch axis encoder outweighing the
KF
RLS
pitch axis drive and any mass differences of the used
camera/lens combinations. Defining r
m
g
sin(yp) =:
tu, the disturbance torque t’ can be written as
Active Multifocal Vision System, Adaptive Control of.
Figure 4 t’ ¼ tu
cosð’Þ þ c
’ ð4Þ
Block diagram control of saccade and smooth pursuit
With (2) and (4) the plant dynamics (1) can be
vision system may lead to a change in the platform summarized to:
dynamics – which is obvious from Fig. 2a and b. The € ¼ km
I  d
’_  c
’  tc
sgnð’Þ
J
’ _  tu
cosð’Þ
parameter estimates of the RLS estimator are used to ð5Þ
adapt the Kalman filter KF as well as the control laws
CSLC and CSSC. Based on the control error a switch The dynamic plant model of the platform pitch axis
determines which control law is applied. can be derived from the yaw axis model by neglecting the
angular position-dependent disturbance torque (4) in (5).
Dynamic Model
Parameter Estimation
The dynamic plant models of platform yaw and pitch
axis differ only slightly. Since the pitch axis model Since the model parameters J, d, c, tc, and tu are
emerges from the yaw axis model, the yaw axis model unknown, a parameter estimator is used to get the
is derived first. Plant dynamics is modeled as estimates ^J ; d^ ; ^c ; ^tc , and ^tu respectively. Discretizing
a nonlinear second-order differential equation as in (5) using the forward differences as suggested in [85]
[83] and [84]:

’_ ¼ ’kþ1  ’k =T ð6Þ
€ ¼ km
I  t’  t’_
J
’ ð1Þ
a discrete time plant model
I is the torque generating motor current, km the
motor torque constant, and J the inertia of the platform ’k ¼  a1
’k1  a2
’k2 þ b2
Ik2
’  ’ 
yaw axis. The angular velocity-dependent disturbance  d1
sgn k1 k2
 d2
cos ð’k2 Þ
torque t’_ is described by T
ð7Þ
_ þ d
’_
t’_ ¼ tc
sgnð’Þ ð2Þ
is derived which is linear in the parameters:
with tc representing Coulomb and d representing vis- a1 ¼ Td
2 2

J  2; a2 ¼ 1  J þ J ; b2 ¼ J
Td cT km T
_ being the signum
cous friction coefficient, with sgnð’Þ ð8Þ
2 2
function. The angular position-dependent disturbance d1 ¼ tc JT ; d2 ¼ tuJT
torque t’ results from twisted cables and from the
unbalanced mass m with respect to the platform yaw Then (7) can be written as linear regression which is
axis: suitable for the application of a RLS parameter estima-
tor. Parameter projection is used to constrain the
t’ ¼ r
m
g
sinðyp Þ
cosð’Þ þ c
’ ð3Þ parameter estimates to stay within reasonable limits
The torsion-spring model c
’ describes the torque as outlined in [86] and [83].
resulting from twisted cables. The term r
m
g
sin(yp)

State Estimation
cos(’) represents the disturbance torque resulting from
the unbalanced mass, with r being the distance from the Both state-space control and sliding-mode control
center of gravity of the unbalanced mass m to the center require full state information. Therefore, analog to
60 Active Multifocal Vision System, Adaptive Control of

[83] and [84], a state estimator is applied, using the can be derived. As a discrete time Kalman filter is used
angular position ’ as measurement input. Since the for state estimation, a discrete time description
plant model (5) is nonlinear, the parameter estimates
k ; yk ¼ Cxk ; with
xkþ1 ¼ Axk þ B u
^tc and ^tu are used to compensate for the nonlinear ð15Þ
disturbance terms in (5) with the disturbance xk ¼ ð’k ; ’_ k ; t d k ; t_ d k ÞT
feedforward:
of (14) is needed. Using (6) for discretization yields
1
Iff ¼ km
ð^tc
sgnð’Þ
_ þ ^tu
cosð’ÞÞ ð9Þ 0 1 0 1
1 T 0 0 0
B  cT J Td  T 0 C B km T C
In [85], a similar approach is suggested for adaptive B C B C
A¼B J J J
C; B ¼ B J C
friction compensation. The plant input current I can be @ 0 0 1 TA @ 0 A ð16Þ
written as: 0 0 0 1 0
I ¼ Iff þ Ifb þ Iref ð10Þ C ¼ ð1; 0; 0; 0Þ

with Iff resulting from feedforward control, Ifb resulting Replacing parameters J, c, and d at every time step k
from feedback control, and Iref resulting from the with their estimates Ĵk, ^c k , and d^k results in Âk and B
^k
reference input r of the axis controller. By inserting as state and input matrices respectively.
feedforward control (9) in the plant model (5) the
equation State-Space Control
€ ¼km
ðIfb þ Iref Þ  d
’_  c
’
J
’ State-space control is used for smooth-pursuit move-
 ðtc  ^tc Þ
sgnð’Þ
_  ðtu  ^tu Þ
cosð’Þ ments that demand a good tracking behavior. There-
fore the reference r is designed to track the desired axis
ð11Þ
angle ’ref up to a ramp signal without steady-state
is derived. error. Thus r is a linear combination of ’ref and its
Summarizing the remaining disturbance torques time derivative ’_ ref with gains w1 and w2:
resulting from parameter estimation errors tu  ^tu
r ¼ w1
’ref þ w2
’_ ref ð17Þ
and tc  ^tc , as well as from unmodeled disturbances
in one disturbance torque td, (11) can be written as Following the controller design in [83] and [84], the
€ ¼ km
ðIfb þ Iref Þ  d
’_  c
’  td
J
’ ð12Þ steady-state error of the discrete time tracking error
’ref ðzÞ  ’ðzÞ is calculated by applying the final
€td ¼ 0 is used as a disturbance model, as this is value theorem. Provided ^c and d^ approach their true
found in [87] to be a good compromise between model values, the steady-state tracking error becomes zero if
complexity and estimation accuracy in the context of w1 and w2 are chosen as follows:
friction compensation.
Using the continuous time state-space formulation ^c d^
w 1 ¼ k1 þ and w2 ¼ k2 þ ð18Þ
km km
x_ ¼ Fx þ G
u; y ¼ Cx; with
ð13Þ
_ td ; t_ d ÞT ; u
x ¼ ð’; ’;  ¼ Ifb þ Iref Using the estimated state vector x^k and the distur-
bance feedforward Iff together with the discrete time
from (12) and disturbance model €td ¼ 0 the state- reference input rk yields the control law:
space description
^td
0
0 1 0 0
1 0
0
1 ^ k k2k ’
uk ¼ rk  k1k ’ ^_ k þ k
km
B c d 1 0 C B km C ð19Þ
B C B C ^tck ^
t
^_ k Þ þ k cosð^
u
F ¼B J J J
C; G ¼ B J C; þ sgnð’ ’k Þ
@ 0 0 0 1 A @0A ð14Þ km km
0 0 0 0 0
Applying the control law (19) to the plant (5),
C ¼ ð1; 0; 0; 0Þ results in a second-order closed-loop system. As
 Active Multifocal Vision System, Adaptive Control of 61

described in [83], the controller gains k1 and k2 are .

s1 s2 s0 e1
determined by the comparison of coefficients with the 30
desired second-order closed-loop characteristic poly- 20
nomial. The derived gains are 10
  0 e1
k1 ¼ ð^J o20  ^c Þ=km and k2 ¼ 2zo0^J  d^ =km −10
−20
ð20Þ −30
−1.5 −1 −0.5 0 0.5 1 1.5
with closed-loop natural frequency o0 and damping z.
Damping is chosen z >1 to avoid overshoot and o0 is Active Multifocal Vision System, Adaptive Control of.
calculated as follows [83, 88]: Figure 5
Phase trajectories and sliding-lines [84]
4:6
o0 ¼ pffiffiffi ð21Þ
z ^J
The system motion of a system under sliding-mode
The constant  is used to adjust the influence of control consists of two modes: the reaching-mode and
the estimated inertia Ĵ on the desired closed-loop the sliding-mode [91]. During the reaching-mode the
dynamics o0. phase trajectory of the system (Fig. 5) starts from
anywhere in the phase plane and moves toward the
Sliding-Mode Control sliding-line s = 0. After reaching the sliding-line s = 0
Saccadic movements are a typical application for time- the phase trajectory will stay on it, provided the
optimal control. Time-optimal control problems can reaching condition is satisfied. Then the system is in
be solved using Pontryagin’s minimum principle [89], the so-called sliding-mode and the system motion is
but the solution is sensitive to unknown plant dynam- only governed by the differential equation s = 0. With
ics or disturbances. Moreover it is difficult to derive the system under control
a solution analytically for nonlinear systems. u ¼ K sgn ðsÞ; K > 0 ð22Þ
Sliding-mode control on the contrary is an inher-
ently robust method to cope with uncertain plant and using the Lyapunov function candidate
dynamics. If the controlled system has reached the 1
sliding-mode, it is insensitive to disturbances and V ðsÞ ¼ s 2 ð23Þ
2
system motion is determined by the sliding-surface
a global reaching condition [91, 94] is given by:
only [90, 91].
: :
There are several approaches that try to utilize the V ðsÞ ¼ s s < 0; s 6¼ 0 ð24Þ
robustness properties of sliding-mode control, while :
simultaneously attaining near time-optimal control That means s > 0 indicates s < 0 and vice versa. Thus
performance, e.g., [92–94]. the phase trajectories will always move toward and stay
These control law design methods are offline on s = 0 after reaching it.
methods and therefore the derived controllers are not In order to apply sliding-mode control, plant
suitable for online adaptation. The design of the dynamics (5) is rewritten as a set of first-order differ-
presented saccade controller is based on the minimum ential equations. Defining the control error ’ – ’ref :=e1
time sliding-mode control law for second-order and noting that ’ref is constant yields
systems derived in [94]. However the sliding-line :
e1 ¼ e2
proposed in [94] is subject to stability constraints : : : ð25Þ
e2 ¼ ðf ð’Þ þ g ð’Þ þ hð’; ’Þ þ uÞ=J
that prevent the control from being time-optimal.
: : : ::
Analog to [84], the proposed sliding-line is designed with e1 ¼ ’ and e2 ¼ ’ . The disturbance torque is
:
to relax these constraints and thus achieve near time separated into three categories: f ð’Þdescribes the fric-
:
optimality. tion torque, with the property f > 0 if ’ < 0 and vice
62 Active Multifocal Vision System, Adaptive Control of

versa. Secondly g(’) is a spring torque and is assumed counteracts the actuator torque and can thus be
to be monotonically decreasing, i.e., ∂g/∂’ < 08’. neglected in the calculation of a. Therefore a is inde-
:
Finally, hð’; ’Þ sums up the remaining disturbance pendent from fmax and gmax, which is a desirable
torques that cannot be described by f or g. u is the property. Figure 5 depicts some switching surfaces of
plant input torque. system (25) with ’ref = 0. The time-optimal switching-
For a second-order system the nonlinear switching line is s0. s1 and s2 are the switching-lines from (26)
surface with a according to (28) or (29) respectively. s2 lies
in between s0 and s1, indicating that using (29) to
s ¼ ae1 þ je2 je2 ¼ 0 ð26Þ
calculate a yields a faster transient response than
is proposed in [94]. Solving the differential equation using (28).
(26) for constant a, the time span t needed to approach Comparing system (25) with the plant dynamics
e1 = 0 is calculated as follows: (5), one derives
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi f ¼ d
’  tc
sgnð’_ Þ
t ¼ 2 je1 ðt ¼ 0Þj=a ð27Þ
g ¼ c
’ ð30Þ
Provided that (24) is satisfied, it is obvious from h ¼ tu
cos ð’Þ
(27) that the resulting transient response is faster for
higher a. Thus a has to be maximized to reach near for the disturbance torques f, g, and h. The maximum
time-optimal control. For a double-integrator plant the control torque K in (22) corresponds to the maximum
sliding-line (26) becomes identical to the time-optimal actuator torque km
Imax.
switching-line for a ! 2K [94], with K being the Since the plant parameters are not known in
maximum control torque. In [94] it is argued that if advance, the parameter estimates derived from the
the plant is not a pure double integrator but subject to RLS parameter estimator are used (cf. section “Param-
disturbances as, e.g., system (25), the maximum values eter Estimation”) for the calculation of the sliding-line.
of the disturbances have to be considered in the Moreover the state estimates are obtained from the
calculation of a. Otherwise there are no resources left Kalman filter introduced in section “State Estimation.”
to compensate for the disturbances. Following this Thus the sliding-line is
argumentation, it can be shown that the reaching    
s¼a ’ ^ k  ’ref þ ’ ^_ k ’
^_ k ð31Þ
condition (24) is satisfied for system (25) with control
(22) and switching surface (26) for with
2 2   
a < ðK  fmax  gmax  hmax Þ ð28Þ a< K  ^c
’
^ k
sgn ’^ k  ’ref  j^tu j ð32Þ
J ^J

But supposing that, e.g., f describes viscous friction Switching Strategy

:
f ¼ d ’ , fmax depends linearly on the largest possible To switch between state-space control and sliding-
:
angular velocity ’ max . Thus a is too conservative for mode control the simple hysteresis
: :
’  ’ max .  
 
The proposed approach utilizes the properties of IF ’ref  ’ > eu THEN saccade ¼ true
the disturbance torques f and g to increase a, leading to  
 
a faster transient response of the control loop: IF ’ref  ’ > el THEN saccade ¼false

2
a < ðK þ g
sgnðe1 Þ  hmax Þ ð29Þ is used, with eu > el > 0. If the control error ’ref ’
J exceeds the upper threshold eu, a saccade is carried
The term g
sgn(e1) is chosen such that a increases if out and sliding-mode control is active. If the control
g reduces acceleration or supports deceleration. On the error goes below the lower threshold el, state-space
contrary a decreases if g supports acceleration or control becomes active until the control error exceeds
reduces deceleration. The friction torque f always eu again.
 Active Multifocal Vision System, Adaptive Control of 63

Experimental Results 10

j [°]
0
Some experimental results described in [84] demon-
−10
strate the validity of the presented saccade and smooth-
0 0.05 0.1 0.15 0.2 0.25
pursuit control laws. The two configurations of the
multifocal vision system shown in Fig. 2 were used 20
for the experiments. These configurations equipped

j [°]
0
with three (Fig. 2b) and five cameras (Fig. 2a) are called
−20
3c- and 5c-configuration respectively. Saccades with 0 0.05 0.1 0.15 0.2 0.25
three different step sizes were conducted. For every
step size the results of the sliding-mode control are 40
20

j [°]
compared with the state-space control, this way dem- 0
onstrating the superior performance of the proposed −20
near time-optimal sliding-mode control over state- 0 0.05 0.1 0.15 0.2 0.25
space control with respect to saccades. The step sizes Time [s]
were 10 , 20 , and 30 (Fig. 6). Every plot of
Active Multifocal Vision System, Adaptive Control of.
Fig. 6 depicts four saccades: two for each platform
Figure 6
configuration, resulting from sliding-mode control
Saccades [84]: Black lines are reference values. Blue lines
and state-space control respectively.
correspond to the 3c-configuration, green lines to the
For fast saccades maximum acceleration followed
5c-configuration. The dashed lines result from the sliding-
by maximum deceleration is required. Sliding-mode
mode control, the solid lines from state-space control
control design is explicitly based on the maximum
achievable acceleration and deceleration. That is why
sliding-mode control suits saccadic movements better 10 ±10° saccade
I [A]

than state-space control. 0

From Fig. 6 it is obvious that saccades under state- −10
space control are either too slow or result in overshoot. 0 0.05 0.1 0.15 0.2 0.25
The results for sliding-mode and state-space control 10 ±20° saccade
are similar only for a small operating range as can be
I [A]

0
seen from the 20 -saccade. −10
Figure 7 depicts the controller output 0 0.05 0.1 0.15 0.2 0.25
corresponding to the saccades of the 5c-configuration 10 ±30° saccade
shown in Fig. 6. For the 20 -saccade the controller
I [A]

0
output of state-space control and sliding-mode control
−10
overlap to a large extent. In particular the switching 0 0.05 0.1 0.15 0.2 0.25
from acceleration to deceleration and vice versa takes Time [s]
place at nearly identical points in time. Therefore the
Active Multifocal Vision System, Adaptive Control of.
resulting saccades are also similar. During the 30 -
Figure 7
saccade the state-space control output changes from
Controller output for saccades of the 5c-configuration [84]:
acceleration to deceleration later than the sliding-mode
Solid lines correspond to state-space control, dashed lines
control output. Also the deceleration phase is finished
to sliding-mode control
later, resulting in considerable overshoot. For the
10 -saccade the state-space controller starts deceler-
ating earlier than the sliding-mode controller and thus performance is close to time-optimal. For practical
the corresponding saccade is finished later. applications, however, it is preferable to avoid
The switching activity (chattering) of the sliding- chattering completely, since it may cause increased
mode controller is low, indicating that the control hardware attrition and induces audible noise.
64 Active Multifocal Vision System, Adaptive Control of

Replacing the signum function in (22) with the satura- saccades and smooth-pursuit movements can be
tion function simulated since it offers fast (saccade like) changes
8 as well as continuous change over time with
< K s for s  jlj
u ¼ K satðsÞ ¼ l ð33Þ a constant slope. The output of the switching hysteresis
:
K sgnðsÞ for s > jlj proposed in section “Switching Strategy” is depicted in
red. Both smooth-pursuit movements and saccades
is a possible alternative. The saturation function intro- show satisfactory control behavior. Furthermore the
duces a boundary layer around the sliding-line s. Out- switching between both controllers is smooth and
side the boundary layer the saturation function equals without noticeable influence on the control
the signum function. Inside the boundary layer the performance.
sliding-mode control is replaced by a high-gain pro-
portional control, thereby avoiding the discontinuity of Vestibulo-Ocular Reflex
the signum function for s = 0.
The use of high resolution telecameras enhances the
The sliding-lines for the 3c- and 5c-configuration
perceptional capabilities of autonomous vehicles. It
are shown in Fig. 8. Note that the slope of the sliding-
enables both the early detection of distant objects and
line of the 3c-configuration is higher than the slope of
provides detailed images of points of interest in the
the 5c-configuration. This results mainly from the
vehicles surrounding. But telecameras are more sensi-
influence of the estimated inertia Ĵ on the calculation
tive to disturbances – especially rotational vehicle
of a in (32). Obviously the inertia of the
motion – than wide-angle cameras.
3c-configuration is smaller than the inertia of the
To obtain stable images and keep objects of interest
5c-configuration.
within camera scope active gaze stabilization is neces-
Moreover the resulting phase trajectories of
sary. In [29, 33], inertial measurements of the rota-
the 20 -saccade for the 3c-configuration and the
tional vehicle motion are used for active image
30 -saccade for the 5c-configuration are depicted in
stabilization in autonomous vehicles.
Fig. 8. After reaching the sliding-line, the trajectories
This is a pure feedforward control, since image-
stay in a small vicinity of the sliding-line, due to the
based information is not used for the stabilization.
nonideal components of a real-world control loop.
Residual image motion may be induced by errors of
A sawtooth signal is used as position reference to
the inertial angular-rate sensor, such as sensor offset,
test the switching between sliding-mode control and
wrong or changing sensor sensitivity, errors in the
state-space control (Fig. 9). With the sawtooth signal

1000
10
500
e1 [°/s]

0 5
j [°]

−500 0
−1000 −5
−60 −40 −20 0 20 40 60 −10
e1 [°] −1 0 1 2 3 4 5 6 7 8 9
Time [s]
Active Multifocal Vision System, Adaptive Control of.
Figure 8 Active Multifocal Vision System, Adaptive Control of.
Sliding surfaces and phase trajectories [84]: The Figure 9
blue line corresponds to the phase trajectory of the Saccades and smooth-pursuit movements [84]: The black
3c-configuration, the green line to the trajectory of line is the reference value. The blue line corresponds to the
the 5c-configuration. The black lines are the corresponding 3c-configuration. The red line states which control law is
sliding surfaces active: state-space or sliding-mode control

mounting position of the sensor, or imperfections in
the control loop of the stabilizing actuator.
Therefore in more recent works visual information
is incorporated in the stabilization process. Either
information from inertial sensors and vision is fused
as, e.g., in [24], or the inertial sensors are calibrated
based on visual information as in [27, 36, 37]. The
latter approach is similar to the vestibulo-ocular reflex
(VOR) that many vertebrates use for gaze stabilization.
The VOR not only stabilizes the image on the retina, but
also continuously adapts to erroneous measurements of
the vestibular organ located in the inner ear [37]. This
adaptation process is based on visual feedback.
In [27, 36, 37] sensor and actuator errors are com-
pensated by means of an adaptive FIR (finite impulse
response) filter as suggested in [95]. The filter adapta-
tion is driven by optical flow. The camera platform in
[27] executes only rotational motions in an indoor
environment. Thus optical flow results solely from
rotational motion. In [36, 37] the camera platform is
mounted behind the windshield of a car. Optical flow
computation is based on distant features. This way the
translational platform motion has little impact on the
resulting optical flow. But calculating the optical flow
by utilizing distant features only is not always possible,
consider, e.g., a car driving on a forest track or in an
inner-city environment.
The presented active inertial gaze stabilization is
inspired by the vestibulo-ocular reflex too. In contrast
to the aforementioned works, however, the sensor and
actuator errors are modeled explicitly. From those error
Embedded PC
RLS
Θ̇r
p̂
z−D
F(p̂)
SP/SAC
Active Multifocal Vision System, Adaptive Control of. Figure 10
Block diagram of the inertial gaze stabilization and its calibration
Active Multifocal Vision System, Adaptive Control of 65
models a filter transfer function is derived that is suit-
able to modify the sensor signal such as to compensate
for the overall stabilization error. The adaptation of the
unknown filter parameters is driven by the residual
camera motion.
Calibrating the Inertial Gaze Stabilization
Analog to [96], the realization of the vestibulo-ocular
reflex includes the open-loop gaze stabilization based
on an MEMS angular-rate sensor and the calibration of
this sensor driven by visual feedback (Fig. 10).
The angular-rate sensor S measures the pitch veloc-
ity Y_ p of the camera platform. After passing the filter
Fð^pÞ the sensor signal Y _ s is the reference value for the
control (SP/SAC) of the gaze stabilizing plant P.
A mirror together with an actuator is used for gaze
stabilization. In fact the reference value for the control
of P is half the size of the filtered sensor signal, since for
a mirror the incident angle equals the emergent angle.
Image correspondences computed from already
stabilized images from camera C together with the
vehicle velocity x_ and the vehicle steering angle l
serve as measurements for the camera motion estima-
tion (CME) based on an unscented Kalman filter
(UKF) [97]. The CME is used to estimate the residual
pitch velocity Y _ r , which is the difference between the
real platform velocity Y _ pl and two times the mirror
_ _
velocity Ym . Yr again is used as an error signal
for a recursive least squares (RLS) parameter estimator
that estimates the parameters ^p of the filter Fð^pÞ in
PC Platform
CME
ẋ, λ, Θm
S C
Θ̇s
P
M
Θ̇m
Θ̇p
66 Active Multifocal Vision System, Adaptive Control of

a way that the residual image motion is minimized. To reduces to

consider the latencies D resulting from image
Y _p
_ m ¼ 0:5 Y ð38Þ
processing in the parameter estimation process, the
sensor signal Y _ s is artificially delayed by D before
which is the desired optimal gaze stabilization.
entering the RLS parameter estimator. To get estimates ^p of the unknown parameters p
with a discrete time RLS parameter estimator
Compensating the Open-Loop Errors (Fig. 10), the filter equation (36) has to be discretized.
Since the sampling interval T is small compared to the
To derive an adaptive filter that compensates for the vehicle’s pitch dynamics, backward differences
open-loop errors of the inertial gaze stabilization, _
xðkÞ ¼ xðkÞxðk1Þ can be used for discretization:
T
the sensor and actuator characteristics are modeled
explicitly as in [96]. ^
_ F ðkÞ ¼ ^ta þT Y
Y _ s ðkÞ  ^ta Y _ s ðk  1Þ  ds
The MEMS angular-rate sensor signal ^k a ^k s T ^k a ^k s T ^k a ^k s
_ s ¼ ks Y
Y _ pl þds ð34Þ ð39Þ

is considered to be biased resulting from an offset ds. The resulting discretized filter equation (39) is
Moreover the sensor is thought to have a sensitivity ks linear in its parameters and can be written as a linear
differing from the nominal sensitivity. Please note regression
that the sensitivity ks may also be influenced by the 2 3
^b
mounting position of the sensor, e.g., when the sensor  T 6 0 7
_ _ _
YF ðkÞ ¼ Ys ðkÞ; Ys ðk  1Þ; 1 4 ^b1 5
is not exactly aligned with the axis of rotation of the
ð40Þ
mirror. d^
The stabilizing mirror and its control are modeled ¼: fT ðk  1Þ^y
as a first-order transfer function in the Laplace domain
which is suitable for the RLS parameter estimation
_ m ¼ 0:5ka Y
Y _s ð35Þ [86].
ta s þ 1 Usually the RLS parameter update is driven by the
with the time constant ta describing the bandwidth of difference of a measured signal y and a prediction ^y of y
the control loop and the gain ka. The gain ka describes based on the estimated parameters ^y. The difference
the steady-state deviation of the mirror velocity Y _m y  ^y is a measure for the RLS parameter estimation
_
from half of the measured sensor value Ys . error. The residual camera pitch velocity Y _ r can also
The goal of the adaptive filter design is to recon- be considered to be a measure for the parameter
struct the real platform pitch velocity Y _ p from the estimation error. Unless Y _ r ¼ 0 the estimated param-
_
sensor signal Ys and to compensate the influence of eters ^p of filter Fð^pÞ are not equal to their true values
the actuator control loop on the gaze stabilization. The p. Therefore equation
filter equation is chosen as follows:
^yðk þ 1Þ ¼ ^yðkÞ þ K ðkÞ Y
_r ð41Þ
_ F ¼ Fð^pÞ ¼ ^ta s þ 1 ðY
Y _ s ^
ds Þ ð36Þ
^k a ^k s can be used for the parameter update, with K ðkÞ being
the parameter update gain of the RLS parameter estimator.
Using the output of filter (36) as input to the mirror
control loop (35) and inserting (34) in (36) results in
Camera Motion Estimation
_ m ¼ 0:5ka ^ta s þ 1 ðks Y
Y _ p þds  ^
ds Þ ð37Þ The estimation of the unknown residual camera
ta s þ 1 ^k a ^k s
motion (CME) is based on image data. This is com-
If the estimated parameters ^p ¼ ½^ta ; ^k a ; ^k s ; ^ds T monly referred to as visual odometry. The estimation of
approach their true values p = (ta, ka, ks, ds)T (37) non-holonomic vehicle motion in real time was

presented, e.g., in [98, 99]. An overview on the relevant
literature is also given therein. Visual odometry relies on
the calculation of image correspondences between
consecutive image frames. A method to calculate
these correspondences in real time is given in [100].
Most works on visual odometry aim at estimating the
camera motion from visual data only. However to enhance
the robustness of the CME the vehicle velocity and the
vehicle steering angle are incorporated in the estimation
process as suggested in [96]. These measurements are
available on most ground-moving vehicles.
One fundamental problem in visual odometry is the
determination of the depth of image keypoints. To
circumvent this, the drivable area in front of the vehicle
is assumed to be an ideal ground plane. Considering
only image correspondences corresponding to that
ground plane implicitly determines the depth of the
image keypoints due to the known camera position
relative to the ground plane.
As a tribute to computation speed, the calculated
image correspondences include some false positives
[100]. Thus a preselection of valid correspondences is
necessary; otherwise the estimation performance may
be corrupted.
The gaze stabilization is independent of the camera
motion estimation. Only the calibration of the angular-
rate sensor is based on the estimated camera motion.
But the calibration can be switched off, whenever the
estimated camera motion seems to be unreliable. This
may be the case, e.g., if too little image correspondences
are available or if the ground plane assumption is
invalid.
The calibration of the gaze stabilization is driven by
the estimated residual camera motion. However this
estimation is not the main purpose of this entry. The
interested reader may refer to [96] for a more detailed
description of the applied camera motion estimation or
to the literature on visual odometry for alternative
approaches.
Experimental Results
The proposed gaze stabilization was evaluated during
several test runs on a bumpy dirt road as described in
[96]. Figure 11 gives some impressions of the test track.
The red rectangles in Fig. 11a–c mark the region of
interest of the feature correspondence algorithm.
Active Multifocal Vision System, Adaptive Control of 67
Feature correspondences are only computed within
this region. According to the ground plane assumption
the ROI corresponds to the drivable area in front of the
vehicle. The green lines in Fig. 11a–c mark correspon-
dences that were accepted as measurements for the
camera motion estimation. The red lines depict corre-
spondences that were considered to be false positives.
The proposed vision-based online-calibration of
the MEMS angular-rate sensor was tested during
three consecutive runs along the test track with acti-
vated sensor calibration and gaze stabilization.
Figure 12 depicts the estimated gain ^k f ¼ ^k 1^k of the
a s
filter Fð^pÞ. For evaluation purposes the filter gain was
artificially reduced from 1 to 0.5. During parameter
estimation the gain rose from ^k f ¼ 0:5 to ^k f 0:88,
which is beneath the nominal gain 1.
The estimated time constant ^ta remained close to
its initial value zero. This indicates that the bandwidth
of the actuator control loop is sufficiently large and no
phase-delay occurs during stabilization. The estimated
residual pitch velocity Y _ r was biased and had to be
high-pass filtered to avoid parameter drift, therefore
the offset ds was estimated to be zero too.
To further evaluate the proposed gaze stabilization,
the estimated residual camera motion Y _ r was com-
pared for the following three runs on the test track:
One with deactivated gaze stabilization, another one
with activated stabilization but uncalibrated sensor
( ^k f ¼ 1), and finally the one with activated gaze stabi-
lization and calibrated sensor ( ^k f 0:88). During this
last run the calibration was active, but the parameters
had already converged.
The stabilization results are plotted in Fig. 13. It
gives an overview of the resulting estimated camera
pitch velocities Y _ r during the test runs. The labels
“pothole,” “dip,” and “washboard” mark peaks in the
pitch velocities Y _ r that originate from passing the
track sections depicted in Fig. 11a–c.
The gaze stabilization based on the uncalibrated
MEMS sensor already showed good results. Therefore
for better evaluation the rms-value of Y _ r was calcu-
lated as a measure for the stabilization error (Fig. 14).
The resulting rms-value was normalized with respect to
the rms-value of Y _ r in case of the unstabilized camera.
The normalized rms-error for the unstabilized camera
is then 100%. It turns out that the rms-error
corresponding to the calibrated gaze stabilization is
68 Active Multifocal Vision System, Adaptive Control of

Active Multifocal Vision System, Adaptive Control of. Figure 11

Selected images from the test track [96]

0.9
Estimated gain

0.8

0.7

0.6

0.5
Calibration off Calibration on Calibration on Calibration on
(1st lap) (2nd lap) (3rd lap)
0.4
0 500 1000 1500 2000
Distance [m]

Active Multifocal Vision System, Adaptive Control of. Figure 12

Estimated gain of the filter Fð^pÞ [96]
 Active Multifocal Vision System, Adaptive Control of 69

6 6

4 4

2 2

Θ̇r [°/s]

ẋ [m/s]
0 0

−2

−4
Pothole Dip Washboard
−6
0 50 100 150 200 250 300 350
Distance [m]

Active Multifocal Vision System, Adaptive Control of. Figure 13

Gaze stabilization results [96]: The blue lines correspond to the deactivated stabilization, the green lines to the activated
_r,
stabilization, and the dark red lines to the calibrated stabilization. Solid lines correspond to the residual camera motion Y
dotted lines to the vehicle velocity x_

100 resulting from the uncalibrated stabilization. This can

100
be considered to be the average reduction of the stabi-
lization error, resulting from an increasing or
80 a decreasing filter gain ^k f . The reduction of the rms-
error corresponds to the magnitude of change of the
filter gain.
60 The test runs were carried out at similar but
not equal velocities (Fig. 13). However, the velocity
40
39.1 during the test run with calibrated stabilization (dark
34.5
red) was almost always higher than the velocity
corresponding to the test run with active but
20 uncalibrated stabilization (green). Although higher
vehicle velocities are assumed to induce higher camera
pitch velocities, the calibrated stabilization still beats
0
the uncalibrated stabilization.
Active Multifocal Vision System, Adaptive Control of.
Figure 14 Future Directions: The Importance of Visual
The normalized rms-error of the gaze stabilization (based Perception
on Y_ r ) [96]. The blue bar corresponds to the deactivated
Perception of the environment for intelligent vehicles
stabilization, the green bar to the activated stabilization, should cover large parts of their surrounding with
and the dark red bar to the calibrated stabilization sufficient accuracy. One possible realization for the
visual sensory system is a multifocal vision system
simultaneously offering a wide field of view and high
resolution. Obviously such a system has to be active,
smaller than the rms-error corresponding to the since the high visual acuity is limited to the small
uncalibrated gaze stabilization. aperture angles of the tele-lenses. Saccades, smooth-
The rms-error for the calibrated stabilization is pursuit movements, and the vestibulo-ocular reflex
approximately 12% smaller than the rms-error are therefore essential capabilities to be able to fully
70 Active Multifocal Vision System, Adaptive Control of
exploit the possibilities offered by an active multifocal
vision system.
However multifocal vision systems are only benefi-
cial if there is a competition for sensing resources, be it
either to observe characteristic parts of an object in
more detail or to observe regions of interest that are
not in the current sensor view. Then, similar to human
selective attention, a selection stage is necessary that
decides which gaze direction is most appropriate to the
task at hand.
The outcome of most current computational
models of visual attention are saliency maps of interest-
ing scene regions. These maps can be used to direct the
telecamera of a multifocal vision system toward this
salient region to obtain high-resolution images thereof.
If top-down cues are incorporated in the map calcula-
tion they can be used, e.g., for object detection of pre-
viously known objects. However, if at all, up to now
saliency maps are only rarely used for object tracking.
Multifocal object tracking requires object recogni-
tion based on the identification and tracking of a set of
characteristic object features, e.g., the lights, the tires,
or the license plate of a car. After identifying these
features in the image frame of a wide-angle camera
they could be successively observed with a high resolu-
tion tele-lense. This kind of object tracking is subject to
current research and therefore it is developed for
a single camera first.
Apart from the computation of saliency maps, there
exist some approaches to gaze control that are used to
track multiple objects, landmarks, or road networks.
These approaches either try to predict the most infor-
mative viewing direction(s) based on information the-
oretic concepts or to plan a sequence of gaze directions
ensuring an optimal viewing behavior considering the
needs of several distributed visual perception modules
with respect to the visibility of interesting scene regions.
However these existing gaze control approaches were
not employed in dynamic environments like inner-city
traffic scenes.
Providing visual perception for intelligent vehicles,
capable of reliable and robust multiple object tracking
together with the detection and tracking of intersec-
tions, roads, traffic signs, and so on, is a major chal-
lenge in future research. Multifocal object tracking and
computational models of visual attention may
contribute to this by no means trivial task. Such
advanced perceptual capabilities give rise to
a competition for sensing resources. Then the intelli-
gent selective gaze control – based on the current
context – choosing the appropriate gaze directions
will be inevitable.
Bibliography
1. Palmer SE (1999) Vision science. MIT Press, Cambridge, MA
2. Kolb B, Whishaw IQ (2009) Fundamentals of human neuropsy-
chology, 6th edn. Worth, New York
3. Kandel ER, Schwartz JH, Jessell TH (1995) Essentials of
neural science behavior, 1st edn. Prentice Hall International,
London
4. Henderson JM (2003) Human gaze control during real-
world scene perception. Trends Cogn Sci 7(11):498–504,
November
5. Treue S (2003) Visual attention: the where, what, how and why
of saliency. Curr Opin Neurobiol 13:428–432
6. Chun MM, Marois R (2002) The dark side of visual attention.
Curr Opin Neurobiol 12:184–189
7. Krauzlis RJ (2005) The control of voluntary eye movements:
new perspectives. Neuroscientist 11(2):124–137
8. Knudsen EI (2007) Fundamental components of attention.
Annu Rev Neurosci 30:57–78
9. Summerfield C, Egner T (2009) Expectation (and attention) in
visual cognition. Trends Cogn Sci 13(9):403–409
10. Land MF, Lee DN (1994) Where we look when we steer. Nature
369:742–744
11. Chapman PR, Underwood G (1998) Visual search of driving
situations: danger and experience. Perception 27:951–964
12. Olson PL, Battle DS, Aoki T (1989) Driver eye fixations under
different operating conditions, University of Michigan, Ann
Arbor, Transportation Research Institute, Technical Report
13. Wann JP, Wilkie RM (2004) How do we control high speed
steering. In: Optic flow and beyond. Kluwer, Norwell,
pp 401–419, ch 18
14. Falkmer T, Gregersen NP (2005) A comparison of eye move-
ment behavior of inexperienced and experienced drivers in
real traffic environments. Optom Vis Sci 82(8):732–739
15. Metz B (2009) Worauf achtet der Faherer? steuerung der
aufmerksamkeit beim fahren mit visuellen nebenaufgaben.
Ph.D. dissertation, Julius-Maximilians-Universität Würzburg
16. Doshi A, Trivedi MM (2009) On the roles of eye gaze and head
dynamics in predicting driver’s intent to change lanes. IEEE
Trans Intell Transp Syst 10(3):453–462
17. Kühnlenz K (2007) “Aspects of multi-focal vision,” ser. Mess-,
Steuerungs- und Regelungstechnik.VDI Verlag 8(1129)
18. Kühnlenz K, Bachmayr M, Buss M (2006) A multi-focal high-
performance vision system. In: Proceedings of the 2006 IEEE
international conference on robotics and automation,
Orlando, pp 150–155
19. Ude A, Gaskett C, Cheng G (2006) Foveated vision systems
with two cameras per eye. In: Proceedings of the 2006 IEEE

international conference on robotics and automation (ICRA
2006), Orlando
20. Roca X, Vitrià J, Vanrell M, Villanueva JJ (1999) Gaze control in
a binocular robot systems. In: Proceedings of the 1999 IEEE
international conference on emerging technologies and
factory automation (ETFA 1999), Spain
21. Peng J, Srikaew A, Wilkes M, Kawamura K, Peters A (2000) An
active vision system for mobile robots. In: Proceedings of the
2000 IEEE international conference on systems, man and
cybernetics (ICSMC 2000), Nashville
22. Yamato J (1999) A layered control system for stereo vision
head with vergence. In: Proceedings of the 1999 IEEE interna-
tional conference on systems, man and cybernetics (ICSMC
1999), Tokyo
23. Panerai F, Metta G, Sandini G (2000) Adaptive image stabiliza-
tion: a need for vision-based active robotic agents. In: Pro-
ceedings of the international conference on the simulation of
adaptive behavior (SAB 2000), Paris
24. Lobo J, Ferreira JF, Dias J (2008) Robotic implementation of
biological Bayesian models towards visuo-inertial image sta-
bilization and gaze control. In: Proceedings of the 2008 IEEE
international conference on robotics and biomimetics,
Bangkok
25. Suppa M, Kielhöfer S, Langwald J, Hacker F, Strobl KH,
Hirzinger G (2007) The 3D-modeller: a multi-purpose vision
platform. In: Proceedings of the 2007 IEEE international
conference on robotics and automation (ICRA 2007), Roma
26. Vilgrattner T, Ulbrich H (2011) Design and control of a compact
high-dynamic camera-orientation system. IEEE-ASME T Mech
16(2):221–231
27. Lenz A, Balakrishnan T, Pipe AG, Melhuish C (2008) An adaptive
gaze stabilization controller inspired by the vestibulo-ocular
reflex. Bioinspir Biomim 3(3):035001
28. Zhang X, Tay ALP (2009) A physical system for binocular vision
through saccade generation and vergence control. Cybernet
Syst 40:549–568
29. Dickmanns ED (2007) Dynamic vision for perception and
control of motion. Springer, London
30. Dickmanns ED (2003) An advanced vision system for ground
vehicles. In: Proceedings of the 1st international workshop on
in-vehicle cognitive computer vision systems, Graz
31. Schiehlen J, Dickmanns ED (1994) A camera platform for intel-
ligent vehicles. In: Proceedings of the 1994 IEEE intelligent
vehicles symposium (IV 1994), Paris
32. Schiehlen J, Dickmanns ED (1994) Design and control of
a camera platform for machine vision. In: Proceedings of the
1994 IEEE international conference on intelligence in robotics
and autonomous systems (IROS 1994), Neubiberg
33. Schiehlen J (1995) Kameraplattformen für aktiv sehende
fahrzeuge ser. Mess-, Steuer- und Regelungstechnik. VDI
Verlag 8(514)
34. Günthner W, Wagner P, Ulbrich H (2006) An inertially stabilised
vehicle camera system – hardware, algorithms, test drives. In:
Proceedings of the 2006 IEEE international conference on
industrial electronics (IECON 2006), Paris
Active Multifocal Vision System, Adaptive Control of 71
35. Günthner W, Glasauer S, Wagner P, Ulbrich H (2006) Enhancing
inertial measurement units by multi-sensor fusion with
visual data for active camera stabilization. In: Proceedings of
joint conference of the 37th international symposium on
robotics ISR 2006 and the German conference on robotics,
Munich
36. Guenthner W, Glasauer S, Wagner P, Ulbrich H (2007) Biomi-
metic control for adaptive camera stabilization in driver-
assistance systems. J Mech Sci Technol 21(6):930–934
37. Günthner W (2008) Enhancing cognitive assistance systems
with inertial measurement units. Springer, Berlin
38. Ducrocq Y, Bahrami S, Duvieubourg L, Cabestaing F (2008) An
effective active vision system for gaze control. In: Advances in
visual computing, ser. Lecture notes in computer science,
vol 5359. Springer, Berlin, pp 267–277
39. Koch C, Ullman S (1985) Shifts in selective visual attention:
towards the underlying neural circuitry. Hum Neurobiol
4:219–227
40. Itti L, Koch C, Niebur E (1998) A model of saliency-based visual
attention for rapid scene analysis. IEEE Trans Pattern Anal
Mach Intell 20(11):1254–1259
41. Driscoll JA, Peters RA II, Cave KR (1998) A visual attention
network for a humanoid robot. In: Proceedings of the 1998
IEEE/RSJ international conference on intelligent robotic
systems (IROS 1998), Victoria
42. Backer G, Mertsching B, Bollmann M (2001) Data- and model-
driven gaze control for an active-vision system. IEEE Trans
Pattern Anal Mach Intell 23(12):1415–1429
43. Breazeal C, Scassellati B (1999) A context-dependent attention
system for a social robot. In: Proceedings of the sixteenth
international joint conference on artificial intelligence (IJCAI
1999), Stockholm
44. Vijayakumar S, Conradt J, Shibata T, Schaal S (2001) Overt
visual attention for a humanoid robot. In: Proceedings of the
2001 IEEE international conference on intelligence in robotics
and autonomous systems (IROS 2001), Hawaii
45. Sprague N, Ballard D (2003) Eye movements for reward
maximization. In: Advances in neural information processing
systems 15. MIT Press, Cambridge, MA
46. Choi SB, Jung BS, Ban SW, Lee M (2004) Trainable attention
model based vergence control for active stereo vision system.
In: Proceedings of the 2004 IEEE international conference on
intelligent sensors, sensor networks and information
processing (ISSNIP 2004), Melbourne
47. Navalpakkam V, Itti L (2005) Modeling the influence of task on
attention. Vis Res 45:205–231
48. Orabona F, Metta G, Sandini G (2005) Object-based visual
attention: a model for a behaving robot. In: Proceedings of
the 2005 IEEE computer society conference on computer
vision and pattern recognition (CVPR 2005), San Diego
49. Ude A, Wyart V, Lin LH, Cheng G (2005) Distributed visual atten-
tion on a humanoid robot. In: Proceedings of the 2005 IEEE-RAS
international conference on humanoid robots, Tsukuba
50. Frintrop S (2006) VOCUS: a visual attention system for object
detection and goal-directed search. Lecture notes in
72 Active Multifocal Vision System, Adaptive Control of
computer sciences ed., ser. (Lecture notes in artificial intelli-
gence), vol 3899. Springer, Heidelberg
51. Begum M, Mann GKI, Gosine RG (2006) A biologically inspired
Bayesian model of visual attention for humanoid robots. In:
Proceedings of the 2006 IEEE-RAS international conference on
humanoid robots (HUMANOIDS 2006), Genova
52. Yu Y, Mann GKI, Gosine RG (2008) An object-based visual
attention model for robots. In: Proceedings of the 2008 IEEE
international conference on robotics and automation (ICRA
2008), Pasadena
53. Vogel J, de Freitas N (2007) Target-directed attention: sequen-
tial decision-making for gaze planning. In: Proceedings of the
2007 IEEE international conference on robotics and automa-
tion (ICRA 2007), Pasadena
54. Borji A, Ahmadabadi MN, Araabi BN (2009) Learning
sequential visual attention control through dynamic state
space discretization. In: Proceedings of the 2009 IEEE interna-
tional conference on robotics and automation (ICRA 2009),
Kobe
55. Xu T, Kühnlenz K, Buss M (2008) Information-based gaze con-
trol adaptation to scene context for mobile robots. In: Pro-
ceedings of the 2008 IEEE international conference on pattern
recognition (ICPR 2008), Tampa
56. Xu T, Wu H, Zhang T, Kühnlenz K, Buss M (2009)
Environment adapted active multi-focal vision system for
object detection. In: Proceedings of the 2009 IEEE international
conference on robotics and automation (ICRA 2009), Kobe
57. Itti L, Baldi P (2009) Bayesian surprise attracts human atten-
tion. Vis Res 49(10):1295–1306, June
58. Xu T, Zhang T, Kühnlenz K, Buss M (2010) Attentional object
detection with an active multi-focal vision system. Int J Hum
Robot 7(2):223–243
59. Michalke T, Kastner R, Adamy J, Bone S, Waibel F, Gayko MKA,
Gepperth A, Fritsch J, Goerick C (2008) An attention-based
system approach for scene analysis in driver assistance.
Automatisierungstechnik 56(11):575–584
60. Aloimonos JY, Weiss I, Bandyopadhyay A (1988) Active vision.
Int J Comput Vision 7:333–356
61. Bajcsy R (1988) Active perception. Proc IEEE 76(8):966–1005,
Invited Paper
62. Aloimonos JY (1990) Purposive and qualitative active vision.
In: Proceedings of the 1990 IEEE international conference on
pattern recognition, Atlantic
63. Ballard DH (1991) Animate vision. Artif Intell 48:57–86
64. Aloimonos Y (ed) (1993) Active perception. Lawrence Erlbaum,
Hillsdale
65. Dickmanns ED (1988) An integrated approach to feature
based dynamic vision. In: Proceedings of the 1988 IEEE com-
puter society conference on computer vision and pattern
recognition (CVPR 1988), Ann Arbor
66. Davison AJ (1998) Mobile robot navigation using active vision.
PhD Thesis, University of Oxford
67. Davision AJ, Murray DW (1998) Mobile robot localisation using
active vision. In: Proceedings of the 5th European conference
on computer vision (ECCV 1998), Freiburg
68. Seara JF, Strobl KH, Schmidt G (2002) Information manage-
ment for gaze control in vision guided biped walking.
In: Proceedings of the 2002 IEEE international conference on
intelligent robots and systems (IROS 2002), Lausanne
69. Seara JF, Strobl H, Schmidt G (2003) Path-dependent gaze
control for obstacle avoidance in vision guided humanoid
walking. In: Proceedings of the 2003 IEEE international confer-
ence on robotics and automation (ICRA 2003), Taipei
70. Seara JF, Strobl KH, Martı́n E, Schmidt G (2003) Task-oriented and
situation-dependent gaze control for vision guided humanoid
walking. In: Proceedings of the 2003 IEEE-RAS international
conference on humanoid robots (HUMANOIDS 2003), Munich
71. Seara JF (2004) Intelligent gaze control for vision-guided
humanoid walking, ser. Mess-, Steuer- und Regelung-
stechnik.VDI Verlag 8(1044)
72. Seara JF, Schmidt G (2005) Gaze control strategy for
vision-guided humanoid walking. Autmatisierungstechnik
53(2):49–58
73. Lidoris G, Kühnlenz K, Wollherr D, Buss M (2007) Combined
trajectory planning and gaze direction control for robotic
exploration. In: Proceedings of the 2007 IEEE international
conference on robotics and automation (ICRA 2007), Rome
74. Kühnlenz K, Lidoris G, Wollherr D, Buss M (2007) On foveated
gaze control and combined gaze and locomotion planning.
In: de Pina ACF (ed) Humanoid robots: new developments.
I-Tech Education and Publishing, Vienna
75. Kühnlenz K, Buss M (2008) Multi-focal vision and gaze control
improve navigation performance. J Hum 1(1):33–44
76. Xu T, Kühnlenz K, Buss M (2008) A view direction planning
strategy for a multi-camera vision system. In: Proceedings of
the 2008 IEEE international conference on information and
automation (ICIA 2008), Zhangjiajie
77. Davison AJ, Kita N (2001) 3D simultaneous localisation and
map-building using active vision for a robot moving on undu-
lating terrain. In: Proceedings of the 2001 IEEE computer
society conference on computer vision and pattern recogni-
tion (CVPR 2001), Lihue
78. Davison AJ, Murray DW (2002) Simultaneous localization and
map-building using active vision. IEEE Trans Pattern Anal
Mach Intell 24:865–880
79. Lidoris G, Kühnlenz K, Wollherr D, Buss M (2006) Information-
based gaze direction planning algorithm for slam.
In: Proceedings of the 2006 IEEE-RAS international conference
on humanoid robots (HUMANOIDS 2006), Genova
80. Pellkofer M, Dickmanns ED (2000) EMS-vision: gaze control in
autonomous vehicles. In: Proceedings of the 2000 IEEE intelli-
gent vehicles symposium (IV 2000), Dearborn
81. Pellkofer M (2003) Verhaltensentscheidung für autonome
fahrzeuge mit blickrichtungssteuerung. Dissertation,
Universität der Bundeswehr München
82. Dickmanns ED (2004) Three-stage visual perception for verte-
brate-type dynamic machine vision. In: Proceedings of the con-
ference on engineering of intelligent systems (EIS 2004), Madeira
83. Unterholzner A, Wuensche HJ (2009) Adaptive state space
control of a camera platform for an autonomous ground

vehicle. In: Proceedings of the 2009 IEEE intelligent vehicles
symposium (IV 2009), Xi’an
84. Unterholzner A, Wuensche HJ (2010) Hybrid adaptive control
of a multi-focal vision system. In: Proceedings of the 2010 IEEE
intelligent vehicles symposium (IV 2010), San Diego
85. Canudas C, Åström KJ, Braun K (1987) Adaptive friction
compensation in DC-motor drives. IEEE J Robotic Autom
RA-3(6):681–685
86. Goodwin GC, Sin KS (1984) Adaptive filtering prediction
and control, Information and system sciences. Prentice-Hall,
Englewood Cliffs
87. Ramasubramanian A, Ray LR (2007) Comparison of EKBF-
based and classical friction compensation. J Dyn Syst Meas
Contr 129:236–242
88. Franklin GF, Powell JD, Emami-Naeini A (2002) Feedback control
of dynamic systems, 4th edn. Prentice Hall, Upper Saddle River
89. Athans M, Falb PL (1966) Optimal control: an introduction to
the theory and its applications. McGraw-Hill, New York
90. Utkin VI (1977) Variable structure systems with sliding modes.
IEEE Trans Autom Control AC-22(2):212–222
91. Hung JY, Gao W, Hung JC (1993) Variable structure control:
a survey. IEEE Trans Ind Electron 40(1):2–22, February
92. Chatchanayuenyong T, Parnichkun M (2008) Time optimal
hybrid sliding mode-PI control for an autonomous underwater
robot. Int J Adv Rob Syst 5(1):91–98
93. You KH, Lee EB (2000) Robust, near time-optimal control of
nonlinear second order systems with model uncertainty. In:
Proceedings of the 2000 IEEE international conference on
control applications, Anchorage, Alaska
Active Multifocal Vision System, Adaptive Control of 73
94. Iliev B, Kalaykov I (2004) Minimum-time sliding mode control
for second-order systems. In: Proceedings of the 2004
American control conference, Boston
95. Dean P, Porrill J, Stone JV (2002) Decorrelation control by the
cerebellum achieves oculomotor plant compensation in sim-
ulated vestibulo-ocular reflex. Proc R Soc Lond Biol Sci
269(1503):1895–1904, August
96. Unterholzner A, Rohland M, Schweitzer M, Wuensche HJ
(2010) Vision based online-calibration of inertial gaze stabili-
zation. In: Proceedings of the 2010 IEEE intelligent vehicles
symposium (IV 2010), San Diego
97. Julier SJ, Uhlmann JK (2004) Unscented filtering
and nonlinear estimation. Proc IEEE 92(3):410–422, Invited
Paper
98. Schweitzer M, Wuensche HJ (2010) Real-time visual
odometry for ground moving robots using GPUs. In:
Proceedings of the international conference on
computer vision theory and applications (VISAPP 2010),
Angers
99. Scaramuzza D, Fraundorfer F, Pollefeys M, Siegwart R (2009)
Absolute scale in structure from motion from a single vehicle
mounted camera by exploiting nonholonomic constraints.
In: Proceedings of the IEEE international conference on com-
puter vision (ICCV 2009), Kyoto
100. Schweitzer M, Wuensche HJ (2009) Efficient keypoint
matching for robot vision using GPUs. In: Proceedings of
the fifth IEEE workshop on embedded computer vision at
ICCV 2009, Kyoto
74 Active Pedestrian Protection System, Scenario-Driven Search Method for

Active Pedestrian Protection System, dangerous situations in specific urban scenarios. The
approach used in this work differs from the ones
Scenario-Driven Search Method for implemented in traditional pedestrian detection sys-
ALBERTO BROGGI1, PIETRO CERRI1, STEFANO GHIDONI2, tems, which are designed to localize all pedestrians in
PAOLO GRISLERI1, HO GI JUNG3 the area in front of the vehicle. Conversely this
1
VisLab – Dipartimento di Ingegneria approach searches for pedestrians in critical areas only.
dell’Informazione, Università degli Studi di Parma, The great advantages of such an approach are that
Parma, Italy pedestrian recognition is performed on limited image
2
Intelligent Autonomous System Laboratory areas – therefore boosting its time-wise performance –
(IAS-Lab), Department of Information Engineering, and no assessment on the danger level is finally
University of Padua, Padova, Italy required before providing the result to either the driver
3
The School of Mechanical Engineering, Hanyang or an onboard computer for automatic maneuvers.
University, Seondong-gu, Seoul, South Korea A further advantage is the drastic reduction of false
alarms, making this system robust enough to control
Article Outline nonreversible safety systems.

Glossary
Introduction
Definition of the Subject and Its Importance
Introduction This paper presents an innovative approach to the
Why Are ADAS Related to Sustainability? detection of pedestrians, explicitly designed to work
Innovation of This Method in a particularly challenging urban scenario, which
State of the Art represents a particularly critical environment for tradi-
Setup tional pedestrian detection approaches. The aim of the
Focus of Attention system is to trigger reversible and nonreversible driving
Vision Fusion assistance system.
Automatic Braking This work was developed by VisLab at Parma
Results University and funded by MANDO Corporation,
Future Directions Korea.
Bibliography
Why Are ADAS Related to Sustainability?
Glossary
In his book May Contain Nuts [1] John O’Farrell
ADAS Advanced Driver Assistance Systems are complex depicts a mother terrified by modern world’s dangers.
systems to help the driver during specific maneu- In order to show how dangerous the road where vehi-
vers, or during normal driving. cles are running can be, she builds a puppet, and throws
Laser scanner Laser scanner is an optical sensor based it onto the road, letting it come out between two
on LIDAR (Light Detection And Ranging) technol- parked cars.
ogy that measures the distance of targets measuring The description of that particular “experiment”
the time delay between transmission of a pulse and intended to highlight two points, among others: first
detection of the reflected signal. of all, the (motivated) worries people have about the
NIR Near InfraRed (0.75–1.4 mm) is the wavelength road environment, especially when dealing with
closest to visible that is often used in image processing
children, which cannot correctly evaluate risks; the
for night vision, together with specific illuminators.
second observation is that road safety relies, in some
cases, on a very strong hypothesis, that is, everyone is
Definition of the Subject and Its Importance following road traffic laws. Of course, this should nor-
This paper presents an application of a pedestrian mally be the case, but an important point is to under-
detection system aimed at localizing potentially stand whether there are margins, in order to recover for

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
 Active Pedestrian Protection System, Scenario-Driven Search Method for
other road users’ illegal behaviors. This applies mainly
to drivers, since they are the strong road users. When
a car hits a pedestrian, even if this is due to a pedestrian
fault, the driver could, in principle, take actions that
may avoid or, at least, mitigate the collision conse-
quences. This is, then, the difference. It is difficult, if
not impossible, for a pedestrian to take actions that can
recover from a driver’s fault; on the contrary, it is often
possible for a driver to cope with pedestrians’ faults.
Road dangers are caused by several factors. At the
very beginning, vehicles were slow, and few of them
were running on roads – traffic also was an unknown
problem! Then, the number of vehicles dramatically
increased, but not enough care was dedicated to the
protection of pedestrians and bicycles mobility, either
because of economic factors, or because the problem
itself is really difficult to face.
Whatever the reason, traffic and related pedestrian
injuries have now become an important issue that has
been on the agenda of several governments and public
institutions for many years. The reasons for this change
can be found, again, in economic factors – every injury
brings a high social cost – but, also, human society
underwent substantial changes during the last few
decades. After more than a century of economic and
technical growth, the capability of mankind to deeply
influence the environment had become so high, and the
available technology so powerful, that it became clear it
is mandatory to keep them under control. In other
words, it became clear that growth could not continue
without limits, but rather, it had to face a new issue: the
capacity to endure, that is, sustainability.
Sustainability is a complex and wide concept, which
includes the capability of reducing as much as possible
the negative impact on the environment that the use of
almost every new technology involves. Therefore sus-
tainability deeply deals also with all transportation
systems, because they pollute, they can be dangerous,
they impact on people’s everyday life, and, above all,
they are essential.
Several studies on countermeasures aimed at reduc-
ing traffic injuries have been undertaken. Car manu-
facturers deeply increased the safety levels available
inside cars. Back to the 1970s, objects like the steering
wheel or the pedals, or, more generally, every part of
a car’s interior, could cause severe injuries in case of
accidents, because they could hit the driver or
75
passengers in an uncontrolled way. Car manufacturers
then invested a considerable amount of money to
reduce fatalities, and studied the problem. Nowadays,
the steering wheel itself is not only able to avoid dan-
gers in case of accidents, but also contains a safety
system, the airbag, that sensibly reduces the conse-
quences of a car accident.
Unfortunately, increasing road safety in the case of
accidents involving pedestrians is sensibly more diffi-
cult. The EuroNCAP institution is involved in
a number of projects related to this field (see www.
euroncap.com for an overview). Some of them aim at
mitigating the consequences for a pedestrian being hit
by a vehicle, for instance by modifying the shape of the
car front, a sort of outer passive safety system. Similar
ideas are under development since decades. Some pat-
ents regarding systems capable of mitigating collisions
with a pedestrian date back to the beginning of the
1980s [2].
Following the idea of mitigating the impact energy,
external airbags were also proposed. Several patents
were submitted by the end of the 1990s [3, 4]. Such
airbags should inflate in case the car hits a pedestrian,
but still the problem of the collision sensor is open and
hard to solve. Such sensors should detect collisions that
are relatively soft, since the mass of a pedestrian is
sensibly lower than that of a vehicle. Moreover, the
sensor is required to be extremely fast and precise,
because each false positive leads to the need of expen-
sive service for restoring the external airbag system.
Together with passive safety systems, active systems
were studied and developed in the last years. Since
active safety systems should predict an imminent col-
lision, they need sophisticated sensors and collision
detection systems. The task of preventing accidents
from happening is extremely difficult, because it
requires that the safety system understands the road
scene at a high level, detecting potentially dangerous
conditions. Such an understanding can be provided
only by a computer system connected to sensors. The
need for safety systems has therefore caused computer
science to deeply enter into the automotive world,
a process that had started with the very first electronic
active safety systems, like ABS, few decades before.
Car manufacturers, together with public institu-
tions, are still investing a huge amount of money in
the study and development of such active systems.
76 Active Pedestrian Protection System, Scenario-Driven Search Method for
Several research groups are addressing the high number
of problems that are related to road safety. Yet a system
capable of a complete understanding of a generic road
scene is not available, but technology is evolving rap-
idly, and now several systems exist that can analyze cars’
or pedestrians’ behavior over time, thus getting a high-
level understanding of the road scene, even if limited to
some aspects.
In order to achieve brilliant results, highly effective
sensors should be used; possible selections include
radars, laser scanners, cameras, or a combination of
them. The number of solutions proposed so far is really
huge, and substantial differences can be found in
a number of design choices. This is clear even by just
having a look at the various prototype vehicles that
have been developed. Some of them are full of sensors,
mounted in a fashion that is not realistic for an off-the-
shelf car. Such vehicles are useful for experimenting
with new solutions, understanding their possible
advantages over more well-known system designs,
without the effort of caring too much about the feasi-
bility of the solution itself. On the other hand, vehicles
designed with a high level of integration are vital to
design systems that can be really installed onto a vehicle
in the future. It often happens, in fact, that restrictions
even on the hardware placement affect software algo-
rithms dedicated to road scene understanding.
An example of the coexistence of these two kinds of
design philosophy was clearly visible during the
DARPA Urban Challenge [5], a competition for
unmanned vehicles. Each vehicle had to complete
a series of tasks, each one consisting in reaching
a destination point driving inside an urban environ-
ment, dealing with traffic and intersections, and
also the other unmanned vehicles. Even if, in principle,
no limitations were imposed on system design and
number and type of sensors, vehicles that attended
the race can be divided into the two categories
described above.
Despite the high number of developed systems, as
noted above, they are all exploiting, with few excep-
tions, the same types of sensor: radars, laser scanners,
and cameras. Cameras can be further divided into three
categories, depending on which part of the electromag-
netic spectrum they are sensing: visible light, near
infrared, or far infrared. They can be considered three
different kinds of sensors since the data they provide
have completely different characteristics, and usually
undergo different processing.
Thanks to the high number of research groups
working on road safety systems, a huge amount of
algorithms has been developed for analyzing all kinds
of data provided by the sensors described above. Also,
due to the fact that those sensors are characterized by
different and somehow complementary features, they
are often used together. This led to the need for sensor
fusion techniques, that is, techniques that, by
exploiting measurements provided by more than one
sensor, derive a single measurement. Such final result is
more reliable than that provided by each sensor, as
sensor fusion techniques know the feature of each
sensor, and their reliability intervals.
As it has been described so far, computer vision
applied to the automotive field is a really hot topic,
and systems are being developed that successfully
achieve a high and rather precise level of road scene
understanding. However, beside system development,
possible applications should be taken into consider-
ation. The introduction of active safety systems based
on such complex equipments into off-the-shelf vehicles
is advancing at a rate that is sensibly slower compared
with the advances in technology gained by research
groups operating worldwide. Beside people acceptance
of such systems, and legal issues, the requirements of
the automotive world are extremely difficult to be met.
If a system is capable of interfering with commands
issued by a human driver, its acceptable fault rate will
be extremely low, if not at no fault level. It is straight-
forward, on the other hand, that such a reliability level
is really hard to be achieved, when dealing with such
complex environment. The scenario is definitely
challenging.
To understand how this kind of technology could
develop, consider that still for a long time systems will
not be able to reliably tackle the whole problem.
Instead, some systems are already on the market, as
equipment of luxury cars, which reliably perform sim-
ple and noninvasive tasks. As an example, some already
available systems are meant to detect possible conflicts
with overtaking vehicles when a lane change maneuver
is undertaken by the driver. In this case the system
issues a warning, like an acoustic signal. Some sporadic
false positives do not have a strong impact; however,
the driver will tend to ignore the system hints if failure
 Active Pedestrian Protection System, Scenario-Driven Search Method for
rate is not extremely low. This kind of systems, that is,
whose failure is not dangerous, but rather may be
annoying, seems to be the next step on the roadmap
of the introduction of active vision systems into the
automotive world.
Given the previous considerations about the need
to reduce accidents and the overview on future active
safety systems, it is now worth analyzing into more
detail which are the road scenarios causing a high
number of fatalities or injuries. Several papers and
reports exist on this topic; moreover, specific studies
addressed the problem of accidents involving children
[6]. This last problem is particularly alarming, since it
involves people of very young age, whose perception of
danger and understanding of road traffic rules is rather
low due to their very young age. The danger comes
from the fact that young children turn out to be often
living in environments that are designed for adults.
However, in the last decades modern societies started
to invest in the safety of new generations, which is
directly related to the sustainability of the societies
themselves.
Describing the causes of pedestrian fatalities is, of
course, an extremely complex task, and it also leads to
subjective conclusions. For instance, the effect of car
speed is debated. It seems to be a key factor in some
studies, like [7]. On the other hand, in other studies
related to children [6] it is stated that “higher vehicular
speeds [. . .] were not associated with child pedestrian
injury indicating that residential streets that had vehi-
cles travelling at higher speeds were not at increased
risk in the absence of other environmental risk factors.”
Both studies point out that a main cause of accidents is
the sudden appearance of the pedestrian in the road
scene, as it often happens when a pedestrian comes
onto the street between two parked cars. This scenario
is quite common, especially in residential areas, where,
despite the low speed limits, a high number of pedes-
trian injuries happen. In most cases, the driver did not
see the pedestrian before the accident occurred.
Investigating on main accident causes leads to a set
of countermeasures that can be undertaken. In [7],
three main actions are reported: “speed control, sepa-
ration of pedestrians from vehicles, and measures that
increase the visibility and conspicuity of pedestrians.”
These involve road infrastructure, including some
important details like the way cars are parked. The
77
implementation of diagonal parking improves the
visibility of pedestrians going toward the street, because
it makes them walk in a diagonal direction, which is
more easily seen from a driver’s perspective. This
last example points out that protecting pedestrians
does not necessarily mean the need for huge and
expensive actions, but rather, that in some cases
simple and well-thought expedients can be extremely
effective.
Besides the modification of road infrastructure,
improvement of pedestrian visibility can also be
achieved by the active safety systems that were men-
tioned above. They seem to be the ultimate solution to
the problem, because they are the only countermeasure
to one of the major issues of the human drivers, that is,
the tendency to get distracted, especially after driving
for a long time. The activity of driving would require,
in principle, a high and constant level of attention
that a human driver can hardly provide. In particular,
a driver usually cares about issues he is expecting,
like a traffic light, but handles with much more diffi-
culty a totally unexpected event, like a child darting
into the street. For this kind of events, only the quick
response of an active safety system can represent the
final solution.
Innovation of This Method
Instead of searching for pedestrians in a large area in
front of the vehicle making no assumption on the
external environment, the system presented in this
work focuses on a specific urban scenario in which
not only the detection of a pedestrian is of basic impor-
tance, but the danger of the situation can be clearly
assessed as well.
In fact, in an advanced driving assistance system,
a correct detection must be followed by the estimation
of its position with respect to the vehicle and the
environment.
As an example, a pedestrian detection system
able to correctly localize all pedestrians present in the
scene provides a huge amount of information that still
needs to be filtered in order to be useful to either the
driver or the onboard computer in charge of automatic
maneuvers.
A possible filter may be implemented by fusing
information coming from other systems, such as lane
78 Active Pedestrian Protection System, Scenario-Driven Search Method for

detection or other situation analysis engines, and CAN
data. A pedestrian exactly in front of the vehicle may or
may not be considered dangerous depending on the
surrounding environment, as shown in Fig. 1.
This work is based on a totally different approach.
Instead of detecting all possible candidates and assess
their danger level, the system first analyzes the scenario,
and then searches for possible pedestrians in specific
positions for that particular scenario. This approach is
called scenario-driven search or SDS. In this way, all
detected pedestrians represent possible threats and no
further filtering is needed (apart from a validation and
a possible final tracking step).
The scenarios that are considered here refer to the
most common urban situations in which the presence of
a pedestrian causes a serious problem that could be
mitigated by an early detection. In particular, when vehi-
cles are moving on an urban road, the most common
threat that a pedestrian may pose – therefore requiring
a successful detection – is road crossing. Stopped vehicles
on the road or on the road edges may occlude visibility,
thus making the detection of the pedestrian more
complex and requiring a real-time reaction.
The application of the SDS approach to this specific
scenario is based on the localization of stopped vehicles
followed by the search for pedestrians in their close
proximity or in the areas partly hidden by them.
Stopped vehicles, whose edges trigger the search for
pedestrians, may be parked cars on the road edge,
vehicles temporarily stopped on the road, or vehicles
queued in a line in front of a traffic light, zebra crossing,
or simply jammed cars.
The first row of Fig. 2 shows some examples of
situations in which the visibility of a crossing pedes-
trian is partly or completely occluded by stopped vehi-
cles. The second row of Fig. 2 highlights, for each
situation, the areas on which the system will perform
a check for the presence of a possible pedestrian.
In other words, the approach is to focus on the
detection of pedestrians appearing from behind

a b

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 1

When a pedestrian is localized but no reference with respect to the environment is provided like in (a), the detector is not
able to assess the danger level. When environmental information are available, the very same pedestrian may become
a threat (b), or may be in a safe position (c)
 Active Pedestrian Protection System, Scenario-Driven Search Method for 79

a b c d

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 2

Situations considered in this work (first row) and areas of interest considered for the detection of a possible pedestrian
(second row). (a) A crossing pedestrian is hidden by a parked vehicle. (b) A pedestrian is crossing the road behind a stopped
bus. (c) A pedestrian is appearing between two parked vehicles. (d) A pedestrian is crossing the road between two vehicles
stopped on the other side of the road

occluding obstacles; pedestrians that are clearly visible in
the frontal area of the vehicle need also to be detected.
This function is also available on other systems [8, 9].
The idea of focusing on a specific scene or scenario
(referring to a dynamic or a static environment, respec-
tively) is not new to pedestrian detection systems. In
2002, Franke and Heinrich [10] developed a module
able to detect balls (which are usually a strong signature
of the presence of a child).
Another example of very specific systems is the one
developed by Curio et al. [11] which was based on the
visual localization of the specific moving pattern of
human legs.
It is known [7] that parked vehicles, blocking the
visibility of pedestrians, are one of the main causes of
accidents. Agran et al. [6] show that the number of
parked vehicles along a street is the strongest risk factor
for pedestrian injuries occurring in residential areas.
Although in these areas parking spaces should be
arranged diagonally, there are situations, as shown in
Fig. 2, in which vehicles stop temporarily on the road
and their position cannot be predetermined and
carefully controlled as in the case of parking lots.
Although some of the situations of Fig. 2 refer to
specific urban areas (i.e., zebra crossings and bus
stops) that could be specifically enhanced by intelligent
infrastructures aimed at warning oncoming vehicles,
other situations can happen in any part of the road
network, making the installation of specific warning
infrastructures impractical.
The main characteristics required by a system like
this are:
● To detect pedestrians quickly, given the short sens-
ing range and the particularly high risk of a collision
with a pedestrian suddenly appearing behind an
obstacle
● To detect pedestrians as soon as they appear, there-
fore even when they are still partly occluded
● To limit the search to specific areas, determined by
a quick preprocessing
Figure 3 shows the coverage of different sensing
technologies (laser and vision) in a specific scenario
considered in this work.
80 Active Pedestrian Protection System, Scenario-Driven Search Method for
Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 3
A pedestrian partially hidden by parked cars may not be detected by a laser scanner positioned in the front bumper but
can be detected using vision, even if partially occluded
The fusion of a laser scanner and vision can provide
a quick and robust detection in case of suddenly
appearing pedestrians. The laser scanner provides
a list of areas in which a pedestrian may appear, while
the camera is able to detect the pedestrian even when
he/she is not yet visible to the laser scanner.
A remarkable improvement of road safety can be
obtained thanks to the use of both reversible and
nonreversible driving assistance systems. Three levels
of intervention are defined:
● Once the pedestrian is detected with a sufficiently
high confidence level, a warning is sent to the driver.
● Should the driver not react promptly to the warning,
the system would issue a second level of warning by
blowing the vehicle’s horn. This second warning is
still considered a reversible system, although it is
much more invasive than the former. The aim of
this loud warning is to attract the attention of both
the pedestrian itself and – once again – the driver.
● In case the danger level is not reduced thanks to
a prompt reaction of the driver (or the pedestrian),
the intelligent vehicle will jump into the third level
and trigger a nonreversible system. In this case it is
automatic braking.
This being a nonreversible and very invasive system,
its triggering must be preceded by an extremely careful
analysis of the danger level and requires the complete
processing to be thoroughly tested with respect to false
detections. In the case of nonreversible systems, the
number of false positives is even more important than
the number of correct detections.
In the literature [12] many different pedestrian
detection systems have been surveyed, and a large num-
ber of test methods for pedestrian detection systems
have been proposed, which are based on single frame
analysis, temporal analysis of continuous video
streams, or event based.
Their performance is still far from the ideal system;
this is the reason why the approach proposed in this
paper is regarded as an alternative to conventional
systems. Pedestrians are searched in situations of clear
danger by limiting the search to specific areas. Besides
being quicker than other systems, this method aims
also at reducing the number of false detections to zero.
The system presented in this work has been devel-
oped and tested by VisLab in Parma on a Hyundai
Grandeur prototype vehicle; Mando is using an exact
replica of this system in Korea (see Fig. 4) to double
testing time.
State of the Art
As long-time pursued advanced driver assistance
systems (ADAS) and active safety vehicles (ASV) such
as lane departure warning (LDW), blind spot
detection (BSD), adaptive cruise control (ACC),
 Active Pedestrian Protection System, Scenario-Driven Search Method for 81

a b

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 4

The Hyundai Grandeur test vehicle equipped in Parma, Italy (a), and the replica in Seoul, Korea (b)

precrash safety (PCS), and collision warning/avoidance
(CW/CA) become commercialized [13–15], interest of
research groups and transportation authorities moves
to newly emerging technologies or candidate projects
such as intersection accident prevention [16], intelli-
gent parking assist system [17], and active pedestrian
protection system (APPS). Particularly, APPS attracts
a great interest because it is directly related with
a vulnerable road user who holds a significant portion
of traffic accident fatalities. The success of ACC, PCS,
and CW/CA does not guarantee the success of APPS
because the pedestrian is relatively small and their
appearance is various and continuously changing.
However, they are expected to provide abundant ele-
mentary technologies and economic backgrounds. In
particular, progress of electronics and sensors, such as
high performance embedded processor, monolithic
microware-integrated circuit (MMIC)-based compact
radar, multilayer scanning laser radar, and high
dynamic range CMOS (HDRC) camera, are regarded
as the enabling technologies for APPS.
Recently, Gandhi and Trivedi provided a great sur-
vey of pedestrian protection systems [12]. They classi-
fied pedestrian protection approaches into four
categories: infrastructure design enhancement, passive
safety system involving vehicle design, active safety
system based on pedestrian detection, and collision
prediction. The pedestrian protection system designed
to be installed on board can be divided into passive and
active approaches according to whether the operation
time is before or after vehicle–pedestrian collision.
If a system is designed to prevent the collision or
mitigate its damage, it is categorized as an active
approach. Contrarily, if a system is designed to mini-
mize the pedestrian’s damage after the collision, it is
categorized as a passive approach. The collision predic-
tion could be differentiated from the active safety sys-
tem in the point that the sensing devices are installed
on infrastructure rather than on board.
Infrastructure enhancement to reduce pedestrian-
related accidents is supposed to be the most traditional
approach. The approach can be divided into three
categories of countermeasures: speed control, pedes-
trian vehicle separation, and measures to increase vis-
ibility and conspicuity of the pedestrian [12].
The first and second could be easily understood by
investigating traffic calming which is a kind of move-
ment or concept [18]. Definitions of traffic calming
vary, but they all share the goal of reducing vehicle
speeds, improving safety, and enhancing the quality of
life. Traffic-calming measures can be separated into two
groups based on the main impact intended. Volume
control measures are primarily used to address cut-
through traffic problems by blocking certain move-
ments, thereby diverting traffic to streets better able
to handle it. Volume control measures also include
divertive or restrictive measures such as full closures,
half closures, diagonal diverters, and median barriers.
Speed control measures are primarily used to address
speeding problems by changing vertical alignment,
changing horizontal alignment, or narrowing the road-
way. Vertical deflection includes speed humps, speed
82 Active Pedestrian Protection System, Scenario-Driven Search Method for
tables, raised crosswalks, raised intersections, and tex-
tured pavements. Horizontal deflection includes traffic
circles, roundabouts, chicanes, and realigned intersec-
tions. Horizontal narrowing includes neckdown,
center-island narrowing, and chokers. US DoT
(Department of Transportation) FHWA (Federal High-
way Administration) reports [19, 20] show that well-
designed and implemented traffic-calming techniques
can have a number of beneficial impacts for bicyclists
and pedestrians. The reduced vehicle speeds associated
with such projects can reduce both the severity and
incidence of motor vehicle/bicycle/pedestrian crashes
and can make bicyclists and pedestrians feel more
comfortable in traffic.
Pedestrian visibility can be increased by improving
roadway lighting since a majority of pedestrian fatali-
ties occur at nighttime. Since parked vehicles block the
vision of drivers as well as pedestrians, removing on-
street parking and implementing diagonal parking in
residential streets would help in reducing accidents
especially involving children [12].
According to the analysis of vehicle impact zones in
pedestrian–vehicle collisions, vehicle front is responsi-
ble for most pedestrian injuries: 71.1% according to US
national highway traffic safety administration
(NHTSA) data. Almost 80% of the fatalities occur
when the pedestrian is struck by vehicle front, whereas
most of the remaining fatalities occur when the pedes-
trian is hurled to the street [21]. Accident analysis also
proved the importance of head injury [22, 23].
Although upper and lower extremities, as well as
head, are the most frequently injured body regions in
passenger car-to-pedestrian accidents, 62% of the
pedestrian fatalities were caused by head injuries.
Pedestrian head suffers both a direct impact with stiff
vehicle structures and a large degree of rotational accel-
erations [24]. In particular, stiff structure of windshield
including A-pillar should not be overlooked. An anal-
ysis of all vehicle parts which caused head injuries
shows that impact on the windshield causes 44.3% of
head injuries. Impact on the windshield causes twice as
often head injuries as at the bonnet, and three times
often for severe head injuries. Head impact on the
A-pillar causes severe injuries in 82% of the cases.
Other vehicle parts mostly cause only slight injuries [24].
Therefore, passive safety measures are being devel-
oped by installing energy-absorbing devices in the
vehicle front structures, on which pedestrian head
hits, such as hood (bonnet), windshield lower part,
and A-pillar. A passive pedestrian protection system
includes an impact-reducing front structure, active
bumper, active hood (hood lifting, active bonnet),
and pedestrian airbag. Active bumper is designed to
provide the appropriate stiffness to cushion the impact
while at the same time providing support of all parts of
the limb to limit knee joint lateral bending [25]. Cush-
ioning method of active bumper includes foam energy
absorber, molded plastic energy absorbers, air-filled
energy absorbers, flexible plastic beam, bumper
deploying, crush cans, outside additional structure,
foam-encapsulated metal, and steel energy absorber
[25]. Active hood system detects vehicle–pedestrian
collision using bumper-installed pedestrian crash sen-
sor and acceleration sensor, and then two actuators lift
the rear part of the bonnet approximately 100 mm
within 60–70 ms after the leg-to-bumper impact [26].
Pedestrian protection airbag deploys an airbag through
a gap between windshield lower part and bonnet lifted
by the active hood system [27]. The airbag is deployed
in U-character shape and covers the windshield lower
part and A-pillar lower part. When up-lifting bonnet and
airbag are used simultaneously, the impact absorbing
effect is significantly larger than when only one measure
is used. It is because vehicle front regions critical to
pedestrian head impact cannot be effectively covered
with only one measure [24]. Intelligent Vehicle Safety
System (IVSS) seems to be a good example: Active
bumper, active hood, and pedestrian protection airbag
is developed simultaneously and they are expected to
operate as a suite of tools [27]. Autoliv also developed
and provided active hood and pedestrian protection
airbag as a pedestrian protection tool chain [28, 29].
Such a development of passive safety measure
is supposed to have been significantly influenced
and encouraged by the reinforced requirement of
pedestrian–vehicle collision. The European Enhanced
Vehicle Safety Committee (EEVC) Working Group
(WG) 10 and WG 17 have developed test procedures
to assess the level of pedestrian protection for vehicle
fronts. Based on the EEVC WG 17 report, legal require-
ments have been derived. The European directive
(2003/102/EC) consists of head impact, upper leg
impact, and lower leg impact. The requirement was
scheduled to be enforced in two phases, that is, 2005
 Active Pedestrian Protection System, Scenario-Driven Search Method for
and 2010 [30]. The EU test levels are 1,000 head-injury
criteria (HIC) for 2/3 of the hood surface and 2,000
HIC for the remaining 1/3. HIC is an internationally
accepted acceleration-based measurement for violence
against the head. HIC values under 1,000 imply that the
risk of life-threatening injury is 15% or less. However,
the curve rises sharply; at 2,000 HIC, the fatality risk is
almost 90% [28].
Active safety systems are based on pedestrian detec-
tion and collision avoidance measures. Considerable
research is being conducted by various groups for
designing pedestrian detection systems. Such systems
can employ various types of sensors and computer
vision algorithms in order to detect pedestrians and
predict the possibility of collisions. The output of the
systems can be used to generate appropriate warnings
for the driver or to perform autonomous braking or
maneuvering in case of imminent collision [12]. Natu-
rally, if impending collision could not be avoided, the
fact can be sent to the passive pedestrian protection
measures and precrash safety system to help their prep-
arations. Even when the autonomous braking cannot
prevent collision, it can mitigate the pedestrian’s dam-
age. The reduction of collision speed leads to
a reduction of the induced head accelerations and
results in a shift of the relevant head impact regions
away from stiff windshield lower parts. The core tech-
nology of active pedestrian protection system is pedes-
trian detection because the active braking function is
already implemented by several active safety systems
such as PCS, CW/CA, and full-range ACC [13, 14].
Pedestrian detection can be divided into five cate-
gories according to the used sensors: thermal infrared
camera-based, visible camera-based including near
infrared camera, millimeter wave radar-based, com-
munication-based, and sensor fusion-based. Even
though scanning laser radar is frequently used by
pedestrian detection system, it is not considered in
a separate category because it is not used as
a standalone sensor but as a participator in sensor
fusion with visible camera.
Thermal infrared camera-based is specialized for
nighttime pedestrian detection, for example, night
vision, and is based on the point that the human
body temperature is higher than ambient temperature.
Although it can easily detect pedestrians even in dark
illumination conditions, it loses the advantage in
83
daytime and its cost is comparatively high for all grades
of vehicle.
David Gernóimo et al. provide a great survey
of visible camera-based pedestrian detection [31].
They propose a general architecture of a visible
camera-based pedestrian detection system consisting
of six phases: preprocessing, foreground segmentation,
object classification, verification, tracking, and appli-
cation. Preprocessing, foreground segmentation, and
appearance-based object classification are very similar
to vehicle detection case [32]. Foreground segmenta-
tion for candidate generation can be performed using
knowledge-based, stereo-based or motion-based
methods. Appearance-based object classification uses
generally proven learning algorithms such as support
vector machine (SVM), AdaBoost, and neural net-
works. Characteristic approaches are contour tree-
based template matching devised by D. Gavrila and
part detection-based to cope with the pedestrian’s con-
tinuously changing contour and appearance. What
kind of feature and classifier is the best for pedestrian
detection is a problem under fierce discussion. To
answer the question, Mohamed Hussein et al. com-
pared performances of two promising features, that is,
histograms of oriented gradients (HOG) and region
covariance (COV), while the total system was
implemented by a rejection cascade of boosted feature
regions [33]. Markus Enzweiler and Dariu M. Gavrila
compared four state-of-the-art pedestrian detection
systems: wavelet-based AdaBoost cascade, linear SVM
with HOG, neural network (NN) with local receptive
field (LRF), and hierarchical shape matching combined
with texture-based NN/LRF classification [34].
Prof. Hermann Rohling’s team of Hamburg
University of Technology developed distinctive pedes-
trian detection systems using only automotive millime-
ter wave radar [35, 36]. They discovered a fact that
radar echo signals from a walking pedestrian contain
characteristic information that differs from other back-
scattering objects like cars and trucks. Automotive
radar sensor can measure not only target range with
high resolution but also velocity with an accuracy of
a few centimeters per second. When the location and
velocity of target are measured, the pedestrian has
a narrow location range and a wide velocity range
because of its moving arms and legs while the vehicle
and pillar have a narrow velocity range with a wide and
84 Active Pedestrian Protection System, Scenario-Driven Search Method for
a narrow location range, respectively. Furthermore, the
tracked velocity of pedestrian shows regular oscillation
with a certain low frequency, which is a strong indica-
tor and a signal feature to distinguish between pedes-
trian and car. Using features extracted from radar
signal, the pedestrian can be classified from moving
vehicle and stationary objects.
Communication-based pedestrian detection is the
newest approach. It is derived from the idea that radio-
based communication does not need line of sight and
could solve hidden pedestrian problems. Although the
feasibility of mobile communications such as global
system for mobile communication (GSM) and univer-
sal mobile telecommunications system (UMTS) was
studies, they have many disadvantages. Every node
should have a precise global positioning system (GPS)
and peer-to-peer communication should be added to
meet latency requirement [37]. EU project WATCH-
OVER and German project AMULETT investigate the
possibility of short-range communication for pedes-
trian protection [38, 39]. In this approach, vehicles
transmit interrogations and transponders embedded
into vulnerable user’s accessories such as school bag,
bicycle, and motor bicycle return identification mes-
sages. With the multi-antenna system, the vehicles can
recognize the directions and distances of the transpon-
ders. Furthermore, the identification message can pro-
vide a user’s information such as agility, age, and
moving pattern. Candidate short communication tech-
nologies include IEEE 802.15.4 (Zig-Bee), radio fre-
quency identification (RFID), and IEEE 802.15.3a
ultra wide band (UWB) radio. Even if this last
approach is very interesting, it is based on the strong
assumption that all road users have transponders: at
present this is unrealistic.
Setup
This section describes the setup, starting from the per-
ception technologies, going through the sensor suite and
actuators, and concluding with the processing system.
Being designed to address an urban scenario in
which the prototype vehicle is running close to stopped
vehicles, a limitation on vehicle speed and detection
range can be accepted. Low-to-medium vehicle speeds
up to 50 km/h and a detection range of about 40 m can
be considered as a safe choice.
One of the constraints of this problem is the detec-
tion of stopped vehicles and other standing obstacles to
be detected; a laser-based solution sufficiently fit this
requirement robustly enough supplying both distance
measurement and shape classification. The speed infor-
mation is commonly missing in laser data. However,
some laser scanners, like the one selected for this appli-
cation, can provide interlaced data. For these devices,
a number of interlaced scans, taken at different times,
are grouped together in a single, high-resolution scan.
Obstacles speed can be estimated by analyzing these
interlaced data together with vehicle inertial informa-
tion. Using obstacle speed a classification between
standing and moving obstacles is straightforward.
Vision, is the key technology to detect pedestrians
even if they are partially occluded. Since distance mea-
surements are already available from the laser and no
further 3D reconstruction is required, a single NIR
camera satisfies the requirements for both daylight
and night scenario.
The camera selected for this system is an AVT
Guppy F-036B. Its sensor of 752
480 pixels has
a wide aspect ratio which is beneficial for automotive
applications since relevant information is often
contained in lateral area. The sensitivity covers both
the visible and the NIR spectra. During the night,
object detection is improved combining the high
response in the NIR domain using and specific illumi-
nation. Additional headlamps are mounted in front of
the vehicle. A NIR LED headlight with 25 of aperture
is mounted in front of the radiator, while the headlamp
blocks customized by SL can illuminate in the visible
domain, using a bifunctional lamp (low-beams and
high-beams), and in the NIR domain using additional
specific lamp.
The laser scanner is a SICK LMS 211–30206. This
laser has a horizontal scanning angle of 100 with
angular resolution selectable from a minimum of
0.25 ; the detection range goes up to 80 m; a fog
correction feature is also available.
All these characteristics well fit the application
requirements, as a lot of commercially available lasers.
But this one also features interlacing. Selecting 1 of
resolution and a phase-gap of 0.25 , four subsequent
rotations are included in the same scan. Four rotations
are performed in 53.28 ms, thus every rotation takes
13.32 ms. This time is comparable with that of the
 Active Pedestrian Protection System, Scenario-Driven Search Method for
events occurring in a driving scenario. This means the
scans deformed mainly depending on the vehicle
movement. This effect becomes more evident using
the interlacing feature.
Since the camera and the laser scanner are not
synchronized, there is a continuously variable time
shift between the two captured sample streams; how-
ever the displacement due to the non-synchronization
is considered negligible due to ratio between the vehicle
speed, which is low, and the processing rate, which is
high.
The laser scanner and NIR headlamps are inte-
grated in the front bumper, as shown in Fig. 5. The
NIR camera is placed inside the driving cabin near the
rearview mirror as shown in Fig. 6.
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 5
A detail of the front bumper showing the laser scanner
integration and the headlights
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 6
How the camera is installed inside the cabin
85
An industrial grade Mini-ITX PC running an
Intel Core 2 Duo–based CPU is fixed in the boot.
The camera is connected through a FireWire
A controller integrated on the motherboard, while
an external RS422 to USB adapter is used to con-
nect laser. Inertial data produced by off-the-shelf
sensors are gathered through the CAN bus using
a USB adapter.
On the actuator side, the braking system installed
on the test vehicle was replaced with Mando’s MGH-40
ESC plus. The braking strength can be controlled
by sending appropriate CAN messages to this ECU. In
this unit is also available a deceleration control inter-
face (DCI) used for high-level functions, such as
adaptive cruise control (ACC) and precrash safety
(PCS). The DCI may receive multiple deceleration
commands. The desired deceleration is selected
depending on priorities and vehicle status. A feed-
forward controller uses the deceleration output as
control signal in a feedback loop to physically
operate on wheel brake pressures. The braking strength
can be controlled in the range [0, 1.0] g with
a resolution of 0.01 g, a stable error lower than 0.05 g,
and response time of 0.3 s. Deceleration is controlled by
the DCI in cooperation with other active control
system Electronic Brake Distribution (EBD), Anti-lock
Brake System (ABS), and Electronic Stability Control
(ESC).
The horn has been modified to receive a control
signal from both the manual command and a USB I/O
board (see Fig. 7a) connected to the processing unit
and detected as a serial port.
A new 100 Hz yaw rate sensor, manufactured by
Siemens VDO and Mando, is connected through
the processing system via CAN bus. This sensor
provides yaw rate with an accuracy of 0.0625 /s
and also lateral acceleration. Figure 7b shows the
sensor.
Focus of Attention
Two methods for laser data classifications are devel-
oped to improve algorithm robustness. Pulses are first
clustered into straight lines for both methods; then the
first one classifies obstacles only considering the last
scan, while the second one verifies the obstacle position
and speed using subsequent scans.
86 Active Pedestrian Protection System, Scenario-Driven Search Method for

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 7

(a) The USB I/O board used to control the horn, and (b) the special yaw rate sensor

Significant shifts between subsequent laser scanner

measurements affect the measured obstacle shape as
a distortion, and cause clustering or classification
problems.
Vehicle movement in terms of rotation and trans-
lation is measured using the ESP CAN box. Having this
information available, each pulse in the scan can be
corrected removing the shape distortion.
Translation is corrected using vehicle speed, while
using the steering wheel angle to compute the yaw rate
allows to completely specify the rotation correction
matrix.
Yaw rate can be computed as follows:

Yawrate ¼ ðVfr þ Vf l Þ=ðtw cos dÞ

where Vfr and Vfl are front right and front left wheels
speeds, respectively, d indicates the wheel angle, and tw
is the vehicle front track.
The additional yaw rate sensor installed on the
vehicle is used to measure directly this value, since
indirect computation is highly affected by noise.
Figure 8 shows raw and corrected laser data refer-
ring to nonmoving obstacles. The four laser rotations
that create a scan are clearly visible before the correc-
tion, while the obstacle appears as a single line after the
correction.
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 8
Laser data: (a) without correction and (b) with correction
After data correction, moving obstacles captured
from interlaced scans are characterized by four parallel
lines, as shown in Fig. 9. Obstacle classification is
performed relying on this information.
 Active Pedestrian Protection System, Scenario-Driven Search Method for
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 9
After data correction a moving obstacle is still represented
by four different lines
Pulses belonging to the same rotation are connected
together as a first step forming chains of segments or
cluster. Points which are not joinable to any chain of
the same rotation are checked for merging with points
of other rotations considering proximity only. Points
not connected to other points or close to the limit of
the laser scanner range (about 80 m) are permanently
discarded.
After clustering, moving obstacles are classified by
identifying four different and parallel clusters, while
standing obstacles are detected when the clusters are
overlapped.
Up to this step, chains of segments are created by
joining pulses without any additional information.
When adjacent segments have approximately the
same orientation, they are merged into a longer seg-
ment, preserving the obstacle shape, but reducing the
amount of information supplied by the sensor. Straight
lines connecting the chain start and end points are
obtained merging a chain of adjacent segments.
The distances between a line and all its internal
points are computed and, if the maximum between
these distances is larger than a threshold, the line is
split into two separate lines.
These steps are iterated while the maximum dis-
tance is larger than a threshold. The result is that each
chain is therefore finally segmented into a polyline.
Figure 10 shows a moving vehicle: The rear bumper
is framed as four parallel lines, while its side – parallel
87
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 10
A moving vehicle and its motion direction: the four parallel
lines correspond to its rear bumper, while the single line
corresponds to its left side. Blue points are polylines start
points, cyan points are corner points, red points are
polylines end points, and violet points are generic line points
to the vehicle movement direction – is marked by
a single line.
After data correction, points clustering and seg-
ment merging, each obstacle detected in the laser scan-
ner field of view, is identified by four lines, one for each
laser rotation that composes the whole high-resolution
scan. Static and moving obstacles can be located and
classified in a robust way using the following observa-
tion. A static object has four lines that are overlapped
on a single shape, while a moving obstacle has the lines
parallel but not overlapped. Figure 11 shows the
algorithm steps.
Although this process is very simple and effective,
when vehicle pitch is significant, false negatives are
possible.
Polylines obtained so far can only be classified
thanks to their size and shape. Obstacles can be divided
in four categories:
● Possible pedestrian
● Road infrastructure
● L-shaped obstacle
● Generic obstacle
Obstacles with limited size are classified as pedes-
trians, while obstacles almost parallel to the vehicle and
with a large size are classified as road edges (guard-rails,
buildings, road infrastructures, etc.).
L-shaped obstacle patterns are detected using
a simple and fast method, based on line orientations.
88 Active Pedestrian Protection System, Scenario-Driven Search Method for

All obstacles not yet classified are tagged as generic

a obstacles; Fig. 12 shows an example.
b
c
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 11
The steps of the algorithm: (a) data clustering,
(b) approximation using polylines, and (c) lines merging: all
the points are merged into a single line; in order to simplify
the following steps, small drifts are ignored
Information about obstacle movement – already
estimated in the previous step – is also stored for
further usage.
The procedure described in the previous subsection
classifies all small obstacles as possible pedestrians.
However a number of false positives may indeed be
present under unfavorable conditions.
Parked vehicles and other fixed obstacles along the
road are used to identify critical areas in front of the
vehicle, to focus the detection in the immediate prox-
imity of these areas where pedestrians can suddenly
appear and become dangerous.
Ego-motion information is used to align and roto-
translate polylines provided by previous scans; a check
on the overlapping between polylines identified in

Pedestrian
LShape

Unknown

LShape Pedestrian

Pedestrian

Pedestrian
Pedestrian
LShape

Road

Unknown

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 12

Obstacle classification: all possible classifications based on shape are represented in the image
 Active Pedestrian Protection System, Scenario-Driven Search Method for
current and in previous frames is then performed to
assign a category to the object between: moving
obstacle, static obstacle, changing shape obstacle, and
new obstacle.
Moving obstacles have small spatial overlapping if
the time window is large.
If no correspondence is found between the current
polyline and those from the past, the object is classified
as a new obstacle.
The areas of interest in which vision will search for
pedestrians is determined using static obstacles;
obstacles whose shape is changing over time are also
of basic importance. They may contain a pedestrian in
a smaller region of their shape.
The centroid of each polyline is computed to track
obstacles over time. Centroids are used as representa-
tive points to simplify the computation. A description
of a more complex laser scanner data processing based
on comparison between each point of polylines can be
found in [40]. A different classification is also made
using distances between the centroid of the last and
previous scans.
Polylines are labeled using the information
contained in the distances between the current centroid
and all the past ones as follows:
● Fixed obstacle: if all distances are below a certain
threshold. This threshold, to detect, is increased
according to the distance of the obstacle (to com-
pensate for a decreasing accuracy) and the time gap
(to partly compensate for errors in ego-motion
reconstruction).
● Moving obstacle: if the distances grow over time.
● Unknown obstacle: this label is given if the object
can neither be labeled as fixed nor moving. This
may happen the first time an obstacle appears or
when the detection is not stable due to bad reflec-
tions of the lasers beam.
The three previous classification results are consid-
ered to compensate for laser data processing inaccu-
rate results. The polyline is classified with the most
frequent one.
Example results are shown in Fig. 13 with an esti-
mation of the object speed.
The following steps of the algorithm use the
processing results on the labeled polylines. This infor-
mation is used to sense the static (obstacles and
89
vehicles) position and moving (pedestrians and vehi-
cles) obstacles in the environment. When obstacles are
identified as static by the second classification stage,
they are used to define the environment structure.
Driving corridor is defined as the area that may be
reached by the vehicle in the near future. Fixed obsta-
cles are used to build the driving corridor, namely, as
shown in Fig. 14.
The driving corridor may have multiple paths or
branches, each with a minimum width larger than
the vehicle one; branches with a width narrower than
this threshold are discarded as closed. The corridor
does not necessarily correspond to the free space
in front of the vehicle, since it is built considering
fixed obstacles only, and may also include moving
obstacles.
This algorithm is focused on the detection of
suddenly appearing pedestrians; the detection is con-
centrated on areas hidden by a static obstacle or
between two static obstacles. These areas are usually
located along the corridor edges. Static obstacles are
used to build the corridor, and also to identify danger-
ous areas, that is, areas in which pedestrians may
appear. Dangerous areas are located behind the furthest
point of each static polyline as shown in Fig. 15.
Obstacles moving inside the corridor, or near its
edges, with a speed and a size typical of a pedestrian, are
tagged as possible pedestrians.
Vision Fusion
The regions of interest are 2D areas in world coordi-
nates; their corresponding areas in the image are then
located thanks to camera calibration and perspective
mapping.
Once the areas of attention are located, the search
for pedestrians is triggered in these areas. Specific
image windows are defined using a perspective map-
ping transformation, considering 90 cm as pedestrian
width and 180 cm as pedestrian height. These image
areas are resampled to a fixed size (24
48 pixels).
In order to reach good classification during both
day and night, two different systems are developed. The
idea is to use a simple and specific classifier for night-
time images, in which pedestrians’ shapes clearly
appear, and to use more complex classifiers (based on
cascade methods) for the day.
90 Active Pedestrian Protection System, Scenario-Driven Search Method for

Moving
V = 4.6 km/h
Fixed
Fixed
Fixed

Moving
V = 5.5 km/h
Fixed
Fixed Fixed

Fixed

Fixed

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 13

Obstacle classification: all possible classifications based on movement are shown. Note that moving polylines change their
position in world coordinates

The nighttime algorithm is based on the assump-
tion that, thanks to NIR illuminators, pedestrians and
other standing objects represent the brightest areas in
the image with the exception of light sources. There-
fore, after removing light sources, it is possible to detect
obstacles with a binarization process and classify them
according to their shape.
At the end of the preprocessing steps, a binarized
image is obtained. The standing objects (including
pedestrians) are remapped as white, while the back-
ground is remapped as black. As the NIR headlamps
illuminate more the legs of the pedestrian, these first
phases are operated separately for the upper and lower
part of the image.
After these operations, the procedure continues to
check for pedestrian trunks and legs by respectively
analyzing the upper and lower half of the image. First
of all open legs are searched for; if open legs are not
found, an algorithm that checks for closed legs is
applied, and, finally a trunk presence check is
performed.
The open legs search is based on striped
skeletonization. For each row of the image, the
white areas present are narrowed to a single point.
 Active Pedestrian Protection System, Scenario-Driven Search Method for 91

Pedestrian
Moving
V = 4.6 km/h
LShape
Fixed Pedestrian
Pedestri
Fixed
Fixed

Pedestrian
Moving
V = 5.5 km/h
LShape
Unknown LShape Fixed
Fixed Fixed

LShape
Fixed

LShape
Fixed

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 14

The pink zone represents the driving corridor. Obstacles moving inside the corridor, or near its edges, with a speed and
a size typical of a pedestrian, are tagged as possible pedestrians

In such a way it is possible to remap legs as lines.
Obtained lines can be transformed to straight lines,
and the position of these lines can be finally studied
to identify legs.
Search for trunk and closed legs are very similar,
differing only in some details and thresholds. The cap-
tured images contain a very low level of detail and it is
often inadequate for a complex analysis. For this reason
a simple procedure is implemented that tries to make
the most of the few details that characterize the pedes-
trian. The algorithm searches for white areas and con-
siders their aspect ratio to confirm the presence of
trunk or closed legs.
At the end of the algorithm four rates are assigned,
which range from 0 to 1. The votes are:
● Vote of the trunk according to the correspondence
with a model and to dimension.
92 Active Pedestrian Protection System, Scenario-Driven Search Method for
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 15
Dangerous areas and possible pedestrians: red circles
represent dangerous areas, while blue boxes identify
possible pedestrians
● Vote of the legs according to the correspondence
with a model and to dimension.
● Open legs rate which is obtained with the ratio
between feet position distance and maximum feet
distance.
● Pedestrian vote which is obtained making an aver-
age of the body and legs vote. The vote of the legs
has more weight if they are open because it is
a dominant characteristic of a pedestrian.
All these ratings will be analyzed at a later stage of
the application to properly determinate if the figure is
recognized as a pedestrian.
AdaBoost (a short for Adaptive Boosting) is
a machine-learning meta-algorithm. It is used to select
the combination of a weak classifier to obtain a precise
classification.
Using an appropriate training set it is possible to
compute the correct weight to be set to each classifier
and obtain a robust classification. AdaBoost is
a technique widely used for the classification of pedes-
trians [41]. Haar features are firstly chosen for the weak
classifier [42]. Different Haar features are selected for
each iteration, as suggested by Viola and Jones [43].
Haar features are very simple features that can be used,
together with a meta-algorithm similar to AdaBoost, to
classify a lot of objects. They consist of a very simple
pattern that can be moved and stretched on the image
to be compared to image integral, in order to obtain an
accurate classification.
As Haar features are very generic, the idea is to
develop other features, specific for pedestrians, in
order to reach a better classification using AdaBoost
again. These features are based on a typical pedestrian
shape and body parts aspect. To detect a pedestrian, the
algorithm provides a vote to each feature of the follow-
ing list:
● Open legs shape
● Left step legs, little mask
● Right step legs, little mask
● Left step legs, big mask
● Right step legs, big mask
● Head top, little mask
● Head top, big mask
● Head, with circle formula
● Pedestrian’s M function
● Horizontal border function
● Tree and pole search function
To increase the potential and the recognition per-
formance, an additional stage, called AdaBoost cascade,
is developed.
In this stage the votes from the classifier
implemented with Haar features, with ad hoc ones,
and nighttime classifier are merged and combined to
obtain an additional level of classification. The votes
provided by the previously described algorithms are
used as input to make a second kind of AdaBoost
classifier.
As the number of outputs from previous classifiers
is very low, and some tests pointed out that they cannot
be used to reach an efficient classification process by
their own, nonlinear combinations of these values are
added as input of AdaBoost algorithm.
 Active Pedestrian Protection System, Scenario-Driven Search Method for 93

a b

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 16

Two subsequent frames of a sequence: (a) the system detects the partly occluded pedestrian and issues an internal alert,
but no warning to the driver. (b) When a detection of a fully visible pedestrian follows the internal alert, a warning is
promptly sent to the driver

According to the classic voting approach, a final
hypothesis resulting formula is implemented: The
final vote is computed as the weighted sum of the
classifiers selected by the AdaBoost training stage.
Instead of using two classes only (pedestrians and
non-pedestrians) three classes are here used:
● Pedestrians
● Non-pedestrians
● Appearing pedestrians
Appearing pedestrians are pedestrians that are ini-
tially not completely visible, that is, partially occluded
by obstacles, so only a part of the pedestrian’s shape can
be framed – the upper or side part only.
AdaBoost was trained using image windows deter-
mined by the previous steps of the algorithm. Candi-
date selection is a complex and critical step for the
AdaBoost training process [16]. Both normal and
flipped samples are used in the training process. If
two or more images are very similar (i.e., the pixel-
wise difference is less than a threshold) only one of
them is used in the training process. Images framed in
Italy, the Netherlands (during tests before the IEEE
Intelligent Vehicle Symposium 2008 demonstration),
and Korea were used in the training process.
Once a pedestrian partly hidden by a vehicle is
detected – by the vision system only – the system
issues an internal alert. In this case no warning is pro-
vided to the driver because the danger level is not yet
determined.
When an alert is issued, the search continues in the
same zone and tracking is started. As soon as a tracked
pedestrian becomes visible also to the laser scanner, the
direction of its movement is considered. If a pedestrian
moving from the corridor’s edge to the corridor’s cen-
ter is detected, a warning is then issued to the driver.
Figure 16 shows an alert and its following warning.
It is of basic importance to note that the driver is
warned only when the pedestrian is completely visible,
like in other systems. However, the system presented in
this work is more reactive than others, since the track-
ing starts when the pedestrian is only partially visible by
one of the sensors. In an urban situation like this,
promptness is an important key to the success of the
system.
Figure 17 shows some results obtained in complex
conditions (bad weather condition, poor light condi-
tion) and a misclassification: a suddenly appearing
pedestrian detected as a normal pedestrian.
Automatic Braking
Previous works have introduced the Region of No
Escape (RONE [44]) as the area that will be anyway
occupied by the vehicle in the near future even if the
driver steers away or brakes.
Like traditional pedestrian detection systems,
pedestrians present in the RONE are here detected by
laser and vision fusion; the presence of a pedestrian in
the RONE triggers the automatic braking, as a Collision
94 Active Pedestrian Protection System, Scenario-Driven Search Method for

a b

c d

e f

Active Pedestrian Protection System, Scenario-Driven Search Method for. Figure 17

Some suddenly appearing pedestrians correctly detected: (a) in an underground parking, (b) in the rain, (c) behind
a misaligned vehicle, (d) behind a wall, (e) at night; and (f) a suddenly appearing pedestrian detected as non-dangerous
pedestrian (false negative)

Mitigation system (i.e., when the accident is
unavoidable).
The RONE is indeed important to detect an inevi-
table crash, but its size and shape are defined assuming
an optimal driver maneuver. Most drivers, anyway,
have (1) longer reaction times and (2) suboptimal
skills; therefore an extension of the RONE is mandatory
in order to act as a driving aid (avoid the crash) rather
than a collision mitigation system.
It has to be understood, though, that any extension
of the RONE will include areas in which the vehicle has
a high probability to move, but it is not 100% sure to be
 Active Pedestrian Protection System, Scenario-Driven Search Method for 95

occupied by the vehicle in the near future (e.g., in case

of a very skilled driver implementing an optimal avoid-
ance maneuver).
The system presented in this work defines the decel-
eration area as a suitable RONE extension designed for
the specific scenario addressed by this project. In fact,
pedestrians appearing from behind an obstacle might
induce a reaction time longer than the time required by
a driver to locate a clearly visible pedestrian. Therefore,
still assuming that the driver is not distracted and has
optimal driving skills, longer reaction times would
enlarge the RONE.
The deceleration area width is defined to be equal to
vehicle width and its length proportional to vehicle
speed, but anyway bounded to 40 m, while yaw is
considered constant for the trajectory computation.
Whenever a pedestrian appearing from behind an
obstacle crosses the road and enters in the deceleration
area, the system will activate the automatic braking,
even if a skilled driver might be capable of avoiding it.
Active Pedestrian Protection System, Scenario-Driven
This is why it is extremely important to be able to
Search Method for. Figure 18
issue a warning to the driver well in advance, and
Driving corridor, RONE, and deceleration area: the pink
provide sufficient time to perform a manual maneuver.
area represents the driving corridor, the red area
The driver has indeed a more subtle and deeper
represents the RONE, and the green area represents the
perception of the pedestrian behavior and a manual
deceleration area
countermeasure might be more effective than an
automatic one.
In this section the concepts of driving corridor,
RONE, and deceleration area are introduced. It is this second aspect is very important because the whole
important to note the difference between these three data processing must be operated real time.
areas in front of the vehicle, which are used for different To evaluate the obtained results the parts of the
purposes. As shown in Fig. 18, the driving corridor is system can be considered separately, in order to assess
a jagged area edged by static obstacles present in front the performance of each part and to plan following
of the vehicle and represents the area where the vehicle developments.
can move. The system can be divided into two parts, the first
The deceleration area is a smoother area that rep- one is the regions of interest localization, while the
resents the positions where the vehicle should move second one is the search for pedestrians inside the
considering a fixed yaw rate. region of interest.
The RONE represents the area where the vehicle will The first part is simpler, and the results are only
move in the near future. qualitatively evaluated. Regions of interest can be cor-
rectly detected with an appropriate precision up to
a distance of about 40 m. Unfortunately some errors
Results
can occur due to strong vehicle pitch variation. If the
During the development of an ADAS it is important to laser orientation varies too much, the framed shape
consider different aspects for the results assessment: The does not correspond to a significant representation of
first aspect is the precision of results, while the second the environment. This problem is solved using a pitch
one is the time that is needed to process the information; rate sensor. If the pitch variation is too big, the whole
96 Active Pedestrian Protection System, Scenario-Driven Search Method for
system is disabled, because the first part of the algo-
rithm is not able to get correct results. Anyway gener-
ally high pitching is time limited, therefore the system
is blinded for a short time interval.
Search for pedestrians inside regions of interest is
tested with three different test sets: one for daytime
scenes, one for nighttime scenes with poorly illumi-
nated environments, and one for nighttime scenes with
proper illumination.
Figure 19 shows ROC curves for ad hoc features,
Haar features, and cascade method. The algorithms
based on Haar features reach more precise results
than algorithms based on ad hoc features, but it is
important to note that the cascade method, that merges
the two methods, can reach a good classification rate.
Results obtained on nighttime scenes with proper envi-
ronmental lighting are similar to the previous one.
Some tests also pointed out that the use of specific
nighttime algorithms produce worse results.
The nighttime search, based on skeletonization,
reaches good results without any further processing,
as shown in Fig. 20.
Even if these results are important to develop the
system, the final goal of the algorithm is not the detec-
tion of all pedestrians present in the scene, but the
detection of dangerous pedestrians only. It is not
important to detect a pedestrian normally walking on
the sidewalk, while it is mandatory to detect
a pedestrian that suddenly appears in front of the
1 1
0.8
". /roc_ah.txt"
". /roc_cs.txt"
0.6 ". /roc_hr.txt"
0.4
0.2
0 0
0 0.2 0.4 0.6 0.8
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 19
ROC curves for daytime classifier algorithm
vehicle; therefore the results must be assessed consid-
ering the dangerous situations only.
A set of sequences were considered for extensive
performance assessment, for a total driving time of
about 10 h in complex urban scenarios. A total of
236 km were driven during day and night, in different
weather conditions (sunny, cloudy, rain, fog). Various
scenarios were included: downtown, large and narrow
roads, underground car parks, highways, and rural
roads. During night tests, scenarios with and without
external illumination were acquired. During night
tests the NIR illuminators were always turned on. To
complete the test sets some specific situations were staged
(such as very dangerous pedestrian crossings), but
a number of dangerous scenarios were anyway framed
in normal driving.
In the tests, the following performance indexes were
considered:
(a) Number of pedestrians suddenly appearing in
front of the vehicle (that must generate warnings
to the driver)
(b) Number of pedestrians appeared in front of the
vehicle that have been successfully tracked and
which can be hit by the vehicle (that must trigger
the automatic brake)
(c) Number of fully visible pedestrians which can be
hit by the vehicle (that must trigger the automatic
brake)
0.8 ". / roc_nt.txt"
0.6
0.4
0.2
1 0 0.2 0.4 0.6 0.8 1
Active Pedestrian Protection System, Scenario-Driven
Search Method for. Figure 20
ROC curves for daytime classifier algorithm
 Active Pedestrian Protection System, Scenario-Driven Search Method for
For each performance index the number of correct
detections, false positives, and false negatives were
computed.
Performance Correct False False
index detections negatives positives
A 24 11 1
B 5 0 0
C 8 0 0
Results of a 10-h drive (236 km, 540,000 images)
A total of 24 suddenly appearing pedestrians were
correctly detected in the tests (case A). Only one false
positive is present.
Only 1 missed warning out of 11 is due to
a pedestrian miss-detection; the others are caused
by either alert misses or delayed detections, but in
these cases, even if the warning signal is not promptly
issued to the driver, the automatic brake would
have stopped the vehicle avoiding the crash if the
pedestrian would have been in a dangerous situation.
All pedestrians into the deceleration area are localized
(no false negatives): five fall into case B, while eight fall
into case C.
No false positives are present, thanks to the special
attention paid to develop this SDS approach.
Anyway it is important to note that, for safety
reasons – given that the tests were performed in real
traffic – the size of the deceleration area was increased
but, nevertheless, the system behaved very satisfactorily.
The limited number of dangerous events requiring the
intervention of the safety system is not surprising. During
normal driving dangerous situations are not frequent;
plus, some of them are staged in order to challenge the
system.
The analysis of the results obtained during the test
highlighted good results also in case of rain, when
pedestrian with umbrellas were detected as well. The
system is anyway not able to discriminate between
multiple pedestrians moving together or in situations
in which the laser scanner is not able to obtain a clear
picture, for example, when pedestrians hold bags or
other large objects; moreover suddenly appearing
pedestrians walking very slowly or pedestrians under
very critical lighting conditions can be missed.
97
A particular attention was paid to the processing
time. A system that it is not able to process the data real
time is useless for automotive applications.
The great benefit of the SDS method is the reduc-
tion of the regions of interest that must be considered
by the vision system. Some tests focused that the num-
ber of considered areas is almost half of the ones con-
sidered using a traditional system. As the image
processing is the most time-consuming part of the
algorithm, the system can reach a significant reduction
of global processing time.
Considering that the system is designed to work in
urban and rural street, where the vehicle speed must be
lower than 50 km/h, a processing frequency of 10 Hz
can be considered sufficient.
Frame processing time changes according to the
chosen method (nighttime method is quicker) and to
the number of regions of interest which are analyzed.
Anyway, the execution time on a good performing PC
is always lower than 70 ms/frame (about 14 Hz), con-
sidering the whole system.
Future Directions
An important concept in road safety aimed at avoiding
collisions with vulnerable road users was presented.
Instead of implementing a traditional pedestrian detec-
tion system followed by both a validation step and the
assessment of the danger level, it was proposed an
innovative approach. Whenever a specific environmen-
tal scenario is detected (i.e., stopped vehicles that may
hide pedestrians), the possible presence of pedestrians
is checked for in well-defined areas only, focusing on
a particularly critical environment, typical of urban
accidents.
Besides directly detecting dangerous situations,
it also boosts timing performance, since the computa-
tionally intensive part – vision-based pedestrian
recognition – is performed only on limited portions
of the image.
The system was developed to tackle a very specific
yet very common scenario, while it does not cover the
whole set of dangerous situations that may occur in an
urban environment.
The search for all the pedestrians present inside the
so-called driving corridor can generalize the system, yet
maintaining a scenario-driven approach, as the search
98 Active Pedestrian Protection System, Scenario-Driven Search Method for
is performed only in a corridor dynamically built using
laser scanner data.
As a conclusion, the proposed approach offers
the following advantages over traditional pedestrian
detection systems.
● The system looks for specific spots in the 3D scene
that might contain pedestrians (even partly
occluded ones); in these areas the search for pedes-
trians is performed by vision; vision is therefore
activated only in selected areas providing a lower
number of false detections and a faster processing.
● Once a partly occluded pedestrian is detected, an
internal alert is issued; when the pedestrian
becomes completely visible, the initial alert is
followed by a warning to the driver. Therefore the
system reacts very quickly to the sudden appearance
of a pedestrian, since tracking already started when
the pedestrian was partly occluded and visible only
to one of the sensors (vision).
This work was developed as a feasibility study
aimed at determining whether the deployment of
nonreversible systems might be a viable solution for
real implementations. Nonreversible systems have
always been considered extremely critical, since their
robustness and effectiveness has to reach 100% to allow
their integration on real vehicles. No sudden stops or
irreversible maneuvers have to be triggered by false
positives in the detection. Unfortunately, reaching
a correct interpretation of all scenes and scenarios is
as desirable as improbable and a lot of testing must be
performed to evaluate possible failures in the detection.
Bibliography
1. O’Farrell J (2006) May contain nuts. Black Swan, London
2. Lucchini E, Kramer M (1981) Safety device for the protection of
pedestrians. Patent
3. Matsuura Y, Maki T (2001) Device for reducing the impact of
pedestrians. Patent
4. Myrholt H, Ericsson M (2002) Deployable A-pillar covers for
pedestrian protection. Patent
5. Buehler M, Iagnemma K, Singh S (eds) (2010) The DARPA
urban challenge. Springer, Berlin
6. Agran PF, Winn DG, Anderson CL, Tran C, Del Valle CP
(1996) The role of the physical and traffic environment in
child pedestrian injuries. Pediatrics 1996(98):1096–1103
7. Retting RA, Ferguson SA, McCartt AT (2003) A review of
evidence-based traffic engineering measures designed to
reduce pedestrian-motor vehicle crashes. J Public Health
2003(93):1456–1463
8. Premebida C, Monteiro G, Nunes U, Peixoto P (2007) A lidar and
vision-based approach for pedestrian and vehicle detection and
tracking. In: Proceedings of IEEE international conference on
intelligent transportation systems, Seattle, pp 1044–1049
9. Hwang JP, Cho SE, Ryu KJ, Park S, Kim E (2007) Multi-classifier
based LIDAR and camera fusion. In: Proceedings of the IEEE
international conference on intelligent transportation
systems, Seattle, pp 467–472
10. Franke U, Heinrich S (2002) Fast obstacle detection for urban
traffic situations. IEEE Trans Intell Transport Syst 3(3):173–181
11. Curio C, Edelbrunner J, Kalinke T, Tzomakas C, Von Seelen W
(2000) Walking pedestrian recognition. IEEE Trans Intell
Transport Syst 1(3):155–163
12. Gandhi T, Trivedi MM (2007) Pedestrian protection systems:
issues, survey, and challenges. IEEE Trans Intell Transport Syst
8(3):413–430
13. Hodzic M (2008) Distronic plus and brake assist plus reduce
rear-end collisions by 20%. http://www.benzinsider.com/
2008/06/distronic-plus-and-brake-assist-plus-reduce-rear-end-
collisions-by-20/. Accessed 30 May 2010
14. The Motor report (2010) Toyota Prius To ship with advanced
safety gear. http://www.themotorreport.com.au/29611/2010-
toyota-prius-to-ship-with-advanced-safety-gear. Accessed 30
May 2010
15. Juliussen E (2010) Driver assist systems. http://www.
autofocusasia.com/electrical_electronics/driver_assist_systems.
htm. Accessed 30 May 2010
16. http://www.prevent-ip.org/en/prevent_subprojects/intersec-
tion_safety/intersafe/
17. BBC News (2010) Self-parking car hits the shops. http://news.
bbc.co.uk/2/hi/technology/3198619.stm. Accessed 31 May
2010
18. www.trafficcalming.org
19. Clarke A, Dornfeld M J (1994) Traffic calming, auto-restricted
zones and other traffic management techniques: their effects
on bicycling and pedestrians. national bicycling and walking
study, Case Study No. 19, Federal Highway Administration
1994
20. Huang HF, Cynecki MJ (2001) The effects of traffic calming
measures on pedestrian and motorist behavior. Research
report FHWA-RD-00-104, Federal Highway Administration,
U.S. Department of Transportation
21. Crandall JR, Bhalla KS, Madeley NJ (2002) Designing road
vehicles for pedestrian protection. Brit Med J
324(7346):1145–1148
22. Yao J, Yang J, Otte D (2008) Investigation of head injuries by
reconstructions of real-world vehicle-versus-adult-pedestrian
accidents. Saf Sci 46(7):1103–1114
23. Koch W, Howard M (2003) Comprehensive approach to
increased pedestrian safety in pedestrian – Car accidents.
Proc Inst Mech Eng Part D J Automobile Eng 217:513–519
24. Fröming R, Schindler V, Kühn M (2006) Requirement engineer-
ing for active safety pedestrian protection systems based on
 Active Pedestrian Protection System, Scenario-Driven Search Method for
accident research. In: Advanced microsystems for automotive
applications, New York, pp 79–106
25. Schuster PJ (2006) Current trends in bumper design for
pedestrian impact. SAE Paper No.: 2006-01-0464
26. Fredriksson R, Håland Y, Yang J (2010) Evaluation of a new
pedestrian head injury protection system with a sensor in the
bumper and lifting of the bonnet’s rear part. http://www.autoliv.
com/wps/wcm/connect/5f997a004ce4f3f8ac68eef594aebdee/
Pedestrian.pdf?MOD=AJPERES. Accessed 30 May 2010
27. http://www.ivss.se/
28. Broge JL (2010) Autoliv actively saving lives. http://www.sae.
org/automag/techbriefs/11-2001/page2.htm. Accessed 30
May 2010
29. Autoliv (2010) Pedestrian protection. http://www.autoliv.
com/wps/wcm/connect/autoliv/Home/What+We+Do/Recent
%20Innovations/Pedestrian%20Protection. Accessed 30 May
2010
30. Kerkeling C, Schäfer J (2005) Structural hood and hinge
concepts for pedestrian protection. In: Proceedings of 19th
international technical conference on enhanced safety,
Washington, DC, pp 379–389
31. Gernóimo D, López AM, Sappa AD, Graf T (2010) Survey of
pedestrian detection for advanced driver assistance systems.
IEEE Trans Pattern Anal Mach Intell 32(7):1239–1258
32. Sun Z, Bebis G, Miller R (2006) On-road vehicle detection:
a review. IEEE Trans Pattern Anal Mach Intell 28(5):1–18
33. Hussein M, Porikli F, Davis L (2009) A comprehensive evalua-
tion framework and a comparative study of human detector.
IEEE Trans Intell Transport Syst 10(3):417–427
34. Enzweileer M, Gavrila DM (2009) Monocular pedestrian detec-
tion: survey and experiments. IEEE Trans Pattern Anal Mach
Intell 31(2):2179–2195
99
35. Folster F, Rohling H, Meinecke M-M (2005) Pedestrian
recognition based on automotive radar sensors. In: 5th
European congress and exhibition on ITS and services,
Hanover
36. Ritter H, Rohling H (2007) Pedestrian detection based on
automotive radar. In: Proceedings of the international confer-
ence on radar systems, Edinburgh
37. David K, Flach A (2010) CAR-2-X and pedestrian safety. IEEE
Veh Technol Mag 3:70–76
38. http://www.watchover-eu.org/
39. http://www.project-amulette.de/
40. Broggi A, Cerri P, Ghidoni S, Grisleri P, Jung HG (2008) Locali-
zation and analysis of critical areas in urban scenarios. In:
Proceedings of IEEE intelligent vehicles symposium, Eindho-
ven, pp 1074–1079
41. Leyrit L, Chateau C, Tournayre C, Lapresté J-T (2008) Associa-
tion of Adaboost and kernel based machine learning
methods for visual pedestrian recognition. In:
Proceedings of IEEE intelligent vehicles symposium,
Eindhoven, pp 67–72
42. Geismann P, Schneider G (2008) A two-staged approach to
vision- based pedestrian recognition using Haar and HOG
features. In: Proceedings of IEEE intelligent vehicles sympo-
sium, Eindhoven, pp 554–559
43. Viola P, Jones M (2001) Rapid object detection using a boosted
cascade of simple features. In: Proceedings of the international
conference on computer vision and pattern recognition, vol 1.
Hawaii, pp 511–518
44. Fuerstenberg K, Schulz R (2006) Laserscanners for driver assis-
tance. In: Proceedings of the international workshop on
intelligent transportation, Hamburg, pp 155–159
100 Advanced Public Transport Systems, Simulation-Based Evaluation
Advanced Public Transport Systems,
Simulation-Based Evaluation
HARIS N. KOUTSOPOULOS1, MOSHE BEN-AKIVA2
1
Department of Transport Sciences, The Royal Institute
of Technology, KTH, Stockholm, Sweden
2
Department of Civil and Environmental Engineering,
MIT, Cambridge, MA, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Advanced Public Transportation Systems (APTS)
Simulation-Based Modeling of APTS
Microscopic Transit Simulation Example: MITSIMLab
Mesoscopic Transit Simulation Example: BusMEZZO
Future Directions
Bibliography
Glossary
APTS Advanced public transportation systems is the
transit component of Intelligent Transportation
Systems (ITS). It refers to technologies for the
collection, communication, and processing of data
useful for the management of transit systems.
AVL Automated vehicle location systems combine
vehicle location and communications technologies
to automatically track the locations of a fleet of
vehicles. AVL is an integral component of auto-
mated vehicle monitoring and control, fleet man-
agement, traffic signal priority, and other transit
applications. AVL can be used to monitor schedule
adherence, estimate arrival times, and communi-
cate location data to a transit operations control
center (TOC) or to field-installed devices that
require real-time vehicle location data.
APC Automatic passenger counters are systems that
count passengers as they board and alight the
vehicle at a stop. APCs may be used with AVL
systems in order to record the spatial distribution
of passenger demand along a vehicle’s route. APC
technologies include treadle mats, infrared beams,
and emerging computer imaging (still at early stage
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
of development). APC records the stop location, the
time and date of arrival at the stop, the time the
doors open and close, the number of passengers
boarding and the number of passengers alighting.
BRT Bus rapid transit is an evolving transit concept
aiming at combining the quality and capacity of rail
transit and the flexibility of bus transit.
Electronic fare payment Electronic fare payment
includes a range of technologies, such as smart
cards, designed to reduce costs associated with fare
collection and to improve customer convenience (also
known as Automated Fare Collection (AFC) systems).
Fleet management Fleet management applications
refer to “vehicle-based” technologies that may be
used to improve vehicle planning, scheduling, and
operations.
ITS Intelligent transportation systems is an umbrella
term referring to sensor, communication, and com-
puting technologies for improved management of
transportation systems. ITS relates to all transpor-
tation modes.
Simulation “. . . the process of designing a model of
a real system and conducting experiments with this
model for the purpose of understanding the behav-
ior of the system or evaluating various strategies for
the operation of the system” [1].
Signal priority Transit signal priority involves the
modification of a traffic signal’s regular timing
plan to give preference to transit vehicles. Signal
priority is designed to reduce transit vehicle delays
at signalized intersections.
Traveler information Traveler information refers to
technologies designed to provide pre-trip and en
route information to travelers to allow them to
make informed trip-making decisions.
Transportation demand management Transportation
demand management refers to systems aimed at
improving the utilization of existing transportation
network infrastructure, through influencing the
demand characteristics.
Definition of the Subject and Its Importance
According to the latest UN report on urbanization,
more that 50% of the world’s population lives in
cities [2]. In Europe and the USA, it approaches 70%.
# 2012, DOI 10.1007/978-1-4419-0851-3

In part due to developments like these, congestion in
urban areas continues to grow and create a number of
negative impacts, including worsening air quality and
noise pollution, consumption of scarce resources, lost
productivity.
Public transportation is an important component
of the transportation system and potentially a critical
element of any strategy toward sustainable mobility in
urban areas. The importance of improved public trans-
portation services toward sustainable and efficient
transportation systems is well recognized. With the
emergence of Intelligent Transportation Systems
(ITS), advanced sensor, communications, and comput-
ing technologies have been introduced to facilitate bet-
ter planning, management, and control of public
transportation systems [3]. Known as Advanced Public
Transportation Systems (APTS), these technologies
aim at developing public transportation systems that
are more reliable and responsive to passenger needs,
and hence, public transportation systems that offer
higher level of service [4, 5].
APTS can be quite complex, with dynamic interac-
tions among the various components. For that reason,
simulation-based models that incorporate all the
dynamic aspects, as well uncertainty inherent in the
operations of public transportation systems, have
drawn a lot of interest for evaluation and design of
APTS.
Introduction
APTS are ITS technologies applied to public transit in
order to improve operational efficiency, cost savings,
safety, quality of service, and other transit measures of
performance. Some APTS applications offer potential
for improving service by providing greater leverage to
service providers for managing and controlling bus
transit operations. Other APTS applications provide
benefits in terms of speed, security, and convenience
directly to the customer. APTS have the potential to
significantly change the way transit services are
provided and the way customers use them. Increas-
ingly, popular technologies such as Automated
Vehicle Location (AVL) systems, Automatic Passenger
Counters (APC), and Electronic Fare Payment
already have a wide range of impacts on bus transit
operations.
Advanced Public Transport Systems, Simulation-Based Evaluation 101
The first examples of APTS date back to the late
1960s and early 1970s with the introduction of
Automated Vehicle Monitoring (AVM) systems. The
majority of these vehicle location technologies were
signpost-based systems (with stationary signposts
along bus routes). The signposts were equipped with
electronic transmitters that emit unique identification
codes. When a bus passed the signpost, an in-vehicle
unit received the signpost’s identification code and
record the time and date, the difference between the
current odometer reading and the last (recorded at the
previous signpost), and the vehicle’s identification
code. The bus sent the information to the TOC via
radio or other medium periodically or when prompted
by the transit operations control center (TOC). Early
implementations of APTS were expensive to install,
operate, and maintain. Since then, new locations
and communication technologies (e.g., GPS) have
emerged and led to both improved performance and
reduced costs.
Advanced Public Transportation Systems
(APTS)
Existing and emerging APTS applications of interest to
this report can be grouped into [4, 5]:
● Fleet Management
● Traveler Information
● Electronic Fare Payment
● Transportation Demand Management
Fleet Management strategies focus on improving the
planning, scheduling, and operations of a fleet of
vehicles [6, 7]. The related strategies aim at improved
service reliability, safety, and operating efficiency
(e.g., reduced nonrevenue time, increased productiv-
ity) and faster service disruption recovery. In general,
fleet management includes technologies that collect
and make available vehicle performance data
(e.g., vehicle location), and technologies that use that
data for real-time control or for planning and schedul-
ing. Important technologies used for fleet management
purposes include: Communications Systems,
Geographic Information Systems (GIS), Automated
Vehicle Location Systems (AVL), Automatic Passenger
Counters (APC), Transit Operations Software, and
Traffic Signal Priority [8].
102 Advanced Public Transport Systems, Simulation-Based Evaluation
Traveler Information in transit applications refers to
technologies used to provide travel information to
passengers in order to assist their trip-making decision.
The information provided may range from static route,
schedule, and fare information to real-time vehicle
location and/or estimated arrival time. AVL systems
are the enabler for the provision of real-time informa-
tion. Traveler information may be disseminated
through various means, for example, web-based itiner-
ary planning services, smart phones, in-vehicle units,
etc. Traveler information is generally expected to
improve the quality of transit service by improving
the passenger experience. Traveler information may
give passengers a better sense of control over their
trip-making decisions and/or enable them to take
action to minimize waiting times at stops, plan transfer
connections, and thus reduce overall travel time.
Information may be provided prior to departure
(e.g., by phone, internet), at the terminal or stop, or
in the transit vehicle. Information influences passenger
trip-making decisions, for example, route, stop,
departure time choice, etc. The attractiveness of
transit alternatives is a function of waiting time,
in-vehicle travel time, transfer time to the connecting
trip, number of transfers, onboard comfort
(crowding), etc. Information systems can provide
timely information regarding many of these attri-
butes [9, 10].
Electronic Fare Payment technologies forego cash
and token payment with the aim of reducing the oper-
ating costs of fare collection systems, increasing safety
and security on the vehicle, improving data collection,
and increasing customer convenience [11, 12]. There
are several electronic fare payment technologies,
including magnetic stripe cards, and smart cards. Elec-
tronic fare payment technologies can have significant
impacts on transit operations. The most obvious of the
potential impacts on operations occur at the bus stop,
where passengers board and alight from the vehicle.
Depending on the type of electronic fare payment
technology, considerable gains can be realized in
terms of lower dwell times at stops through increased
transaction speeds. Contact card technologies, where
the card is physically swiped through a card reader, and
contactless card technologies, where the card and card
reader communicate without physical contact but
rather via an electromagnetic signal, affect passenger
boarding rates differently. Boarding rates increase to
a greater extent with contactless card technologies.
Transportation Demand Management is the applica-
tion of technology to alter the usage patterns of the
transportation network, with an emphasis on encour-
aging users to travel by transit. There is a broad range of
technologies designed to better coordinate various
transit services, and provide forums for organized
carpooling and carsharing. In general, approaches to
managing transportation demand aim at increasing the
number of high occupancy vehicles in congested trans-
portation networks, promote travel in off-peak hours,
and implementing transit incentive programs. Exam-
ples include dynamic ridesharing and automated ser-
vice coordination [13].
Simulation-Based Modeling of APTS
Overview
From the previous discussion, it is clear that evolving
information, communications, and sensor technolo-
gies and innovative transit operations control strategies
are becoming critical elements of viable, competitive
public transit systems. APTS, through a variety of data
collection and communication capabilities, support
improved operations planning and real-time transit
operations management. APTS, for example, provide
valuable input to planning applications that may lead
to better transit system design (e.g., improved sched-
uling and route planning). APTS also enable a variety
of real-time operational strategies (e.g., holding and
dispatching) that aim at schedule maintenance and
schedule restoration (after major disturbances). Addi-
tionally, APTS allow the sharing of real-time perfor-
mance information with travelers to influence demand
and improve passenger level of service.
As innovative technological solutions are integrated
with transit services and becoming more common, it is
important to have appropriate tools for evaluating and
refining the corresponding operating strategies prior to
deployment. Because of the complexity of the system,
both from the supply and demand sides, simulation is
an appropriate tool to evaluate, refine, and design
APTS strategies. Simulation allows the representation
of the dynamic aspects of transit operations as well as
capturing of the stochastic behavior of the system.
Hence, simulation enables representation of complex

interactions between the network components: general
traffic, transit vehicles and passengers, and comparison
of various scenarios and alternatives under realistic and
consistent conditions.
In the context of general traffic operations, simula-
tion models have been developed and used for evalua-
tion and design since early on [14]. Traffic simulation
models, based on their level of detail, are characterized
as macroscopic, mesoscopic, or microscopic. Macro-
scopic models use a fluid approximation to capture
traffic dynamics with a continuous representation of
traffic flows [15, 16]. Mesoscopic models represent
individual vehicles (which have travel behavior associ-
ated with them, e.g., route choice) [17–19]. However,
the modeling of traffic dynamics is rather aggregate
through speed-density relationships and queuing the-
ory considerations. Microscopic models represent indi-
vidual vehicles which in addition to travel behavior also
have driving behavior. The network is represented in
detail, and vehicles make acceleration/deceleration and
lane-changing decisions based on traffic conditions,
information, and overall travel plans [20, 21]. Hybrid
simulation models have also been developed that
represent different parts of the network and the asso-
ciated operations at different levels of details. They
combine micrsoscopic simulation at small areas of
interest, while the remaining network is represented
at a mesoscopic level of detail [22].
Transit simulation has not been developed at the
same pace as pure traffic simulation models. Although
simulation models can have many advantages for pub-
lic transportation research and applications, compre-
hensive transit simulation models are still at an early
stage of development. An extensive review of 32 micro-
scopic traffic simulation models report that only 52%
represent public transportation at all, 42% model tran-
sit priority, and only 6% model transit traveler infor-
mation systems. At the same time, the majority of users
they interviewed were interested in large-scale applica-
tions at the urban or regional context, and that these
users ranked modeling of public transportation the
second most important capability in traffic simulation
models [23]. Another review of a representative group
of microscopic simulation models concluded that
a detailed representation of bus operations and the
interactions between buses and other vehicles has
been largely ignored or unrealistically simplified [24].
Advanced Public Transport Systems, Simulation-Based Evaluation 103
In the context of microscopic simulation models,
most of the early advances related to transit systems
have focused on implementation of transit signal pri-
ority [25–28] or operations of bus stops [24, 26, 29].
For example, in [26], the microscopic simulation
model DRACULA was enhanced, mostly from the traf-
fic point of view. The transit modeling was limited, and
it did not include the representation of schedules
(arrival and departure times at each stop). Passenger
arrival rate was drawn from a normal distribution with
a stop-specific average value and fixed variance. The
dwell time was a function of the number of boarding
passengers only. Targeted applications included the
evaluation of roadside vs. bay stops, reserved bus
lanes vs. guide ways (special ways for guided buses),
and bus signal priority.
Recently, several microscopic traffic simulation
models have significantly enhanced or extended their
transit capabilities, indicating a shift away from tradi-
tional, simplistic bus transit modeling and an increased
emphasis on realistic bus transit operations simulation.
The microscopic traffic simulation model MITSIMLab
was systematically extended to represent in detail APTS
operations [30]. Earlier work with DRACULA was
extended by integrating a transportation planning
tool [31, 32]. The simulation includes supply and
demand submodels that interact according to
a learning algorithm, where the demand (in terms of
drivers’ route choice) for each day is affected by the cost
experienced in the preceding day. The simulation is
based on two loops: external (day-to-day) iterations
of the demand loop and internal (within-day) itera-
tions of the supply loop.
Similarly, commercial packages such as
Transmodeler [33] and VISSIM [34] have added exten-
sive transit simulation capabilities. A number of more
specialized systems have also been built on top of
available commercial traffic simulation tools. For
example, as Bus Rapid Transit (BRT) systems gain in
popularity, simulation models of their operations have
also appeared in the literature [35]. SmartBRT,
a microscopic simulation model for evaluating BRT
performance, is built on top of PARAMICS [36].
PARAMICS is a microscopic traffic simulation model
[37]. The level of transit integration is rather low since
the SmartBRT entities, including bus stops and passen-
gers, do not interact directly with corresponding
104 Advanced Public Transport Systems, Simulation-Based Evaluation
PARAMICS entities. This can create considerable com-
plications and inaccuracies in transit modeling and is
indicative of the difficulties in integrating successfully
transit capabilities on top of existing tools. The model
has the capability to represent bus signal priority, var-
ious fare collection mechanisms, and detailed dwell-
time functions.
MISTRANSIT (Microscopic Simulation Transit) is
a PARAMICS API for modeling transit operations
[38, 39]. In general, the movement of the buses is run
by PARAMICS, while MISTRANSIT operates control
strategies and collects transit-related measures of per-
formance. Detailed representation of dwell times at
stops as a function of vehicle design parameters,
and boarding and alighting passengers are included.
Demand is represented through origin–destination
flows between stops resulting to individual passengers
with attributes, such as stop arrival time, and
transfer parameters. MISTRANSIT was used to evalu-
ate holding strategies, operations at bus stops
nearby traffic signals, bus signal priority, and
capacity of busways with skip–stop operations and
interchanges between various public transportation
components.
MILATRAS is another API that adds dynamic tran-
sit assignment capabilities to PARAMICS [40–42]. It
aims at analyzing transit system performance and level
of service under real-time information, as well as eval-
uation of BRT and LRT systems. Given that the under-
lying platform is a microscopic simulation model,
MILATRAS is suited for the analysis of medium-size
networks with medium to low frequencies. The model
adds dynamic representation of individual passengers
and their learning process as they adjust their behavior
in response to changes in transit system’s performance.
The transit path choice problem is modeled as
a Markov decision process with each path decision
(departure time, boarding and alighting decisions)
expressed as a state-action pair. While the overall
model does not seek an equilibrium solution, the
Markov decision process fulfills the conditions for con-
vergence to a unique and optimal solution (in terms of
passenger state-action choices).
Microscopic models are limited when applied to
large-scale applications because of the level of detail
and extensive computational effort they require. In
contrast, mesoscopic simulation models, which
represent individual vehicles but avoid detailed model-
ing of their second-by-second movement, are more
appropriate for system-wide evaluation of transit oper-
ations and APTS, as they are for general
traffic. Research in this area is also limited, especially
with respect to APTS applications with large-scale net-
works [43].
However, activity has recently increased with
research focusing on both representation of APTS in
mesoscopic simulation models and development of
demand models that capture transit-related travel
behavior. In contrast to microscopic models,
mesoscopic models have a greater need for detailed
representation of passenger demand. In this respect,
dynamic transit assignment capabilities are an impor-
tant feature of mesoscopic transit models. In order to
evaluate the long-term effects of bus signal priority
strategies on drivers’ route choice and modal shift,
a simulation-based dynamic traffic assignment model
within DYNASMART was developed [44]. As a
mesoscopic traffic model, DYNASMART represents
individual vehicles and uses speed-density relation-
ships to capture traffic dynamics [18]. Transit vehicles
are generated deterministically, and dwell times are not
calculated explicitly but taken as a fixed delay on the
link capacity. Passengers are generated according to
time-dependent origin–destination zone demands.
Passengers have individual preferences and character-
istics that impact their behavior.
A comprehensive transit modeling framework was
proposed in [45]. The framework uses a multi-agent
approach to transit simulation, focusing on improving
the representation of travelers’ behavior. The public
transportation system consists of three components:
passenger behavior, road traffic dynamics, and specific
bus-network operations, which include the interac-
tions between the buses, passengers, and road
traffic. The network representation is multimodal
including pedestrian, road, and bus networks. The
multi-agent approach considers the roles that each
agent plays: a bus plays two roles simultaneously (vehi-
cle and transportation service), while a traveler plays
two roles alternately (as a pedestrian and as a bus
passenger). While the vehicles’ schedules and travelers’
routes are predetermined, their progression is deter-
mined by the interaction between the three compo-
nents of the system. The simulation implementation
 Advanced Public Transport Systems, Simulation-Based Evaluation 105

uses a hybrid approach, where buses and travelers are General Requirements for Effective Transit
simulated at a microscopic level of detail and Simulation
remaining vehicles are simulated at a macroscopic
In order to simulate APTS applications, it is necessary
level of detail.
to represent bus transit operations at a level of detail
Another comprehensive framework for mesoscopic
that supports the operational characteristics of the
transit simulation is BUSMezzo [46, 47], capable of
technology or system of interest. For example, AVL
modeling transit operations, and evaluating and
systems that monitor bus performance and determine
designing APTS at the network level. The transit sim-
holding and dispatching solutions to schedule devia-
ulation model is fully integrated within the platform of
tion problems cannot be simulated in a model that
Mezzo, a mesoscopic traffic simulation model [48].
does not represent explicitly bus schedules. Figure 1
Mezzo is an event-based simulator and is designed
summarizes various APTS technologies and the impli-
with hybrid capabilities (to interface with microscopic
cations they have on bus transit operations and their
models).

APTS technologies

Geographic Automated Electronic Automatic Comm-

information vehicle fare passenger unications
systems location payment counters systems

Applications

Automated Transportation Transit Traveler

Transit signal Dynamic
service management operations information
priority ridesharing
coordination center software systems

Operations
planning
Improved scheduling
Real-time operations Route planning
Cost reduction
Control & dispatching
Passenger Revenue increases
Information provision
Emergency response travel behavior
Incident response
Pre-trip mode choice
Service restoration
Bus route choice
Intermodal & service
Bus stop choice
coordination
Departure time choice Transit
Transfer connection
demand
protection
Dynamic re-scheduling &
Attract ridership
re-routing
Influence OD patterns

Simulation Input

Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 1

APTS impacts on transit operations and implications for simulation (source: [30])
106 Advanced Public Transport Systems, Simulation-Based Evaluation

representation in a simulation framework. In general, Some transit system variables, such as routes, schedules
APTS and passenger demand, are strongly influenced by
operations planning or passenger trip planning appli-
● Enable a variety of real-time operational strategies
cations and do not generally vary during the course of
(e.g., holding and dispatching) that directly affect
a simulation period. These variables are typically
transit vehicle movements
represented as static inputs to the model. APTS that
● Provide valuable input to planning applications
interact with real-time operations, on the other hand,
that lead to better transit system design (e.g.,
should be represented using models that capture inter-
improved scheduling and route planning)
actions between:
● Allow the sharing of real-time performance informa-
tion with travelers to influence demand and improve ● Buses and other modes
passenger level of service (e.g., route choice) ● Buses and passengers
● Buses and field-installed control devices
In order to simulate APTS applications, the simulation
● Bus operators and the operations control center
model should be based on a detailed representation of
● Passengers and traveler information systems
bus transit operations. Bus operations are subject to
a number of incidental and controlling forces, includ-
ing service schedule design, passenger demand, and
Microscopic Transit Simulation Example:
dispatcher control and intervention. A simulator capa-
ble of capturing the behavior of these interacting com- MITSIMLab
ponents should have the following elements: Microscopic simulation models are very detailed and
appropriate for evaluation of APTS strategies at the
1. Transit Supply representation at the appropriate
operational level, including the detailed representation
level of detail. Transit supply refers broadly to the
of various design aspects and parameters. MITSIMLab
system level components of bus transit operations
is an example of a general microscopic traffic simula-
that are generally under the control of the transit
tion laboratory that models APTS in detail [30].
service provider. The transit supply includes the
transit network, schedules, and fleet assignments.
It also includes the vehicle movements and interac- Transit System Simulation
tions with the remaining traffic, as well as the rep-
MITSIMLab was originally developed for the design
resentation of APTS. APTS representation is crucial
and evaluation of advanced traffic management sys-
and includes surveillance and monitoring systems
tems (ATMS) and advanced traveler information sys-
that generate and distribute real-time information,
tems (ATIS) [49, 50]. It is based on sophisticated
the application of that data to real-time control
behavioral models, including driving behavior and
strategies, etc.
travel behavior (e.g., route choice), that capture the
2. Demand representation. Transit demand refers to
complex decisions that drivers make before departing
the passengers and their behavior with regard to
and en route to their destination. MITSIMLab has
use of the system, including en route and pre-trip
a modular structure, which facilitates the addition of
decisions, such as route choice, as well as behavior
new functionality to the model.
at bus stops and transfer points.
The objectives of the overall bus transit model are to
3. Measures of effectiveness (MOE). MOEs include the
simulate, at a high level of accuracy and detail, the
indicators, levels of service, and other measures that
continuous and complex interactions between vehicles
are used to evaluate the performance of a transit
of various types, surveillance technologies, traffic man-
system or an APTS strategy.
agement strategies and control, and information dis-
The model requirements may be incorporated into semination devices in the network. The modular
a simulation model by providing system variables and framework of MITSIMLab, enhanced with bus transit
parameters as input to the model and/or modeling simulation, is illustrated in Fig. 2 with its two main
internally the effects of APTS on system variables. components: MITSIM and TMS.
 Advanced Public Transport Systems, Simulation-Based Evaluation 107

TMS
- Traffic Management Center & Traffic
Control Strategies
- Transit Operations Control Center &
Real-Time Transit Control Strategies

- Traffic Surveillance Systems MITSIM - Traffic Control & Routing

- Transit Surveillance & Devices
Monitoring Systems - Traffic Flow Simulator - Transit Control and
(AVL, APC) - Transit Operations Simulator Traveler Information Devices

Graphical User Interface (GUI)

Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 2

Bus transit and APTS modeling framework (source: [30])

MITSIM represents the physical road network and
simulates the movements of individual vehicles of all
types through the network. As such, MITSIM
represents the physical portions of the transit network
(e.g., bus stops and bus lanes) and simulates the move-
ments of transit vehicles and the interactions between
transit and non-transit vehicles. MITSIM requires
detailed input describing the transit and traffic
network, bus operations, and travel demand (transit
and car).
Transit Supply Representation The components of
the transit system (transit network, schedule design,
and fleet assignment) are generally considered to be
static information, provided as input. A detailed rep-
resentation of routes and schedules is used such that
transit and traffic operations in the simulated network
are sensitive to the variation in the route and schedule
inputs. The transit system representation is based on
the following inputs: network, vehicle movements,
transit surveillance and monitoring, and transit opera-
tions control center.
Transit Network. The transit network representation
describes the elements of a transit network, including
bus stops and road segments that are used by transit
routes. Routes are defined as sequence of links that
make up the path and a list of bus stops along the
path. Additionally, the transit network representation
supports the simulation of deadheading trips, trips
between depots and terminals, and other vehicle
movements.
● Schedule Definition. The schedule definition input
describes trips on a route and their timing. When
the schedule definition input is read, trips are
created. Trips have a single assigned route and
a series of scheduled arrival times at the stops or
time points along that route. The representation of
the schedule is flexible to allow the simulation of
different types of routes (e.g., high frequency,
express, etc.).
● Run Definition. The run definition input specifies
the series of trips to which a bus may be assigned.
When the run definition file is read, runs are created
as a sequence of trips to which the bus is assigned.
The representation of runs is generic to allow the
simulation of interline, deadheading, and other
nonrevenue trips.
● Fleet Assignment. The bus assignment input defines
the run to which each bus in the fleet is assigned, the
type of bus (e.g., articulated), and the departure
rate, or time, at which buses enter the network to
begin their assignments.
Transit vehicle movements. A set of models capture bus
operator behavior which may be different than other
drivers in the network. For example, bus operator
travel behavior is a function of the route and schedule
input rather than merely an origin and a destination.
108 Advanced Public Transport Systems, Simulation-Based Evaluation
Similarly, bus operator driving behavior includes
maneuvers and decision-making processes that are
unique to bus operators. Bus operators may accelerate,
decelerate, change lanes, and perform other maneuvers
in order to arrive at a bus stop, reenter the traffic stream
from a bus stop, and achieve a variety of other objec-
tives that do not pertain to other drivers. Furthermore,
other drivers may change their behavior in the presence
of buses. Some of the more important elements of
vehicle movement and interactions that are modeled
include:
● Behavior between stops. Operator behavior between
stops refers to driving behavior that controls the
vehicle’s trajectory from one stop to the next. Some
of the tasks of bus operators between stops include
mandatory lane changing to arrive at bus stops,
discretionary lane changing when the bus is far
from the downstream stop, and lane choice when
bus lanes are present. Furthermore, some bus oper-
ator characteristics capture the fact that bus opera-
tors generally know their routes very well and are
familiar with the network.
● Behavior approaching stops. This pertains to both
car drivers and bus drivers. The tendency of private
auto drivers to overtake buses when traveling
behind a bus in a lane that contains stops down-
stream as an important element of general driving
behavior near stops was reported in [24]. It is pos-
tulated that drivers anticipate the bus’ eventual
deceleration to a stop to serve passengers and thus
will change lanes or “squeeze” around the bus in the
same lane in order to pass. This suggests that gen-
eral car-following acceleration models do not cap-
ture the interactions between buses and other
modes correctly. Accordingly, in MITSIMLab,
drivers have a high probability of changing lanes
when traveling behind a bus in a lane that contains
bus stops downstream. That probability increases
when the bus is actively decelerating to pull into
a stop.
● Behavior at Stops. Behavior at stops deals
mostly with dwell time, the time the bus operator
spends stopped with the doors closed, the time the
operator takes to open and close the doors, and
the time it takes to serve passengers while
the doors are open. The most common dwell-
time models assume dwell time to be a linear
function of the number of boarding and alighting
passengers. A wide variety of dwell-time models
have been proposed in the literature, including
models that account for fare collection type, bus
type (e.g., low-floor, number of doors), and other
factors [51–53]. The default dwell-time model in
MITSIMLab is the one suggested in [53], the Tran-
sit Capacity and Quality of Service Manual
(TCQSM).
T ¼ g þ aA þ bB þ e;
where A is the number of alighting passengers, B is
the number of boarding passengers, and e is
a random term. The values of parameters a, b, and
g depend on bus type and design, fare payment
strategy, and passenger characteristics, etc. A
crowding factor may also be added when there are
standing passengers on the bus [52].
● Behavior departing from stops. The critical bus oper-
ator maneuver near wayside bus stops is the depar-
ture from the stop. When a steady traffic stream
occupies the general traffic lane into which the bus
must enter, the operator must locate a gap into
which it may accelerate from a standstill. General
lane-changing models might not capture this
behavior realistically. In most lane-changing
models, the acceptable gap into which the subject
vehicle will merge is a function of its position rela-
tive to the lead and lag vehicles and its speed relative
to the lag vehicle in the target lane. For a bus
departing from a stop, where its speed is zero,
such a lane-changing model may not be adequate
and could lead to unrealistic delays. Specially
adjusted gap acceptance models are used to capture
this behavior.
Traffic surveillance and monitoring. The systems in
place to collect information on the status of the transit
operations are simulated in detail (to the level of func-
tional characteristics, e.g., measurement errors). Sur-
veillance is a generic term that refers to various
detection and sensing technologies located in the trans-
portation network or onboard vehicles. Some APTS
surveillance technologies include automatic vehicle
location (AVL) and automatic passenger counters
(APC). The communication between MITSIM and

TMS is used to mimic the transmission of vehicle
location and passenger load information between vehi-
cles (MITSIM) and the transit operations control cen-
ter or field-installed devices (Fig. 2).
There are three issues related to real-time bus tran-
sit surveillance data that are taken into consideration:
ownership, storage, and distribution. Ownership refers
to the entity (e.g., bus) that has possession of the data.
For example, to emulate APC technologies, buses store
passenger load information as passengers board and
alight, and the user may specify input parameters to
capture the random error in APC counts. Storage refers
to the length of time an entity keeps the data before
purging it, deleting it, or sending it to another entity.
For instance, buses keep track only of their most cur-
rent schedule information, which is updated each time
they arrive at a stop. When a bus arrives at a sensor
upstream of a signal equipped for priority, the sensor
collects the schedule adherence information and relays
it to the signal controller in TMS to determine whether
priority should be granted. Finally, data distribution
deals with when and between which entities the infor-
mation is shared. If bus load and location information
is collected periodically, as through AVL systems, then
this information is reported to TMS according to the
system logic. Through an accurate representation of
data ownership, storage, and distribution, MITSIMLab
is able to represent any underlying AVL or other bus
surveillance system.
Transit operations control (TOC) center activities.
The operations control strategies as well as
decentralized field-deployed strategies are simulated
in TMS. TMS mimics the logic of the strategy under
evaluation, and may use real-time traffic and transit
data from the surveillance system as input to that
logic. As a TOC simulator, TMS handles incoming
transit surveillance data and executes APTS operations
that take place in a TOC.
Device-based control strategies, such as signal pri-
ority, are also simulated. The information regarding
approaching buses received from the surveillance sys-
tem is used to trigger the corresponding strategy which
calculates appropriate actions and sends the signal
timing data to the signal controller in TMS. TOC-
based strategies, on the other hand, can be simulated
by direct control of vehicle movements. For instance,
stop-based control strategies, like holding of a bus if it
Advanced Public Transport Systems, Simulation-Based Evaluation 109
is ahead of schedule, are simulated by placing condi-
tions on bus departures from a stop. APTS control
strategies can vary widely in design and implementa-
tion from agency to agency and from application to
application. The modular structure of the simulation
model facilitates the testing of a large number of
control strategies.
Demand Representation The representation of the
passenger demand is an important element of a simu-
lation model. Passenger demand in MITSIMLab is
modeled at two levels of detail (in increasing order of
data requirements):
● Dwell times are randomly generated independently
for each stop according to probabilistic distribu-
tions. However, such a simplified representation of
demand, although efficient in terms of input
requirements, ignores the impact that dwell times
at one stop may have on the headway and, hence,
dwell times at downstream stops.
● Arrival rates and alighting rates (e.g., as a percent-
age of the bus load) at stops. The model assumes
that passengers arrive according to some probabi-
listic distribution and generates randomly the num-
ber of passengers waiting to board the incoming bus
based on the actual headway from the previous
bus. Likewise, the percentage of the bus load
alighting at a stop determines the number of
alighting passengers. In this representation,
passengers are generated in aggregate and have
no identifying characteristics, origins, or
destinations.
For the last representation of demand, it is important
to model the way passengers arrive at bus stops. Three
categories of transit passengers have been reported in
the literature: those who arrive coincidentally with the
bus and thus have no waiting time, those who are
familiar with the schedule and arrive close to the vehi-
cle arrival time, and those who arrive randomly [54].
The fraction of the passengers in each category depends
on the frequency of service. For example, for high-
frequency routes, it is reasonable to assume that pas-
sengers arrive randomly. The model uses an aggregate
representation of demand with time-dependent arrival
rates. The default arrival model assumes that passenger
arrivals have a Poisson distribution:
110 Advanced Public Transport Systems, Simulation-Based Evaluation
ðlðtÞT Þk e lðtÞT
P ½Bðt; t þ T Þ ¼ k  ¼
k!
where P[B(t, t + T) = k] is the probability that k
passengers arrive in the interval of length T and g(T)
is the average arrival rate during the corresponding
interval.
Measures of Effectiveness An important require-
ment for any transit simulation model is the generation
of appropriate measures of effectiveness (MOEs). In
general, simulation allows the collection of a consider-
able amount of output data, including vehicle trajecto-
ries, sensor readings, and segment and link statistics.
MITSIMLab collects detailed information (e.g., pas-
senger load, speed, mileage) about each bus and at
stops (e.g., dwell times, passengers boarding, alighting
and left behind) at user-specified intervals during
a simulation. From the detailed data, a number of
MOEs can be generated to characterize the perfor-
mance of the system as dictated by the nature of the
application. These MOEs are organized around five
levels: system, route segment, stop, vehicle, and
passenger.
Case Study: Evaluation of Transit Signal Priority
The purpose of this study is to evaluate the effectiveness
of transit signal priority using the MITSIMLab labora-
tory (more information can be found in [30] and [55]).
ITS, with the associated advances in sensor technolo-
gies, has enabled the development of interesting transit
signal priority strategies [56–59]. Traffic signal priority
is designed to reduce transit vehicle delays at signalized
intersections. Reduced delay at intersections may also
reduce overall travel time, aid schedule adherence and
headway maintenance, and increase person throughput
at the intersection.
Transit signal priority strategies are varied: they can
be passive or active, and unconditional or conditional.
Passive priority requires no communication between
vehicle and controller, and involves the development of
a signal timing plan that reduces delay on the transit
vehicle’s approach. Passive priority can be achieved by
allotting more green time to the transit vehicle’s
approach, coordinating signals to improve progression
along a bus corridor. Active signal priority dynamically
adjusts the signal timing when the transit vehicle is
detected. This is accomplished by extending the green
interval in the current phase, terminating the current
phase to start an early green interval for the transit
vehicle’s approach, or by inserting an extra green
phase on the vehicle’s approach. As such, active priority
requires sensors that identify approaching buses.
Depending on the implementation, a communications
system may also be needed to convey the sensor infor-
mation to the traffic signal controller. The signal con-
troller uses the information to adjust the timing of
the signal plan to favor the movement of the
approaching bus.
Active traffic signal priority can be unconditional or
conditional. Unconditional strategies give priority to
every qualified vehicle that approaches the intersection
regardless of the vehicle’s schedule or the impacts on the
conflicting approaches. Conditional priority grants pri-
ority to approaching transit vehicles only if the vehicle
meets some predetermined condition(s). In general, the
condition for priority depends on the vehicle’s current
performance. Based on the available information, it can
take various forms. For example, in the simplest case, the
condition is based on the schedule (i.e., whether
the vehicle is ahead of or behind schedule according to
the timetable). However, as APTS facilitates the moni-
toring of the system in greater detail, other, more
advanced, conditions can also be envisioned (based for
example on the passenger load on the vehicle, headway
from preceding bus, or other measurements or combi-
nation of criteria). Thus, communications and control-
ler technologies that support conditional signal priority
must be able to transmit and manipulate different pieces
of data for evaluating priority eligibility.
In this case study, a number of conditional and
unconditional signal priority strategies in an urban
network in Stockholm, Sweden, are considered. The
study network is located at the western end of
Södermalm, one of a cluster of islands in inner Stock-
holm. The simulated time period of interest is the
7:30–8:30 a.m. morning peak hour. The study network
is dominated by the Hornstull intersection that
joins three major arterials: Liljeholmsbron,
Långholmsgatan, and Hornsgatan (Fig. 3). There are
six signalized intersections and one signalized
pedestrian and bicycle crossing in the network.
Three local bus routes and one BRT route operate
 Advanced Public Transport Systems, Simulation-Based Evaluation 111

Långholmsgatan Bus rapid transit route (Blue Buses)

N
Local routes

Signalized pedestrian/bicycle crossing

Signalized

Bus stop Hornsgatan

Hornstull
intersection

Liljeholmsbron

Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 3

Bus routes in the study network (source: [30])

within the study area. Buses that serve the local bus
routes have 15-min headways during the peak periods,
and the BRT buses operate at 7.5-min headways. BRT
and local buses share all bus stops. The cross-street
traffic in the study network is relatively low compared
to traffic on the three arterials due mostly to street
closures that prevent drivers from using alternate
routes to avoid congestion.
Only BRT buses are equipped to communicate with
sensors to request priority. The objectives of the study
are to evaluate the effects of various transit signal
priority strategies on transit service and network per-
formance. Four priority implementations are com-
pared in the case study:
● No priority (base case)
● Unconditional priority
● Headway-based conditional priority
● Load-based conditional priority
For headway-based conditional priority, the time head-
way between the bus requesting priority and the bus
before it must be equal to or greater than a minimum
headway threshold in order for priority to be granted.
A load-based conditional priority strategy gives prior-
ity to buses only when the number of the passengers
onboard (load) exceeds a threshold. With the use of
automated passenger counters (APC), such a strategy is
feasible under APTS.
The analysis considers two benefits from each strat-
egy: improved person throughput and reduced transit
travel times. However, while it is of interest to assess the
extent to which different signal priority strategies
achieve these objectives, it is also important to evaluate
other impacts, especially to the vehicles entering the
network from the side streets. Hence, the measures of
effectiveness include average vehicle travel time and
total person travel time for all vehicles, BRT buses
alone, and side-street vehicles alone. Impact on travel-
time variability is also considered.
A sensitivity analysis designed to explore the effects
of increased side-street demand is also conducted. Fur-
thermore, given that APC technologies may suffer from
measurement errors, the impact of such errors on the
performance of signal priority was examined.
112 Advanced Public Transport Systems, Simulation-Based Evaluation
Vehicle travel time, total person travel time, and
travel-time variability results for the priority strategies
as well as the base case were collected using the outcome
from several independent simulation replications.
A priori, it is expected that BRT vehicles should experi-
ence greater travel-time savings, while non-priority
movements (i.e., side-street traffic) should experience
higher travel times as the conditions for signal priority
become less restrictive.
Figure 4 summarizes the main results. Average BRT
travel times decrease as the priority conditions become
less restrictive, with the lowest average travel times
occurring under unconditional priority. Since the
demand on the side streets in the study network is
very low, the impact on non-transit, non-priority vehi-
cles entering the network from the side origins is not
substantial.
Interestingly, the results also indicate that condi-
tional priority strategies can achieve similar benefits as
the unconditional priority strategy without granting
priority quite as often. Under conditional priority
about 70% of the buses receive priority. This tends to
indicate that conditional priority strategies also have
less impacts on the side traffic. However, due to the low
demand on the side streets in this case, the change in
average travel time for side-street vehicles for different
priority strategies is too low to draw general
350.0
300.0
Average travel time (s)
250.0
200.0
150.0
100.0
50.0
0.0
All SB BRT
Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 4
Signal priority impact on travel times (source: [55])
conclusions about the trade-off between transit travel-
time savings and side-street travel-time penalties.
Total person travel time is the total time spent by
all commuters in the system. The results show that
under conditional priority, particularly load-based
conditional priority, the difference between
conditional and unconditional priorities is very small
(compared to the difference with respect to
average travel time per vehicle). For example, the total
passenger travel time for the load-based conditional
priority is only 1.5% higher than the unconditional
priority case.
Travel-time variability is an important measure of
performance from the standpoint of traveler level of
service. The results from the simulation study indicate
that travel-time variability for BRT vehicles was also
reduced under signal priority. Furthermore, condi-
tional priority yielded almost similar benefits in that
regard, as unconditional priority.
A major advantage of simulation is that it allows
testing of a large number of different scenarios and
facilitates sensitivity analysis with respect various oper-
ating and design parameters. In this case study, the
transit simulation laboratory is used to evaluate the
impact of increased demand levels as well as the impact
of the accuracy of the APC equipment used to count
boarding and alighting passengers.
No priority
Unconditional
Load-based
Headway-based
NB BRT Side traffic

450.0
400.0
350.0
Average travel time (s)
300.0
250.0
200.0
150.0
100.0
50.0
0.0
All
Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 5
Signal priority impact on travel times with increased side-street demand (source: [55])
Effective signal priority strategies should balance the
benefits for the transit vehicles with the negative impacts
on the side traffic. Since the side-street volumes in the
study network are not substantial to show marked
changes in travel time for non-priority vehicles, three
priority strategies (no priority, unconditional priority,
load-based conditional priority) were also evaluated
assuming a 40% increase in the demand entering the
network from the side arterial Liljeholmsbron. Figure 5
summarizes the results with respect to average travel
times. Load-based conditional priority strategies yield
similar benefits as unconditional priority. More impor-
tantly, the improvements in transit vehicle perfor-
mance under conditional priority can be achieved
with considerably less adverse impact on the side-street
traffic. The results are consistent with those reported in
a field study of schedule-based conditional priority in
Eindhoven, the Netherlands [56]. The study, using data
collected in the field, showed increases in side-street
travel times of 100–300% at some intersections under
unconditional priority, while the schedule-based con-
ditional priority strategy produced significant reduc-
tion in transit travel times with only a minor impact on
non-priority vehicles.
Automatic passenger counters (APC) may have
measurement errors. For example, studies indicate
that infrared APC systems on buses in Oregon have
the tendency to overestimate passenger load, citing
errors as high as 10% [60]. Thus, another scenario is
simulated to determine whether passenger loads
Advanced Public Transport Systems, Simulation-Based Evaluation 113
No priority
Unconditional
Load-based
SB BRT NB BRT Side traffic
measured by APC with error might affect the perfor-
mance of load-based conditional priority. Assuming
that the APC count varies between 5% below and
15% above the true load, the rate of priority (% of
buses receiving priority) increases from 66% (with no
error) to 68% (with error). BRT buses experience travel
times which on the average are slightly lower than with
the same load-based strategy assuming no APC error.
Furthermore, there was no effect on the side-street
travel times. Hence, the strategy appears robust even
under the assumed measurement errors. This analysis
illustrates another useful application of simulation.
A simulation model, at the level of detail described in
this paper, can be used to refine APTS strategies by
explicitly modeling all aspects and characteristics of
the technologies involved (including operating charac-
teristics, such as measurements errors).
Mesoscopic Transit Simulation Example:
BusMEZZO
Mesoscopic transit simulation models are more appro-
priate for the analysis of the operations of transit sys-
tems at the network level. BusMEZZO is an example of
such a model. The transit simulation component is
directly integrated within the platform of Mezzo,
a mesoscopic traffic simulation model [48]. Mezzo is
an event-based simulator, which models vehicles indi-
vidually. Links are divided into two parts: a running
part, which contains vehicles that are not delayed by the
114 Advanced Public Transport Systems, Simulation-Based Evaluation
downstream capacity limit, and a queuing part, which
extends upstream from the end of the link when capac-
ity is exceeded. The boundaries between the running
and queuing parts are dynamic and depend on the
extent of the queue. Vehicles have travel behavior and
make route-choice decisions based on prior experience,
available traffic information, etc.
Simulating Transit Operations
Mesoscopic transit simulation requires similar func-
tionalities as microscopic models. However, the level
of detail in the representation of some components,
especially on the supply side, is lower.
Supply Representation The supply representation
includes the transit network, the movement of vehicles
and interactions with other traffic, the representation
of the surveillance system, and the functionality of the
transit operations control center.
Transit network, schedules, and runs are defined in
a similar fashion as in the microscopic model presented
in the previous section. However, vehicle movements
and interactions are represented in simplified way.
Buses enter the network according to the provided
schedule, interact with the car traffic, and move
according to prevailing conditions on the link (based
on speed-density relationships). They stop at stops
according to a dwell time that is calculated using
models similar to the ones in microscopic simulation.
Buses depart from stops according to the calculated
dwell time or other considerations, such as timetable
constraints for the specific route, or real-time control
strategies that are implemented to improve the reliabil-
ity of the route. Upon arrival at the end terminal, a bus
either enters the next trip in its schedule or exits the
network. With this process, delays encountered by
buses on one trip influence its ability to perform the
next trip on time. Hence, the model captures the prop-
agation of delays along the route and is capable of
producing the occurrence of bunching.
An important aspect in modeling transit supply is
the representation of APTS both in terms of the sensor
technology in place to collect data and the operating
strategies. As in micrsoscopic simulation, the model
keeps track of data routinely collected by the APTS in
place (AVL, APC, etc.). This data is processed in
combination with schedule information preceding
and following vehicles’ data in order to generate infor-
mation about the bus performance (e.g., schedule
adherence and actual headways). This information, in
turn, is used to trigger real-time strategies in order to
improve operations. The model is capable of simulat-
ing a wide range of transit operations control strategies
so that they can be evaluated and refined.
Strategies that are used to improve transit service
reliability are of particular importance. Service reliabil-
ity is often connected to headway variability. Headway
variability impacts a number of important measures of
level of service for the passengers, such as waiting
times, uncertainty with respect to arrival times at
destinations, and passenger loads. Reliability also
impacts utilization of resources. In this context, the
model can be used to both assess the causes of
unreliable service and develop strategies for effectively
dealing with it. Service reliability has been the subject
of many studies, including the comprehensive work
reported in [61, 62]. A number of studies also
exist that investigate the manifestation of service
reliability using data collected through APTS technolo-
gies (e.g., [63–65]).
Real-time operations control has always been used
as means to improve service reliability. Figure 6 illus-
trates a taxonomy of the various types of real-time
control strategies [66]. A strategy may be applied at
the station, route/intersection, or at the fleet level.
Station-based strategies are either holding (vehicles
held at stops for service restoration) or skipping strat-
egies (where stations may be temporarily skipped dur-
ing a particular trip). Skipping stops may take several
forms, such as deadheading or expressing.
Holding is a particularly popular strategy because
of the relative ease of implementation. A number of
theoretical models have appeared that cover various
aspects of the problem and apply them to both real
and idealized scenarios. Originally, most of the work
focused on simple network configurations with limit-
ing assumptions and simplifications to make the prob-
lem tractable. Early research attempted to provide an
understanding of the impact of holding strategies on
passenger waiting times, with classic papers such as
[67–69] leading the way in this direction. Recent
work in the area includes [66, 70–73] and focuses on
both evaluation of impacts and development of

Station
Holding Skipping Speed control
Deadheading Expressing Short turning
Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 6
Real-time operations control strategies (source: [66])
methods to design optimal strategies. Real-time control
of buses for schedule coordination at terminals is stud-
ied in [74].
Because of the difficulties in obtaining analytical
solutions to the problem, many studies relay on simu-
lation to evaluate the developed strategies. For exam-
ple, a simulation model of a single direction bus route
was used to test several strategies, including holding,
skipping, loading (but not unloading) passengers, and
signal priority in [74]. Similarly, in [75], simulation
was used to evaluate different real-time control strate-
gies. BusMEZZO provides a platform where existing
strategies can be evaluated, and new approaches
developed.
Demand Representation Passenger demand is
modeled at various levels of detail (more information
can be found in [76]):
● Arrival and alighting rates as discussed in section
“Transit System Simulation.”
● Passenger origin–destination flows and related
travel choices and travel behavior. The model can
simulate individual passengers with attributes
(e.g., age, income), origins, and destinations
(assuming that he required information is avail-
able). A demand simulator component has been
developed to deal with the diverse requirements of
transit demand representation. Passengers are
Advanced Public Transport Systems, Simulation-Based Evaluation 115
Real-time control
Route/intersection Fleet
Traffic signal Vehicle addition
preemption
explicitly represented. Their behavior is captured
dynamically as they make station and route (bus
line) decisions based on experience, real-time tran-
sit information, and other service attributes. As
mentioned in the previous section, in the context
of mesoscopic models, transit assignment function-
ally is an important requirement. A dynamic transit
assignment component is implemented based on
two submodels: choice-set generation and path
choice process. The generation of the choice set
aims at reproducing the set of alternatives consid-
ered by passengers. The choice-set generation
model is based on the transit network attributes –
transit routes, timetables, transferring stops and
links, as well as access and egress links. However,
in order to narrow the set of all available paths to
a reasonable choice set, a number of filtering rules
are used which apply logical and behavioral
constraints.
Once the set of alternative paths has been generated,
the path choice model determines how passengers
choose between their alternatives. The path choice
model takes into account traveler preferences and deci-
sion rules. A dynamic representation of the transit
system facilitates the explicit modeling of capacity con-
straints. Moreover, passenger travel experiences and
real-time information are important inputs to the
path choice decisions. The dynamic assignment
116 Advanced Public Transport Systems, Simulation-Based Evaluation
model considers the path choice as a decision process
that consists of a sequence of dependent decisions. In
general, there are two distinctive types of decisions:
(a) stop-related decisions (origin stop choice, transfer
stop choice, and destination stop choice) and
(b) route-related decisions (choice among alternative
transit routes). A discrete choice framework is used to
model the various decisions [77].
Measures of Effectiveness Outputs from the simula-
tion include stop level statistics, such as early and late
arrivals, dwell times, numbers of boarding and alight-
ing, bus loads, and travel times between stops. Aggre-
gations at the level of the trip, the vehicle, or the line,
such as schedule adherence, headway and passenger
wait time distributions, load profiles, and other level
of service measures, are also computed.
Case Study: Evaluation of Real-Time Holding
Strategies
The mesoscopic model described in the previous sec-
tion is used to evaluate holding strategies to improve
service reliability of a busy urban route in Tel-Aviv,
Israel (for more information see [78]). This high-
demand route connects a dense suburb to the CBD.
The scheduled headway during the peak period is
8 min, and the average running time is 49 min inbound
and 41 min outbound.
The study compares the base case of no holding to
two scenarios with holding strategies. In designing
holding strategies, the main aspects of interest are the
amount of time the bus is held, and the location of the
stops (time points) where holding may be applied. In
general, holding strategies are either schedule-based or
headway-based. Schedule-based strategies apply con-
trol in order to ensure that buses run according to the
timetable. Headway-based strategies aim at improving
service by regulating the headways at stops.
Headway-based control is more appropriate for
lines that have very frequent service (since in those
cases, according to [54], passengers arrive at the stop
randomly and do not plan their arrival according to
schedule). Schedule-based strategies are more appro-
priate for low frequency routes with passengers coor-
dinating their arrival with the schedule. Another
important difference between the strategies is the
amount of information required for their implemen-
tation. Schedule-based strategies require minimum
information (a static timetable), while headway-based
strategies require real-time information (that may vary
depending on the exact nature of the strategy). In
general, it is expected that more information should
help the system perform better. A comparison between
holding control strategies that rely only on local infor-
mation to APTS-based strategies that incorporate more
information was conducted in [75] in the context of
coordination at transfer points. Not surprisingly, the
study concluded that the best holding control strategy
was the one that used all available information, includ-
ing predicted arrival times of connecting buses, num-
ber of transferring passengers, and expected number of
boarding passengers at downstream stops.
Schedule-based strategies hold buses that arrive
early at a time point until a predefined time from
their scheduled departure time is reached. Depending
on the available information, more or less sophisticated
headway-based control strategies can be implemented.
For example, if only information regarding the preced-
ing bus is available, the strategy takes the form of
holding a bus until a certain threshold with respect to
the scheduled headway is reached. If information is
available with respect to both the preceding and fol-
lowing buses, then the holding strategy may take into
account the headway with both buses and hold the
subject vehicle until headways between them are equal
(mean-headway strategy) or until a maximum holding
time is reached [79, 80].
Regarding the location of the time points, strategies
are typically applied at a subset of stops along a route
[81]. Several studies (e.g., [82–85]) suggest that time-
point stops should be located at the beginning of
a sequence of high-demand stops. In contrast, [72]
found that there are no significant benefits from defin-
ing time points along bus routes, except the origin
terminal.
Following the recommendations in the literature,
the design of the holding strategies that were tested had
the following characteristics. The time points were
located at three stops upstream of a sequence of high-
demand stops. The holding time at each time was based
on two strategies: a strategy that aims at equalizing the
headway (when possible) between the subject bus and
the preceding and following buses (mean headway),

14000
12000
10000
Time [s]
8000
Time [s]
6000
4000
2000
0
0 2000 4000 6000 8000 10000 12000 14000
Distance [m]
Bus 1 Bus 2 Bus 3
a b
Advanced Public Transport Systems, Simulation-Based Evaluation. Figure 7
Time-space diagram of buses with no control (a) and mean-headway control (b) (source: [78])
and a schedule-based holding. Therefore, the two strat-
egies represent two rather opposite cases of informa-
tion availability: minimum information (timetable)
and relative high (arrivals of preceding and following
buses at the time point).
Figure 7 illustrates the impact of control using
a time-space diagram of the trajectories of three
buses, each performing four chained trips. Sloped seg-
ments represent running time between stops, and ver-
tical segments indicate dwell times and holding times at
time points. Recovery times between trips at both ter-
minals are also shown. Figure 7a illustrates the base
case of no control, while Fig. 7b corresponds to the case
of mean-headway control. Buses are experiencing
bunching when there is no control, while the holding
strategy is effective in preventing this by creating even
headways at time points. For example, on the third trip,
bus 2 is getting increasingly bunched with bus 1 under
(a), a situation that is largely remedied when holding is
applied.
At the system level, several measures of perfor-
mance were calculated for each scenario, including
average passenger waiting time, travel-time, and on-
time performance. The schedule-based strategy
resulted in very small benefits with respect to average
waiting time for the passengers. In general, it had
practically the same performance as the base case of
no control with respect to all measures. The reason for
this is that, with the timetable used for the route, only
a small number of trips were ahead of schedule, and
Advanced Public Transport Systems, Simulation-Based Evaluation 117
14000
12000
10000
8000
6000
4000
2000
0
0 2000 4000 6000 8000 10000 12000 14000
Distance [m]
Bus 1 Bus 2 Bus 3
therefore the schedule-based strategy was used very
infrequently. The mean-headway strategy reduced
headway variability, as measured by the coefficient of
variation, by as much as 30%. The average waiting time
deceased by 5%. However, the mean-headway strategy
results in service regularity gains at the expense of
higher travel times and deterioration of on-time
performance.
Future Directions
Transportation systems are highly complex, and
researchers, planners, and engineers often use simula-
tion to study their performance, analyze and design
possible system improvements, and develop decision
support systems to aid with their management. Over
the past several years (motivated in part by the intro-
duction of Intelligent Transportation Systems and
APTS), transit simulation models have seen a lot of
advances. As technology and policy innovations are
constantly introduced, the required functionalities of
transit simulation models also need to be expanded to
satisfy the diverse needs of operations, planning, and
policy evaluation. Information technology, for exam-
ple, has created and will continue to create many
opportunities for transit systems. These opportunities
are twofold: (a) advances in sensor technologies allow
operators to collect a lot of data and information about
the system and (b) advances in communications tech-
nologies facilitate the dissemination of useful
118 Advanced Public Transport Systems, Simulation-Based Evaluation
information to passengers, even personalized informa-
tion on demand. In order to provide the functionality
needed to support the new application foci, develop-
ments in the context of both microscopic models and
mesoscopic models are needed.
On the microscopic side, seamless integration of traf-
fic and transit driving behavior (at the individual vehicle
level) needs to be further developed to reliably model
APTS at the operational level. On the mesoscopic side,
transit simulation needs to be closely integrated with
developments in the area of simulation-based dynamic
traffic assignment. Integrated, multimodal models that
represent demand and supply at a consistent, disaggre-
gate level, and capture the demand–supply interactions
are important for the evaluation of future strategies to
deal with sustainable transportation systems. Multi-
modal models need to be sensitive to changes in the
system introduced by policy and technological innova-
tions. In this respect, better integration of demand and
supply representations at a consistent level of detail is
important and should be organized in modules with the
following characteristics: (a) a demand module with
disaggregate, dynamic, activity-based representation of
the choices of individual travelers; (b) a supply module,
with detailed, disaggregate, dynamic modeling of multi-
modal network performance through vehicle-based traf-
fic simulation; and (c) a demand–supply interactions
module with integration between the demand and sup-
ply modules through appropriate algorithms.
Integrated, simulation-based, dynamic, and disag-
gregate models have the potential to provide increased
realism in capturing both travel behavior and network
performance and so provide considerable flexibility in
representing a wide range of systems and predicting
their impacts. A number of challenges, both theoretical
and practical, need to be resolved before such models
become operational. Important challenges include the
development of appropriate demand–supply algo-
rithms that are able to predict the resulting behavior
of the system, data and approaches for development
and estimation of the constituent models, and the
systematic validation and calibration of the overall
model, and approaches to deal with the computational
requirements of such large-scale models (more infor-
mation about such a disaggregate, dynamic multi-
modal modeling of transportation systems can be
found in [86]).
Bibliography
Primary Literature
1. Shannon RE (1975) Systems simulation: the Art and science.
Prentice-Hall, Englewood Cliffs
2. United Nations Population Fund (2010) State of world
population 2007. http://www.unfpa.org/swp/2007/english/
introduction.html
3. Miles J, Chen K (2004) ITS handbook, 2nd edn. PIARC, Paris
4. Casey RF, Labell LN, Moniz L et al (2000) Advanced public
transportation systems – the state of the art: update 2000.
U.S. Department of Transportation, Federal Transit Adminis-
tration (FTA), Washington, DC
5. Hwang M, Kemp J, Lerner-Lam E, Neuerburg N, Okunieff P
(2006) Advanced public transportation systems: state of the
art update 2006, FTA-NJ-26-7062-06.1. Federal Transit Admin-
istration (FTA), Washington, DC
6. Furth P, Hemily B, Muller T, Strahman J (2006) Using archived AVL-
APC data to improve transit performance and management, TCRP
report 113. Transportation Research Board, Washington, DC
7. Furth P (2000) Data analysis for bus planning and
monitoring, TCRP synthesis 34. Transportation Research
Board, Washington, DC
8. Smith H, Hemily B, Ivanovic M (2006) Transit signal priority
(TSP): a planning and implementation handbook. ITS America
and US Department of Transport, Washington, DC
9. Hickman MD, Wilson NHM (1995) Passenger travel time and
path choice implications of real-time transit information.
Transp Res C 3(4):211–226
10. Nuzzolo A, Russo F, Crisalli U (2001) A doubly dynamic
schedule-based assignment model for transit networks.
Transp Sci 35(3):268–285
11. Wilson NHM, Zhao J, Rahbee A (2004) The potential impact of
automated data collection systems on urban public transport
planning. In: Wilson NHM, Nuzzolo A (eds) Schedule-based
dynamic transit modeling: theory and applications. Academic,
Dordrecht, pp 74–99
12. Vandebona U, Richardson AJ (1985) The effects of fare-
collection strategies on transit level of service. Transp Res
Rec 1036:79–87
13. Transportation Demand Management User Service (2005)
National ITS architecture. US Department of Transportation,
Washington, DC. www.iteris.com/itsarch/html/user/usr18.htm
14. Barceló J (ed) (2010) Fundamentals of traffic simulation,
vol 145, International series in operations research & manage-
ment science. Springer, New York
15. Daganzo C (1994) The cell transmission model: a dynamic
representation of highway traffic consistent with the hydro-
dynamic theory. Transp Res B 28(4):269–287
16. Messmer A, Papageorgiou M (1990) METANET: a macroscopic
simulation program for motorway networks. Traffic Eng
Control 31(9):466–470
17. Ben-Akiva M, Bierlaire M, Burton D, Koutsopoulos HN,
Mishalani R (2001) Network state estimation and prediction

for real-time transportation management applications. Netw
Spat Econ 1(3/4):293–318
18. Ben-Akiva M, Koutsopoulos HN, Antoniou C, Balakrishna R
(2010) Traffic simulation with DynaMIT (chap 10). In:
Barcelo J (ed) Fundamentals of traffic simulation. Springer,
New York
19. Mahmassani HS (2001) Dynamic network traffic assignment
and simulation methodology for advanced systems manage-
ment applications. Netw Spat Econ 1(3/4):267–292
20. Yang Q, Koutsopoulos HN (1996) A microscopic traffic simula-
tor for evaluation of dynamic traffic management systems.
Transp Res 4C:113–129
21. Ben-Akiva M, Koutsopoulos HN, Toledo T, Yang Q,
Choundhury C, Antoniou C, Balakrishna R (2010) Traffic simu-
lation with MITSIMLab (chap 6). In: Barcelo J (ed) Fundamen-
tals of traffic simulation. Springer, New York
22. Burghout W, Koutsopoulos HN, Andreasson I (2005) Hybrid
mesoscopic-microscopic traffic simulation. Transp Res Rec
1934:218–225
23. Algers S, Bernauer E, Boero M, Breheret L, Dougherty M, Fox K,
Gabard JF (1997) Review of microsimulation models. In:
SMARTEST project deliverable D3, EU, Institute for Transport
Studies, University of Leeds, Leeds. Available at: http://www.
its.leeds.ac.uk/projects/smartest/deliv3f.html
24. Silva PCM (2001) Modelling interactions between bus opera-
tions and traffic flow. Doctoral dissertation, University College
London
25. Khasnabis S, Karanti RR, Rudraraju RK (1996) NETSIM-based
approach to evaluation of bus preemption strategies. Transp
Res Rec 1554:80–89
26. Liu R, Clark S, Montgomery F, Watling D (1999) Microscopic
modelling of traffic management measures for guided bus
operation. In: Proceedings of eighth world conference on
transport research, Antwerp
27. Chang J, Collura J, Dion F, Rakha H (2003) Evaluation of service
reliability impacts of traffic signal priority strategies for bus
transit. Transp Res Rec 1841:23–31
28. Lee J, Shalaby A, Greenough J, Bowie M, Hung S (2005)
Advanced transit signal priority control using on-line
microsimulation-based transit prediction model. In: Preprints
of the transportation research board 84th annual meeting,
Washington, DC
29. Ding Y, Chien S, Zayas A (2001) Simulating bus operations with
enhanced corridor simulator. Transp Res Rec 1731:104–111
30. Morgan DJ (2002) A microscopic simulation laboratory for
advanced public transport system evaluation. Master thesis,
Massachusetts Institute of Technology
31. Liu CL, Pai TW (2006) Methods for path and service planning
under route constraints. Int J Comput Appl Technol 25 (1):40–49
32. Liu R, Van Vliet D, Watling D (2006) Microsimulation models
incorporating both demand and supply dynamics. Transp Res A
40:125–150, Washington, DC
33. http://www.caliper.com/
34. http://www.ptvag.com/index.php?id=345
35. BRT. http://www.cfte.org/trends/brt.asp
Advanced Public Transport Systems, Simulation-Based Evaluation 119
36. Werf JV (2004) SmartBRT: a tool for simulating, visualizing and
evaluating bus rapid transit systems. California PATH Research
Report, Institute of Transportation Studies, University of
California, Berkeley
37. http://www.paramics.com/
38. Cortes CE, Fernandez R, Burgos V (2007) Modeling passengers,
buses and stops in traffic microsimulators. The MISTRANSIT
approach on the PARAMICS platform. In: Proceedings of the
86th transportation research board annual meeting,
Washington, DC
39. Fernandez R, Cortes CE, Burgos V (2010) Microscopic simula-
tion of transit operations: policy studies with the MISTRANSIT
application programming interface. Transp Plann Technol
33(2):157–176
40. Wahba M, Shalaby A (2006) A general multi-agent modelling
framework for the transit assignment problem – a learning-
based approach. In: Innovative internet community systems.
Springer, Berlin, pp 276–295
41. Wahba M, Shalaby A (2006) MILATRAS: a microsimulation plat-
form for testing transit-ITS policies and technologies. In:
Proceedings of the IEEE ITSC, Canada
42. Wang J, Wahba M, Miller EJ (2010) A comparison of an agent-
based transit assignment procedure (MILATRAS) with conven-
tional approaches. In: Proceedings of the 89th transportation
research board annual meeting, Washington, DC
43. Boxill SA, Yu L (2000) An evaluation of traffic simulation
models for supporting ITS development. Center for Transpor-
tation Training and Research, Texas Southern University
44. Abdelghany KF, Abdelghany AF, Mahmassani HS, Abdelfatah AS
(2006) Modeling bus priority using intermodal dynamic network
assignment-simulation methodology. J Public Transp 9(5):1–22
45. Meignan D, Simonin O, Koukam A (2007) Simulation and
evaluation of urban bus-networks using a multiagent
approach. Simul Model Pract Theory 15:659–671
46. Cats O, Toledo T, Burghout W, Koutsopoulos HN (2010)
Mesoscopic modeling of bus public transportation. In: Pro-
ceedings and CD of the 89th transportation research board
(TRB) annual meeting, Washington, DC
47. Toledo T, Cats O, Burghout W, Koutsopoulos HN (2010)
Mesoscopic simulation for transit operations. Transp Res C
18(6):896–908. doi:10.1016/j.trc.2010.02.008
48. Burghout W (2004) Hybrid microscopic-mesoscopic traffic
simulation. Doctoral dissertation, Royal Institute of Technol-
ogy, Stockholm
49. Yang Q (1997) A simulation laboratory for evaluation of
dynamic traffic management systems. Ph.D. thesis, Depart-
ment of Civil and Environmental Engineering, Massachusetts
Institute of Technology
50. Yang Q, Koutsopoulos HN, Ben-Akiva M (2000) A simulation
laboratory for evaluating dynamic traffic management sys-
tems. Transp Res Rec 1710:122–130
51. Levinson HS (1983) Analyzing transit travel time performance.
Transp Res Rec 915:1–6
52. Lin T, Wilson NHM (1992) Dwell time relationships for light rail
systems. Transp Res Rec 1361:287–295
120 Advanced Public Transport Systems, Simulation-Based Evaluation
53. TCRP Report 100 (2003) Transit capacity and quality of service
manual (TCQSM), 2nd edn. Transportation Research Board,
Washington, DC
54. Jolliffe JK, Hutchinson TP (1975) A behavioral explanation of
the association between Bus and passenger arrivals at a bus
stop. Transp Sci 9(3):248–282
55. Morgan D, Koutsopoulos HN, Ben-Akiva M (2003) Simulation-
based evaluation of advanced public transportation systems.
In: Wilson NHM, Nuzzolo A (eds) Schedule-based dynamic
traffic modeling: theory and applications. Series in operations
research/computer science interfaces, vol 28, chap 6.
Academic, Boston, pp 95–102
56. Furth PG, Muller THJ (2000) Conditional bus priority at signal-
ized intersections: better service quality with less traffic
disruption. Transp Res Rec 1731:23–30
57. Chang J, Collura J, Dion F, Rakha H (2003) Evaluation of service
reliability impacts of traffic signal priority strategies for bus
transit. Transp Res Rec 1841:23–31
58. Kim W, Rilett LR (2005) An improved transit signal priority
system for networks with nearside bus stops. Transp Res Rec
1925:205–214
59. Lee J, Shalaby A, Greenough J, Bowie M, Hung S (2005)
Advanced transit signal priority control using on-line
microsimulation based transit prediction model. Transp Res
Rec 1925:185–194
60. Kimpel T, Strathman J (Apr 2002) Automatic passenger counter
evaluation: implications for national transit database reporting.
http://www.upa.pdx.edu/CUS/PUBS/docs/PR124.pdf
61. Abkowitz M, Slavin H, Waksman R, Englisher L, Wilson N (1978)
Transit service reliability. Report UMTA-MA-06-0049-78-1.
USDOT Transportation Systems Center, Cambridge
62. Abkowitz M, Tozzi J (1987) Research contributions to manag-
ing transit service reliability. Adv Transp J 21:47–65
63. Strathman JG, Dueker KJ, Kimpel T, Gerhart R, Turner K,
Taylpr P, Callas S, Griffin D, Hopper J (1999) Automated bus
dispatching, operations control and service reliability. Transp
Res Rec 1666:28–36, Washington, DC
64. Strathman JG, Kimpel TJ, Dueker KJ, Gerhart RL, Turner K,
Griffin D, Callas S (2001) Bus transit operations control: review
and experience involving Tri-Met’s automated bus
dispatching system. In: Transportation Research Board, 80th
Annual Meeting, Washington, DC
65. Strathman JG, Kimpel TJ, Dueker KJ, Gerhart RL, Callas S
(2002) Evaluation of transit operations: data applications of
Tri-Met’s automated bus dispatching system. Transportation
29:321–345
66. Eberlein XJ (1995) Real time strategies in transit operations:
models and analysis. Ph.D. dissertation, Department of Civil
Engineering, MIT
67. Osuna EE, Newell GF (1972) Control strategies for an idealized
public transportation system. Transp Sci 6:52–72
68. Barnett A (1974) On controlling randomness in transit opera-
tions. Transp Sci 8(2):102–116
69. Barnett A (1978) Control strategies for transport systems with
nonlinear waiting costs. Transp Sci 12(2):119–136
70. O’Dell S, Wilson N (1997) Optimal real-time control strategies
for rail transit operations during disruption. In: Wilson N (ed)
Computer aided transit scheduling, vol 471, Lecture note in
economics and mathematical systems. Springer, Berlin,
pp 299–323
71. Hickman M (2001) An analytic stochastic model for the transit
vehicle holding problem. Transp Sci 35(3):215–237
72. Eberlein XJ, Wilson NHM, Bernstein D (2001) The holding
problem with real-time information available. Transp Sci
35(1):1–18
73. Shen S, Wilson NHM (2001) Optimal integrated real-time dis-
ruption control model for rail transit systems. In: Voss S,
Daduna J (eds) Computer-aided scheduling of public trans-
port, vol 505, Lecture notes in economics and mathematical
systems. Springer, Berlin, pp 335–364
74. Koffman D (1978) A simulation study of alternative real-
time Bus headway control strategies. Transp Res Rec
663:41–46
75. Dessouky M, Hall R, Zhang L, Singh A (2003) Real-time control
of buses for schedule coordination at a terminal. Transp Res A
37:145–164
76. Cats O, Koutsopoulos HN, Burghout W, Toledo T (2011) Effect
of real-time transit information on dynamic passenger
path choice. Transportation Research Record. J Transp Res B
(in press)
77. Ben-Akiva M, Lerman S (1985) Discrete choice analysis.
Theory and application to travel demand. MIT Press,
Cambridge, MA
78. Cats O, Burghout W, Toledo T, Koutsopoulos HN (2010) Evalu-
ation of real-time holding strategies for improved bus service
reliability. Submitted for presentation and publication at the
2010 13th international IEEE conference on intelligent trans-
portation systems (ITSC 10), Madeira Island
79. Koutsopoulos HN, Wang Z (2007) Simulation of urban rail
operations. Transp Res Rec 2006:84–91
80. Daganzo CF (2009) A headway-based approach to eliminate
bus bunching: systematic analysis and comparisons. Transp
Res B 43:913–921
81. Abkowitz M, Lepofsky M (1990) Implementing headway-
based reliability control on transit routes. J Transp Eng
116(1):49–63
82. Liu G, Wirasinghe SC (2001) A simulation model of reliable
schedule design for a fixed transit route. J Adv Transp
35(2):145–174
83. Abkowitz M, Engelstein I (1984) Methods for maintaining
transit service regularity. Transp Res Rec 961:1–8
84. Turnquist MA, Blume SW (1980) Evaluating potential effective-
ness of headway control strategies for transit systems. Transp
Res Rec 746:25–29, Washington, DC
85. Wirasinghe SC, Liu G (1995) Determination of the number and
locations of time points in transit schedule design – case of
a single run. Ann Oper Res 60:161–191
86. Ben-Akiva M, Bottom J, Gao S, Koutsopoulos HN, Wen Y (2007)
Towards disaaggregate dynamic travel forecasting models.
Tsinghua Sci Technol 12(2):115–130

Books and Reviews
Ceder A (2007) Public transit planning and operation: theory,
modeling and practice. Elsevier, Burlington
Chung E, Dumont A (eds) (2009) Transport simulation: beyond
traditional approaches. EPFL Press, Lausanne
Sussman J (2005) Perspectives on intelligent transportation
systems (ITS). Springer, New York
Advanced Public Transport Systems, Simulation-Based Evaluation 121
Vuchik V (2005) Urban transit: operations, planning, and
economics. Wiley, Hoboken
Vuchik V (2007) Urban transit systems and technology. Wiley,
Hoboken
Wilson NHM, Nuzzolo A (eds) (2004) Schedule-based dynamic
transit modeling: theory and applications. Academic,
Dordrecht
122 Batteries, Battery Management, and Battery Charging Technology
Batteries, Battery Management, and
Battery Charging Technology
ROBERT S. BALOG1, ALI DAVOUDI2
1
Department of Electrical and Computer Engineering,
Texas A & M University, College Station, TX, USA
2
University of Texas, Arlington, TX, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Battery Chemistries
Lead Acid
Battery Charging Circuits
Battery Management
Future Direction
Bibliography
Glossary
Alternating current (AC) Usually in the form of a sine
wave, the bidirectional current flow continually
reverses. Typically, AC will have a zero mean
(average) and a nonzero RMS value. In the USA,
most “grid” electricity is generated and distributed
in the form of 60 Hz sinusoidal waveform. In other
areas of the world, 50 Hz is commonly used.
Amp-hour (Ah) A unit of electric charge, the amount
of current delivered for 1 hour.
Battery A group of one or more cells electrically
connected in series and/or parallel combinations
to achieve higher voltage or current than what is
capable from a single cell.
Battery management system (BMS) An electronic
device or system that monitors and controls
a rechargeable battery. Parameters measured may
include cell temperature, voltage, and current. From
this data, the BMS can compute the state of charge of
the battery and estimate the state of health, remaining
cycle lives, or remaining service life. Typically, a BMS
system will include the ability to communicate with
a host system and battery charger. The BMS may also
contain sensors and circuitry for protection such as
over-current, over-temperature, or over-voltage.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Battery pack A group of batteries or individual cells
electrically connected in series and/or parallel com-
binations along with the required electrical inter-
connections, mechanical packaging, thermal
management, and sensing circuitry. Because it is
a self-contained assembly, a battery pack can be
easily swapped in and out of the application.
Battery string Series-connected batteries used to pro-
duce a higher voltage. The same current passes
through all the cells, but each cell voltage can vary.
Charge balancing becomes a significant issue for
a long string of 50 or more cells.
C-rate The rate at which a battery can deliver or accept
current, stated in terms of the rated capacity of the
cell in amp-hours. This may also be referred to as
the hour rate, such as the 1-h rate.
Cell An individual electrochemical device that con-
verts between chemical energy and electrical energy.
The cell construction, open circuit voltage, energy,
and power density, all depend on the chemistry of
the cell (see section “Battery Chemistries”).
Coulombic efficiency Ratio of charge delivered by
a rechargeable battery during discharge cycle to
the charge stored during charge cycle.
Depth of discharge (DoD) An alternate method to
indicate the state of charge of the battery; it is the
reciprocal of SoC.
Direct current (DC) Unidirectional current that con-
tinually flows only in one direction. Sources such as
batteries, fuel cells, and solar cells produce electric-
ity in the form of direct current.
Memory effect An effect observed particularly in
nickel cadmium batteries in which the cells gradu-
ally lose capacity when subject to repeated partial
discharges followed by complete recharges cycles.
This electrochemical effect is different than the loss
of capacity due to aging and use. In some cases,
a series of full discharge/recharge cycles can restore
some or all of the lost capacity due to the memory
effect but not due to cycle life or aging.
Proportional-integral (PI) controller A type of linear
feedback control in which an error signal is calcu-
lated as the difference between a measured signal
and the desired reference set point. The controller
attempts to adjust the operation of the process
using a weighted combination of the present error
# 2012, DOI 10.1007/978-1-4419-0851-3

(proportional term) and accumulated past error
(integral term).
Primary battery One-time-use batteries not capable
of being recharged. They are discarded or preferably
recycled once the stored energy has been depleted.
These types of batteries will not be considered in
this article.
Root mean square (RMS) A scalar quantity that is
computed from the square root of the sum of the
squares of a dataset and is commonly used to rep-
resent a voltage or current over one cyclical period.
Average power is the product of the RMS current
and RMS voltage.
Secondary battery Commonly referred to as
rechargeable batteries, they can be repeatedly
recharged electrically by passing current through
them in the opposite direction to that of the dis-
charge current. Secondary batteries are of signifi-
cant interest for their ability to store and supply
energy and are the focus of this article.
Self-discharge rate The rate at which a battery dis-
charges, or looses stored energy, due to internal cell
reactions.
SLI batteries Starting, lighting, and ignition (SLI)
batteries are used in automobiles. These batteries
are usually characterized by a high discharge rate.
State of charge (SoC) Defined as the capacity left in
a battery expressed as a percentage of some refer-
ence. SoC of a battery is usually expressed as
a percentage of the current battery capacity when
it is fully charged.
State of health (SoH) A metric that reflects the general
condition of a battery and its ability to deliver
the specified performance compared with a fresh
battery. It takes into account factors such as
charge acceptance, internal resistance, voltage, and
self-discharge. It is an estimate rather than a
measurement.
Watt-hour (Wh) A unit of energy, the amount of
power delivered for 1 h. It is equivalent to 3,600
Joules of energy.
Definition of the Subject
Batteries, both primary and rechargeable, are impor-
tant energy storage devices ubiquitous in our daily,
modern lives. Whether in our handheld portable
Batteries, Battery Management, and Battery Charging Technology 123
electronics, conventional or hybrid/electric cars, or in
the electrical “grid,” battery technology will continue to
evolve as technology improvements increase storage
capacity and lifetime and reduce cost.
Introduction
Batteries are perhaps most synonymous with portable
power such as flashlights and personal electronics like
cell phones, music players, and laptops to name a few
modern examples. We have become trained to monitor
the battery “bars” to determine how much more we can
talk, e-mail, or surf the “net.” Batteries are far more
prevalent, however, and they tend to be underappreci-
ated, such as the ones in our automobiles which per-
form dutifully until suddenly, usually after years of
neglect, they fail to start our car. An electrical power
system is the ultimate in just-in-time manufacturing in
that electricity generation and consumption must be
perfectly balanced at all times. This imposes difficult
operating constraints on the system when variability on
both the load side and source side exists, such as with
renewable energy sources like wind and solar. There-
fore, energy storage is important to ensure that electric-
ity can be generated and energy stored generated when
the source is available, and the load can continue to be
supplied when the variable, renewable energy source is
insufficient. While there are many technologies for
energy storage, batteries play an important role to
improve the electric power system asset utilization,
energy availability, reliability, and performance. Several
large-scale, high-energy battery technologies hold
promise of providing economical energy storage for
a wide range of these power system and energy manage-
ment applications.
This chapter will discuss issues related to batteries,
battery charging, and battery management. The first
section will provide an overview of the different types
of battery chemistries. The focus in this chapter is on
rechargeable batteries which can accept, store, and then
deliver energy at a future point in time. Subsequent
sections will discuss circuits to charge and manage the
batteries.
A detailed description of internal construction of the
individual cell is beyond the scope of this chapter, but it
is sufficient to consider that a battery is typically com-
prised of electrodes, electrolyte, and some form of case
124 Batteries, Battery Management, and Battery Charging Technology
CATHODE ANODE
(+) (-)
Electrolyte
Parallel interleaved Battery
Plates Case
Batteries, Battery Management, and Battery Charging
Technology. Figure 1
Conceptual illustration of a lead-acid battery with parallel,
interleaved plate that comprises the anode and cathode.
The electrolyte can be a liquid or a gel, and the enclosure
can be vented or sealed
or enclosure as shown in Fig. 1. Various materials are
commonly used, each with advantages and disadvan-
tages. The voltage, current rating, and storage capacity
are all functions of the material and construction.
Battery Failure
Some of the most common causes for battery failure
are short circuiting of battery terminals, improper
charge control while charging secondary batteries,
improper polarity connections while charging
a battery, and a long period of use. Internal failure
modes include thermal runaway which is a major prob-
lem in lead-acid batteries, loss of electrolyte by gassing,
mechanical damage, and warping of plates due to an
excessively high rate of discharge.
Methods of Charging
Float charge is most commonly used for non-sealed
lead-acid batteries, which can tolerate sitting at an
elevated voltage for prolonged periods of time. The
original meaning of the term stems from the telecom
industry’s jar batteries in which the liquid electrolyte was
exposed to the atmosphere and the batteries “sat on
float” at a continuous elevated voltage to be sure
they were fully charged when called upon to back up
the telephone equipment. In this mode of operation,
the charging current depends on the state of charge
of the cells, and the current gradually declines to a
minimum value as the cells become fully charged.
Float charging could result in continuous out gassing
which reduces the level of water in the electrolyte.
Hence, float charging of lead-acid batteries often
requires “watering” of the batteries to maintain the
level of electrolyte.
Trickle charge is used to maintain a battery at
a full state of charge by supplying a small current
to offset the self-discharge characteristics of the bat-
tery. Some battery chemistries, like the lead-acid
varieties, have relatively high self-discharge rates
and can lose their ability to hold a charge when
stored discharged for a long period of time due to
sulfation, the crystallizing of the lead sulfate.
A trickle charger, sometimes called a “battery tender”
is often used in automobile, boats, motorcycles, RV and
other seasonal equipment where the battery can sit for
months without use.
Bulk charge involves restoring the majority of the
energy storage capability of a battery. Hence, a bulk
charge may be followed by an equalizing charge.
Equalization charge is the final stage of charging for
a series-connected battery string in which the SoC of
individual cells are forced to the same value. For lead-
acid batteries, this can be as using a higher charging
voltage. The cells with higher SoC begin to outgas while
the weaker cells “catch up.” Other battery chemistries
cannot tolerate overvoltage charging. Charge balancing
techniques will be described in detail in the next
section.
Rapid charge involves quickly recharging a battery
at a high C-rate. Depending on the battery chemis-
try, this mode of charging may only replace
a portion of the full capacity. It is usually imperative
to monitor the cell temperature rise during rapid
charging to avoid excessive outgassing or thermal
runaway.
Regardless of the mode of charging, the charging
types include constant-current charging and constant-
current–constant-voltage charging (constant-current
charge until current attains a flat profile, then
constant-voltage charging takes over).

Battery Chemistries
A wide variety of battery electrochemistries are available
today. While there is a clear evolution of technological
progress, each technology has advantages and disadvan-
tages that make them more or less suited for different
applications. This section presents an overview of com-
monly available technologies and explores their general
characteristics, advantages and disadvantages, and charg-
ing considerations.
Lead Acid
General Characteristics
Lead-acid batteries are by far the most common battery
type and represent approximately 40–45% of the total
global battery sales. Lead-acid batteries are available in
large quantities and in a variety of sizes and designs.
They are manufactured in sizes from smaller than 1 Ah
to several thousand Ah. Design variations to electrode
plates enable them to perform as batteries with high
discharge rate and low-energy (automobile starting
applications) and also for deep-cycle applications
with moderate discharge rates.
The world’s largest battery-based energy storage sys-
tem is a 40-MWh battery located in Chino, California.
It uses individual industrial-size lead-acid cells in series
and parallel connection to make a 10-MW system capa-
ble of delivering energy into the utility grid at 2,000V
and 8,000A for 4h.
Advantages and Disadvantages
Advantages include:
● One of the main reasons for the popularity of lead-
acid batteries is that they can be manufactured on a
local basis worldwide with high rates of production.
● Traditionally, they have been used as starting-
lighting-ignition (SLI) batteries in automobiles,
though now they are being slowly phased out by
high-performing nickel cadmium and nickel metal
hydride batteries.
● They have a high cell open-circuit voltage of 2.0 V
which is one of the highest for aqueous electrolyte
systems.
● SoC estimation is very easy in the case of lead-acid
batteries. The specific density of electrolyte bears
a direct relation to the SoC of the battery.
Batteries, Battery Management, and Battery Charging Technology 125
● It provides good charge retention for intermittent
charge applications.
● Cell components can be easily recycled.
● They are very cheap compared to other secondary
battery designs.
● They are also available in maintenance-free designs.
Disadvantages include:
● Lead-acid batteries require frequent maintenance.
These batteries lose water while in operation, and
the water level needs to be replenished.
● As they employ lead in their construction, they are
usually heavy and not conducive for portability.
● Lead-acid batteries have a relatively low cycle life
(50–500 cycles). However, up to 2,000 cycles have
been achieved with cell design improvements,
heavily derating the discharge capacity and using
active charge equalization techniques (see section
“Cell Charge Equalization”).
● One of the major drawbacks is the thermal runaway
problem. It can occur due to improper design of
battery or charger.
Application Examples Lead-acid batteries are com-
monly installed in uninterruptible power supply (UPS)
systems, in renewable and distributed power systems.
Traditionally, they were used as SLI batteries in auto-
mobiles. Other applications include telephone systems,
power tools, communication devices, emergency light-
ing systems, and as the power source for mining and
material-handling equipment.
Charging Characteristics Lead-acid batteries do not
exhibit memory effect. However, it is very vital to have
the correct charge and float voltages to achieve a long
life. Charging up to the rated voltage level is a must; any
deviation from this voltage level leads to electrode cor-
rosion or negative plate sulfation in the long run. Lead-
acid batteries used for deep-cycle applications usually
have a short life of about 300 discharge/charge cycles.
Lead-acid batteries are typically charged in three
stages, which are constant-current bulk charge, equali-
zation final charge, and float charge. The constant-
current charge provides bulk of the charge and takes
up about half of the required charge time. The equal-
izing charge continues charging at a lower charge
current and ensures all cells are fully charged and the
126 Batteries, Battery Management, and Battery Charging Technology
state-of-charges are similar. This may be done through
active techniques or simply overcharging some cells to
ensure the undercharged ones receive a full charge.
The float charge compensates for the self-discharge
loss when a battery sits unused for a period of time.
Valve-Regulated Lead Acid (VRLA)
General Characteristics The valve-regulated lead-
acid (VRLA) battery is a completely sealed cell, which
prevents electrolyte loss due to outgassing, and hence
alleviates the need to water the cells. The VRLA battery
incorporates the gas recombination principle which
allows the oxygen generated at normal overcharge rates
to be reduced at the negative plate, eliminating oxygen
outgassing. If the rate of outgassing exceeds the rate of
recombination, the partial pressure of hydrogen builds up
and can cause the vent to open, which destroys the cell.
Advantages and Disadvantages
Advantages include:
● VRLA batteries do not require “watering” as is the
case with other types of lead-acid batteries since the
water level is maintained due to a regenerative cell
reaction. Thus, VRLA batteries do not require the
frequent maintenance for battery operation, which is
an advantage over the conventional lead-acid battery.
● Battery cells can be packaged more tightly because
of sealed construction and immobilized electrolyte,
hence the footprint and weight of the battery are
reduced.
● SoC is measured by measuring the voltage (The
open-circuit voltage can therefore be used to approx-
imate the state of charge). The measurement of the
open-circuit voltage to determine the SoC is based on
the relationship between the electromotive force and
the concentration of the sulfuric acid in the battery.
● One of the major advantages with VRLA batteries is
that they do not experience memory effect which is
common with nickel cadmium batteries.
Disadvantages include:
● These batteries have a shorter life span than other
types of lead-acid batteries. In UPS applications,
a life span of 5–10 years is typical since their life
reduces due to harsh operating conditions.
● They should not be stored in discharged condition.
● These batteries are sensitive to higher temperature
environment. As with most batteries, the capacity is
dependent on the discharge rate and temperature;
the capacity decreasing with decreasing tempera-
ture and increasing discharge rate.
● Electrolytes act as heat sinks in batteries.
However, VRLA batteries are “acid starved” as
they use much less electrolyte, hence they do
not have an effective heat-sink mechanism
which makes them more prone to thermal
runaway condition than compared to flooded
lead-acid batteries.
● They have low cell resistance; hence immense short-
circuit current is available.
Application Examples Because the VRLA batteries are
sealed, they are ideal for many consumer applications.
They are found in many uninterruptable power supply
systems for computers, and as backup power for other
electronics’ light security systems, and emergency light-
ing. Because high discharge rates are possible, they are
also advantageous for engine starting.
Charging Characteristics Constant-voltage charg-
ing is the most efficient and fastest method of
charging the VRLA battery. To carry out fast
charging of VRLA battery, the charger must be
capable of charging the battery at the 2C-rate.
Fast charging is usually possible with a full charge
achieved within 4 hours, and some batteries that
accommodate even higher C-rates can charge so that
they are returned to full capacity within 1 h. When float
charging is used, the charger maintains each cell at
2.3–2.4V in order to ensure a complete charge and
maximum battery life.
Nickel Cadmium (Ni-Cd)
General Characteristics Nickel-cadmium battery is
a very reliable, sturdy, long-life battery. Little mainte-
nance is needed on nickel-cadmium batteries. These
batteries have two major variants based on the intended
application. Cells with thin sintered plates have low
internal resistance and have a high power to volume
ratio, while cells with thick electrode plates have a high
energy storage capacity. Nickel-cadmium batteries are
manufactured as unsealed open/semi-open batteries

in which electrolyte/gas can escape through a vent
(vented Ni-Cd batteries) and as fully sealed batteries
which do not require topping up of the electrolyte with
water. Nickel-cadmium batteries exhibit voltage sup-
pression, also called as memory effect. The memory
effect causes the battery to deliver only the capacity,
which was used during the repeated charge/discharge
cycles before. Because of this, the whole capacity of Ni-
Cd batteries should be used for each discharge cycle to
avoid a decrease of the maximum capacity. “Memory
effect” makes this battery less suited for applications that
don’t allow a complete discharge.
Advantages and Disadvantages
Advantages include:
● They have a longer life than compared to lead-acid
batteries.
● Extended operating temperature range compared to
lead-acid batteries. This makes them more flexible in
heavy duty applications in regard to temperature range.
● Extremely rugged batteries.
● Good charge retention. They can be stored for
long periods without significant deterioration of
SoC.
● These batteries require less maintenance.
Disadvantages include:
● Cadmium is an environmentally hazardous sub-
stance. Hence, disposal of Ni-Cd batteries is
a major issue.
● They exhibit strong voltage depression (memory
effect) phenomenon. This is a major drawback of
Ni-Cd cells.
Application Examples Because of their favorable
electrical properties, excellent reliability, low main-
tenance, rugged design, and long life, nickel-
cadmium batteries are used in a large variety of
applications. Unsealed Ni-Cd with higher capacity
ranges are used in high-energy applications such as
traction and large emergency power installations.
Industrial nickel-cadmium batteries are used in
train lighting and air conditioning for rail cars,
emergency and standby systems such as emergency
brakes and lighting in mass-transit and subway cars,
diesel-engine cranking in locomotives and com-
muter cars, railroad signaling, communication
Batteries, Battery Management, and Battery Charging Technology 127
along tracks, as well as standby power in rail sta-
tions and traffic control systems. The nickel-
cadmium battery is also used in power-generating
stations and power distribution. Sealed Ni-Cd bat-
teries are usually of a low energy storage type
(30 Ah). These are extensively used in consumer
electronics, domestic markets, and low-energy
applications.
Charging Characteristics Constant-voltage charging
is the preferred method for unsealed nickel-cadmium
batteries.
Nickel-Metal Hydride (Ni-MH)
General Characteristics Nickel-metal hydride
(Ni-MH) cells share many of the same characteristics
as nickel-cadmium cells and reduce some of the disad-
vantages. A major difference between nickel-metal
hydride and nickel-cadmium cells is that cell reaction
of nickel-metal hydride cells is exothermic, and hence,
internal temperature of Ni-MH cell rises when it is in
operation. On the other hand, nickel-cadmium cells
have an endothermic cell reaction.
Advantages and Disadvantages
Advantages include:
● These batteries do not pose environmental hazards
like Ni-Cd batteries.
● They possess higher specific energy and energy den-
sity than nickel-cadmium cells.
● Bipolar battery pack designs are popular as the
energy storage in hybrid vehicles.
Disadvantages include:
● Overcharging leads to serious consequences for the
battery; it leads to battery heating which in turn
releases hydrogen gas and increases fire hazard. As
a result, Ni-MH batteries require complex charging
circuitry to ensure overcharging does not take
place.
● Complex circuitry for charging implies Ni-MH cells
cannot be charged on conventional battery char-
gers. This necessity contributes to an increased
cost for Ni-MH battery.
● Limited service life of about 200–300 cycles if
repeatedly discharged at high load currents.
128 Batteries, Battery Management, and Battery Charging Technology
● Higher self-discharge rate than Ni-Cd. Hence,
the shelf-life of a fully-charged Ni-MH battery is
much shorter. The shelf life is improved by
adding chemical additives, which lowers the energy
density.
● Memory effect is present but to a much lesser extent
than Ni-Cd battery.
Application Examples The nickel-metal hydride bat-
tery used to be incorporated in computers, cellular
phones, and other consumer electronic applications.
In recent years, lithium-ion battery has replaced
nickel-metal hydride batteries as it possesses a superior
specific energy and energy density.
Charging Characteristics Repeated high load cur-
rent discharging reduces the service life of nickel-
metal hydride batteries to about 200–300 cycles. Ideal
discharging of these batteries usually occurs at load
currents of 0.2–0.5C. Ni-MH battery suffers from
a high self-discharge rate. Self-discharge rates are
reduced by adding chemical additives to the cell. How-
ever, this results in a trade-off, as chemical additives
that slow down the self-discharge rate decrease the
energy density of the cell. For a particular make of
battery, it has been observed that it loses about 32%
of charge if stored for about 80 h.
Lithium Ion (Li-Ion)
General Characteristics Lithium-ion cells are often
specified for use in high-performance applications
where small size and high performance are of para-
mount importance. They have higher specific energy
and energy density than nickel-metal hydride cells.
High energy density of Li-ion cell also makes it less
stable. These cells pose a potential fire hazard if not
managed properly.
Advantages and Disadvantages
Advantages include:
● Li-ion cells have a long cycle life (3,000 cycles at
80% depth of discharge). Hence, they are suited for
applications which require a large number of
charge-discharge cycles.
● They can operate over a wide temperature range.
● Low self-discharge rate leads to a long shelf life for
these batteries.
● Rapid charge capability endears these batteries with
consumer electronics applications.
● High energy efficiency. These cells can be up to 94%
efficient in terms of energy over a cycle.
● No “memory effect.”
● These batteries do not pose an environmental
hazard as they are not made up of toxic
materials.
● High energy density (300–400 kWh/m3, 130 kWh/
ton).
● One of the salient qualities of Li-ion cells is that
they have a very high coulombic efficiency as com-
pared to lead-acid cells. A new Li-ion cell exhibits
up to 94% coulombic efficiency as against a lead-
acid battery which has only 80% coulombic
efficiency.
Disadvantages include:
● Li-ion cells have a moderately high initial cost.
● High energy density of Li-ion cells also gives rise to
instability and fire hazard issues. This behavior of
Li-ion cells makes it mandatory to have a complex
protective circuitry. The battery management sys-
tems for Li-ion cell must be able to prevent
overcharging of these cells. Overcharging in Li-ion
cells leads to thermal runaway and subsequently
a fire hazard.
● Venting and thermal runaway may take place when
cells are crushed.
● Li-ion batteries composed of series strings of cells
require charge equalization modules. Li-ion cells do
not support low cost, easy to implement passive
cell-balancing schemes.
Application Examples Li-ion batteries are employed
in high-performance applications where size and
energy are important. They are the preferred choice
for consumer electronics such as laptop computers,
cellular telephones, and personal media player where
long battery life is required. The suitability for use in
military, aerospace, and automobile applications is also
being explored.
Charging Characteristics Li-ion cells are fabricated
in discharged state and thus must be charged before

use. The cells are typically charged using either
a constant current (CC) or a constant current-
constant voltage (CCCV) with a taper charge
regime. Exceeding the maximum voltage is
a potential safety hazard and can result in irrevers-
ible damage to the battery. At the same time, charg-
ing to a lower voltage reduces the capacity of the
battery. The construction of these batteries tends to
make them susceptible to internal short circuits due
to impurities. With the demand for higher energy
density, an internal short can lead to catastrophic
consequences including the risk of fire. Hence,
internal temperature rise of the battery pack needs
to be monitored as well as cell voltage. These con-
siderations make it crucial to incorporate special
control circuitry for lithium-ion batteries for man-
agement of charge and discharge. Although Li-ion
cells have a flat voltage profile, it has been
established in literature that cell open-circuit volt-
age provides a good estimate of the state of charge
(SoC) ranging from 20% to 100%.
Sodium Sulfur (Na-S)
General Characteristics A sodium-sulfur battery is
a type of molten metal battery constructed from
sodium (Na) and sulfur (S). This battery has
a high energy density, high efficiency of charge/
discharge (89–92%), and long cycle life, and is fab-
ricated from inexpensive materials. Sodium sulfur
batteries usually have operating temperatures of 300–
350
C. They are most prominently found in power grid
operations, electric trains, and wind generation
operations.
Advantages and Disadvantages
Advantages include:
● These batteries are constructed from inexpensive
materials.
● Per unit volume, sodium sulfur batteries deliver
three to five times more energy than lead-acid bat-
teries. They have a high specific energy in the range
of 160 kW/h.
● Designed to last 15 years through 2,500 full charge/
discharge cycles.
● It has up to 89% cell DC efficiency with no self-
discharge.
Batteries, Battery Management, and Battery Charging Technology 129
● Insensitive to ambient temperatures which make it
suitable to be housed outdoors.
● Extremely fast charge cycling; it can transit from full
charge to full discharge in milliseconds.
● Sodium sulfur battery facilitates remote operation
and monitoring along with the added benefit of
minimum maintenance.
Disadvantages include:
● These batteries require an elaborate thermal man-
agement system to ensure energy efficiency.
● Cells need to be hermetically sealed when operated
in corrosive outdoor environments.
Application Examples Sodium-sulfur batteries have
operating temperatures of 300–350
C and extensively
employ sodium polysulfides which are highly corro-
sive. These considerations make Na-S batteries suitable
for large-scale nonmobile applications such as grid
energy storage. Sodium-sulfur batteries have many
applications in power systems. They are used to pro-
vide additional power during voltage sags. They pro-
vide prompt spinning reserves for grid frequency
control and reactive power support. Sodium-sulfur
batteries also perform admirably in stabilization of
wind energy fluctuations by providing the necessary
additional power. Na-S batteries are a possible energy
storage technology to support renewable energy gener-
ation, specifically wind farms and solar generation
plants. They are also a better option as against pumped
hydroelectric storage schemes. In Europe, motivation is
toward implementing Na-S batteries for automotive
traction applications.
Lithium-Iron Monosulfide
General Characteristics Practical energy densities of
100 Wh/kg are achieved by secondary lithium-iron
monosulfide cells. Lithium-iron sulfide batteries
require a thermal management system to maintain
optimum operating temperatures. Precise voltage
monitoring is needed from the charger. This battery
system requires a thermal management system to
maintain the rated operating temperature for the bat-
tery and a specialized charger in which careful voltage
control is maintained. This battery technology was
under active research in the 1970s for electric vehicles.
It is yet to be commercialized.
130 Batteries, Battery Management, and Battery Charging Technology
Nickel Zinc (Alkaline Rechargeable System)
General Characteristics The nickel-zinc (zinc/
nickel-oxide) battery is an alkaline rechargeable sys-
tem. Currently the nickel-zinc system is capable of
delivering about 50–60 Wh/kg and 80–120 Wh/L
depending on the specific design. Nickel-zinc batteries
are suited for high discharge rate applications such as a
camera flash.
Advantages and Disadvantages
Advantages include:
● Good capacity retention has made it a competitor
of silver-zinc and silver-cadmium systems. Fully
charged nickel-zinc batteries lose about 20% of
their original capacity within a month when left
on open-circuit at 25
C. Nickel-zinc batteries can
be stored for up to 3 years.
● It has a wide range of temperature tolerance
(39
C to +81
C operating temperature).
● One of the major plus points is that it has a flat
discharge voltage profile for a significant portion of
the discharge.
● Ni-Zn battery possesses a fast recharge capability.
● It has a sealed maintenance-free design.
Disadvantages include:
● The main stumbling block for the success of Ni-Zn
battery technology is that Ni-Zn battery possesses
a poor cycle life (up to 150 cycles).
● It has a relatively low volumetric energy density.
Application Examples Nickel-zinc provides a very
low-cost option for a long-cycle-life alkaline-
rechargeable system. The nickel-zinc system is suited
for mobile applications such as electric bicycles, electric
scooters, electric and hybrid vehicles, or other deep-
cycle applications. Deep-cycle applications for nickel-
zinc batteries include trolling motors, electric bicycles
and scooters, wheelchairs, golf carts, electric
lawnmowers, electric vehicles, and similar uses.
Nickel-zinc batteries have the required discharge-rate
capability, high cycle life at deep depths-of-discharge,
lightweight rugged construction, and are capable of
being stored over the winter months without any main-
tenance charging.
Charging Characteristics The battery should be
protected from excessive overcharge and over
discharge. Extreme abuse may reduce battery perfor-
mance and cycle life or cause a potential safety hazard.
Metal Air
General Characteristics The most popular metal-air
battery is the zinc-air electrode battery. Zinc-air sec-
ondary batteries are being developed which can store
three times as much as lithium-ion batteries but only
cost half as much. The intended applications include cell
phones, portable consumer electronic items, and electric
vehicles. Lithium-air, calcium-air, and magnesium-air
batteries have also been studied, but cost and problems
such as anode polarization or instability, parasitic cor-
rosion, nonuniform dissolution, safety, and practical
handling have so far inhibited the development of com-
mercial products.
Advantages and Disadvantages
Advantages include:
● Zinc-air battery has high energy density. Suitable
for electric vehicle technologies.
● Zinc-air cells intake air with CO2 removed. Earlier,
this was seen as a disadvantage. However, deploy-
ment of this cell on electric vehicles with a CO2 filter
will lead to drastic reduction in CO2 emissions.
Hence, this cell is being proposed as a likely candi-
date for automotive applications.
● Zinc-air cells have a flat discharge voltage profile.
● These cells have a long shelf life (dry storage).
● Zinc-air battery has a safe battery chemistry which
reduces fire hazards greatly.
Disadvantages include:
● Main difficulty in the development of recharge-
able zinc-air battery is the deactivation in perfor-
mance of air electrode after a few charge/discharge
cycles.
● Internal short circuit is a major problem in
Zn-Air secondary cells. Internal short circuit
usually occurs after several charge/discharge
cycles due to zinc dendrite growth through the
separator.
● Depending on the discharge rate, the energy capac-
ity of the cell drops drastically.

● Drying out of electrolyte limits shelf life once
opened to air.
● Limited operating temperature range.
Application Examples Future applications for
rechargeable zinc-air batteries include cell phones and
electric vehicles. The high energy density, low-cost
constituent materials, and relatively safe cell chemistry
are some of the positive application advantages of
this battery type. However, more efforts are required
to increase the number of charge/discharge cycles
of this battery before it becomes practical for commer-
cial use.
Zinc-Bromine
General Characteristics The zinc-bromine battery is
a high-energy-density battery. There are no toxic mate-
rials in this battery system. Zinc-bromine battery is
constructed from low-cost materials; the battery cells
are made up of plastic. It provides a viable energy
storage option for a variety of generation sources and
is ideally suited for applications requiring 2–10 h of
energy storage. The energy density of this advanced
battery makes it attractive for applications where tra-
ditional energy storage batteries, such as lead-acid bat-
teries, are unacceptable due to size and weight
restrictions. The zinc-bromine battery is a flow battery.
Auxiliary systems are required for circulation and tem-
perature control of the electrolyte. The zinc-bromine
battery has a good shelf life. These batteries are more
prominent in the high energy range: 10–500 kWh.
These battery systems are custom built for specific
applications.
Sodium-Beta
General Characteristics Sodium-beta batteries are
high-temperature operating systems and hence have
high energy density compared to ambient temperature
systems.
Advantages and Disadvantages
Advantages include:
● Sodium-beta batteries are characterized by high
energy density, lack of required maintenance, and
performance independent of ambient temperature.
Batteries, Battery Management, and Battery Charging Technology 131
● They have a potential for low cost (relative to other
advanced batteries) and are cheap as raw materials.
Disadvantages include:
● Thermal management is critical. Requires thermally
insulated battery enclosure.
● These batteries operate in the region of 160–500
C.
Hence, these batteries must not be affected by ambi-
ent temperatures.
Application Examples Sodium-beta battery technol-
ogies have been developed for use in relatively large-scale
energy storage applications (10–1,000 kWh). Sodium-
beta batteries can be used at generation facilities for load
leveling, spinning reserve, and area regulation. They are
also used for line stability and voltage-regulation pur-
poses at distribution facilities. This technology is also
suitable for peak demand reduction and power quality
maintenance at a customer site. Sodium-beta batteries
have also been investigated for their suitability in trac-
tion applications since, for a given physical envelope,
sodium-beta batteries can provide significantly greater
energy (range) at a reduced weight while meeting the
vehicle power requirements compared with conven-
tional battery technologies.
Battery Charging Circuits
Battery Charging Techniques
Proper battery charging techniques can significantly
improve battery performance and life cycles. Thus,
several factors such as fast charging, good quality of
charging current, and avoiding under and over charg-
ing are considered. In particular, over charging can
damage battery’s physical components while under
charging can reduce its energy capacity. Therefore, an
appropriate control technique for charging process
should be adopted. Different battery charging tech-
niques are presented in this section. Operating princi-
ple of the battery charge controller is discussed for
each technique, and the block diagram of the controller
is depicted. Depending on the selected charging tech-
nique, output voltage or current of the converter are
fed back to the charge controller circuit. Each charging
technique has a charging characteristic curve which is
defined as the voltage-current trajectory of the battery
during the charging period (vbat versus ibat ).
132 Batteries, Battery Management, and Battery Charging Technology

Constant-Current Charging (CC) Constant-current
charging is the most conventional battery charging
technique. In the charging characteristic curve shown
in Fig. 2a, the battery current is kept constant while
battery voltage increases during charging until it
reaches the maximum allowed value (i.e., rated voltage
of the battery). Figure 2b shows the voltage and current
profiles of the battery during the charging period.
When the battery voltage has reached its final value,
the charge controller circuit disconnects the battery
from the converter to avoid overcharging and conse-
quent battery failure. The voltage profile of the battery
during the charging and discharging intervals is not
necessarily linear, as shown in Fig. 2b, and depends on
the type of the battery.
Figure 3 shows the associated control block dia-
gram. The output current is compared with the refer-
ence current to produce the error current. The error
current is fed to a proportional-integral (PI) controller
to eliminate steady state error and generate the
command signals for switching of the charger power
electronic circuits.
The main advantage of this method is its simplic-
ity. The converter circuit can be realized by an AC-DC
or DC-DC converter based on the input type. More-
over, the output capacitor can be dropped as the
output voltage regulation is not required, and regu-
lating merely the output current is sufficient. Elimi-
nation of the output capacitor simplifies the converter
circuit to a first-order current filter which can be
analyzed simply and implemented inexpensively.
Fast charging can be realized by large current
pumping; however, this has the potential to degrade
the internal chemical reactions in the battery. Alter-
natively, the low charging current prolongs the charg-
ing time. Therefore, proper choice of the charging
current is crucial for enhancing the charging perfor-
mance and increasing the lifetime of the battery
charging system.
Constant-Voltage Charging (CV) Constant-voltage
charging is another conventional battery charging tech-
nique. The charging characteristic curve of this tech-
nique is illustrated in Fig. 4a. In this method, the

vbat vbat, ibat

Vmax
vbat
Vmax

Iconst ibat

Iconst ibat t
a b

Batteries, Battery Management, and Battery Charging Technology. Figure 2

(a) Charging characteristic curve of the constant-current technique; (b) Current and voltage profiles of the battery during
the charging process

+ PWM
Signal Gate
+_ Controller Driver
_
Iref /Vref (e.g. PI) Circuit
Semiconductor
Switch

Ibat /Vbat Sawtooth wave

Batteries, Battery Management, and Battery Charging Technology. Figure 3

Block diagram of the controller circuit useful for constant-current or voltage charging
 Batteries, Battery Management, and Battery Charging Technology 133

vbat vbat, ibat

vbat
Vmax

Vconst
I0

ibat

ibat
Istart t
a b

Batteries, Battery Management, and Battery Charging Technology. Figure 4

(a) Charging characteristic curve of the constant-voltage technique; (b) Current and voltage profiles of the battery during
the charging process

vbat vbat, ibat

Stage1 Stage2 Stage3
Vconst
vbat
Vconst
Vth Vreduced
Vreduced
Iconst
ibat
Imaint.
ibat
Imaintenance Iconst t
a b

Batteries, Battery Management, and Battery Charging Technology. Figure 5

(a) Charging characteristic curve of the CC-CV technique; (b) Current and voltage profiles of the battery during the
charging

battery voltage remains constant while the battery cur-
rent is decreasing and eventually becomes very low. The
current and voltage profile during charging is shown in
Fig. 4b. This initial high current (Io ) temporarily
increases the battery temperature and can degrade the
battery, depending on the chemistry. The battery
charge controller disconnects the battery from the
charger circuit when the current reaches a specific
amount (Imin ), as shown in Fig. 4b. The control block
diagram is shown in Fig. 3, where the sensed and
reference variable is the battery and reference voltage,
respectively.
Similar to the CC methodology, the CV charging
strategy is straightforward and easy to implement. As
opposed to the CC charging strategy, a capacitive out-
put is required to regulate the output voltage. This
output stage increases the model order, and therefore,
the controller design process is slightly more complex
than CC technique, yet simple enough for low-cost
applications.
The main disadvantage of the CV charging meth-
odology is the temperature rise of the battery during
initial charging stages. While the internal voltage of the
battery is very low, the applied voltage at the terminal
of the battery is relatively high. Thus, considering the
small internal resistance of the battery, a large current
will flow which can damage the battery. Another draw-
back of CV charging is the long charge time due to very
low currents at the final stages of the charging.
Mixed Constant-Current and Constant-Voltage
Charging (CC-CV) The CC-CV charging technique
utilizes both CC and CV charging. Figure 5a, b
illustrates the charging characteristic curve and
134 Batteries, Battery Management, and Battery Charging Technology
current-voltage profiles of this method, respectively.
The charging is performed in three stages. First, the
battery current is kept constant until the battery voltage
reaches a predetermined value (Vconst ). Then, the bat-
tery voltage is kept constant while the battery current
drops below a specific value (Imin ). The final stage is the
float charge stage where the battery voltage is reduced
to compensate for the loss caused by self-discharge of
the battery.
Since the current is kept constant at the first charg-
ing stage, the current rush disadvantage of CV charging
technique is avoided. Also, since the voltage is kept
constant at the second stage, the overcharging disad-
vantage of CC charging is avoided. Moreover, the float
charging stage keeps the battery in full charge for a long
time without degrading the battery. These advantages
of CC-CV charging technique make it perfect for appli-
cations with high performance requirement. However,
the design complexity makes this option more expen-
sive compared to either CC or CV techniques.
Pulse Charging Pulse-charging technique is inher-
ently a CC charging method that uses a pulsating
charging current. This pulsating current causes the
ions to distribute evenly throughout the battery and
electrodes. Consequently, even distribution of ions
prolongs the lifetime of the battery and enhances the
charging performance. Figure 6 shows the current pro-
file of the battery using pulse-charging technique. Due
to the internal battery resistance, the battery voltage
ibat
Iconst
t
DTs
Ts 2Ts 3Ts
Batteries, Battery Management, and Battery Charging
Technology. Figure 6
Current profile of the battery using pulse-charging
technique
increases slightly during the charging subintervals. DC
value of the charging current can be controlled by the
duty cycle of applied current between zero and Imax .
The charging rate is controlled by the DC value of
charging current; thus, charging time of the battery
can be controlled by the duty cycle:
tcharge ¼ f ðIDC Þ ¼ f ðDImax Þ; ð1Þ
where D, IDC , and Imax are the duty cycle, DC value, and
the maximum of the battery current, respectively. Large
duty cycles are applied during the initial stages of charg-
ing to ensure fast charging. After battery voltage is
increased sufficiently, the duty cycle gradually decreases
to prevent overcharging and consequent battery failure.
Pulse current charging needs an extra control loop
to regulate the duty cycle of the reference current (Iref )
by the output voltage. As the output voltage decrease,
this additional control loop decreases the duty cycle
and consequently the DC value of the charging current.
Following a pulsating reference current requires a fast
control loop. The bandwidth of the control loop should
be sufficiently higher than the switching frequency.
This bandwidth is limited by the effective bandwidth
of the current sensor. Thus, increasing the bandwidth
of the controller requires high-bandwidth current sen-
sors, which can be costly.
Taper Characteristic Charging In taper charging,
neither battery current nor battery voltage is kept con-
stant. Instead, a linear combination of battery voltage
and current is kept constant:
k1 Vbat þ k2 Ibat ¼ c ¼ constant; ð2Þ
where k1 and k2 are constant coefficients which are
determined by design requirements. CC and CV charg-
ing techniques are special cases of the taper character-
istic charging in which k1 ¼ 0 or k2 ¼ 0, respectively.
Figure 7a, b illustrates the charging characteristic curve
and voltage-current profiles of the taper charging tech-
nique, respectively.
The block diagram of the charge controller using
this technique is depicted in Fig. 8. Sensing of both
battery voltage and current is required and is the main
drawback of this strategy. Sensors are usually the most
expensive components of the converter circuits, and
increasing the number sensor may drastically increase
the total cost of the system.
 Batteries, Battery Management, and Battery Charging Technology 135

vbat vbat , ibat

Vmax
Vmax vbat
I0

ibat
ibat t
Istart
a b

Batteries, Battery Management, and Battery Charging Technology. Figure 7

(a) Charging characteristic curve of the taper-characteristic-charging technique; (b) Current and voltage profiles of the
battery during the charging

+ PWM
Gate
Controller signal
C +_ driver
(e.g. PI) _
circuit
Semiconductor
switch
k1vbat + k2ibat
Sawtooth wave

Batteries, Battery Management, and Battery Charging Technology. Figure 8

Block diagram of the controller for taper characteristic charging

Assembled Characteristic Charging Assembled or
modified charging is a technique whose characteristic
curve is a combination of CC, CV, or taper character-
istic curves. In this technique, when a certain threshold
is reached (e.g., certain voltage), the characteristic
curve is changed from one technique to another. Two
conventionally assembled characteristic charging
methods are assembled constant-current charging and
assembled taper characteristic charging.
Assembled Constant-Current Charging Figure 9a
illustrates the charging characteristic curve of the
assembled constant-current charging technique. As
seen in Fig. 9a, battery is charged by a high constant
current (I1 ) until the output voltage reaches the thresh-
old voltage (Vth ), after which the CC charging will
continue using a lower constant current (I2 ). Then,
charging with low current will continue until the
output voltage reaches its maximum value (Vmax ),
which is usually the rated voltage of the battery.
The current and voltage profiles of the battery are
shown in Fig. 9b.
As the high constant current reduces the charging
time but increases the risk of overcharging, one may
trade off between fast charging and overcharging pro-
tection. This drawback of CC charging has been over-
come in assembled constant-current technique. Short
charging time can be achieved by using high constant
current (I1 ) at the first stage (v  Vth ), while
overcharging can be avoided by using low constant
current at the second stage (v > Vth ).
Assembled Taper Characteristic Charging Assem-
bled taper characteristic charging utilizes two different
taper charging characteristic curves, as shown in
Fig. 10a. The characteristic equation of this technique
can be expressed as:
(
k1 Vbat þ k2 Ibat ¼ c; Vbat  Vth
0 0 0 ð3Þ
k1 Vbat þ k2 Ibat ¼ c ; Vbat  Vth
136 Batteries, Battery Management, and Battery Charging Technology

vbat vbat , ibat

Vmax Vmax
vbat
Vth
Vth
I1

ibat
I2
ibat t

a b

Batteries, Battery Management, and Battery Charging Technology. Figure 9

(a) Charging characteristic curve of the modified constant-current technique; (b) Current and voltage profiles of the
battery during the charging process

vbat vbat, ibat

Vmax Vmax
vbat
Vth
Vth
I1

I2 ibat
ibat t
I2 I1
a b

Batteries, Battery Management, and Battery Charging Technology. Figure 10

(a) Charging characteristic curve of the modified taper characteristic technique; (b) Current and voltage profiles of the
battery during the charging

where c > c 0 . Figure 10b depicts current and voltage
profiles of the battery during the charging time interval.
Enhanced charging performance and prolonged
battery life can be achieved using this technique. The
main disadvantages of this technique are implementa-
tion complexity and using two sensors (current and
voltage sensors).
Qualitative Comparison Between Different Charging
Techniques Table 1 summarizes different charging
techniques and their main properties:
energy during the high radiation hours during the day
and supply load during the low-radiation hours and
nights. The battery is an effective energy storage tech-
nology for a PV systems. Lead-acid batteries are com-
mon due to their high energy density, prevalences and
low cost. Different structures that can be used to con-
nect the battery stack to the PV panel are presented in
the next sections. In all of these structures, a series
diode is connected between the PV panel and the bat-
tery to protect the battery against short circuit during
the low-radiation conditions and nights.

Series Charge Regulators Figure 11 illustrates a typ-

Battery Charging Circuits for Photovoltaic (PV)
Systems
Energy storage components are an essential part of
a stand-alone photovoltaic (PV) system as they store
ical charge regulator structure which can be used for
both series and parallel charging. In series charging,
switch S2 is always open. Switch S1 disconnects the PV
panel from the battery when state of charge (SoC) of
 Batteries, Battery Management, and Battery Charging Technology 137

Batteries, Battery Management, and Battery Charging Technology. Table 1 Qualitative comparison between different
charging techniques

Charging technique Advantages Disadvantages

Constant current (CC) ● Simplicity of implementation ● Overcharging risk if high charging
● First order open-loop transfer-function current is used
● Simplicity of the controller ● Prolonged charging time if low current
is used
Constant voltage (CV) ● Simplicity of implementation ● Battery temperature rise during the
initial stages of charging
● Prolonged charging time due to low
current during the final charging stages
● Second order open-loop transfer-
function
Constant current – ● Protecting the battery against overcharging ● Requires both voltage and current
constant voltage ● No current rush at the beginning stage of sensor for closed-loop control
(CC – CV) charging
● Enables float charging after the battery is fully
charged
● High performance
Pulse current ● Even distribution of ions throughout the ● Requires fast control-loop
electrodes and inside the battery ● Requires high bandwidth current-
● Prolonged battery life due to maximizing sensor
chemical performance ● Expensive
● Charging rate can be controlled by the duty cycle
of the battery current
Taper characteristic ● Gradual current decrease during the charging ● Requires both voltage and current
● Protection against overcharging sensor for closed-loop control
● Complexity of implementation
Assembled constant ● Eliminates overcharging property of CC charging ● More complicated control scheme than
current ● Eliminates prolonged charge time property of CC CC charging
charging
● Takes other advantages of CC charging
Assembled taper ● Different adjustment of characteristic curve is ● Requires both current and voltage
characteristic possible to satisfy different design requirements sensors for closed-loop control
● Enhanced charging performance ● Most complex control scheme

battery reaches 100% to protect the battery against
overcharging. Switch S3 disconnects the battery from
the load when the battery SoC is less than a predefined
value to avoid deep discharging mode, which can
degrade the battery and shorten its life. All switches
should have very low loss which means they should
have very low forward voltage and on-time resistance.
Thus, mechanical relays are preferred to fast power
semiconductors such as IGBTs with high on-time volt-
age drops.
Shunt Charge Regulators In a typical shunt charge
regulator, the switch S1 in Fig. 11 is always closed, and
switch S3 operation is similar to series charge regulators
to disconnect the load from the battery when battery
SoC is less than a predefined value to prevent the
battery from deep discharging. Switch S2 shorts the
output of the PV panel during low-radiation condi-
tions and nights. When the output of the PV panel is
shorted, series diode protects the battery against short
circuit condition.
138 Batteries, Battery Management, and Battery Charging Technology

Pulse-Width Modulation Charge Regulators Supe- Figure 12 shows a buck-boost-type PWM charge con-
rior charging functionality can be achieved using troller with voltage and current feedback. The charge
pulse-width modulation (PWM) charge regulators. controller also has the ability to control switch S2 to
PWM-type charge regulators are able to eliminate disconnect the load under the condition in which
external disturbances and can be programmed to fol- battery SoC is less than a predefined value to protect
low a specific charging technique that has been the battery against deep discharging.
discussed in section “Battery Charging Techniques”.
Battery Charging Circuits with AC Sources
Several factors should be considered when designing
a charging circuitry with AC input voltage, e.g.,
Charge nominal output DC voltage, allowable ripple in output
Controller current, type of the input AC voltage source (single-
phase or three-phase), AC voltage source ratings (volt-
S1 S3
age and frequency), charger efficiency and power factor,
charger current limiting capability, and recharge dura-
tion time. In this section, the conventional battery
PV Battery charger’s configurations with an AC input source are
Panel +
S2 Vout
discussed.
Load
_
Battery Charger Configurations Given the DC
nature of batteries, a rectifier circuit is needed. In
a conventional configuration, shown in Fig. 13,
a phase-controlled rectifier is adopted. Another config-
uration is a rectifier bridge in series with a buck DC-DC
Batteries, Battery Management, and Battery Charging converter as shown in Fig. 14. If an electrical isolation is
Technology. Figure 11 required, a DC-DC converter with a high-frequency
Circuit diagram of series and shunt PV charging system isolation transformer can be used as shown in Fig. 15.

Charge
Controller Vbat

L1 L2 S2

PV
Battery
Panel +
Duty Cycle
Load Vout
_
S1

Batteries, Battery Management, and Battery Charging Technology. Figure 12

Circuit diagram of the PWM-PV charging system
 Batteries, Battery Management, and Battery Charging Technology 139

Phase Controlled
Rectifier
AC
Source Battery

Batteries, Battery Management, and Battery Charging Technology. Figure 13

Phase-controlled rectifier

Diode Rectifier

AC Buck Battery
Source Converter

Batteries, Battery Management, and Battery Charging Technology. Figure 14

Rectifier bridge in cascade with a buck converter

Diode Rectifier

Switching DC-DC
AC Converter with High Battery
Source Frequency Isolation
Transformer

Batteries, Battery Management, and Battery Charging Technology. Figure 15

DC-DC converter with a high-frequency isolation transformer

2
Single-Phase, Phase-Controlled Rectifiers The cir- Vbat  0:9VS cos a  oLS Ibat;min ð4Þ
p
cuit configuration of a single-phase thyristor-
controlled rectifier is illustrated in Fig. 16. The battery where Ibat;min is the minimum value of battery current
is presented as a DC voltage source in series with which occurs at t ¼ oa .
a resistor (rd). Considering a continuous battery cur- The average value of battery current can be calcu-
rent, the voltage and current waveforms are illustrated lated using (5):
in Fig. 17. Due to the presence of AC side inductance, Vbat ¼ rd Ibat þ Ed : ð5Þ
there is a commutation interval, u. In the continuous
mode of battery’s current, the approximate average Figure 18 shows the input line current (is).
value of the battery voltage is: Neglecting Ls, AC source current is a square wave
140 Batteries, Battery Management, and Battery Charging Technology

α +
Ls is

rd
Vs Vbat Battery
Ed

Batteries, Battery Management, and Battery Charging Technology. Figure 16

Single-phase thyristor rectifier

Vs
Vbat
Ed
ibat

0
ωt

u
α

Batteries, Battery Management, and Battery Charging Technology. Figure 17

Battery’s voltage and current waveforms

Vs

is1 is
0
ωt

Batteries, Battery Management, and Battery Charging Technology. Figure 18

Input line current

current with only odd harmonics. These harmonics can 48.43%. The power factor of the single-phase configu-
be expressed as (6): ration is expressed in (7):

Is1 PF ¼ 0:9 cos a: ð7Þ

Ish ¼ ð6Þ
h At low values of ibat , the battery current waveform
where Is1 is the RMS value of the fundamental fre- may become discontinuous as shown in Fig. 19.
quency of Is . The total harmonic distortion of input Through the appropriate choice of firing angle a, one
line current in a single-phase configuration is around can control the average value of battery voltage and
 Batteries, Battery Management, and Battery Charging Technology 141

Vs
Vbat
Ed
ibat

0
ωt

Period of
α discontinuity

Batteries, Battery Management, and Battery Charging Technology. Figure 19

Battery voltage and current in discontinuous mode of operation

can be expressed in terms of (8). Compared with the

Vbat
single-phase configuration, the ripple in the output
voltage of three-phase rectifier is less:
pffiffiffi
3 2 3
Vbat  VLL cos a  oLS Ibat ð8Þ
α=0 p p
Increasing α where VLL is the RMS value of the line voltage of AC
source, and Ibat is the battery current.
Figure 23 shows the input line current of phase “a”
(iS ). Neglecting LS , AC source current harmonics can
be expressed as (9):
ibat Is1
Ish ¼ where h ¼ 6n  1 ð9Þ
h
Batteries, Battery Management, and Battery Charging
Technology. Figure 20 The total harmonic distortion of input line current
Rectifier output profile for different values of a in three-phase configuration is around 31.08%, less
than that of a single-phase configuration. The power
factor of three-phase configuration is:
3
thus battery current and, therefore, the power deliv- PF ¼ cos a: ð10Þ
p
ered. Figure 20 shows the variations of Vbat versus ibat
for different values of a. For a given value of a, if ibat Rectifier Bridge in Cascade with a Buck DC-DC
becomes less than a specific threshold, Vbat increases Converter In this configuration a rectifier is in cas-
sharply. In such a case, a has to increase to keep Vbat cade with a buck DC-DC converter. In the buck con-
constant. verter, which is shown in Fig. 24, the average output
voltage is lower than DC input voltage. Figure 25 shows
Three-Phase, Phase-Controlled Rectifiers Figure 21 the waveforms of Vd, Vbat, VL, and iL for a resistive load
shows the circuit configuration of a three-phase phase- and continuous conduction of inductor. ton denotes
controlled rectifier connected to a battery. Similar to the time that the switch is on in a switching period.
the single-phase configuration, the battery is presented Because of the fact that the integral of the inductor
as a DC voltage source in series with a resistor (rd). voltage should be zero over a time period, the output
Assuming a continuous current for battery, battery’s voltage is:
voltage waveform is illustrated in Fig. 22. In this figure,
u and a present the commutation interval and delay Vbat ton
¼ ¼ D: ð11Þ
angle, respectively. The average value of battery voltage Vin Ts
142 Batteries, Battery Management, and Battery Charging Technology

Ibat

α +
Ls is
a
rd
b Battery
n Vbat
Ed
c

Batteries, Battery Management, and Battery Charging Technology. Figure 21

Three-phase phase-controlled rectifier

Vbat
α

0
ωt

Batteries, Battery Management, and Battery Charging Technology. Figure 22

Battery’s voltage waveform

Vcn Van Vcn

iS
iS1
ωt
α u

Batteries, Battery Management, and Battery Charging Technology. Figure 23

Input line current

Ignoring losses, the expression for output current is: primary DC voltage is converted from one DC level
to another. This task is done by using a high-frequency
Ibat Vin 1
¼ ¼ : ð12Þ transformer. The output of this configuration is con-
Iin Vbat D
trolled by adopting a PWM feedback control. Figure 26
DC-DC Converter with a High Frequency Isolation shows the simplified block diagram of this configura-
Transformer In this configuration using a diode rec- tion with its control system. Flyback, forward,
tifier, the input AC voltage is rectified into a primary push-pull, half bridge, and full bridge are the main
DC voltage. Using high-frequency switching, the types of this battery configuration.
 Batteries, Battery Management, and Battery Charging Technology 143

iin iL ibat

+ + + -
VL +

Vd Vbat Battery
Vin

- - -

Batteries, Battery Management, and Battery Charging Technology. Figure 24

Buck converter

Vd Flyback Converter Figure 27 shows the circuit

Vin topology of this converter. This converter is derived
from buck-boost converter. Defining the switch duty
ratio D as the ratio of the time the switch is on ton to the
total switching period Ts , D ¼ tTons , the DC output
Vbat voltage is:

a 0 t
Vbat N2 D
¼ : ð13Þ
Vin N1 1  D
VL Forward Converter Figure 28 shows the circuit
Vin−Vbat topology of this converter. This converter is derived
from a step-down (buck) converter. The expression
for the DC output voltage level is:
t Vbat N2
¼ D: ð14Þ
−Vbat Vin N1

b Push-Pull Converter Similar to forward converters,

push-pull converters also have the same operation as
buck converters. Circuit configuration of these con-
iL verters is shown in Fig. 29. The expression for the DC
TS output voltage level is:
Vbat N2
¼2 D; 0 < D < 0:5: ð15Þ
Vin N1
IL=Ibat
In practice, in order to avoid simultaneous conduc-
tion of both switches, D should be less than 0.5.
0
t
Ton Toff
Half-Bridge Converter Figure 30 shows the circuit
c TS configuration of this converter. This converter is also
derived from step-down converters. The expression for
Batteries, Battery Management, and Battery Charging
the DC output voltage level is:
Technology. Figure 25
(a) Diode voltage; (b) Inductance voltage; (c) Inductance Vbat N2
¼ D; 0 < D < 0:5: ð16Þ
current Vin N1
144 Batteries, Battery Management, and Battery Charging Technology

HF Isolation
Transformer

Rectifier
+ Battery
AC Source Diode Terminal
Rectifier Filter

PWM PI
Generater Controller
Battery
Voltage/Current
Reference

Batteries, Battery Management, and Battery Charging Technology. Figure 26

DC-DC converter with a high-frequency isolation transformer and its control system

+ +

Vin N1 N2 Battery
Vbat

- -

Batteries, Battery Management, and Battery Charging Technology. Figure 27

Flyback converter

+ +

Vin N1 N2 Vbat Battery

- -

Batteries, Battery Management, and Battery Charging Technology. Figure 28

Forward converter

+
N1 N2
+ Vbat Battery
Vin N1 N2 -
-

Batteries, Battery Management, and Battery Charging Technology. Figure 29

Push-pull converter
 Batteries, Battery Management, and Battery Charging Technology 145

+
N2
Vin N1 Vbat Battery
N2
Vin/2
−

Batteries, Battery Management, and Battery Charging Technology. Figure 30

Half-bridge converter

+
N1 N2
Vbat Battery
N2
Vin
−

Batteries, Battery Management, and Battery Charging Technology. Figure 31

Full-bridge converter

Full-Bridge Converter Full-bridge converters also Bidirectional Configurations for Battery Charging
have the same operation as buck converters. Their with AC Sources
circuit configuration is shown in Fig. 31. The expres-
Bidirectional configurations for battery charging are
sion for the DC output voltage level is:
used in multifunctional battery energy storage systems
Vbat N2 (BESS) and plug-in hybrid electric vehicles (PHEV). In
¼2 D; 0 < D < 0:5: ð17Þ
Vin N1 the case of BESS, considering normal operation of the
146 Batteries, Battery Management, and Battery Charging Technology

3 phase bidirectional converter

Utility
Battery Grid
bank

za zb zc

Batteries, Battery Management, and Battery Charging Technology. Figure 32

Bidirectional configuration for BESS

Bidirectional
Voltage Source
Converter

Bidirectional
AC DC/DC Battery
Source Converter

Batteries, Battery Management, and Battery Charging Technology. Figure 33

Bidirectional configuration for PHEV

power system, the battery package can be charged and
BESS can be controlled to shave the pick load. In the
power failure condition, the battery pack is discharged
and BESS provides an uninterruptible power for the
utility grid. Figure 32 shows a three-phase configura-
tion of a BESS system. Sophisticated control systems
provide required commands for the switching system
of the bidirectional converter which is a current-forced
sinusoidal pulse-width modulated switching.
In the case of PHEVs, the battery pack should have
this capability to store energy from an external power
source and regenerative braking as well as sending back
the stored energy to the utility grid. Figure 33 shows the
PHEV battery charger configuration. This battery char-
ger consists of a bidirectional AC-DC converter in
cascade with a bidirectional DC-DC converter.
Contactless Battery Charging
In recent years, a new approach for battery charging,
named contactless battery charging, has emerged. This
new charging scheme alleviates the wiring require-
ments and provides an easy charging process for cell
phones, laptops, cameras, PHEVs, etc. Currently, there
are two dominant technologies for contactless battery
charging named inductive contactless battery charging
and capacitive contactless battery charging.
Inductive Contactless Battery Charging The planar
contactless battery charging platform is the predomi-
nant technology for contactless chargers. It uses induc-
tive interface between energy transmitter and energy
receiver and has the potential to unify all battery charg-
ing protocols related to portable electronic devices such
as cell phones, music players, etc. Generally, there are
two main schemes for implementing inductive
contactless battery charging. The first scheme adopts
horizontal flux in which magnetic flux moves horizon-
tally onto the planar charging surface and flows on both
sides of the planar. To achieve this requirement, a thick
layer of soft magnetic material such as ferrite or amor-
phous alloy as the flux guide is used. Figure 34a shows
the charging platform with horizontal magnetic flux
trajectories. Figure 34b shows the secondary harvesting
device which should pick up the fluxes from the surface
 Batteries, Battery Management, and Battery Charging Technology 147

of the charging platform. This energy-harvesting device flow to the charging platform surface. The magnetic
consists of a magnetic core and a winding and flux has uniform magnitude over the charging plat-
is mounted into the portable electronic device. form. This perpendicular flux trajectory allows a slim
The cross-sectional area of this device should be large energy harvester device which is mounted into the
enough to collect enough flux and energy. portable electronic device. Figure 35 shows the charg-
The second scheme adopts perpendicular flux in ing platform with perpendicular magnetic flux
which magnetic flux has a perpendicular trajectory trajectories.
Figure 36 shows the block diagram of an inductive
contactless charger. The harvested energy is rectified,
Flux limited, and regulated to generate a DC voltage across
trajectories
the battery’s terminals. Under and over discharge
Charging platform protection, setting of the battery voltage and current
profiles, and implementing battery charging control
techniques can be achieved by using an appropriate
control system. Conventional configurations for bat-
a tery charging circuits, explained before, can be used for
the battery charger.
Harvesting device
Capacitive Contactless Charger Thick electromag-
netic flux guide is an important requirement in
b implementing efficient inductive contactless chargers.
Batteries, Battery Management, and Battery Charging The confined electric field between conductive plates
Technology. Figure 34 can alleviate this requirement in capacitive contactless
(a) Charging platform with horizontal magnetic flux chargers, as shown in Fig. 37. Figure 38 shows the
trajectories; (b) Energy-harvesting device circuit configuration of a capacitive contactless charger
based on a series resonant architecture. Two coupling
capacitors transfer power from Vin to battery. Consid-
ering a DC power supply, using an inverter, the input
Flux
trajectories DC voltage is converted to an AC output voltage to
Charging platform
enable current flow through the coupling capacitors.
Electromagnetic
This inverter is not required if supplied from an AC
shield source. Inductors are used to enable soft switching. At
the final stage, a diode rectifier provides the required
Batteries, Battery Management, and Battery Charging DC voltage for battery charging. The circuit parameters
Technology. Figure 35 should be designed in such a way that a circuit with the
Charging platform with perpendicular magnetic flux least possible value of capacitor and the desired effi-
trajectories ciency is achieved.

Inductive
Link

Rectifier Regulator Battery Battery

Limiter charger

Batteries, Battery Management, and Battery Charging Technology. Figure 36

Inductive contactless charger scheme
148 Batteries, Battery Management, and Battery Charging Technology

Electromagnetic Interference (EMI) Conductive Noise Conductive noise addresses

disturbances introduced by coupling of two or more
Electromagnetic interference (EMI) is defined as deg-
circuits through interconnecting wires. For example,
radation in performance of an electronic system caused
when a utility transformer supplies a rectifier, it is
by electromagnetic disturbances. Electromagnetic
exposed to the same current as of the rectifier. There-
compatibility (EMC) is the ability of that system to
fore, if the input current of the rectifier has harmonic
operate satisfactorily in an electromagnetic environ-
content that is intolerable for the transformer, it can
ment without introducing intolerable electromagnetic
cause the transformer to malfunction.
disturbances to other devices in that environment.
Battery charging circuits are power electronic con-
Inductive Coupling Inductive coupling refers to dis-
verters in nature. Therefore, EMC rules and regulations
turbances that are caused by unwanted coupling of
exist that apply to the battery charger circuits. Power
windings in different circuits. Leaking flux of an induc-
electronic converters must emit electromagnetic
tor inside a circuit may encircle a winding or a closed
disturbances below the maximum permitted amount
path in another circuit and results in inductive cou-
(emission) and be able to operate properly when
pling. Figure 39 illustrates the modeling circuit of
exposed to electromagnetic disturbances below the
inductive coupling. Mutual inductance, M, indicates
maximum permitted amount (susceptibility).
the degree to which two circuits are coupled. The
There are four types of electromagnetic distur-
winding voltage set is denoted in (18):
bances: conductive noise, inductive coupling, capaci-
8
tive coupling, and radiation. > di1 di2
< v1 ¼ L 1 þM
dt dt ð18Þ
>
:v ¼ M di 1 di 2
2 þ L2
dt dt
The noise is represented as a mutual term using the
mutual inductance
di2
vnoise;12 ¼ M : ð19Þ
AC dt
Battery
Capacitive Coupling Figure 40 portrays the capaci-
tive coupling circuit. Capacitive coupling happens when
there is a small capacitance between different parts of
different circuits, and the medium is electric field. The
disturbance voltage caused by Vg can be expressed as:
Batteries, Battery Management, and Battery Charging
Technology. Figure 37 ZðsÞCs
Vnoise ðsÞ ¼ Vg ðsÞ: ð20Þ
Capacitive contactless power transferring ZðsÞCs þ 1

L C
Vin
Battery
L C

Batteries, Battery Management, and Battery Charging Technology. Figure 38

Capacitive contactless charger circuit configuration
 Batteries, Battery Management, and Battery Charging Technology 149

M emission of the converter circuit. Radiation disturbance

is trivial in power electronic converters because switching
L1 i1(t) L2 i2(t) frequency is relatively low, which means that the wave-
v1(t)
length of the electromagnetic wave is much longer than
v2(t)
the circuit dimensions. Susceptibility of the converter to
Vg(t) Z1(s) capacitive and inductive noises can be diminished by
Z2(s)
shielding the converter. Air core inductors should be
avoided in the converter circuit to meet electromagnetic
interference (EMI) and electromagnetic compatibility
Batteries, Battery Management, and Battery Charging
(EMC) emission requirements. Using air core inductors
Technology. Figure 39
causes long flux paths through the air that can encircle
Modeling circuit of inductive coupling
other windings and cause inductive coupling. On the
other hand, using magnetic core inductors defines
a specific path for flux through the core and prevents
Stray Capacitor inductive coupling. The most challenging EMC issue of
C the power electronic converters is conductive noises. As
discussed in the previous sections, battery charging cir-
cuits are mainly AC-DC or DC-DC circuits. In DC-DC
battery charging circuits, pulsating input current should
be prevented to meet EMC emission requirements. Thus,
Vg(t)
Z(s) Vnoise(t) buck and buck-boost converters are not suitable candi-
dates. As an alternative, Boost, Ćuk, and SEPIC con-
verters can be used as they have built-in input current
filters. The most suitable choice is a Ćuk converter as it
has output current filter that makes it perfect for con-
stant-current charging applications. There are two types
of AC-DC battery charging circuits, namely phase-
Batteries, Battery Management, and Battery Charging
controlled rectifiers and PWM rectifiers. Phase-
Technology. Figure 40
controlled rectifiers draw square shape input currents
Modeling circuit of capacitive coupling
that have a very high harmonic content. Thus, the
designed input current filter for phase-controlled recti-
Thus, decreasing C reduces the magnitude of induced fiers should have lower cutoff frequencies which requires
noise voltage. larger and bulkier filter components. On the other hand,
the input current of PWM rectifiers has less harmonic
Radiation Radiation disturbance is caused by the content that are located at higher frequencies. This char-
absorption of electromagnetic waves where it is not acteristic of the PWM converters enables one to design
intended. When the circuit dimensions are very small input current filters by smaller inductors.
compared to one fourth of the wavelength of an elec-
tromagnetic wave, that electromagnetic wave cannot be
Regulations and Standards
fully absorbed by the circuit. Thus, radiation distur-
bance usually occurs in high-frequency waves with Several standards have been developed by different insti-
shorter wavelengths. tutes to regulate electromagnetic emission and suscep-
tibility of electrical equipment, namely: MIL-STD 461;
IEEE/ANSI SA-C63.14; IEEE 519, IEC 555-2, FCC, title
EMC in Battery Charging Circuits
47, part 15; German VDE; and Japanese standards.
Different guidelines should be followed in battery charg- MIL-STD 461 establishes requirements for control
ing circuits to reduce electromagnetic susceptibility and of EMI emission and susceptibility of electronic and
150 Batteries, Battery Management, and Battery Charging Technology
electromechanical equipment which are produced for
use by agencies of the Department of Defense.
IEEE/ANSI SA-C63.14 provides the definitions for
specific terms that are related to EMC, electromagnetic
pulse (EMP), and electrostatic discharge (ESD).
IEEE-519 regulates the injection of harmonic cur-
rents to the power system due to nonlinear loads such
as static power converters, arc discharge devices, satu-
rated magnetic devices, and, to a lesser degree, rotating
machines.
The IEC 555-2, IEC 1000-3-2, and IEC 1000-3-4 are
emission standard related to operation of power elec-
tronic converter. These standards usually set an upper
limit for harmonic content in each frequency.
Battery Management
Need for Battery Management
Two critical parameters for battery performance are
battery voltage and operating temperature. A battery
usually consists of a pack of cells connected in series.
Manufacturing processes lead to imperfections in cells,
as a result of which, all the cells in a pack are not
identical. Electrical imbalances occur during charging
and discharging of battery packs. Some cells in
a battery will have different voltage levels for the same
charging. This mismatch needs to be monitored to
improve efficiency and safety of battery pack. Higher
than rated temperatures in the batteries lead to unde-
sirable chemical effects. Self-sustaining internal
temperature buildup and gas pressure buildup are
common. These conditions lead to safety hazards. It is
vital to have a mechanism to monitor cell temperature
and voltage and keep them within limits.
The main role of battery management systems
(BMS) is to monitor cell voltage/current, state of
charge/state of health, and the internal battery temper-
ature and ambient temperature. The monitoring cir-
cuitry provides signals to the protection unit as well.
Battery management systems differ on the basis of their
primary functions, which depend upon the intended
application. BMS for standby batteries in a power plant
deal with monitoring of various battery parameters,
maintaining readiness to deliver full power in the
event of a system failure, and ensuring equal charging
to increase battery life. On the other hand, a BMS in an
electric vehicle must communicate with other controls
in the automobile such as temperature controls, engine
management, and safety systems.
There are three main objectives common to all
battery management systems:
● Protect the cells or the battery from damage.
● Prolong battery life via smart control.
● Maintain battery in a state in which it can fulfill the
functional requirements of the application for
which it was specified.
State of Health Measurement
The state of health is a “measurement” that reflects
the general condition of a battery and its ability to
deliver the specified performance compared with a
fresh battery. It takes into account such factors as
charge acceptance, internal resistance, voltage, and
self-discharge.
It is an estimate rather than a measurement. Battery
manufacturers do not specify the SoH because they
only supply new batteries. The SoH only applies to
batteries after they have started their aging process
either on the shelf or once they have entered service.
The SoH definitions are therefore specified by test
equipment manufacturers or by the user.
Any parameter which changes significantly with
age, such as cell impedance or conductance, can be
used as a basis for providing an indication of the SoH
of the cell. Changes to these parameters will normally
signify that other changes have occurred which may be
of more importance to the user. These could be changes
to the external battery performance such as the loss of
rated capacity or increased temperature rise during
operation, or internal changes such as corrosion.
Methods of Determining the State of Charge
State of charge (SoC) is defined as the capacity left in
a battery expressed as a percentage of some reference.
SoC of a battery is usually expressed as a percentage of
the current battery capacity when it is fully charged.
This definition may lead to an erroneous result when
applied to a battery that has been in service for a long
time. The maximum capacity of a battery reduces sig-
nificantly with prolonged use. Considering this value to
be the reference for state of charge measurement can
lead to false results.

There are several techniques to determine SoC.
A majority of them depend on measuring some conve-
nient parameter which varies with the state of charge.
Some are specific to particular cell chemistry. Some of
the important ones are listed below:
Specific Gravity Measurement In lead-acid batte-
ries, the electrolyte (sulfuric acid) is used up as the
battery is subjected to a discharge cycle. As more and
more electrolyte is used up, the specific density of
sulfuric acid reduces. Thus, the specific gravity of the
sulfuric acid is an indication of the state of charge for
lead-acid batteries. This method is not feasible for
VRLA batteries.
Voltage Measurement Cell voltage typically
decreases for most cells during a discharge cycle.
There is an almost linear relation between the state of
charge and cell terminal voltage. Results may vary
depending upon cell terminal voltage, temperature,
discharge rate, and age of cell. A compensation factor
needs to be included for SoC calculation by this
method.
Li-ion cells have a flat voltage profile during
a discharge cycle. The cell voltage drops abruptly
when they are close to a complete discharge. As the
cell voltage variation with SoC is not linear, cell voltage
cannot be selected as a parameter to indicate SoC of
Li-ion cells. In general, the voltage measurement tech-
nique fails for cells which have a flat voltage profile
during a discharge cycle.
Current-Based SoC Estimation Charge in or out of
the cell is calculated by integrating the current delivered
by the cell for certain duration. This charge is com-
pared with the charge contained by a fully charged cell.
This method provides a high accuracy for SoC mea-
surement as it measures the charge directly. Three
popular current sensing methods are used.
● Current shunt: The simplest method of determin-
ing discharge current is by measuring the voltage
drop across a low ohmic value, high precision,
series, sensing resistor. This resistor, connected
between the battery and the load, is called
a current shunt. This method of measuring current
causes a slight power loss in the current path and
Batteries, Battery Management, and Battery Charging Technology 151
also heats up the battery and is inaccurate for low
currents.
● Hall-effect transducers: The Hall-effect transducers
avoid the power loss problem, but they are more
expensive. Unfortunately, they do not tolerate high
currents and are susceptible to noise.
● Giant magneto resistance (GMR) sensors: These
sensors are even more expensive than Hall-effect
transducers, but they have higher sensitivity. They
also have better high temperature stability than
Hall-effect devices.
Kalman Filter Technique A Kalman filter is an algo-
rithm to estimate the inner states of any dynamic system.
The cell is represented by a mathematical state-space
model, and SoC is considered as a state of the system.
The Kalman filter accurately estimates the value of SoC
and uncertainty. Armed with this information, the bat-
tery pack can be subjected to a more complete use,
without fear of over- or under-charging cells. Nonlinear
models of cells provide a highly accurate result.
The Kalman filter model for a rechargeable battery
used in automobiles is initialized with a priori state
estimate when the vehicle is turned on. The priori state
estimate is provided on the basis of open-circuit volt-
age readings and a look-up table, self-discharge rate
data from the cell model and the prior SoC when the
vehicle was turned off. The algorithm then repeatedly
updates the state estimate and state-uncertainty (error
bound) estimate with each set of new measurements, as
the system runs. Laboratory tests on real cells have
shown that excellent SoC estimation with very tight
error bounds are obtained whether or not the initial
SoC estimate is accurate.
SoC Estimation from Internal Impedance Measure-
ments During charge-discharge cycles, the composi-
tion of active chemicals changes in the battery, and this
is reflected in the changes to internal impedance of the
battery. Plots for variation of internal cell impedance
with state of charge may be obtained for different cells.
Thus, measuring internal impedance can indirectly
provide information about SoC of the cell. However,
on the flip side, cell internal impedance is also affected
by temperature, which may lead to erroneous
measurements.
152 Batteries, Battery Management, and Battery Charging Technology

Cell Balancing ● High efficiency: Implementation of this technique

should result in minimum power loss of the system.
When a battery pack is charged and discharged as
● Small volume: Bulky volumes restrict portability
a unit, individual cell temperature and internal chem-
and increase cost.
istry characteristics can cause capacity imbalances in
● Simple wiring schemes: Complex wiring schemes
the form of voltage variations. Imbalances in cell volt-
increase cost and decrease the reliability.
ages are caused by differences in cell capacities, internal
● In automobile applications, the cell must also be
resistances, chemical degradation, and inter-cell and
equipped to handle pulse charging from regenera-
ambient temperatures during charging and
tive braking.
discharging. Any capacity imbalance between the mod-
ules can threaten the long-term reliability of the string
as overall pack capacity is brought to the upper and
Passive Balancing Passive-charge balancing tech-
lower limits of charge. Imbalances in cell voltages can
niques involve bleeding off of charge from the cells
cause cell overcharging and discharging, decreasing the
with the highest voltage (hence, state of charge) in
total storage capacity and lifetime of the unit.
a battery pack (Fig. 41). A cell with a high charge is
In a battery pack consisting of series connected
indicated by its higher cell voltage. Excess energy is
cells, some cells have a diminished capacity owing to
removed through a bypass resistor until the voltage or
slight differences in manufacturing. When such
charge matches the voltage on the weaker cells. Some
a battery is subjected to charging cycle, the reduced
passive balancing schemes stop charging the battery
capacity cells reach full charge earlier than the other
pack at the instant when any one of the cells in the
cells in the battery, and there is a danger of
overcharging these degraded cells. The capacity of the
anomalous cells reduces even further with every suc- Vbat +
cessive charge/discharge cycle. The cumulative result is
a temperature and pressure buildup which paves the
way for an early failure of the cell. I1
v1
Once a cell has failed, the entire battery must be
replaced, and the consequences are extremely costly.
Replacing individual failed cells does not solve the
problem since the characteristics of a fresh cell would
I2
be quite different from the aged cells in the chain, and v2
failure would soon occur once more. Some degree of
refurbishment is possible by cannibalizing batteries
of similar age and usage, but it can never achieve the Control
level of cell matching and reliability that is possible I3
v3
with new cells.

Cell Charge Equalization

Battery management systems (BMS) for charge equal- I4
v4
ization monitor the state of charge of each cell. In low-
cost applications, a circuit to monitor individual cell
voltage may be employed. Switching circuits then con-
trol the charge applied to each individual cell in the Vbat −
series during the charging process to equalize the
charge on all the cells in the pack. Batteries, Battery Management, and Battery Charging
A good charge/discharge equalization technique Technology. Figure 41
should ideally have the following features: Passive charge balancing with a dissipative resistor
 Batteries, Battery Management, and Battery Charging Technology 153

pack reaches full charge. During discharge operation, Vbat +

they discharge the fully charged cells into a load until
I1
these cells reach the same charge level as the weaker
cells. These schemes lead to underutilization of the
A
battery pack.
Other schemes are designed to continue charging
v1
until all the cells are fully charged but to limit the
B
voltage which can be applied to individual cells and
to bypass the cells when this voltage has been reached. I2 IC1
The shortcomings of this technique are that this C1
method uses low bypass currents, and hence, equaliza- A
tion times are very long. Also, the battery pack perfor-
mance is governed by the weakest cell. This process also v2
causes power loss due to the bypass resistors which
could drain the battery if operated continuously. It is
I3 B
however very easy to implement and is a low-cost IC2
option. Passive balancing is adequate for both lead- C2
acid and nickel-based batteries, incorporating series A
string up to about six cells. Passive equalization cannot
be employed for charge equalization in Li-ion-based v3
battery strings.

Switched Capacitor The switched capacitor charge
equalization method, also called a flying capacitor
method, uses a capacitor to transfer charge between
cells in a battery pack. Typically, these are adjacent cells
in the series string as shown in Fig. 42. When the
capacitor is connected to a cell with the higher voltage,
the capacitor is charged. When the charged capacitor is
connected to a cell with a lower voltage, the capacitor
discharges and transfers charge into the new cell. When
switched between two cells repeatedly, this movement
of charge results in current from the cell with higher
voltage to the cell with lower voltage. Finally, cell volt-
ages of all cells in the string reach the same value and
voltage equalization is achieved. High-frequency
switching can be accomplished using a single-pole
double-throw switch implemented using two sets of
MOSFETs. The MOSFETs are switched with frequen-
cies ranging in several kHz. An important feature in the
switching scheme is that switches break before make to
prevent momentary short circuits of the respective
cells. Accordingly, the switch control signals incorpo-
rate a dead-band interval. This method is easy to
implement and it results in reduced losses. A major
drawback is that equalization process takes a long time
if voltage differences between adjacent cells are small.
I4 B
IC3
C3
A
v4
B
Vbat−
Batteries, Battery Management, and Battery Charging
Technology. Figure 42
Schematic of a switched capacitor equalizer
Multi-Output Transformer This method uses
a transformer with its primary winding connected
across the battery and one secondary winding which
can be switched across individual cells. Both flyback
and forward topologies are employed in practice.
Figure 43 illustrates a flyback-based circuit to carry
out charge balancing for a battery string. In the flyback
structure, when the main switch is switched on, some
154 Batteries, Battery Management, and Battery Charging Technology
Vbat +
v1
v2
v3
Controlled
switch
v4
Vbat −
Batteries, Battery Management, and Battery Charging
Technology. Figure 43
A schematic of a flyback-type inductive shuttle charge
distribution
energy is stored in the transformer magnetic field.
When the switch is turned off, the energy is transferred
to the secondary of the transformer, similar to the
flyback concept. The bulk of the energy is taken up by
cells with lowest cell voltage. The main difference of
this method over the switching capacitor scheme is that
this method involves taking pulses of energy from the
full battery, rather than small charge differences from
a single cell, to top up remaining cells. Averaging of the
charge level is similar to flying capacitor method, but it
avoids the problem of small voltage differences in cell
voltage and is consequently much faster. The trans-
former must have well-balanced secondary windings,
otherwise voltage imbalances occur. Also, care must be
taken to prevent transformer saturation.
DC-DC Converter Methods This technique involves
using power electronics-based converters to transfer
energy from higher charged cells to charge-efficient
cells in a battery string. Different converter topologies
such as the flyback, buck-boost, and Ćuk have been
investigated to transfer energy between the batteries.
Energy transfer may be unidirectional or bidirectional.
In unidirectional flyback mode, excess energy in the
higher charged cell is transferred to the magnetized
inductor of the transformer during turn-on operation
of the converter. During the turn-off mode, this energy
is returned to charge-deficient cells in the battery bank,
and thus, charge leveling is achieved. Bidirectional
converters are employed to facilitate local energy trans-
fer between two neighboring cells independent of their
respective cell voltage. The converters are operated in
discontinuous mode to minimize switching losses. The
disadvantage is that energy transmission efficiency is
low when energy transfer occurs from the first cell to
the last with many intermediate stages in a long string.
To avoid power loss in intermediate stages, non-
isolated buck-boost topologies have been explored in
which energy is transferred from the first cell to all
other cells in a bank without any intermediary.
Future Direction
Batteries and battery technologies are expected to
become even more important in the future as con-
sumers demand longer battery life from consumer
electronics; variable energy sources, such as wind and
solar, increase in prevalence in the electrical grid; and
hybrid and all-electric vehicles become commonplace,
to name a few applications currently driving technol-
ogy development. The emphasis will be to develop
batteries with low cost, high energy density, low
weight/volume, completely safe, environmentally
friendly, easily disposable, and made from abundant
raw materials. Auxiliary battery management systems
will become increasingly integrated into the battery
pack to create a true “plug and power” energy storage
module.
Lithium-based cells hold great promise for the
future. Lithium-ion cells have nearly proliferated the
consumer electronics market. These batteries are also
expected to find a prominent role as ideal electrochem-
ical storage systems in renewable energy plants, as well

as power systems for sustainable vehicles, such as
hybrid and electric vehicles. However, scaling up the
lithium battery technology for these applications is still
problematic since issues such as safety, costs, wide
operational temperature, and materials availability are
yet to be adequately resolved. Safety is an important
issue with Li-ion technology, and hence, it forms one of
the most important aspects of future research.
Lithium-copper cells are seen as the future prospect
for lithium-based batteries, although the technology
is still in a nascent state. Traditionally, cathodes of
Li-based cells have not been reusable and hence the
motivation for Li-Cu system which makes provision
for reusable electrodes.
Li-air batteries in which oxygen from the air serves
as the cathode and lithium as the anode are also being
explored and hold promise to increase energy density.
Estimates indicated that the lithium-air battery could
hold 5–10 times as much energy as a lithium-ion bat-
tery of the same weight and double the energy for the
same volume. In theory, the energy density could be
comparable to that of liquid fuels such as gasoline.
However, rechargeable Li-air batteries are still some
years away. Flow batteries hold promise for stationary,
high-energy applications such as power stations to
provide peak loads, backup energy in case of emergen-
cies, and source leveling to mitigate variable produc-
tion from renewable sources. Zn-Bromine, metal air,
vanadium redox are some of the batteries which are
being explored for UPS and grid storage applications as
an alternative to VRLA batteries, which is currently the
standard.
On the battery management side, mobile energy
platforms are driving push toward wireless charging
technologies and standards for contactless charging.
The Wireless Power Consortium has recently devel-
oped the standard for contactless charging called “Qi”
for low-powered electronic devices (i.e., less than
5 watts). This standard is suitable for cellular phones
and contains interface definition, performance require-
ments, and compliance testing. Further progress is
expected for more powerful chargers for higher-
power-consumption devices such as laptops. In the
future, all compatible devices with the “Qi” standard
will carry the “Qi” logo to show compatibility with
“Qi” transmitters. The final goal is to have portable
electronic devices charged with the stations with “Qi”
Batteries, Battery Management, and Battery Charging Technology 155
logo in airports, railway stations, hotels, etc. Currently,
the inductive contactless chargers can transmit power
in short distance and with the same speed as wired
chargers (e.g., cell phone applications). However,
capacitive battery chargers can transmit higher levels
of power in shorter time frame. Once the associated
costs are alleviated, capacitive contactless charges can
dominate the market. It has already started in applica-
tions such as a camera flash, where the ability to trans-
fer high power over a short period of time is important.
Inductive contactless chargers, however, will remain
prime candidates for vehicular-charging applications.
Intermittency of renewable energy systems intro-
duces some integration problems especially for high
penetration levels of wind and solar energy systems.
The battery energy storage systems are expected to play
an important role in addressing intermittency. Appro-
priate control strategies are under development for
charging systems to make wind and solar plant hourly
dispatched based on forecasted conditions. Bidirec-
tional converters are a key element here; the battery is
charged by the grid during low-demand hours or by
wind and solar plants during high-irradiance hours
and windy periods. The battery is then discharged to
the grid during shading of solar panels or lower wind
speeds.
Charging circuitry is also used in stations for elec-
tric and plug-in hybrid vehicles (EVs and PHEVs).
Three standardized charging levels are defined by the
National Electric Code for EVs and PHEVs. Level one
method uses 120 V, 15 A (12 A usable) power outlets
which are common in residential and commercial units
in the USA. However, they prolong the charging period
due to limited maximum power (about 1.44 kW). Level
two chargers use 240 V, single-phase, 40 A, outlets.
They are known as the “primary” or “preferred”
method of charging for PHEVs. There are two types
of level two chargers, namely, inductive and conductive
(wired). Although level two chargers enable faster
charging, they require more safety restriction. Level
three chargers use 480 V, 60–150 kW, three-phase
power lines. They are intended to charge battery in
about 10–15 min, also known as “fast charging”
method. One may argue that PHEV batteries can be
charged in 1–2 h using level two chargers, and thus,
there are less incentives for higher power level three
chargers in charging infrastructure for PHEVs.
156 Batteries, Battery Management, and Battery Charging Technology
Integrated battery chargers are an important aspect
when considering system-on-chip devices. System-on-
chip devices combine different parts of a system, which
may conventionally require their own chip, into a single
chip. For example, a typical system-on-chip sensor is
designed to sense and transmit the data from an inac-
cessible location to the main controller. It is comprised
of an integrated sensor, analog-to-digital converter,
micro-controller, memory, and RF transmitter. Usually
these system-on-chip sensors must be autonomous
systems and thus generate their own power for an
extended period of time. The required power for
these micro-sensors can be supplied by solar cells or
piezoelectric materials. Given the intermittency in the
output of solar cell, micro-batteries are required to
power the system during the low-light conditions.
Therefore, integrated battery charging circuits are one
of the essential components of system-on-chip sensors.
Integrated battery chargers must meet specific criteria,
namely, high efficiency, simple structure, and, most
importantly, small size. Size is the prevalent factor in
determining the total cost of the system in system-on-
chip application. Therefore, in the applications that
require extreme integration and low cost, battery is
also integrated in the chip together with charger circuit.
Bibliography
1. Aditya JP, Ferdowsi M (2008) Comparison of NiMH and Li-Ion
batteries in automotive applications. In: Proceedings, IEEE
vehicle power and propulsion conference. IEEE, Piscataway
2. Available online: http://www.eetimes.com/electronics-news/
4196006/Analyst-Super-capacitor-aided-wireless-charging-the-
key-to-powering-mobiles
3. Available online: http://www.technology.am/wireless-charg-
ing-022522.html
4. Available online: http://www.wired.com/autopia/2010/11/
ditch-the-cord-let-the-road-charge-your-ev/
5. Available online: http://www.wired.com/autopia/2011/03/
google-installs-a-wireless-ev-charging-station/
6. Barnes FS, Levine JG (2011) Large energy storage systems
handbook. CRC Press, Boca Raton
7. Beart P, Cheng L, Hay J (2006) Inductive energy transfer
system having a horizontal magnetic field. UK Patent
GB2,399,225
8. Bito A (2005) Overview of the sodium-sulfur battery for the
IEEE stationary battery committee. In: Proceedings, IEEE power
engineering society general meeting. IEEE, Piscataway, p 1232
9. Bullis K (2009) High-energy batteries coming to market:
rechargeable zinc-air batteries can store three times the
energy of a lithium-ion battery. http://www.technolo-
gyreview.com/business/23812/
10. Chia-Hsiang L, Chun-Yu H, Ke-Horng C (2010) A li-ion battery
charger with smooth control circuit and built-in resistance
compensator for achieving stable and fast charging. IEEE
Trans Circuits Syst 57(2):506–517
11. Cong-Ling H, Jie W, Miao Z, Jin-Ming Y, Jin-Peng L (2004)
Application of adaptive algorithm of solar cell battery charger.
In: Proceedings of the 2004 IEEE international conference on
electric utility deregulation, restructuring and power technol-
ogies (vol 2). IEEE, Piscataway, pp 810–813
12. Cope RC, Podrazhansky Y (1999) The art of battery charging.
Battery conference on applications and advance, the four-
teenth annual, California, pp 233–235
13. Crompton TR (1995) Battery reference book, 2nd edn.
Butterworth-Heinemann Ltd, Oxford/Boston
14. Hua CC, Lin MY (2000) A study of charging control of lead-acid
battery for electric vehicles. Industrial electronics ISIE, pro-
ceedings of the 2000 IEEE international symposium (vol 1),
Puebla, pp 135–140
15. Hua CC, Lin MY (2000) A study of charging control of lead-acid
battery of electric vehicles. IEEE proceedings. of ISIE (vol 1),
Dallas, pp 135–140
16. Hui SYR (2006) Planar inductive battery charger. UK Patent
GB238,972
17. IEEE Std 519 (2009) IEEE recommended practices and require-
ments for harmonic control in electrical power systems. IEEE,
New York
18. Jaganathan SH (2009) Battery charging power electronics and
control for plug-in hybrid electric vehicle. Thesis, Tennessee
Technological University, pp 94–103
19. Kamibayashi M, Tanaka K (2001–2002) Recent sodium sulfur
battery applications. In: IEEE transmission and distribution
conference and exposition (vol 2). IEEE/PES, Atlanta,
pp 1169–1173
20. Köhler U, Kümpers J, Ullrich M (2002) High performance nickel-
metal hydride and lithium-ion batteries. J Power Sourc
105:139–144
21. Kiehne HA (2007) Battery technology handbook, 2nd edn.
Chapter 12, Charging methods and techniques: general
requirements and selection of chargers, Dekker
22. Kline M (2010) Capacitive power transfer. M.S. thesis, Depart-
ment of Electrical Eng. and Computer Science, University of
California, Berkeley
23. Kodali VP (1996) Engineering electromagnetic compatibility.
IEEE Press, New York, pp 345–348
24. Koenig AA, Rasmussen JR (1990) Development of a high spe-
cific power sodium sulfur cell. In: Proceedings IEEE 34th inter-
national power sources symposium, Cherry Hill, pp 30–33
25. Krein PT, Balog RS (2002) Life extension through charge equal-
ization of lead–acid batteries. In: Proceedings, IEEE interna-
tional telecommunications energy conference (INTELEC). IEEE,
Piscataway, pp 516–523
26. Kuhn BT, Pitel GE, Krein PT (2005) Electrical properties and
equalization of lithium-ion cells in automotive applications.

In: Proceedings, IEEE vehicle power and propulsion confer-
ence. IEEE, Piscataway, p 5
27. LeFevre BW (2003) Integrated microbattery charger for auton-
omous systems. Birgham Young University, Provo
28. Lex P, Jonshagen B (1999) The zinc/bromine battery system for
utility and remote area applications. Power Eng J 13:3
29. Li P, Bashirullah R (2010) A wireless power interface for
rechargeable battery operated medical implants. IEEE Trans
Circuits Syst 54(10):912–916
30. Liawt CM, Chiangt SJ, Huang SC (1994) A three-phase multi-
functional battery energy storage system. Proceedings of
IECON 94 Conference. IEEE, New York, pp 458–463
31. Linden D, Reddy TB (2002) Handbook of batteries, 3rd edn.
McGraw-Hill, New York
32. Liu X, Ron Hui SY (2007) Equivalent circuit modeling of a
multilayer planar winding array structure for use in a universal
contactless battery charging platform. IEEE Trans Power
Electron 22(1):21–29
33. Marshman C (1992) Guide to the EMC directive. 89/336/EEC.
IEEE Press, Piscataway
34. Mohan N, Undeland TM, Robbins WP (2003) Power electronics
converters, applications and design, 3rd edn. Wiley, New York
35. Montrose MI (1996) Printed circuit design techniques for emc
compliance. IEEE Press, New York
36. Moore SW, Schneider PJ (2001) A review of cell equalization
methods for lithium ion and lithium polymer battery systems.
SAE Technical Paper Series. pp 1–5
37. Morrow K, Karner D, Francfort J (2008) Plug in hybrid electric
vehicle charging infrastructure review. US department of energy
vehicle technologies program – advanced vehicle testing activ-
ity. Battelle Energy Alliance, The Idaho National Laboratory
38. Palanisamy T (1990) Charging techniques for a universal lead-
acid battery charger. Proceedings of the 34th international
power sources symposium. IEEE, New York, pp 72–76, 25–28
Batteries, Battery Management, and Battery Charging Technology 157
39. Plett G (2002) Kalman-Filter soc estimation for LipbHev cells.
In: Proceedings of the 19th international battery, hybrid and
fuel cell electric vehicle symposium & exhibition (EVS19).
Busan, pp 19–23
40. Poon NK, Pong MH, Tse CK (2003) A constant-power battery
charger with inherent soft switching and power factor correc-
tion. IEEE Trans Power Electron 18(6):1262–1269
41. Pop V, Bergveld H, Notten P, Regtien P (2005) State-of-the-art
of battery state-of-charge determination. Meas Sci Technol
16(12):93–110
42. Rashid MH (2006) Power electronics handbook, 2nd edn.
Academic, California
43. Redl R (1996) Power electronics and electromagnetic compat-
ibility. Power electronics specialists conference, PESC ’96
record. 27th annual IEEE (vol 1). IEEE, New York, pp 15–21,
1:23–27
44. Tamyurek B, Nichols DK, Demirci O (2003) The NAS battery:
a multifunction energy storage system. In: Proceedings IEEE
power engineering society general meeting (vol 4). IEEE,
Piscataway, pp 1991–1996
45. Tamyurek B, Nichols DK, Vollkommer HT (2003) Sodium
sulfur battery applications. In: Proceedings IEEE power engi-
neering society general meeting (vol 4). IEEE, Piscataway,
pp 2323–2324
46. Vrankovic Z, Abedini AZ, Nasiri A (2007) A novel battery char-
ger for automotive applications. IEEE conference of PESC’07.
IEEE, Piscataway, pp 1465–1470
47. Wiesener K, Ohms D, Benczúr-Ürmössy G et al (1999)
High power metal hydride bipolar battery. J Power Sourc
84(2):248–258
48. Zhi-Guo K, Chun-Bo Z, Ren-Gui L, Shu-Kang C (2006) Compar-
ison and evaluation of charge equalization technique for
series connected batteries. In: Record, IEEE 37th IEEE power
electronics specialists conference (PESC), Jeju, pp 1–6
158 Battery Technologies

Battery Technologies limitations of the batteries to drive them have not

enabled them to compete in the general consumer’s
RAMAKRISHNAN MAHADEVAN1, MOHIT CHHABRA2, market with the internal combustion engine. Automo-
PUNEET PASRICH3, FRANK BARNES3 tive parts are limited by space and weight. Therefore
1
Telecommunications, University of Colorado, EVs and hybrid electric vehicles (HEVs) require
Boulder, CO, USA a battery with high energy density. Specific energy is
2
Electrical Engineering, University of Colorado, defined as the energy per kilogram of the battery while
Boulder, CO, USA energy density is the energy per unit volume. Early
3
Electrical, Computer, and Energy Engineering designs (1990–1995) of lithium-ion batteries only had
Department, University of Colorado, Boulder, a specific energy of 0.2 kWh/kg. EVs also require a bat-
CO, USA tery with high power output for large power draw, such
as quick acceleration. The General Motors EV1 had
Article Outline a battery pack weighing almost 600 kg while the car
weighed 1,350 kg. The battery would then account for
Definition of the Subject
more than 44% of the car’s weight. A marked improve-
Introduction
ment in energy cell density of batteries was required for
Basic Characteristics of Batteries
practical EVs.
Electrical Modeling for Batteries
Lead Acid Batteries
Nickel-Cadmium Battery Introduction
Nickel Metal Hydride Batteries
In this paper, the first section covers the relationship
Lithium-Ion Batteries
between the internal resistance of the cell, maximum
Metal–Air Batteries
power output, energy efficiency in charging and
Zinc Bromine Battery
discharging, and reduction of cell heating. This is followed
Supercapacitors
by a review of the characteristics of lead acid, nickel metal
Recommendations for Future Work
hydride, lithium-ion batteries, and supercapacitors.
Bibliography
Recent advances in power electronics provide com-
petitive performance for electric cars so they provide
Definition of the Subject
the necessary acceleration for wide spread adoption
The demand for high performance and long life storage [1]. However batteries still lack the energy density of
systems has lead to numerous research initiatives, gasoline or diesel fuels and thus cannot compete in
aimed at the development of such systems. Develop- terms of the overall range that can be driven between
mental paths are aligned with the requirements of the recharging. Additionally, recharging may be slow.
applications of these systems. A detailed understanding However, ongoing research and improvements in
of technical characteristics and cost considerations is manufacturing may reduce both energy density limita-
provided in this paper for various battery chemistries: tions and costs. Concerns about greenhouse gases and
contemporary, vintage, and prospective. A clear under- other air pollutants are the other driving factors for the
standing of battery characteristics is essential for guid- renewed interest in EVs and HEVs, which may reduce
ing the selection of such batteries. Indeed, because of or eliminate the emission of NOx, CO2, and hydrocar-
the critical functions fulfilled by batteries as well as bon pollution from mobile vehicles depending on how
the substantial costs of advanced batteries, a realistic the electricity used to charge the battery was generated.
appraisal of candidate battery performance and costs EVs and HEVs also reduce noise levels when running
against requirements is key to judging the prospects of on electric mode. In 2009, 1.6 million HEVs were
new applications such as EV, HEV, and plug-in hybrid registered in the USA [2].
electric vehicles (PHEV). Although electric vehicles This section concentrates on a variety of battery
(EVs) have been around since before 1900 the technologies that have shown significant promise for

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
 Battery Technologies 159

long life and high performance. A brief, generic Basic Characteristics of Batteries
list of battery characteristics are provided in
An electrical battery is a power source composed of one
section “Electrical Modeling for Batteries”.
or more electrochemical cells which convert stored
Section “Lead Acid Batteries” introduces various
chemical energy into electrical energy. In this section,
types of lead acid battery chemistries. Section “Nickel
rechargeable batteries which can be used multiple times
Metal Hydride Battery” introduces the nickel metal
is discussed. A battery is composed of one or more
hydride chemistry, while lithium-ion is discussed in
electrochemical cells, each consisting of two electrodes
section “Lithium-Ion Batteries”. The upcoming
and one (or two) electrolyte-containing anions and
technology of supercapacitors is also introduced and
cations.
a comparison to conventional electrochemical
Figure 1 depicts the typical electrochemical cell volt-
counterparts is provided.
age behavior during charging (labeled “Charge curve V”)
This entry first gives a general overview of
and discharging (labeled “Discharge curve V”). Cell volt-
characteristics of batteries, followed by the electrical
age drops and rises during discharging and charging
modeling used for batteries in EVs and concludes by
respectively, in proportion to the cell’s internal voltage.
presenting the different battery technologies used
The difference between the terminal open-circuit voltage
today.

(1) High power output

Low internal
(2) Good energy efficiency
resistance

(3) Suppression of cell heat generation

Fully charged
region of increased
resistance near end of
V charging Charge curve V
Terminal open-circuit voltage V
Charge curve V
Internal resistance

ΔV = I·R

Discharge curve V Current value

Region of increased internal

resistance during discharging

Q
Δq

Battery Technologies. Figure 1

Cell voltage profile and internal resistance in charging and discharging operations with respect to cell charge Q [3]
160 Battery Technologies
and the discharge and charge voltage is attributed to the
cell’s internal resistance. An ideal cell provides constant
terminal voltage until empty. The internal resistance of
real cells prevent ideal performance during discharge
resulting in lower terminal voltage.
The essential characteristics of batteries are maxi-
mum power output, charging/discharging efficiency,
and temperature effects. These are key factors to be
considered when designing or choosing batteries for
particular applications. These characteristics are fur-
ther elaborated for each battery in the device specific
sections of the paper. The basic definitions are shown
below.
Maximum power output: A lower limit is set for each
cell’s voltage. The maximum dischargeable current value
is determined such that the cell voltage does not fall to
this limit. The maximum current is determined as
Imax ¼ ðOpen Circuit Voltage
 Lower Limit VoltageÞ=Internal Resistance
Then, maximum power output of a cell can be
calculated as Pmax = Imax ● (Open Circuit – Lower
Limit Voltage). High power output requires low inter-
nal cell resistance.
Charging/discharging efficiency: Electrical energy
transferred during a charging/discharging cycle can be
computed as
E ¼ DQ  V
where DQ is the transferred charge and V is the termi-
nal voltage at that time.
In Fig. 1, the area between the charge curve and the
horizontal axis is the total electrical energy transferred
to the battery. Accordingly, the area between the dis-
charge curve and horizontal curve is the total energy
drained from the battery. Battery efficiency is a ratio of
discharge to charge energy values. It is imperative to
reduce the difference between these two curves to
improve battery efficiency.
Temperature effect: Charging and discharging cycles
can cause a cell’s temperature to rise. For lithium-ion
cells, operating temperatures above 50 C can lead to cell
degradation. Ozawa [4] attributes the difference in charg-
ing and discharging curves in Fig. 1 to heat lost due to
different operating temperatures. To suppress this dif-
ference, the cell’s internal resistance must be reduced.
Electrical Modeling for Batteries
A simple model of a battery can help alleviate issues of
understanding of a battery’s behavior. The simplest
electrical equivalent model of a cell is a voltage source
with a resistor in series, as shown in Fig. 2a. This is the
internal resistance of the cell. Davide [5] provides
corresponding voltage and current response curves
for each proposed model.
A more complex model is described in Fig. 2b. This
model emulates the behavior of a battery when instan-
taneously loaded. The initial voltage drop across bat-
tery terminals is small, due to R1. The voltage then
drops exponentially to the level of both resistances
with a time constant T = R2 ● C2 (1 min). This
phenomenon is known as relaxation.
A variation on 2b is 2c, where the time constant
T = R3 ● C3 is 1 ms. This model correctly emulates AC
impedances on the order of 1 kHz. This property can
however be misleading to the user since such imped-
ances are usually measured with no load by the manu-
facturer. The user may also misread this impedance as
the cell’s DC resistance. Figure 2(d) combines 2(b) and
2(c) and thus equally satisfies both the manufacturer
and the user.
Internal resistance of a cell results from a series com-
bination of resistances due to bulk metal, chemical pro-
cesses, and terminals. Varying loads can have dynamic
effects on these properties of a battery. Thus the battery’s
internal resistance (Ri) also has dynamic properties.
Davide explains that internal resistance can vary with
the following battery characteristics:
State of Ri is high for both high and low state of
charge: charge levels
Temperature: Ri is higher at cold temperatures
Current: Ri is higher at higher currents when
charging (compared to the same current
when discharging)
Usage: Ri increases over time with cell usage
Figure 3 shows how the internal resistance of a lith-
ium iron phosphate (LiFePO4) battery varies versus the
above four battery characteristics.
For preliminary research purposes, it is entirely
possible to create a generic model and vary key
 Battery Technologies 161

R1
Voltage
+
Current
Time
a
C2
R1
Voltage
+ R2
Current
Time
b
C3
R1
Voltage
+ R3
Current
Time
c
C2 C3
R1
Voltage
+ R2 R3
Current
Time
d

Battery Technologies. Figure 2

Electrical modeling schemes for a battery [5]

Resistance Resistance Resistance Resistance

Charging Discharging Charging

Discharging

@ 25°C @ 50% SOC @ 25°C @ 25°C

50% SOC 1C discharging 1C rate 50% SOC
new cell new cell new cell 1C discharging

−1C 0 1C −20 0 60 0 100 0 1000

a Current [A] b Temperature [°C] c SOC [%] d Cycles

Battery Technologies. Figure 3

Internal resistance variation for a LiFePO4 battery [5]

parameters to simulate the performance of various
battery chemistries. Table 1 provides a list of key
parameters for three batteries chemistry as well as
three types of supercapacitors.
Detailed reviews of various battery chemistry com-
positions are provided in the following sections. We
consider not only relevant chemical reactions, but also
applications, cost, and efficiency concerns associated
with every chemistry type. The following section details
lead acid batteries, their cell construction, and the
electrochemistry, for several types of lead acid batteries
used in EVs and their characteristics.
Lead Acid Batteries
Lead acid (PbA) batteries are inexpensive and have
characteristics somewhat suitable for energy storage
in EVs. Design is possible for short, high-rate
162 Battery Technologies

Battery Technologies. Table 1 Performance parameters

Nickel metal
Lead acid hydride Lithium-ion Super capacitors
Electrolyte H2SO4 + KOH LiPF6 Aqueous Organic Aqueous
H2O electrolyte electrolyte electrolyte
Anode PbO2 M (Metal) Graphite Carbon Carbon Metallic
Cathode Pb Ni(OH)2 LiMn2O4 Carbon Carbon Metallic
Cell voltage (V) 2.4 1.2 4.0 1 3 1
Specific energy (Wh/kg) 35 50–80 250–400 0.2–1.3 3–6 1
Relaxation time constant (s) 5 2 1 N/A N/A N/A
1 3 1 3 3
AC impedance time constant (s) 10 –10 10 –10 10 N/A N/A N/A
Recommended operating 25 20 to 50 50 25 to 85 40 to 85 30 to 70
temperature ( C)

discharges or for bulk energy storage. These batteries
are commonly used to power the starter motor of
conventional internal combustion engines.
Lead acid batteries are classified into two broad
categories, traditional vented lead acid (VLA) batteries
and valve-regulated lead acid (VRLA) batteries. The
latter have dominated the stationary battery market
over the past 20 years [1]. VRLA batteries have
immobilized electrolyte and a means to recombine
charged gas, thus conserving water in the cell. They
are typically smaller than their vented counterparts
and require less maintenance. Most of the VLA and
VRLA batteries use pasted plate construction, in which
a mixture of active materials is pasted onto lead alloy
grids. The characteristics of the final battery are heavily
influenced by the alloy composition. Tubular plate
construction is also quite common, in which the active
material is held in non-woven fibrous tubes with
a center conductor of lead alloy.
Lead acid batteries have less energy density and
specific energy as compared to the more recent lith-
ium-ion and NiMH batteries but these batteries gener-
ally have high ampere rating. Since lead acid batteries
do not have the deep cycling capability and energy
density compared to other batteries used in EVs, they
are being replaced by NiMH and Li-ion batteries for
traction purposes. However earlier EVs used lead acid
batteries like the GM EV1 (Generation 1). The Delphi
704 VRLA-Prismatic battery has an ampere rating of
53 Ah and a specific energy of 35.3 Wh/kg. These
batteries have a good cycle life of 15,000–30,000 cycles.
In the following section, the cell construction and elec-
trochemistry of lead acid batteries are discussed [4].
Cell Construction and Electrochemistry
All lead acid designs share the same basic chemistry. The
positive electrode (cathode) is formed of porous lead
dioxide (PbO2) while the negative electrode (anode) is
composed of sponge lead (Pb). Sulfuric acid (H2SO4)
in water serves as gelled electrolyte between the two
electrodes. Lead dioxide at the cathode is supported by
a thin lead grid. Figure 4 provides a cross-sectional view
of a standard lead acid battery cell.
Pure lead is generally too soft to be used as a plate
material and usually requires addition of an alloy. Alloy
candidates vary and are responsible for positive plate
subtypes such as lead-antimony, lead calcium, and pure
lead. Lead-antimony cells are recommended for appli-
cations requiring very long life and discharging to
depths greater than 20% of rated capacity. Pure lead
cells are recommended for float and shallow cycling
service where average discharge depth is less than
20%. Finally, pure lead alloy cell types are used when
very low charged stand loss is a requirement in the
application and occasional deep cycles are expected [6].
During charging, as the cell approaches full charge,
a majority of the PbSO4 is converted to Pb and PbO2.
Beyond this point and under high charging voltages,
cell voltage becomes greater than the gassing voltage
(about 2.39 V per cell) and the overcharge reactions

Lead-acid cell charging
+ –
Gen
I a cycle from approximately 2.1 V at the top of charge to
+ –
PbO2
electrode Pb electrode
electrons
electrolyte:
H2SO4 + H2O
Battery Technologies. Figure 4
Charging lead acid cell schematic [3]
begin. This results in the electrolysis of the water
contained in the electrolyte, and the production of
hydrogen and oxygen. In sealed lead acid cells, this
reaction is controlled to minimize hydrogen evolution
and loss of water by recombining evolved oxygen with
the negative plate.
The discharge chemical reactions are as follows:
Positive: PbO2 þ 4Hþ þ SO4 2 þ 2e
! PbSO4 þ 2H2 O
Negative: Pb þ HSO4  ! PbSO4 þ Hþ þ 2e
Overall: PbO2 þ Pb þ 2H2 SO4 $ 2PbSO4 þ 2H2 O
As the cell discharges, lead at the anode reacts with
hydrogen sulfate ions to form lead sulfate, along with
hydrogen ions and electrons. These electrons then
travel through the external circuit. Note that sulfuric
acid is a strong electrolyte and breaks down into hydro-
gen and hydrogen sulfate ions even before being placed
in the battery. This breakdown is independent of the
battery plates. At the cathode, lead dioxide is reduced
by the two electrons released at the anode and the lead
ion then reacts with sulfate to form lead sulfate along
with water. Discharge continues until all of the sulfuric
acid is converted to water. Discharging is an exothermic
process and charging is an endothermic process.
Battery Technologies 163
As discharging progresses, the concentration of the
sulfuric acid electrolyte falls. As a result, the no-load
(open circuit) voltage of a lead acid cell varies during
approximately 2 V at the bottom of discharge. In addi-
tion, the decreasing concentration of electrolyte leads
to an increasing internal resistance that reduces overall
efficiency of the cell. All these factors come together to
signify for a given current, the terminal voltage of a lead
acid cell is lower at the end of discharge than at the
beginning [3].
Lead is insoluble and stays at either plate during
charging or discharging. This property allows
recharging of the battery. In contrast, non-rechargeable
batteries contain reaction products that either convert
to an insoluble precipitate or are lost as gas.
Two side reactions can occur in lead acid cells,
primarily at the top of charge: hydrogen production
from the lead negative plates and oxygen generation
from the lead dioxide positive plate. In flooded cells,
these side reactions lead to a loss of water that must be
made up periodically. The loss of water is particularly
high during overcharge and equalization of batteries of
cells, which must be performed each full charge and
once every 10 or 20 cycles, respectively. Overcharge and
equalization both involve charging the lead acid cells
somewhat beyond the ampere-hour capacity that has
been discharged, so as to ensure that all the electrodes
of all the cells of the battery are fully and equally
recharged [3].
The next section gives a brief description of the
different types of lead acid batteries.
Types of Lead Acid Batteries (by Electrode Material)
Lead Calcium Batteries Lead acid batteries with
electrodes that are modified by the addition of calcium
are more resistant to corrosion, overcharging, gassing,
water usage, and self-discharge. All of these factors
shorten the life of a battery. The increased resistance
increases the battery life of lead calcium batteries.
Additionally, they have a larger electrolyte reserve area
above the plates, higher cold cranking ampere ratings,
and require little maintenance [4].
Lead-Antimony Batteries Lead acid batteries with
electrodes modified by the addition of antimony
improve the mechanical strength of electrodes which
164 Battery Technologies
is important for EV and deep discharge applications.
Water loss and the internal heat are also minimized.
They have a longer life than lead calcium batteries.
These batteries are also easy to recharge when they are
completely discharged and are comparatively less
expensive compared to other types of lead acid batte-
ries. However, lead-antimony batteries have a higher
self-discharge rate typically 2–10% per week, whereas
lead calcium batteries have a self-discharge rate of
1–5% per month [4].
Advanced Lead Acid Batteries In the early part of the
decade, the lead acid battery community formed the
Advanced Lead–Acid Battery Consortium (ALABC) in
a concerted effort to make electric vehicles a reality by
overcoming the shortcomings of the VRLA battery.
Much has been achieved by the global research and
development program of the ALABC [7].
The ALABC originally developed advanced lead
acid batteries as a replacement for the more expensive
NiMH batteries used in hybrid electric vehicles. These
advanced lead acid batteries exhibit better performance
as compared to VRLA batteries, in terms of stationary
energy storage [8]. Some of the results of research in
advanced lead acid batteries are discussed below.
Swelling of the positive active material in the direc-
tion normal to the plane of the plate remains a serious
concern for VRLA batteries. The tendency for the pos-
itive active material to expand with repeated deep
cycling is, however, beyond dispute. Experimental
work at the University of Brno demonstrated a clear
correlation between loss of capacity with cycling and
increase of active-material resistance which is pre-
sumed to arise as a result of swelling [7].
An in-depth study of positive-grid alloys demon-
strated that the introduction of tin brings additional
benefits in the form of enhanced corrosion resistance
and reduced electronic resistance. Positive grids that
contain no antimony but have 1–1.5% by weight of
tin do not suffer expansion in the plane of the plate
and as a result of the lower corrosion rate, it is
possible to contemplate a substantial reduction in
grid thickness and weight in order to increase the
specific energy [7].
Also, the ALABC development work discovered that
adding a small amount of carbon (2–4% by weight) to
the negative active material of the electrode minimized
negative plate sulfation and enhanced VRLA battery
cycling performance in partial-state-of-charge cycling.
Battery developers/suppliers who are currently involved
in research and development of advanced lead acid
batteries include East Penn, North Star, Exide technolo-
gies, Axion Power, and Furukawa [8].
Types of Lead Acid Batteries (by Construction)
Flooded/Vented Lead Acid Batteries These batteries
have electrodes or plates which are immersed in the
electrolyte. The level of the electrolyte reduces as
a result of charging and the gases formed which are
vented to the atmosphere, so distilled water is added to
bring the electrolyte back to its initial level. The most
familiar example of a flooded lead acid cell is the 12 V
automobile battery.
Valve Regulated Lead Acid (VRLA) Batteries VRLA
batteries are also called sealed lead acid (SLA) batteries.
They are designed to prevent electrolyte loss through
evaporation, spillage, and gassing, which in turn pro-
longs the life of the battery and reduces maintenance
cost. Instead of simple vent caps on the cells to let gas
escape, VRLA batteries have pressure valves that open
only under extreme conditions. Hydrogen and oxygen
are generated by galvanic action of the battery during
charging. VRLA batteries need an electrolyte which
reduces gassing by impeding the release of these gases
to the atmosphere. A catalyst is usually involved that
causes the hydrogen and oxygen to recombine into
water. This system is called a recombinant system.
Since acid electrolyte spillage is eliminated, these bat-
teries are safer [4]. There are two categories of VRLA
batteries.
Types of VRLA
AGM (Absorbed Glass Mat) Batteries AGM batteries
are similar to VRLA batteries. A boron silicate
fiberglass mat acts as a separator between the
electrodes and absorbs the free electrolyte. The main
purpose of this type of battery is to recombine the
hydrogen and oxygen generated during the charging
process. The fiberglass mat absorbs and immobilizes
the electrolyte, but keeps it in a liquid rather than a gel
form. In this way, the electrolyte is more readily available
to the plates allowing faster reactions between the

electrolyte and the plate material which gives higher
charge/discharge rates as well as deep cycling
capability. AGM batteries are very robust and can
withstand severe shock and vibrations. Also, the self-
discharge rate is low, typically 0.25–0.75% per month.
They are also called “starved electrolyte” or “dry”
batteries because the fiberglass mat is 95% saturated
with the electrolyte and there is no excess liquid [4].
Gel Cell The gel cell is similar to the AGM battery,
except that the electrolyte is gelled to immobilize it. The
electrolyte is mixed with a silica compound to create
the gelled solution. It uses an alternative recombinant
technology to promote recombination of the gases
produced during charging. It also reduces the
possibility of spillage of the electrolyte. Charging rates
are limited, because overcharging may cause excess of
gases to be released causing damage. The gel cell cannot
be fast charged on a conventional automotive charger
since it can cause permanent damage [4].
Choosing a Battery
The ideal EV battery should have a constant output
voltage at any value of current drawn over the entire
state-of-charge (SOC) range and it should also accept
high charge rates at the same voltage. No battery has
such ideal characteristics but the suitability of a battery
for EV use can be determined by the ratio of charging
voltage to discharging voltage over the range of SOC
and current [9].
Given below are the important factors to consider
while choosing a battery for EV use.
Power Battery power performance is generally spec-
ified by manufacturers as W/kg at either 2/3 or 1/2 of
the open circuit voltage. This ignores the ratio of the
terminal voltage under load to that during charge or
regenerative braking which is the relevant parameter
for EV applications. A more useful figure of merit for
comparing the power performance of EV batteries is
the power density (W/kg) at 75% end-of-charge volt-
age (Veoc). For example, if the battery voltage of an EV
drops below 75% of Veoc during acceleration, it means
that the battery is operating at low cycle efficiency.
Another consideration is the cost of the vehicle drive
train which is proportional to the product of the max-
imum voltage rating and the current rating. A low
Battery Technologies 165
power battery needs more expensive drive train com-
ponents to achieve a given level of performance because
it needs higher current to extract the same power at
a lower voltage. A high power battery not only
increases vehicle performance but it reduces the cost
of the motor and electronics for equivalent
performance.
Energy Energy capacity of an EV battery determines
the vehicle range. The two parameters which are com-
monly used to specify energy capacity are specific
energy and energy density. Reducing energy consump-
tion is often a better way to increase EV range than
increasing battery capacity. EV batteries cannot yet
match the energy density of gasoline but provide
power to exceed the acceleration of conventional vehi-
cles. One of the unique features of EVs is the capability
to provide high peak power and excellent acceleration
without sacrificing operating efficiency [9]. The Tesla
Roadster sports car demonstrates this by operating at
180 Wh/mile in normal driving and offering accelera-
tion from 0 to 60 mph in 3.7 s, which is quicker than
a gasoline-powered Mercedes-Benz SLR McLaren. The
first-generation EV1 used Delphi’s valve-regulated
lead acid batteries, which is capable of operating at
164 Wh/mile. Nickel metal hydride batteries (NiMH)
batteries (See section “Nickel-Cadmium Battery”) have
twice the energy density and specific energy compared
to lead acid batteries. An EV fitted with a NiMH battery
which is of the same size and weight as a lead acid
battery has twice the range. However, due to power
considerations, it is not necessary that an EV could be
built to match the range of a lead acid battery powered
EV using a NiMH battery with half the size and
weight [9].
Cost Although lead acid batteries have less energy
density and specific energy and are bulkier compared
to NiMH and lithium-ion batteries, they are typically
inexpensive. Today, a lead acid battery pack for
an EV costs $215/kWh, a NiMH battery costs
$350/kWh, and a lithium-ion battery costs
$400/kWh [9]. The lower cost of the lead acid battery
makes it an affordable technology and also decreases
the overall cost of EVs. Axion Power, a firm based in
Pittsburgh, Pennsylvania, has come up with a new
166 Battery Technologies

technology called lead-carbon battery which is derived cooling is therefore required to maintain the battery
from traditional lead acid battery technology. Axion within a specific temperature range to achieve opti-
Power has managed to convert a pickup truck to run on mum performance [4].
a pack of the lead-carbon battery for around $8,000
[10]. If further cost reductions are achieved in the Life Battery life represents an important parameter
production of batteries, then EV pricing can become in battery performance that affects customer satisfac-
competitive with conventional vehicles. tion and overall cost. Battery life depends mainly on the
number of cycles. Technical factors such as temperature
Temperature Effects on Batteries Batteries function control, charging procedures, and exposure to abusive
because of electrochemical reactions (charging and conditions also play an important role in determining
discharging) taking place in the cell. These chemical the battery life. Lead acid batteries in EVs may last
reactions are dependent on temperature. Nominal bat- 15,000–30,000 cycles if used regularly. However, they
tery performance is usually specified for working tem- start deteriorating after 2 years regardless of the kind of
peratures between 20 C and 30 C. However the actual use. Lead acid batteries respond well to battery man-
performance can deviate substantially if the battery is agement systems that control temperature and main-
operated at higher or lower temperatures. The opera- tain SOC equalization. A factor related to battery life is
tion of any battery generates heat due to the I2R losses the battery residual value. Although virtually all EV
as current flows through the internal resistance of the batteries are recycled for economic and environmental
battery whether it is being charged or discharged. This reasons, EV batteries may retain value greater than
is also known as electrical heating or Joule heating. their salvage value after their useful life as a traction
During discharge, since the total energy within the battery is complete. AC Propulsion regularly sells spent
system is fixed, the temperature rise is limited by the traction modules as starter batteries with a 2-year war-
available energy. Another factor which affects the per- ranty for $25 or about $50/kWh. These modules are
formance of a battery is the ambient temperature. If the retired because their capacity is diminished by 30% or
ambient temperature is higher than the temperature of more, but they serve well as starter batteries [9].
the battery, the battery gains heat from the surrounding Table 2 below shows some typical specifications of
and if the ambient temperature is lower, the battery VRLA batteries used in EVs.
losses heat to the surrounding. The difference in the
ambient and the optimal operating temperature of
a battery reduces its performance. Overheating may
also be caused in batteries because of highly exothermic
reactions, which generate a lot of heat and may ulti-
Battery Technologies. Table 2 Specifications of VRLA
mately damage the batteries [4].
batteries in EVs [11, 12]
Lead acid batteries are exothermic during charg-
ing, which makes their temperature rise dramatically. 1998 Ford GM EV 1,
If the temperature exceeds beyond a limit, the battery Electric vehicle Ranger EV first model
may be damaged. Also, being a galvanic cell, its Manufacturer Delphi Delphi
internal resistance is temperature dependent. Internal Number of modules 39 26
temperature decreases as the temperature rises due to
Weight of module (kg) 19.3 18.8
the increase in electron mobility. The cell is very
inefficient at low temperatures but the efficiency Nominal module voltage (V) 8 12
improves at higher temperatures due to the lower Nominal system voltage (V) 312 312
internal impedance and also because of the increased Nominal capacity (Ah) 60 (100% 53 (50%
rate of chemical reactions. However, the lower inter- capacity) capacity)
nal resistance also causes the self-discharge rate to Specific energy (Wh/kg) 23.7–25.2 26.3–31.9
increase. Additionally, the cycle life deteriorates at
Efficiency (Wh/mile) 237–356 115–164
high temperatures. Some form of heating and
 Battery Technologies 167

Lead Acid Battery Circuit Model opposite direction when compared to charging, leading
to a minus sign before the integral. Division by 3,600 is
Consider Fig. 5 as a lead acid cell model. We construct
needed to ensure that cell capacity is in units of watt-
this circuit using PSpice software. Consider the cell to
hours, Wh. Alternate charging and discharging opera-
provide 2 V nominal voltage, 10 Wh capacity,
tion is conducted by applying AC current IAC at 4 A,
round-trip efficiency of 90%, and 70% initial state of
100 Hz, as shown in Fig. 16. Resistor R1 is chosen
charge. Here, available energy during operation is com-
arbitrarily high to ensure there is no voltage drop at
puted as
the cell’s output terminals.
ð Figure 6 shows the state of charge (SOC) of the cell.
1 kV1 Ibat
SOCðtÞ ¼ SOCinitial þ dt Note that the Y-axis provides state of charge as
CapacityðWhÞ 3600
a percentage. Figure 7b provides cell voltage V1,
where k is cell efficiency, V1 is cell voltage, and Ibat is cell whereas IAC is shown in 7a.
current. During discharge, current flows in the Note that in addition to the nominal 2 V, cell
voltage is a function of state of charge of the cell. For
example, a certain degree of overvoltage can occur with
U22 R1 high state of charge. A 70% state of charge leads to a cell
0 2
1 3 1 (over) voltage of 2.1 V in 7b.
IOFF = 0 + I1 2 100k
IAMPL = 4 – BAT
When IAC is positive, current travels into the cell,
FREQ = 100 simulating charging. This is evident in Fig 18b where
V1 drops while IAC increases in the positive direction.
0
Negative IAC implies cell discharge, with V1 above the
nominal 2 V. An immediate advantage of lithium-ion
Battery Technologies. Figure 5 batteries is their diversity in operation. Figure 19 shows
Lead acid cell model using PSpice only a few possible chemical combinations.

Battery Technologies. Figure 6

Cell state of charge
168 Battery Technologies
Battery Technologies. Figure 7
Applied AC current and cell voltage
Nickel-Cadmium Battery
The nickel-cadmium battery commonly abbreviated as
NiCd battery is a type of rechargeable battery which
works by oxidizing nickelic hydroxide (Ni(OH)2) into
nickelous hydroxide (NiOOH) and has metallic cad-
mium as electrodes. The charge/discharge reaction can
be given as follows:
2NiðOHÞ2 þ CdðOHÞ2 ! 2NiOOH þ Cd þ H2 O
The features of this battery are that the electrolyte
does not take part in any of the chemical reactions
and the active materials are insoluble in the electrolyte.
Nickel-cadmium batteries have a reputation for their
robustness, reliability, and service life. They can also
operate under severe weather conditions, with
operating temperatures ranging from 40 C to +60 C
(because the electrolyte has a very low freezing point),
excellent cycling capability (up to 2,000 cycles at 80%
depth of discharge), long storage life, and low or zero
maintenance [10, 11].
NiCd batteries have been used extensively in con-
sumer electronics and power tools. In the recent past,
NiCd batteries have also been used in electric vehicles,
an example being the Peugeot 106 electric, which used
NiCd batteries manufactured by SAFT [12].
Nickel-cadmium cells have a nominal cell potential
of 1.2 V. This is lower than the 1.5 V of many popular
primary cells, and consequently they are not appropri-
ate as a replacement in all applications. NiCd
batteries have an energy density of 40–60 Wh/kg,
which is greater than that of lead acid batteries,
but less when compared to NiMH and Li-ion
batteries [14].
Advances in battery-manufacturing technologies
throughout the second half of the twentieth century
have made batteries increasingly cheaper to produce;
about 1.5 billion NiCd batteries were produced annu-
ally up until 2000. NiCd batteries never became widely
used as a replacement for lead acid batteries in the areas
where those batteries dominate, mainly because of the
toxic nature of the battery and the reduction in its
capacity as the battery ages. However, NiCd batteries
had an overwhelming majority of the market share for
rechargeable batteries in consumer electronics up until
the mid-1990s [10].
 Battery Technologies 169

Nickel Metal Hydride Batteries
Nickel metal hydride (NiMH) batteries were first devel-
oped in the late 1980s and ever since have been used in
portable electronic devices like camcorders, power
tools, and cell phones. NiMH batteries have high stor-
age capacity and relatively high ramp rates. They have
succeeded lead acid batteries owing to higher number
of cycles and higher energy density [14]. A study by
EPRI in 2004 showed that NiMH batteries are accept-
able for full-function EVs, city EVs, and plug-in hybrid
electric vehicles (PHEVs) [8].
Although the technology was largely driven by mil-
itary and government research in its earlier days,
recently, significant contributions have been made by
EV manufacturers. Applications of NiMH EV batteries
include all PHEVs manufactured by General Motors,
Honda, Ford, Toyota, and the Vectrix scooter, among
others. The town of Nice, France, now operates its low
floor tram, manufactured by Alstom using NiMH
batteries. Honda’s humanoid prototype robot ASIMO
is another application.
Nickel Metal Hydride Cell Construction and
Chemistry
The basic representation of a NiMH cell is shown in
Fig. 8. The positive electrode of a NiMH cell consists of
a spongy mass of nickel hydroxide Ni(OH)2. During
charge, the positive electrode releases hydrogen into the
electrolyte, which in turn combines with hydroxide
(OH) ions. This reaction results in nickel oxyhydroxide
(NiOOH) on the positive electrode and water (H2O)
in the electrolyte plus one free electron, e.
The equation for the positive electrode can be
given by:
Charge
NiðOHÞ2 þ OH ! NiOOH þ H2 O þ e
discharge
ð1Þ

Metal Hydride Nickel Electrode

Electrode (Positive)
(Negative)
overcharge
O2 Ni
4OH− 2H2O + O2 + 4e−
MH

charge e−
xOH− H2O
Charge Capacity

MHx NiOOH
Separator

M M
xH2O Ni(HO)2
xe− OH−
discharge

overdischarge
H2 Ni
2e− + 2H2O 2OH− + H2
MH

Separator Impregnated with KOH Solution

Battery Technologies. Figure 8

Schematic representation of a NiMH cell [7]
170 Battery Technologies
The negative electrode contains a metal alloy–
hydride complex. In an exothermic reaction, the
water (H2O) from the electrolyte combines with the
free electron to form metal hydride (MH) and hydrox-
ide (OH). The metal alloy absorbs and desorbs the
hydrogen, allowing the electrochemical reaction to
occur without producing hydrogen gas. Equation 2
shows the net chemical reaction that occurs at the
negative electrode during charge and discharge.
Charge
M þ H2 O þ e ! MH þ OH ð2Þ
discharge
The most commonly used electrolyte is a solution
of aqueous potassium hydroxide, KOH, in which the
KOH concentration ranges from 25% to 40% by mass.
The two equations show that there is no net change in
electrolytic quantity. This result contrasts with other
electrolyte systems like NiCd, where water is generated
at both electrodes during charge and consumed at both
electrodes during discharge. Although transient elec-
trolyte concentration gradient occurs in NiMH batte-
ries, its constant average concentration has good overall
performance in gas recombination, kinetics, high and
low temperature operation, and resistance to cycle life
limitations caused by corrosion and swelling [7].
Overcharge and Over-discharge Reactions
The NiMH battery has overcharge and over-discharge
reactions that allow the battery to handle abuse condi-
tions of overcharge and over-discharge without adverse
effects. The reaction that takes place at the two elec-
trodes during overcharge is an oxygen recombination
reaction. It can be shown as follows:
At the positive (Ni) electrode OH ions are oxi-
dized generating oxygen:
4OH $ 2H2 O þ O2 þ 4e
At the negative (MH) electrode the oxygen is reduced:
2H2 O þ O2 þ 4e $ 4OH
The net result is generation of heat which is pro-
portional to the energy input. This occurs at the
expense of increasing the stored charge in the battery.
If the rate of charge input exceeds the rate of recombi-
nation, the cell pressure increases, which may lead to
cell damage.
During over-discharge, hydrogen is released at the
positive electrode and recombined at the negative elec-
trode. The reactions are as follows:
At the positive electrode, water is reduced and
hydrogen gas is released:
2H2 O þ 2e $ H2 þ 2OH
The hydrogen gas is then oxidized at the negative
(MH) electrode:
H2 þ 2OH $ 2H2 O þ 2e
The ability of the NiMH cell to tolerate over-
discharge is very important for large series strings of
batteries since capacity mismatches may cause some
cells to over-discharge. To ensure that the overcharge
and over-discharge reactions function properly and
thus control the buildup of cell pressure, the NiMH
battery is constructed with the Ni electrode as the
capacity limiting electrode and the MH electrode with
excess capacity to allow recombination reactions to
take place during overcharge and over-discharge [7].
Types of NiMH Batteries
The earliest pioneering work on NiMH batteries was
performed at the Battelle-Geneva Research Center
starting after its invention in 1967. These batteries
showed high specific energy up to 50 Wh/kg, power
density up to 1,000 W/kg, and a reasonable deep cycle
life of 500 cycles (depth of discharge: 100%). Due to the
inherent deep cycling capability and the high specific
energy, NiMH batteries have been used in EVs. The
two major types of NiMH batteries used today are
the cylindrical batteries and the prismatic batteries.
The next section describes the cylindrical NiMH
batteries in detail giving the battery structure and the
battery characteristics. The later section describes pris-
matic NiMH batteries and gives the comparison
between the two of them. Currently, there are over two
million HEVs running worldwide, which use NiMH
batteries [15].
Cylindrical NiMH Cell
Battery Structure The single cell shown in Fig. 9 is
a sealed cylindrical battery of D-size with a diameter of
32 mm and a height of 60 mm. A sealing plate is
equipped with a valve to prevent bursting with an
 Battery Technologies 171

Cap (+) Vent

Sealing plate
Insulating gasket Insulating ring

Current collector (+) Negative electrode

Separator

Positive electrode
Case (–)

Current collector (–)

Battery Technologies. Figure 9

Structure of a cylindrical battery

Battery Technologies. Table 3 Characteristics of a single 1000

Normal capacity: 6.5Ah
cylindrical cell and battery consisting of six cells [13]
800
Specific power [W/kg]

2 seconds
Battery module
Single cell (six cells)
600
Output 625 W/kg, 2,160 W/l 600 W/kg, 2,070 W/l 10 seconds
power 400
Input power 500 W/kg, 1,720 W/l 480 W/kg, 1,660 W/l
6.0V/module cut-off
Energy 45 Wh/kg, 172 Wh/l 43 Wh/kg, 161 Wh/l 200
density SOC50 SOC100
0
Nominal 1.2 V 7.2 V 0 1 2 3 4 5 6 7
voltage Charge capacity [Ah]

Battery Technologies. Figure 10

Specific power of a cylindrical battery at different state of
charge levels [13]
increase in internal pressure. The battery case is made
of steel and both positive and negative electrodes are
Discharge Power Characteristics Figure 10 shows the
coiled and separated by a separator. This battery
specific power characteristics of a battery module that
optimizes the reaction area of the electrodes, reducing
consists of six cells at different state of charge (SOC).
resistance for current collection and improving
The module provides 800 W in 2 s and 650 W in 10 s. As
electrolyte composition to obtain high power
seen in the graph, for a given range of charge capacity,
characteristics. The nominal battery capacity is 6.5 Ah
the specific power is almost constant. Also, for the 50%
and the maximum output power is more than 100 W
SOC and 100% SOC, there is not much variation in
per cell. To reduce the internal resistance and shield
specific power, giving 650 W/kg in 2 s and 600 W/kg for
against vibrations, disk plates are inserted for cell
a 10 s duration. Relatively constant input and output
connection [13].
power is highly desired by EV manufacturers. It can be
Battery Characteristics Table 3 gives the characteristics clearly seen from the graph that NiMH prismatic
of a single cylindrical cell and battery consisting of six batteries deliver an almost constant power for any
cells. SOC, which is highly desirable in batteries for EVs.
172 Battery Technologies
110
30A, 10sec. Pulse Charge
Pulse Charge Efficieny [%]
100
90
80
0°C
70 25°C
40°C
60
3.0 3.5 4.0 4.5 5.0 5.5
Capacity [Ah]
Battery Technologies. Figure 11
Pulse charge efficiency of cylindrical battery module at
different temperatures [13]
Charge Efficiency For practical use, high charge
efficiency is required over a wide temperature range.
Figure 11 shows the battery has a very high charge
efficiency, which provides high regenerative
acceptability over the SOC range, mostly used during
normal vehicle operation. The ampere-hour charge
efficiency is nearly 100%. Also because of the very
small energy loss, heat generation of the battery is
minimized.
Life Cycle Characteristics To preserve an active battery
system, flow of power in and out of the battery needs to
be maintained, even if minimal. The battery’s
controller needs to ensure that the battery is neither
fully charged nor completely discharged. To test such
operating conditions, a life test was conducted at
Panasonic’s test facility in Shizuoka, Japan. One
example of the results is shown in Fig. 12 [13]. The
test provides results in which the input and output
power simulate real vehicle driving at 35 C. As
a result, the durability was found to be equivalent to
more than 100,000 km driving without deterioration of
battery characteristics.
Prismatic NiMH Battery
Structure To improve power, reduce the number of
connected parts. Figure 13 shows the appearance of
a prismatic module design consisting of six cells in
10 50
Internal resistance [mΩ /module]
Battery temperature: 35°C
8 40
Capacity [Ah]
Capacity
6 30
4 20
Internal resistance
2 10
0 0
0 50,000 100,000 150,000 200,000
Corresponding driving distance [km]
6.0 6.5 7.0
Battery Technologies. Figure 12
Life cycle characteristics of cylindrical battery module with
simulated actual driving pattern [13]
series. The conventional cylindrical module has
connecting plates to connect cells and plastic
insulating rings to prevent external short circuits.
The prismatic module adopts internal connection;
therefore, connecting plates and plastic insulating
rings are not required. As a result, a shorter current
path is accomplished. The investigation of electrode
dimensions and current collector design shows that
the current flow path is shortened. By thinning the
positive and negative electrodes, the number of
electrodes are increased, which increases the reaction
area and decreases the current density. Resin material is
used to make the battery case, which provides increased
safety and reliability. The resin material makes it easy to
design bumps and ribs on the side of the battery case.
The bumps and ribs provide cooling and also reduce
the pack volume by 40%. The resin also reduces 20%
of the prismatic battery’s weight compared to a
cylindrical battery pack. The six cells are internally
connected to form one module. External connection
parts like bus bars and connecting plates are also not
required [13].
Figure 14 shows the improved power characteristics
and reduced internal resistance of this battery, as com-
pared to the cylindrical battery. The main source of the
reduction in internal resistance is the components.
A high specific power of 1,000 W/kg, which is higher
than that of the conventional cylindrical battery, is also
achieved.
 Battery Technologies 173

Battery Technologies. Figure 13

Prismatic NiMH battery [13]

5.0 Battery Technologies. Table 4 Comparison between

at 25°C prismatic and cylindrical battery modules [13]
Internal Resistance [mΩ /cell]

4.0mΩ /cell
4.0 Reaction (Positive Electrode)
Cylindrical
Prismatic module module
3.0 Reaction (Negative Electrode) 2.8mΩ /cell
Dimensions (mm) 19.6  106  275 35  384
(width  height  (diameter 
2.0 Electrolyte length) length)

Parts Nominal voltage (V) 7.2 7.2

1.0
Nominal capacity (Ah) 6.5 6.5

0.0 Specific power (W/kg) 1,000 600

Conventional Cylindrical New Prismatic Specific energy 46 43
(Wh/kg)
Battery Technologies. Figure 14
Internal resistance of prismatic and cylindrical battery cell Weight (g) 1,020 1,090
before activation [13]

Table 4 shows the module characteristics of both

prismatic and cylindrical battery types. It is clear that
the prismatic design is superior to the cylindrical
design.
Discharge Power Characteristics Figure 15 shows the
power characteristics of the prismatic module at
different SOC. This battery provides 1,000 W of
power per kilogram for 10 s and 1,150 W of power for
2 s with the SOC maintained in a range 40–60%. It also
has high power density at low SOC; about 800 W/kg for
10 s and 1,000 W/kg for 2 s at 20% SOC.
Figure 16 shows the power characteristics at
different temperatures for the cylindrical battery and
the prismatic battery. As seen, the characteristics of
174 Battery Technologies

1600 the prismatic battery are better (490 W/kg at 0 C

1400 Nominal Capacity 6.5Ah and 120 W/kg at 30 C). Also the temperature
Specific Power [W/kg]

1200 2 seconds where power discharge of 200 W/kg is possible is reduced

1000 by about 10 C as compared to the cylindrical battery.
800 10 seconds
Life Characteristics of Prismatic Batteries In Fig. 17,
600
Pattern A represents a normal driving profile, with
400
a maximum SOC deviation of 15% and pattern
200 6.0V/module cut-off
B represents a high load driving profile with
0 a maximum SOC deviation of 30%. After tests
0 20 40 60 80 100
SOC [%] corresponding to a driving distance of 300,000 km
with pattern A at 35 C, there was no deterioration
Battery Technologies. Figure 15
observed in battery performance. Also, with pattern
Specific power of prismatic battery at different SOC [13]
B at 55 C, the battery was able to achieve a driving
distance of 200,000 km [13].
1400 Figure 18, above, shows the EV-95 (95 Ah) and
1200 EV-28 (28 Ah) units developed by Panasonic for EVs.
Specific Power [W/kg]

EV-95 has long life characteristics of more than 1,000

1000
Prismatic cycles and 4 years of onboard driving [13].
800 The batteries shown above have the following
600 characteristics:
Cylindrical
400
1. Optimization of additives (Y2O3) for excellent
10-second pulse
200 charge efficiency of the positive electrodes
6.0V/module cut-off
0 2. MnNi5 system of hydrogen-absorbing alloy for the
−40 −30 −20 −10 0 10 20 30 40 50
negative electrodes
Temperature [°C]
3. High performance, hydrophilic-treated polypro-
Battery Technologies. Figure 16 pylene separator
Specific power of prismatic and cylindrical battery module
at different temperatures [13]

50 10
Pattern A: Battery Temp. 35°C
Internal Resistance [mΩ /module]

Pattern B: Battery Temp. 55°C

40 8
Capacity
Capacity [Ah]

30 6

20 Pattern B 4

10 Internal Resistance Pattern A 2

0 0
0 50000 100000 150000 200000 250000 300000 350000
Corresponding Driving Distance [km]

Battery Technologies. Figure 17

Life cycle characteristics of prismatic batteries [13]
 Battery Technologies 175

Ev-95 Ev-28

Battery Technologies. Figure 18

EV-95 and EV-28 batteries by Panasonic

Battery Technologies. Table 5 Basic characteristics of the Battery Technologies. Table 6 Battery characteristics [8]
EV-28 and EV-95 batteries [13]
Cell Specific Specific
EV-28 EV-95 Battery size energy power
Manufacturer design (Ah) (Wh/kg) (W/kg)
Dimensions (width  height  75  110  116  175 
length) mm  mm  mm 388 388 Texaco High 7.5 40 650
Ovonic power
Nominal voltage (V) 12 12 battery
Medium 28 48 440
Nominal capacity (Ah) 28 95 power
Weight (kg) 6 18.7 Medium 45 71 390
Specific energy (Wh/kg) 58 65 power
Specific power at 80% depth 300 200 VARTA High 10 30 630
of discharge (W/kg) power
Self discharge at 45 C, 1 20 20 Medium 45 50 220
month (%) power
Cycle life at 25 C ambient >1,000 >1,000
temperature, 80% DOD
(cycles)
The second-generation GM EV1 used NiMH batte-
ries. Table 7 shows the specifications of that battery.

The electrode groups consist of alternately stacked Lithium-Ion Batteries

positive electrodes and negative electrodes interleaved
with separators. Inserting these electrode groups into
a resin battery case and sealing with a cover equipped
with a valve after filling with alkaline electrolyte forms
the cell. A battery module consists of ten cells
connected in series by metal plates and these are con-
figured to permit airflow between the cells to ensure
a uniform temperature distribution.
Table 5 gives the characteristics of the EV-28 and
EV-95 batteries manufactured by Panasonic.
Below is Table 6, which gives characteristics of
NiMH batteries, produced by Texaco Ovonic Battery
Systems and VARTA.
Lithium-Ion Battery Chemistry
Lithium is the least dense of all solid metals, has the
greatest electrochemical potential, and provides the
largest energy density by weight. For many years, lead
acid had been the predominant battery for large-scale
equipment. G. N. Lewis pioneered early work with
lithium-ion battery technology but it was not until
the early 1970s before it became commercially viable.
Because of the inherent instability of lithium metal,
early attempts at rechargeable batteries failed. Research
then shifted to lithium-ion-based composites which,
although less dense, are safer to use. Lithium-ion
176 Battery Technologies

Battery Technologies. Table 7 GM EV 1 (generation 2) internal resistance of lithium-ion cells is comparatively

battery specification [16] low, which means they do not lose a considerable
amount of energy as heat (I2R loss) [8]. In general,
Ovonic energy
Manufacturer products the internal resistance of a cell is modeled as a resistor
in series with an ideal voltage source. Maximum effi-
Type Nickel metal hydride
ciency occurs with a low internal series resistance and
Number of modules 26 a large load resistance. Maximum power can be
Weight of module 18.3 kg extracted from a cell when load resistance is equal to
Weight of pack(s) 481 kg the cell’s internal resistance. Under this condition
half the total power is expended as heat, while the
Nominal module voltage 13.2 V
other half is available for use. The availability of high
Nominal system voltage 343 V
power output is a useful property for lithium-ion
Nominal capacity (50% charge 85 Ah cells, especially when instantaneous power is required.
capacity) Consider the example of a race car looking to set a new
quarter-mile speed record. This car would need high
instantaneous power without regard to cell degrada-
tion due to overheating.
Battery Technologies. Table 8 Redox potential [7]

Electrode Potential Electrode Potential Operation

reaction (V) reaction (V)
Numerous lithium-based alloys are available for use as
Li+ + e ↔ Li 3.01 Cu2+ + 2e ↔ Cu 0.34
anode. Lithium polymers are considered commercially
Na+ + e ↔ Na 2.71 Ag+ + e ↔ Ag 0.80 safe, while metallic lithium is still at the developmental

Ni 2+
+ 2e ↔ Ni 0.23 Cl2 + 2e ↔ 2Cl 1.36 stage [17]. Charge flow for any of the combinations in
Pb 2+
+ 2e ↔ Pb 0.13 F2 + 2e ↔ 2F 
2.87 Fig. 19 is typically the same, as shown in Fig. 20.
A common electrolyte is lithium hexafluoro-
phosphate (LiPF6). It was chosen for its high redox
potential. Ehsani et al. [17] mentions that LiPF6 is
batteries emerged during the early 1990s. Today, unstable under high heat and moisture. Hydrofluoric
lithium-ion is the fastest growing and most promising acid (HF) is produced as a by-product to the electro-
battery chemistry. lyte’s reaction with water. For this reason, there was
As mentioned earlier in section “Basic Characteris- significant opposition to its use in the early 1990s. Since
tics of Batteries,” a battery functions as a combination then however, it has been revealed that a small amount
of chemical reactions, oxidation at the anode (where of HF increases life cycle, because of the formation of
electrons are generated in discharge mode), and a strong passive layer on the cathode.
reduction at the cathode (at which electrons enter The addition of HF prevents formation of lithium
the cell). Depending on the direction of the current dendrite. (Most metals are combined with alloys for
(charge or discharge) each electrode is either the cath- dexterity and strength. Due to the presence of alloys,
ode or the anode. The electrolyte can serve as the metal tends to exist in a nonuniform, snowflake-
a reaction propellant or even as an integral part of like distribution called dendrite.) LiPF6 is also easily
the reaction. This combination of reactions is referred soluble in the solvent, making it a good solid interface
to as redox. Table 8 provides redox reaction potentials on the surface of anode materials. Lithium-ion is con-
for eight elements. sidered to be the current generation technology for EV
Lithium and fluoride generate the highest poten- batteries, in part because of it high energy density. Also,
tial during oxidation and reduction, respectively [7]. lithium has the highest redox reaction potential of any
A significant advantage of lithium-ion batteries is their metal. It is also necessary to consider the fact that key
efficiency in charge retention and energy supply. The elements of lithium-ion batteries are facing exhaustion.
 Battery Technologies 177

Anodes Electrolytes Cathodes

Metallic Li1-xNi1-y-zCoyMzO4
Liquid Organic
Lithium (M=Mg, Al, etc)
Electrolytes
Primary
Batteries Li1-xCo1-yMyO2
Lithium
Alloys
Li1-xMn2-yMyO4
Solid Polymer
Graphites
Electrolytes

Lithiated
Carbons
Polyanionic Compounds
Other
Carbons Polymer Gel Li1-xVOPO4:LixFePO4
Electrolytes

Composite
Alloys
Sn(O)-based
Other Sn(M)-based
Lithiated
Materials Ionic Liquids Li1-xMn1-yMyO2
3d Metal
Oxides: (M=Cr, Co, etc)
Nitrides

Battery Technologies. Figure 19

Numerous chemical combinations for lithium-ion batteries

Cobalt metal, used for cathode, has been in shortage for

e− e−
quite some time [17]. Most metals are combined with
V alloys for dexterity and strength. Due to the presence of
Discharge Charge alloys, the metal tends to exist in a, nonuniform, snow-
flake-like distribution called dendrite [17].
Li+
Li+ Li+ Efficiency
Charge
Lithium-ion batteries are practically 100% charge effi-
Li+ Li+ cient, according to Brodd et al. [5]. In essence, all the
charge lost during discharging is recovered during
Li+ charging. Note that there is a net loss in energy during
Li+ Li+ each complete charging/discharging cycle. This is
Discharge
because the cell voltage is lower during discharging as
compared to the charging cycle. Figure 21 provides
Anode Cathode
LixC Li1–xMyOz
discharge curves for voltage and current of a lithium-
ion cell. Note that the voltage continues to drop during
Battery Technologies. Figure 20 discharge before stagnating at a constant level toward
Charge flow for lithium-ion batteries [17] the end of charge availability. This constant voltage
178 Battery Technologies

[V]
4
3.6
Constant current, 0.3 C
3
Constant current, 3 C
2.5
Constant
2 voltage

SOC [%]
0
100% 50% 20% 0%

[A]
Constant current, 3 C
3C

Constant
2C voltage

1C
Constant current, 0.3 C
0.3 C SOC [%]
0
100% 50% 20% 0%

Battery Technologies. Figure 21

Voltage and current during discharge [18]

can be attained by discharging at a lower current, as
shown in the current curve of Fig. 21. As the current
drops to zero, the total charge from the cell adds up to
the same value regardless of the rate of discharge
(whether the discharge was at a low current or at
a high current) [5].
Applications that require constant current will be
unable to extract the last bit of energy out the battery.
Other applications, such as for EVs, have more flexi-
bility. Some of today’s EVs can operate at lower torque,
which implies lower current, to allow the driver to
reach a safe location before running out of battery
power. Such applications can use the entire charge
contained in a battery.
Although not obvious, both the graphs in Fig. 21
are related. As a cell ages, it appears to gradually lose
capacity. Davide [18] explains that this capacity is
actually unused, but not lost. Internal resistance of
a cell increases with age, therefore the cell gets
undercharged and under-discharged along the process.
This issue can be resolved by allowed fixed cutoff volt-
ages to vary internally. Due to fixed cutoff voltages, the
cell charges/discharges less and less each time due to
increasing internal resistance, resulting in an apparent
loss of capacity. By raising the upper cutoff voltage
during charge and lowering the lower cutoff voltage
during discharge, we can recover some of this lost
capacity. A battery management system that is able to
measure a cell’s internal resistance and compensate its
cutoff voltage accordingly can make better use of the
cell’s capacity [18].
Reliability
To minimize operation and maintenance (O&M) costs
for the customer, it is essential for a battery manufac-
turer to conduct rigorous stress tests on their battery

DYNAMIC STRESS TEST (DST) CYCLE
50%
Regen Power
Percent of Peak Discharge Power
25%
0%
−25%
−50%
−75%
−100%
Discharge Power
−125%
0 60
Battery Technologies. Figure 22
Dynamic stress test by USABC [19]
systems. A common test adapted by manufacturers in
the USA is known as the dynamic stress test (DST),
developed by the United States Advanced Battery Con-
sortium (USABC). Figure 22 shows an example of the
battery charging and discharging power profiles that
must be successfully followed to acquire DST confor-
mation. The test can be performed at various tempera-
tures to determine ideal operating conditions. The test
is designed to gauge battery performance characteristics
such as partial discharge and hold, sustained climb,
thermal performance, and fast charge response. One
such battery was tested by the Electric Power Research
Institute (EPRI). Figure 23 shows DST results of that
battery, specifically energy storage capacity versus num-
ber of DST test cycles at 80% depth of discharge. As
shown, a 12.5% drop in capacity is attained after 3,000
cycles. This is in contrast to a 37% drop in capacity for
NiMH batteries subject to a similar test [20]. Longer life
expectancy combined with larger power density make
lithium-ion batteries an ideal candidate for EVs.
Twenty watt-hours of energy storage capacity is lost
after 3,000 charge–discharge cycles, as seen in Fig. 23.
Thermal Runway
Thermal runaway (TR) is defined as a phenomenon
where an increase in temperature creates conditions for
further increases in temperature. Viable conditions for
Battery Technologies 179
120 180 240 300 360
Time (seconds)
TR exist when a battery is producing more heat than it
can dissipate. Thermal runaway can lead to leaks,
venting of gas, and possibly a fire. According to Saft,
temperature in a cell must be kept below 145 C to
prevent conditions for TR from arising.
As battery temperature typically increases during
operation, resistance of the ion-transferring electrolyte
decreases. This leads to an increase in the current
passing through the battery, raising the temperature
of the battery even further. If the electrolyte
becomes too hot, its resistance will become negligible
and allow enough current to flow for chemical com-
pounds on the cathode and anode to breakdown and
short circuit.
Excessive ambient temperatures are another cause
of TR, according to Saft. Computers, cell phones, and
other portable devices are often subject to prolonged
exposure to direct sunlight. Additionally, portable elec-
tronics could be operated in constricted spaces, with
insufficient room for heat dissipation. High ambient
temperatures can also harbor conditions for TR.
EPRI identifies overcharging as yet another source
for TR [20]. Overcharging can lead to build up of
lithium deposits that can penetrate the film separating
the anode and the cathode, leading to a short circuit.
All lithium-based batteries are equipped with protec-
tive circuitry to avoid overcharging. Failure of
180 Battery Technologies

180

160

Energy storage capacity (Wh)

140

120

100

80

60

40

20
Temperature 20°C; DST test cycles
0
0 500 1000 1500 2000 2500 3000 3500
Number of 80% DOD cycles

Battery Technologies. Figure 23

DST results for Saft’s lithium-ion battery [20]

100
SOC = 50% (3.65V)
SOC = 100% (4.0V)
(years to 20% peak power fading)
Projected calendar life

10

1
0 10 20 30 40 50 60
Storage temperature (C)

Battery Technologies. Figure 24

Projected calendar life as a function of storage temperature [20]

protective circuitry can lead to build up of lithium-ions
on the graphite anode, forming lithium dendrite.
(Most metals are combined with alloys for dexterity
and strength. Due to the presence of alloys, the
metal tends to exist in a nonuniform, snowflake-like
distribution called dendrite.) With increasing
formation of this dendrite, useful anode surface
area diminishes, leading to reduced battery capacity.
If charging continues even further, the dendrite will
penetrate the separator and react with the cathode,
creating a short circuit. Metallic lithium is still at the
developmental stage because it is prone to dendrite
formation. Figure 24 depicts EPRI’s projection of
effects of storage temperature on battery life.

Irrespective of the cause, TR is a dangerous pros-
pect. Potential large-scale use will require development
of adequate prevention of conditions leading to TR
[20]. USABC and EPRI have proposed various solu-
tions, albeit with negative impacts on power density
and capacity.
Capacity Fading
Capacity fading (CF) refers to reduction in battery
capacity after repeated charge or discharge cycles.
Yunjian et al. [21] uses a lithium-ion battery, with
lithium manganese oxide (LiMn2O4) as cathode and
graphite as anode to study capacity fading characteris-
tics. Lithium hexafluorophospate (LiPF6) is present as
an electrolytic salt. The battery was charged to 4.2 V
and then stored for a period of 96 h at 55 C. A notable
reaction was the formulation of hydrofluoric acid (HF)
and a consequent dissolution of manganese (Mn). The
following reactions were noted:
LiPF6 þ 4H2 O ! PO4 3 þ LiF þ 5HF
2LiMn2 O4 þ 4Hþ ! 3MnO2 þ Mn2þ
þ 2Liþ þ 2H2 O
Mn2+ enters the electrolyte and manganese dioxide
(MnO2) is deposited on the cathode. A reported
8 mAh/g of capacity was lost on the first day. Only
2 mAh/g of capacity was lost during each of following
3 days. MnO2 formation on the cathode protected it
from engaging in further dissolution of manganese
(Mn), which explains the dwindling in CF. Charge
retention of the battery was found to be 83.3%,
85.8%, 86.9%, and 88.6% for those 4 days.
Elevated temperatures can accelerate the rate at
which CF occurs, as evident from work done by Xia
et al. [22]. When operated above 65 C, the number of
operational cycles is well below the EV requirement of
1,000–1,500. Degradation of electrode materials has
been found to be the main reason for CF at high
temperatures [22].
Cost
Lithium-based batteries are subject to aging even if not
being used. Also, lithium batteries are dependent on
protective circuits to maintain safe levels of voltage and
Battery Technologies 181
Battery
systems
Battery labor
systems 2%
Module
hardware
labor
17%
14%
Module
hardware
23%
Module
chemical
44%
Battery Technologies. Figure 25
Component costs as percentage for Saft’s EV lithium-ion
battery [18]
current. These factors propagate into a 40% (average)
higher manufacturing cost for lithium-based batteries
compared to NiMH [20]. Figure 25 provides a break-
down of costs to manufacture an EV lithium-ion
battery.
Several automotive manufacturers have announced
plans to incorporate lithium-based batteries for use in
HEVs, although NiMH batteries currently predomi-
nate. Lithium-ion batteries store more than twice as
much energy per kilogram (250–400 Wh/kg) compared
to the next-best battery technology, NiMH.
After having purchased their EV, customers are now
responsible for operating costs for as long as they own
the vehicle. Operating costs can vary based on battery
capacity, efficiency, environmental conditions, and
usage. Nevertheless, once gas savings balance out the
cost of the EV, long-term savings in operating costs
become apparent. A lab experiment conducted by
EPRI in 2010 [23] employed various EV types, each
with a lithium-ion battery system. Operating costs
(loosely termed “fueling costs”) are shown in Fig. 26.
Here, the following EV types were used: Fuel Cell
Vehicle (FCV), Plug-in Fuel Cell Vehicle (PFCV),
Range Extender Fuel Cell Vehicle (FCV), Battery Elec-
tric Vehicle (BEV), and Plug-in Hybrid Electric Vehicle
(PHEV). A “20” (or “40”) in the name indicates 20 (or
40) miles on a full charge.
182 Battery Technologies

0.350
Near Term Cost Scenario ($/mi)
0.300 Long Term Cost Scenario ($/mi)

Fueling costs (2010$ ml−1)

0.250

0.200

0.150

0.100

0.050

0.000
V

20

40

20

40
FC

BE
V

PF

EV

EV
C

E-

PH

PH
PF

PF

Battery Technologies. Figure 26

Operational costs per mile for various Li-ion EV configurations [23]

In the near term, FCVs are more expensive due to
the cost of hydrogen. The FCV fitted with a range
extender module uses considerably lesser hydrogen, as
consistent with the lower near term operating cost. As
expected, predominant battery use in EVs results in
lower near term costs. In the long term, the difference
in cost of hydrogen versus electricity became smaller, as
apparent in the near identical long-term costs for all
vehicle types.
To make lithium-ion batteries practical for mass-
produced electric-drive vehicles, new technologies
must increase the energy the batteries store and the
speed with which they can discharge it. They must
also lengthen cycle life to 15 years or 241,000 km
(150,000 miles) – the average life of a vehicle [24].
Perhaps it would be appropriate to compare lith-
ium ion to other chemistries in terms of cost of
a battery pack for EVs. Figure 27 provides such a com-
parison for a PHEV-40, as per 2010 cost estimates made
by EPRI’s Duvall et al. [23] and the Electric Auto
Association.
Notably, Pb-A and NiMH (best case) appear to
carry the lowest and highest overall costs respectively.
It is important to mention that a typical EV lead acid
battery pack only has a life of approximately
14–16 months (assuming the vehicle is driven
20–40 miles daily). On the other hand, the NiMH
pack has a life span of roughly 5.5 years.
Metal–Air Batteries
Metal–air batteries get their name from their reactants
which undergo electrochemical reactions to release
energy. They are one of the most compact and poten-
tially the least expensive batteries available. They are
also environmentally friendly. However, their main
disadvantage so far has been that electrical recharging
of these batteries is very difficult and inefficient.
Although many manufacturers offer units that can be
refueled, where the consumed metal is mechanically
replaced and processed separately, there still have not
been many developers who offer rechargeable batteries.
Rechargeable metal air batteries that are under produc-
tion have a life of only a few hundred cycles and an
efficiency of about 50% [1, 6].
Current research in lithium–air batteries have
shown that they can provide an energy density of
1,700 Wh/kg, which is more than the energy density
 Battery Technologies 183

Thermal Management,
Housing and
Assembly ($)
Battery Management
Unit ($)
Cells ($)

NCA: Lithium Nickel Cobalt

Aluminium Oxide
NCM: Lithium Nickel
Manganese Aluminium
Oxide

Battery Technologies. Figure 27

Cost breakdown for complete battery pack used for typical PHEV40

Li2O2
O2 Discharge reaction
2(Li+ + e−) + O2 Li2O2
Li2O2

Li+

Battery Technologies. Figure 28

Li–air battery [6]

of any other battery technology present today and is
comparable to the practical energy density of gasoline
[6] (Fig. 28).
Air electrodes and metal–air battery technologies
have already been used in fuel cell systems. Earlier, zinc
was the predominant metal used in metal–air batteries.
Recently, lithium instead of zinc as the metal has
started being used which has increased the energy out-
put eightfold. As seen in the figure above, the lithium–
air battery’s porous carbon cathode (gray) is flooded
with the electrolyte (blue). The Li+ ions (small dots)
react with oxygen molecules (pairs of dots) at catalyst
sites (yellow) to form lithium oxides (orange). The
oxygen electrode proceeding in tandem with lithium
according to the reaction 2Li + O2 ! Li2O2 can deliver
a capacity of 1,200 mAh/g. The first lithium–air cell was
successfully assembled and discharged in 1996, but
attractive rechargeability was demonstrated only
recently [3].
It could be argued that such a system unites within
the same device the two most prominent failures of
battery and fuel-cell technologies, namely, the inability
to master lithium and oxygen electrodes. These per-
ceived issues have prevented the practical use of lith-
ium–air batteries. Improving energy storage and
preventing Li2O2 from clogging the electrode require
a better understanding of the reaction mechanism of
the oxygen electrode. Engineering and chemical
184 Battery Technologies
advances are also required to prevent the ingress of
either CO2 or H2O, which could react with either
Li2O2 or lithium metal [3].
Zinc Bromine Battery
The zinc bromine battery is based on the chemical
reactions between two commonly available materials,
zinc and bromine. This battery was developed by Exxon
in the early 1970s. It is a hybrid flow battery system
since zinc is deposited on the negative electrode during
the charge cycle. Unlike conventional flow batteries, the
energy storage capacity of the zinc–bromine hybrid is
constrained by the amount of surface area available for
deposition [4, 9].
One of the main advantages of zinc–bromine batte-
ries is that it can be fully discharged without any damage
to the battery and has a life of at least 1,500 charge/
discharge cycles. This battery is ideally suited in appli-
cations that require deep cycle and long cycle life energy
storage. These batteries are made with low-cost, recycla-
ble plastics and manufactured with techniques suitable
for mass production and at low production costs [4].
As seen in the figure below, the battery consists of
a zinc negative electrode and a bromine positive elec-
trode separated by a microporous separator. An aque-
ous solution of zinc/bromide is circulated through the
ZINC
DEPOSIT
ANOLYTE
RESERVOIR
PUMP
Battery Technologies. Figure 29
ZnBr battery [4]
two compartments of the cell from two separate reser-
voirs. The electrolyte stream in contact with the posi-
tive electrode contains bromine which is maintained at
the desired concentration by equilibrating with
a bromine storage medium. The bromine storage
medium is immiscible with an aqueous solution
containing zinc bromide. All battery components are
made from a bromine inert plastic [4] (Fig. 29).
During discharge, zinc is converted to zinc ion with
the release of two electrons and bromine is converted to
bromide, the zinc ion and bromine combine to form
zinc bromide. The chemical process used to generate
the electric current increases the zinc-ion and bromide-
ion concentration in both electrolyte tanks. During
charge, metallic zinc is deposited as a thin film on one
side of the plastic electrode. Meanwhile, bromine is
evolved as a dilute solution on the other side of the
membrane, reacting with other compounds in solution
(organic amines) to form viscous, dense bromine-
adduct oil that sinks to the bottom of the tank. The
bromine oil is allowed to mix with the rest of the
electrolyte during discharge. The net efficiency of this
battery is about 75% [9].
The zinc–bromine battery uses electrodes that do
not take part in the reactions but merely serve as sub-
strates for the reactions. Therefore, there is no loss of
performance, as in most rechargeable batteries, from
POROUS
SEPARATOR
BROMINE
ELECTRODE
PUMP
VALVE
CATHOLYTE
RESERVOR
COMPLEX PHASE

repeated cycling causing electrode material deteriora-
tion. When the zinc–bromine battery is completely
discharged all the metal zinc plated on the negative
electrodes is dissolved in the electrolyte and again pro-
duced the next time the battery is charged. In the fully
discharged state the zinc/bromine battery can be left
indefinitely [4].
The zinc–bromine battery has an energy density
of 75–85 Wh/kg. The power characteristics of the
battery can be modified for selected applications, mak-
ing them extremely useful for multipurpose energy
storage [4].
Supercapacitors
In the field of energy storage, two parameters are fun-
damental for storage devices: the energy density and
the power charging and discharging rates. The first
parameter defines the amount of energy that can be
stored in a given volume. The lower the maximum
charging and discharging rates, the more time that is
required for loading and unloading the required
amount of energy into the storage device. The ideal
storage device should have both a high energy density
together with high power charging and discharging
rates. This is unfortunately not the case and compro-
mises have to be made. Current battery technologies
have relatively high energy densities but relatively poor
power densities. Supercapacitors are a compromise
between batteries and conventional capacitors. Their
main characteristic is that they possess both a relatively
high energy density and power density [25].
Integrating with an Onboard Energy Source
for an EV
The integration of the onboard energy source for an
electrically propelled vehicle with a supercapacitor
bank (SB) as a peak power unit can lead to substantial
benefits in terms of EV performances, battery life, and
energy economy. An SB made up of single cells
connected in series and in parallel, featuring appropri-
ate parameters of energy density, power density, with
high charging–discharging efficiency and affordable
cost can be beneficial as a suitable device to support
the energy source of an electric vehicle, thus providing
optimized energy management [26].
Battery Technologies 185
The potential benefits of an integrated system SB-
electrochemical energy source are the following:
● To improve the vehicle efficiency and energy econ-
omy over variable power driving conditions
● To assure high performance and good vehicle
behavioral response independently from the status
of the energy source (including age)
● To improve the endurance of the energy source to the
extent of its dependence on the high rate power demand
● To extend the vehicle’s range at full performance as
a consequence of the combined effect of load level-
ing of the onboard energy source and of better
efficiency of energy recovery
Second to the range of the EV, high acceleration
performance is a highly desired feature to achieve inte-
gration with the majority of today’s vehicles. The inte-
gration of an SB with a battery allows a system design
optimized according to driving requirements, by mak-
ing the specific power and specific energy parameters of
the energy source independent of each other. As
a consequence of this consideration, it will be possible
to design the storage battery with respect to energy
storage capability and life requirement without taking
into account the power requirement [26].
Difference Between Capacitors and Supercapacitors
A supercapacitor differs from a conventional capacitor
both in the physical phenomena and the materials from
which it is made. In today’s supercapacitors, the dielec-
tric is an electrolyte interposed between two electrodes.
When a voltage is applied, a double layer of charges is
formed at the interface between the electrodes and the
electrolyte. In the case of supercapacitors, the distance
between the charges corresponds to the thickness of the
double layer, that is, only a fraction of a nanometer. This
accounts for the difference in terms of capacitance per
square centimeter between a conventional capacitor in
the (order of magnitude nF/cm2) and a supercapacitor
(order of magnitude 50,000 nF/cm2) [26].
The way to increase the energy stored in a conven-
tional capacitor is to operate at high voltages (up to
3,000 V) consistent with the dielectric breakdown volt-
age. In today’s supercapacitors, the voltage to be
applied must be limited by either the solvent or the
organic electrolyte decomposition voltage (1.23 V and
186 Battery Technologies

Battery Technologies. Table 9 Characteristics of Electrolyte

Separator
supercapacitors based on the electrode and electrolyte
+ – + –
material [26] + –
+ –
+ –
+ –
+ – + –
+ –
Electrode Metallic + –
+ – + –
+ – + –
material Carbon Carbon oxides + – + –
+ – + –
+ – + –
Electrolyte Aqueous Organic Aqueous + –
+ –
+ –
+ –
electrolyte electrolyte electrolyte +
+ –
+ –
+ –
+ – –
+ – + –
Maximum 1 3 1 + – + –
+ – + –
voltage (V) + – + –
+ –
+ –
+ – + –
+ –
Specific power 0.8–2.6 1.5–5 0.5 + –
+ – + –
(kW/kg) + – + –
+ –
+ –
+ – + –
Specific energy 0.2–1.3 3–6 1 + –
+ –
+ –
+ –
+ – + –
(Wh/kg) + – + –

3.5 V, respectively). In this case, the only way to
increase the stored energy is to raise the capacitance
value by adopting electrode materials with very high
specific area. In particular, active carbon may reach,
through suitable chemical processes, a specific area of
103 m2/g, and a specific capacitance of 102 F/g. In
addition, it is possible to reach very low values of the
internal resistance that allows the device to provide
high output power.
So far, only supercapacitors operating on the basis
of charge separation phenomena due to the application
of an external voltage between the electrodes have been
considered. There is another type of supercapacitor,
called pseudocapacitor, in which reduction and oxida-
tion reactions occur during charge and discharge
phases. The electrodes are made of metallic oxides
(typically ruthenium or iridium) and electrolytes are
liquid. The characteristics of supercapacitors are sum-
marized as in Table 9 [26, 27].
Working Principle of a Supercapacitor
Supercapacitors are electrochemical double layer
capacitors (EDLC). Their technology is similar to
that of batteries, but the main difference is that they
involve only electrostatic phenomenon (non-faradic).
This is the first difference between batteries and
supercapacitors. Their power density is higher because
there are no chemical reactions during charging and
discharging. Also, for this reason, charge storage in
Current
Electrodes
collector
Battery Technologies. Figure 30
Principle of supercapacitors [25]
EDLCs is highly reversible, which allows them to
achieve very high cycling stabilities. EDLCs generally
operate with stable performance characteristics for
many charge–discharge cycles, sometimes as many as
106 cycles. On the other hand, electrochemical batteries
are generally limited to only about 103 cycles. In dou-
ble-layer capacitors, the energy is stored by charge
transfer at the boundary between electrode and elec-
trolyte. The amount of stored energy is a function of
the available electrode surface, the size of the ions, and
the level of the electrolyte decomposition voltage [25].
Supercapacitors contain two electrodes, a separator
and an electrolyte, as shown in Fig. 30. The two elec-
trodes provide a high surface area, defining the energy
density of the component. On the electrodes, current
collectors with a high conducting part assure the inter-
face between the electrodes and the connections of the
supercapacitor. The two electrodes are separated by
a membrane, which allows the mobility of the charged
ions and forbids electronic contact. The electrolyte
supplies and conducts the ions from one electrode to
the other. As the dissociation voltage of the electrolytes
used is less than 3 V, this limits the maximum voltage
that can be reached by the supercapacitor. Another
disadvantage is lower ionic conductivity, which reduces
the power capability [25].

The performance of an electrochemical double
layer supercapacitor (EDLC) can be adjusted by chang-
ing the nature of its electrolyte. As mentioned earlier, an
EDLC can utilize either an aqueous or organic electrolyte.
Aqueous electrolytes generally have lower equivalent
series resistance (ESR) compared to organic electrolytes.
However, aqueous electrolytes also have lower break-
down voltages. Therefore, in choosing between an aque-
ous or organic electrolyte, one must consider the trade-
offs between capacitance, ESR, and voltage. Because of
these trade-offs, the choice of electrolyte often depends
on the intended application of the supercapacitor [27].
While the nature of the electrolyte is of great impor-
tance in supercapacitor design, the subclasses of EDLCs
are distinguished primarily by the form of carbon they
use as an electrode material. Carbon electrode materials
generally have higher surface area, lower cost, and more
established fabrication techniques than other materials
such as conducting polymers and metal oxides.
Electrode Materials Used in EDLC
Activated Carbon Activated carbon is the most com-
monly used electrode material in EDLCs. It is less
expensive and possesses a higher surface area than
other carbon-based materials. Activated carbons utilize
a complex porous structure composed of different
sized micropores (<200 nm wide), mesopores
(200–5,000 nm) and macropores (>5,000 nm) to
achieve high surface areas. Although capacitance is
directly proportional to surface area, evidence suggests
that, for activated carbons, not all of the high surface
area contributes to the capacitance of the device. This
discrepancy is believed to be caused by electrolyte ions
that are too large to diffuse into smaller micropores,
thus preventing some pores from contributing to
charge storage [27]. Research also suggests an empirical
relationship between the distribution of pore sizes,
energy density, and power density of the device. Larger
pore sizes correlate with higher power densities and
smaller pore sizes correlate with higher energy densi-
ties. As a result, the pore size distribution of activated
carbon electrodes is a major area of research in EDLC
design. In particular, research has been focused on
determining the optimal pore size for a given ion size
and upon improving the methods used to control the
pore size distribution during fabrication [27].
Battery Technologies 187
Carbon Aerogels There is an increasing interest in
using carbon aerogels as an electrode material for
EDLCs. Carbon aerogels are formed from
a continuous network of conductive carbon
nanoparticles with interspersed mesopores. Due to
their continuous structure and their ability to bond
chemically to the current collector, carbon aerogels do
not require the application of an additional adhesive
binding agent. As a result of this, carbon aerogels have
lower ESR than activated carbons. (The internal com-
ponents of the capacitor (current collectors, electrodes,
and electrolyte) contribute to a resistance, which is
measured in aggregate by a quantity known as the
equivalent series resistance (ESR). The voltage during
discharge is determined by this resistance.) The
reduced ESR yields higher power and this is the pri-
mary area of interest in supercapacitor research involv-
ing carbon aerogels [27].
Carbon Nanotubes Electrodes made from this mate-
rial are grown as an entangled mat of carbon nanotubes
with an open and accessible network of mesopores.
Unlike other carbon-based electrodes, the mesopores
in carbon nanotube electrodes are interconnected,
allowing a continuous charge distribution that uses
almost all of the available surface area. Thus, the surface
area is utilized more efficiently to achieve capacitance
comparable to those in activated-carbon-based
supercapacitors, even though carbon nanotube elec-
trodes have a smaller surface area compared to acti-
vated carbon electrodes [27]. As the electrolyte ions can
easily diffuse into the mesoporous network, carbon
nanotube electrodes also have a lower ESR than acti-
vated carbon. In addition, several fabrication tech-
niques have been developed to reduce the ESR even
further. Carbon nanotubes can be grown directly onto
the current collectors, subjected to heat-treatment or
cast into colloidal suspension thin films [27]. The effi-
ciency of the entangled mat structure allows higher
energy densities [27].
Pseudocapacitors
In contrast to conventional EDLCs, which store charge
electrostatically, pseudocapacitors store charge
faradically through the transfer of charge between elec-
trode and electrolyte. This is accomplished through
188 Battery Technologies
electrosorption, reduction–oxidation reactions, and
intercalation processes. These faradic processes may
allow pseudocapacitors to achieve greater capacitances
and energy densities than EDLCs. There are two electrode
materials that are used to store charge in pseudoca-
pacitors, conducting polymers and metal oxides [27].
Electrode Materials Used in Pseudocapacitors
Conducting Polymers Conducting polymers have
a relatively high capacitance and conductivity and a
relatively low ESR and cost compared to carbon-
based electrode materials [27]. In particular, the
n/p-type polymer configuration, with one negatively
charged (n-doped) and one positively charged
(p-doped) conducting polymer electrode, has the
greatest potential energy and power densities. How-
ever, a lack of efficient n-doped conducting polymer
materials has prevented these pseudocapacitors from
reaching their potential. Additionally, it is believed that
the mechanical stress on conducting polymers during
reduction–oxidation reactions limits the stability of these
pseudocapacitors through many charge–discharge cycles.
This reduced cycling stability has hindered the develop-
ment of conducting polymer pseudocapacitors [27].
Metal Oxides Metal oxides like ruthenium oxide
have high conductivity, which makes them a good
electrode material for supercapacitors [27]. The
capacitance of ruthenium oxide is achieved through
the insertion and removal of protons into its amor-
phous structure. In its hydrous form, the capacitance
exceeds that of carbon-based and conducting polymer
materials. Furthermore, the ESR of hydrous ruthenium
oxide is lower than that of other electrode materials.
As a result, ruthenium oxide supercapacitors may be
able to achieve higher energy and power densities than
similar carbon electrode supercapacitors. However,
despite this potential, the success of ruthenium
oxide has been limited by its prohibitive cost. Thus,
a major area of research is the development of fabrica-
tion methods and composite materials to reduce
the cost of ruthenium oxide, without reducing the
performance [27].
Composite Electrode Capacitors Composite elec-
trodes integrate carbon-based materials with either
conducting polymer or metal oxide materials and
incorporate both physical and chemical charge storage
mechanisms together in a single electrode. The carbon-
based materials facilitate a capacitive double layer of
charge and also provide a high-surface-area backbone
that increases the contact between the deposited
pseudocapacitive materials and electrolyte. The
pseudocapacitive materials further increase the capac-
itance of the composite electrode through faradic reac-
tions [27]. Composite electrodes constructed from
carbon nanotubes and polypyrrole, a conducting poly-
mer, have been particularly successful. This electrode is
able to achieve higher capacitances than either a pure
carbon nanotube or pure polypyrrole polymer-based
electrode [27]. This is attributed to the accessibility of
the entangled mat structure, which allows a uniform
coating of polypyrrole and a three-dimensional distri-
bution of charge. Moreover, the structural integrity of
the entangled mat limits the mechanical stress caused
by the insertion and removal of ions in the deposited
polypyrrole. Therefore, unlike conducting polymers,
these composites have been able to achieve a cycling
stability comparable to that of EDLCs having carbon-
based electrodes [27].
Model for Supercapacitors
A conventional model describing the supercapacitor
modeled as an equivalent electrical circuit is useful for
developments of applications where supercapacitors
can be used. Figure 31, below, shows such a model.
The first parameter is the capacitance of the com-
ponent. It defines the capacitive behavior of the
supercapacitor and the energy that can be stored. This
capacitance is not constant. It is dependent on the
voltage across and is the reason why the capacitance
of the supercapacitor is modeled as a constant capaci-
tor Co, with a parallel capacitor Cv, which has a linear
dependence on the voltage V [25].
C ¼ Co þ Cv ; where Cv ¼ K  V ð3Þ
This defines the capacitance C of the component.
Following this definition, the current capacitance
can be derived from (3).
Q ¼ CV
Ic ¼ dQ=dt
 Battery Technologies 189

iC

R5

VC R1 r1 rn

Co Cv C1 Cn

Battery Technologies. Figure 31

Equivalent circuit model for supercapacitors [25]

Battery Technologies. Table 10 Energy capability versus The current capacitance Ci shows that it is possible
the capacitance [25] to store 16.55% more energy than expected considering
Capacitance C = 1,800 F W = 5,625 J
only the constant capacitance given by the manufac-
turer. Another consequence of this voltage-dependent
Current Ci = 2,100 F Wi = 6,562 J +16.65%
capacitance is that the voltage will not increase linearly
capacitance increase
during the charge with a constant current. Two other
Energy Cw = 1,850 F Ww = 5,781 J +2.77% parameters are conventional for a supercapacitor
capacitance increase
model. The first one is the series resistor Rs that induces
voltage drops during charge and discharge. Its value
influences the energy efficiency of the component and
its power density. The second conventional parameter
Therefore, is the leakage resistor Rl that induces load losses when
the component is in a standby mode. However, this
Ic ¼ ðCo þ 2K  VÞdV=dt ð4Þ resistor is not the only parameter that influences the
Currentcapacitance Ci ¼ Co þ 2K  V voltage variation across a supercapacitor between
a charge and discharge operation. This is illustrated in
Similarly, energy capacitance can also be defined. Fig. 32 [25].
The upper curve shows the typical voltage variation
Power ðPÞ ¼ Ic  V ¼ ðCo þ K  V2 ÞdV=dt ð5Þ
Vc across a 800 F/2.5 V supercapacitor during the end
Energy ðWw Þ ¼ 1=2ðCo þ 4=3ðK  VÞÞV2 ð6Þ of its charge with a constant current. Once the charge is
ended, the voltage decreases rapidly. This could be due
Energycapacitance ðCw Þ ¼ Co þ 4=3ðK  VÞ ð7Þ
to the leakage resistor but this is not strictly the case if
The amount of energy that can be stored into the model in Fig. 31 is considered again. It makes n
a supercapacitor can be expressed as a function of the parallel RC circuits appear. This is illustrated in the
capacitance that is considered. This is summarized for lower part of Fig. 32 where the relaxation phenomena
a 1,800 F/2.5 V supercapacitor, where measurements are presented as the charge of RC circuits with different
for C = 1,800 F and K = 150 F/V in Table 10. constant times. The result is a decrease of the main
190 Battery Technologies

Loading of a 800F/2.5V supercapacitor

2.5

2.49

2.48
u [V]

2.47

2.46

2.45
0 200 400 600 800 1000 1200 1400 1600 1800 2000
t (s)

2.5

uc
Voltages in the model [V]

2.49

uc1
2.48

2.47
uc2

2.46

2.45
0 200 400 600 800 1000 1200 1400 1600 1800 2000
t (s)

Battery Technologies. Figure 32

Relaxation phenomena [25]

voltage, not due to a dissipative component, but due to
a homogenous arrangement of the loads and ions on
the electrodes and in the electrolyte. Once all the
voltages across all the equivalent capacitors are equal,
then relaxation phenomena are supposed to be ended.
The decrease of the main voltage is then linked to
a leakage current. An ideal supercapacitor model
should propose an infinite number of RC branches.
For practical reasons linked to measurement, most of
the models identify only two or three of these equiva-
lent subcircuits [25].
Sizing Method of a Supercapacitive Tank
For a supercapacitor, the total amount of energy WM
that can be stored is expressed as a conventional
capacitor:
WM ¼ 1=2ðC  VM 2 Þ
where C is the capacitance of the component and VM is
the maximum voltage. The voltage across the compo-
nent has to be decreased from its maximum allowed
value to 0 V for the use of the total amount of stored
energy. This is not possible because the current pro-
vided by the supercapacitor should be infinite. For this
reason, the minimum voltage when discharging the
component has to be limited and all the energy stored
in the component is not used [25]. It is then necessary
to define the parameter d, which is the ratio between
the minimum allowed voltage Vm for the discharging,
and the maximum voltage VM that defines a full charg-
ing of the component. d is expressed in percent and is
called the discharge voltage ratio:
d ¼ ðVm =VM Þ  100 ð9Þ
Under this condition, the usable energy Wv that
ð8Þ a supercapacitor can provide is given by the equation:
 Battery Technologies 191

Battery Technologies. Table 11 Sizing of

a supercapacitor tank L Rs
3
Ns d (%) Volume (m ) Weight (kg) E (kWh)
C
4,872 50 1.46 1,943 7.6
IF
5,709 60 1.71 2,283 8.92 Vc
Vc
7,164 70 2.15 2,865 11.2
Battery Technologies. Figure 33
Interfaces for supercapacitors [25]

 2 during charge and discharge, internal losses are due

Wv ¼ WM 1  ðd=100Þ2 ð10Þ
If d = 50%, the voltage is half of the maximum
voltage at the end of the discharge, then the usable
energy Wv will be 75% of the total stored energy WM.
For a given usable energy, the last two equations can be
combined to identify the number of supercapacitors in
the supercapacitor bank.
 
Ns ¼ 2Wv =CVM 2 1  ðd=100Þ2 ð11Þ
Let us take an example where the supercapacitor
is 1,800 F/2.5 V and the usable energy Wu = 0.55 MJ =
5.7 kWh. The corresponding sizing of a supercapacitive
tank is given for three different values of d according to
(10) and this supercapacitor in Table 11 [25].
Wv ¼ 0:55MJ; C ¼ 1800; FUM ¼ 2:5; VIM ¼ 200A
The best choice should define a SB with d = 50%
because of the reduced number of supercapacitors and
because the usable energy Wv is 75% of the maximum
stored energy WM. On another hand, the choice of
Ns = 7,164 (d = 70%) leads to a strong oversizing of
the SB where the usable energy Wv is only 51% of
the maximum stored energy. This offers an interest-
ing versatility in the management of the energy that
has to be provided. If that tank and its associated
power converters are designed for d = 70%, it
will be possible to vary the voltage more than
that limit in case of stronger random power
constraints [25].
Energy Efficiency and Power Availability
One of the main parameters of a supercapacitor is
its series resistance. Independent of voltage falling
to this element. Even if manufacturers succeed in
reducing the value of the series resistor, its value
induces an energy efficiency lower than unity with
the consequence of a reduction of the power
availability [25].
The energy efficiency of supercapacitors is taken
into consideration while sizing a supercapacitive tank
because the way supercapacitors are charged and
discharged influence their performance. One of the
only ways to control the charging/discharging process
is to develop power electronics interfaces in order to
control the current for loading/unloading the
supercapacitors [25]. This is illustrated in Fig. 33. The
charge of supercapacitors is realized with a constant
current, as shown in Fig. 34, where supercapacitors are
loaded from a dc voltage source. The charging/
discharging current is adjusted to maintain the product
of vcI constant. The value of the charging/discharging
current or power will define the energy efficiency of the
supercapacitive tank together with the value of the
equivalent series resistor. This is illustrated in Fig. 35
where the voltage discharge ratio d is 50%. In this case,
75% of the total used energy is loaded and unloaded for
a 2,600 F/2.5 V/0.7 mO supercapacitor. The energy
efficiency is presented versus the time, the current,
or the power needed for loading/unloading the
component [25].
The time for loading energy is chosen to be 10 s. Its
energy efficiency is expected to be over 90%. This
means a charging current limited to 320 A or a charging
constant power limited to 700 W. The same can be
applied to the current or the power during discharge.
The constant power has to be kept lower than 420 W to
obtain more than 90% energy efficiency. This has
a consequence on the power density of the component
192 Battery Technologies
0.98
0.96
0.94
0.92
Efficiency
0.9
0.88
0.86
0.84
0.82
0.8
101
325
702
Battery Technologies. Figure 34
Charging characteristics [25]
which is 4,300 W/kg. Taking into account the energy
efficiency, the power density will only be 806 W/kg
if the energy efficiency is to be kept higher than 90%.
This leads to an increased number of supercapacitors
for the design of a storage tank. Therefore, the
energy efficiency and the power availability both
have to be taken into account for the design of
a supercapacitor storage bank. The final parameter to
be taken into account is the efficiency of the power
converter needed for interfacing the supercapacitors
with their load [25].
Power Train System Architecture
Architecture conceived with the electronic interface
between the SB and the traction battery are shown in
Fig. 36 below.
The main characteristics of this architecture are:
● The DC-to-DC converter has to be dimensioned
for the maximum power flow into and out of the
SB [26].
● The SB voltage may be kept always lower than the
minimum traction battery voltage. As a
102 103 t (s)
32.5 3.25 Ice (A)
62 6 P (W)
consequence, a simpler DC-to-DC converter design
may be obtained. The DC-to-DC converter may
operate only in step-up mode during the traction
phase and only in step-down mode during the
braking phase [26].
● A lower operating voltage implies a lower number
of cells connected in series and a reduced complex-
ity of the electronic module of the SB [26].
● The possibility to add an auxiliary generator (aux-
iliary power unit) configuration without modifying
the existing structure [26].
Future Developments in Supercapacitors
Currently, there is work in progress on the develop-
ment of supercapacitors using solid dielectrics.
The energy density stored in a capacitor is
given by:
1
W ¼ eo er E 2
2
eo is the dielectric constant of free space, er is the relative
dielectric constant, and E is the electric field strength.
Relative dielectric constants of 200,000 have been mea-
sured for 500 nm particles of BaTiO3 coated with 5 nm
 Battery Technologies 193

0.98

0.96

0.94

0.92
Efficiency

0.9

0.88

0.86

0.84

0.82

0.8
101 102 103 t (s)

−325 −32.5 −3.25 Icd (A)

529 60 6 P (W)

Battery Technologies. Figure 35

Discharging characteristics [25]

SUPERCAPACITOR
BANK SYSTEM

Current control DC/DC

CONVERTER
Drive train system

BATTERY
INVERTER Motor
SYSTEM

Battery Technologies. Figure 36

Powertrain architecture with DC/DC converter and series supercapacitor bank series connected [26]
194 Battery Technologies

of SiO2 [28]. If these particles can be fabricated Recommendations for Future Work
into supercapacitors with breakdown electric field
Various applications and options discussed in this entry
strengths of the order of the breakdown strength of
are still in the development phase and therefore are not
SiO2, around 6  106 V/cm, then energy densities of
“grid-ready.” Current progress in the energy storage
the order of 100 KJ/cm3 may be possible. If very high
sector is limited to fulfillment of functional specifica-
energy densities are achieved, these supercapacitors
tions only. Grid storage solutions are a long term goal
may replace batteries in EVs. There are a variety of
for this sector. To develop these solutions, short-term
materials with which very high dielectric constants
goals need to include fundamental research, analysis,
can be obtained but there is a great deal of work to be
testing, evaluation of risk, and demonstration of tech-
done before these supercapacitors become available for
nology. Table 12 is considered a starting point for the
EVs [28].
fulfillment of short- and long-term goals.

Battery Technologies. Table 12 Recommendations for future work

Action Short term 2011–2013 2012–2016 2015–2020
Develop value Conduct assessment of Simulation and cost- Large-scale Value and risk of
propositions for utility and private benefit analysis of demonstrations renewable integration
storage to renewable integration renewable plants and verified by the utility
enable more private systems and understood by the
renewable end user
integration
Development Develop target for Persistent enforcement Monitor progress of Identification and
and development of of regulations in all regulations and provide continued
enforcement of regulations to address types of regions (high feedback for enforcement of key
regulations issues and risks and low renewable improvement in regulations
associated with storage concentration) methodologies of local
systems utility
Scenario Develop models for Identify incremental Demonstration of Identification of
modeling accurate forecasting of (smart grid, nuclear) and effects of best- and relevant scenarios that
resource accessibility, detrimental (EV, worst-case scenarios on will help guide the
demand, dispatch storage) technologies, in supply/demand, future development of
capabilities, pricing, and terms of grid technological potential storage
reliability operational risk development, and technologies
capital costs
Contribution: Development of Assessment and Assessment of potential Identification of low-
utility and customer and utility regulated enforcement customer actions (e.g., risk actions for the
public actions for propagation of utility enforced use of energy efficient utility and the
of the renewable agenda actions (e.g., price devices) customer
incentive for customers,
storage management)
Reliability Develop current, short- Identify key factors Develop plan to Studious development
assessment and long-term affecting reliability of proportionally increase of reliability risks and
assessment roadmap for storage systems (e.g., integration of low-risk solutions for potential
risks associated with variability, ramping storage systems, storage technologies
storage integration requirements, peak with increasing
mismatch) demand, to achieve
higher base-load
penetration

Bibliography
Primary Literature
1. Hatziargyriou N et al (2000) Cigre technical brochure on
modeling new forms of generation and storage. TF 38.01.10
Nov 2000
2. HEV sales by models (2010) Alternative fuels and advanced
vehicles data center. Retrieved Sept 2010 from http://www.
afdc.energy.gov/afdc/data/vehicles.html
3. Lithium-Ion Technology Status Report (2005) EPRI
4. Lead acid batteries Woodbank Communications Ltd (2005)
Retrieved 20 Nov 2010 from http://www.mpoweruk.com/
leadacid.htm
5. Brodd RJ, Ozawa K, Yoshio M (2009) Lithium-ion rechargeable
batteries, materials, technology and new applications. Wiley-
VCH, Weinheim
6. Solar Energy Report (2005) Arizona Solar Center. Retrieved 12
Nov 2010
7. Kopera JJC (2005) Considerations for the utilization of NiMH
battery technology in stationary applications. In: The Battcon
2005 proceedings, Albercorp, Florida, 2005
8. Duvall M (2004) Advanced batteries for electric-drive vehicles:
a technology and cost-effectiveness assessment for battery
electric vehicles, power assist hybrid electric vehicles, and
plug-in hybrid electric vehicles. Final report 1009299. EPRI,
Palo Alto, CA. Retrieved 30 Sept 2010
9. Gage TB (2000) Lead acid batteries: key to electric vehicle
commercialization experience with design, manufacture, and
use of EVs. Proceedings of the 5th Annual Battery Conference
on Applications and Advances, 11–14 January 2000,
pp 217–222
10. Lead Acid Batteries Recharged. The economist. http://www.
economist.com/node/12924031?story_id=12924031
11. Ford Ranger EV Vehicle Specifications (1998) Electric
transportation applications. http://avt.inel.gov/pdf/fsev/eva/
ford_eva.pdf
12. General Motors EV1 Vehicle Specifications. Electric transporta-
tion applications. Retrieved 8 Oct 2010 from http://avt.inel.gov/
pdf/fsev/eva/genmot.pdf
13. Taniguchi A, Fujioka N, Ikoma M, Ohta A (2001) Development
of nickel/metal-hydride batteries for EVs and HEVs. J Power
Sources 100(1–2):117–124, Elsevier Science
14. Notten PHL, Van Beek JRG (2000) Nickel-metal hydride batte-
ries: from concept to characteristics. Chem Ind 54(3):102–115,
CEON/CEES
15. Avicenne Conference, Nice 2008, M. A. Fetcenko/ECD
16. General Motors EV1 Vehicle Specifications with NiMH batte-
ries. Electric transportation applications http://avt.inel.gov/
pdf/fsev/eva/ev1_eva.pdf
17. Schalkwijk WAV, Scrosati B (2002) Advances in lithium-ion
batteries. Plenum, New York
18. Davide A (2010) Battery management systems for large lith-
ium battery packs. Artech House, Boston
19. General Outline for Advanced Battery Testing. USABC
Battery Technologies 195
20. Alexander M, Duvall M (2005) Batteries for electric drive
vehicles – status 2005. EPRI
21. Huajun G, Qiyang H, Wenjie P, Xinhai L, Yong Y, Yunjian L
(2009) Performance and capacity fading reason of LiMn2O4/
graphite batteries after storing at high temperature. J Rare
Met 28(4):322–327
22. Xia Y, Yoshio M, Zhou Y (1997) Capacity fading on cycling of
4 V Li/LiMnO cells. J Electrochem Soc 144:2593
23. Duvall M, Rogers A (2010) Plug-in fuel cell vehicle technology
and value analysis. EPRI
24. Voelcker J (2007) Lithium batteries for hybrid cars. IEEE Spec-
trum. Retrieved from http://spectrum.ieee.org/green-tech/
advanced-cars/lithium-batteries-for-hybrid-cars
25. Barrade P (2003) Energy storage and applications with
supercapacitors. Laboratoire d’Electronique Industrielle (LEI),
Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne
(Switzerland)
26. Faggioli E, Rena P, Danel V, Andrieu X, Mallant R, Kahlen H
(1999) Supercapacitors for the energy management of electric
vehicles. J Power Sources 84(2):261–269, Elsevier
27. Halper MS, Ellenbogen JC (2006) Supercapacitors: a brief
overview. MITRE, Nanosystems Group, Virginia
28. Yang Cheng-Fu (1996) Improvement of the sintering and
dielectric characteristics of surface barrier layer capacitors by
CuO addition. Jpn J Appl Phys 35:1806–1813
Books and Reviews
Advanced Lead Acid batteries with Carbon enhanced
electrodes. http://www.electricitystorage.org/ESA/technolo-
gies2/_advanced_lead-acid_batteries_with_caron-enhanced_
electrodes_ala-cee_batter/. Accessed йил Sept 2010
Chan CC, Wong YS (November/December, 2004) Electric vehicles
charge forward. IEEE Power Energ Mag 2(6):77–79
Everything about NiCd batteries. http://www.batterychronicle.
com/2009/10/everything-about-nicd-batteries-2/. Accessed
йил 26 Oct 2010
Jacoby M (2010) Rechargeable metal air battery. http://pubs.acs.
org/cen/science/88/8847sci1.html. Accessed йил 22 Nov 2010
Metal-air batteries. http://www.electricitystorage.org/ESA/
technologies/metal-air_batteries/. Accessed йил Sept 2010
Moseley PT (2000) Improving the valve-regulated lead-acid bat-
tery. J Power Sources 88(1):71–77, Elsevier, 10th October 1999
Putois F (1995) Market for nickel-cadmium batteries. J Power
Sources 57(1–2):67–70, Elsevier
Riezenman MJ (1998) Engineering the EV future. IEEE Spectr
35(11):18–20
Tarascon JM, Armand M (2008) Building better batteries. http://
www.nature.com/nature/journal/v451/n7179/full/451652a.html.
Accessed йил 6 Feb 2008
Zinc-Bromine battery. http://energystoragedemo.epri.com/cec/
zbb/tech_desc.asp
ZnBr batteries. http://www.electricitystorage.org/ESA/technolo-
gies2/_zinc_bromine_zn_br_batteries/. Accessed йил 2009
April
196 Bicycle Integration with Public Transport
Bicycle Integration with Public
Transport
JOHN PUCHER1, RALPH BUEHLER2
1
Bloustein School of Planning and Public Policy,
Rutgers University, New Brunswick, NJ, USA
2
School of Public and International Affairs, College of
Architecture and Urban Studies Virginia Tech,
Alexandria, VA, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Trends in Bike-Transit Integration
Extent of Bike-and-Ride Facilities in North America
Case Studies of Bike-Transit Integration
Future Directions
Bibliography
Glossary
Bike lane A special road lane reserved for bikes only,
usually about 4–6 ft wide, normally located on the
right-hand side of the road.
Bike locker A box-like metal or plastic container for
secure bike storage, often at rail stations, usually
rented on a monthly basis.
Bike rack A device on which bikes can be mounted for
transport (on buses) or attached for storage and
security (on sidewalks).
Bike station Full-service, secure bike parking facility,
usually providing repair and rental services, acces-
sories, and touring advice.
Bus Motorized coach services providing communal
transport on roads, usually in mixed traffic, but
sometimes as part of special bus rapid transit
(BRT) systems with separate rights of way.
Light rail Type of rail system similar to streetcars
(trams) but often with separate right of way over
at least part of its route, special boarding stations,
and pre-paid fares.
Metro Urban rail system with separate rights of way,
frequent service, and high carrying capacity, usually
found only in large cities.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Public transport/transit General term for communal
transport in cities, including buses, trams, trolleys,
light rail, metros, suburban rail, ferries, and funiculars.
Suburban rail Short-distance passenger rail systems
connecting large cities with their surrounding met-
ropolitan areas, with longer distances between stops
and less frequent service than metro systems.
Definition of the Subject and Its Importance
Coordinating bicycling with public transport is mutu-
ally beneficial, enhancing the benefits of both modes
and encouraging more bicycling as well as more public
transport use. Bicycling supports public transport by
extending the catchment area of transit stops far
beyond walking range and at much lower cost than
neighborhood feeder buses and park-and-ride facilities
for cars. Access to public transport helps cyclists make
longer trips than possible by bike. Transit services can
also provide convenient alternatives when cyclists
encounter bad weather, difficult topography, gaps in
the bikeway network, and mechanical failures.
Introduction
The overall importance of both bicycling and public
transport in northern Europe has provided a strong
rationale for coordinating these two modes of urban
transport in recent decades [1]. European studies find
that coordinating bicycling with public transport is
mutually beneficial, enhancing the benefits of both
modes and encouraging more bicycling as well as
more public transport use [2–8]. In the Netherlands,
35% of all rail passengers reach their stations by bike –
compared to 25% in Denmark and 9% in Sweden [1].
The main form of bike-transit integration in
Europe is bike parking at suburban train and metro
stations [1]. In the Netherlands, there are 350,000 bike
parking spaces at train stations. German cities also have
extensive bike parking at rail stations. The City of
Berlin, for example, has over 32,000 bike parking spaces
at its metro and suburban rail stations [9]. By compar-
ison, no city in North America has more than 15,000
bike parking spaces of any kind, let alone at public
transport stops.
The most impressive bike-and-ride facilities in
Europe are state-of-the-art “bike stations,” which
# 2012, DOI 10.1007/978-1-4419-0851-3

provide secure, sheltered parking as well as a wide range
of services such as bike repairs, accessories, washing,
rentals, and travel advice. In 2007, there were 67 bike
stations in the Netherlands, 70 in Germany, and 20 in
Switzerland, almost always located adjacent to train
stations to facilitate bike-and-ride [1, 5, 10–12]. As
noted later in this entry, there are only ten bike stations
in all of North America, and they are much smaller
than those in Europe.
North America obviously lags behind Europe in the
integration of bicycling and public transport. In the
past, bike-and-ride in North America was limited by
low overall levels of cycling and public transport use in
most cities, just the reverse of the situation in northern
Europe [1, 5, 12, 13]. In recent years, however, both
cycling levels and public transport use have risen
sharply in the USA and Canada, and bike-and-ride
trips have been increasing as well. Indeed, in some
cities it has been so successful that the demand for
bike-and-ride facilities exceeds the available supply
[14, 15].
This entry describes the programs and policies cur-
rently being implemented in North America to inte-
grate bicycling with public transport. It starts off with
a brief overview of the various kinds of integration and
the extent of their implementation. Most of the entry,
however, is devoted to case studies of bike-transit inte-
gration in two large cities in Canada (Toronto and
Vancouver) and six large cities in the USA (San
Francisco, Portland, Minneapolis, Chicago, Washing-
ton, DC, and New York City). The case study analysis
compares the type and extent of integration measures
undertaken in the various cities, noting the strengths
and weaknesses of each city’s integration policies. The
entry concludes by identifying the most innovative and
successful policies in the eight cities and offers policy
recommendations for future improvements.
Trends in Bike-Transit Integration
In recent years, levels of cycling and public transport
use have reached record highs in both the USA and
Canada. In 2008 public transport use in the USA was at
its highest level since the early 1960s [16, 17]. Between
1995 and 2008, public transport trips rose by 38% in
the USA and by 46% in Canada [17, 18]. Similarly,
levels of cycling have increased considerably since
Bicycle Integration with Public Transport 197
1990. In the USA the total number of bike trips to
work increased by 32% from 1990 to 2005–2007 (aver-
aged) [19, 20]. Over the shorter period 1996–2006, the
number of bike trips to work in Canada rose by an even
larger 42% [21].
While rising public transport use and increased
cycling have provided the rationale for more bike-
transit integration, federal funding in the USA has
provided the necessary financing for a wide range of
projects implemented in recent years at the state and
local government levels [14, 15, 22, 23]. Indeed, the
federal government finances some categories of bike-
transit integration projects with an especially high 95%
federal share. The integration of public transport with
cycling has been endorsed by the Federal Transit
Administration, the Federal Highway Administration,
and the Transportation Research Board [14, 15, 24].
The Transport Association of Canada [25] and
Transport Canada [26] have also endorsed bike-transit
integration, but there is no federal funding for
urban transport in Canada. There is no federal
funding for urban transport in Canada, but provincial
and local governments have provided large increases in
funding for public transport and bicycling in recent
years, including projects aimed at better integration
[27, 28].
As documented in this entry, virtually every large
city in the USA and Canada has been undertaking
a range of measures to promote bike-and-ride. There
are five main categories of measures to promote bike-
transit integration:
(a) Provision of bike parking at rail stations and bus
stops, with different degrees of shelter and security
(b) Multifunctional bike stations providing not only
parking but also a range of services such as bike
rentals, repairs, parts and accessories, bike wash-
ing, showers and lockers, and touring advice
(c) Bike racks on buses, usually exterior, but occasion-
ally interior storage
(d) Bikes on board vehicles, usually rail vehicles,
sometimes with special bike racks, hooks, or even
bike cars on trains
(e) Bike paths, lanes, and on-street routes that lead to
public transport stations and stops, thus facilitat-
ing the bike’s role as feeders and collectors for
public transport
198 Bicycle Integration with Public Transport

Extent of Bike-and-Ride Facilities in Canadian transit systems surveyed by the Transporta-

North America tion Research Board [15], 14 systems provided bike
lockers at some of their rail and bus stops, but the
The only available national statistics on bike parking at
actual number of lockers was not reported. The same
public transport stops are from recent surveys of 272
TRB survey reported eight staffed bike parking stations
American and Canadian transit systems by the American
in 2005, mainly on the West Coast. A few more bike
Public Transportation Association [29, 30]. In the USA,
stations have opened since then [10].
the supply of bike parking spaces in 2008 was 24,178 at
By far the most important form of bike-transit
rail stations, 9,005 at bus stops, and 176 at ferry termi-
integration in North America is bike racks on buses.
nals. For the same year, Canadian systems reported
Indeed, on this dimension, North America is far ahead
2,892 bike parking spaces at rail stations and 481 at
of Europe, where very few buses come equipped with
bus stops. Between 2006 and 2008, the supply of bike
bike racks. That is not surprising since 60% of all public
parking increased by 67% in Canada and 26% in the
transport trips in the USA are by bus [18]. Bike racks
USA [29, 30]. That is impressive progress, but it is
are inexpensive to install, easy to operate, and do not
striking that in 2008, the total supply of bike parking
take up space on the vehicles themselves [15]. The 2005
at public transport systems in the entire USA was
TRB survey found that systems throughout the USA
only slightly more than the 32,000 spaces in Berlin,
and Canada provide bike racks on buses, and that most
Germany [9].
systems have eliminated fees they had previously
Not only is there much less bike parking in North
charged for rack use. As shown in Fig. 1, the percentage
America, but it is far less likely to be sheltered and/or
of buses with bike racks almost tripled in the USA in
guarded. Throughout northern Europe, there has been
only 8 years, from 27% in 2000 to 72% in 2009 [30, 33].
a trend toward sheltered, guarded bike parking, usually
Since 60% of all public transport trips in the USA are by
outdoors but increasingly in bike stations [1, 5, 9, 12,
bus, it is understandable that the focus of bike-transit
31, 32]. By comparison, unattended bike lockers are the
integration efforts in North America has been on bus-
main form of secure bike parking at North American
bike racks [18, 33]. They are inexpensive to install, easy
public transport stops. Of the 56 large American and

80%

71% 72%
70%
63%
Percent of Buses with Bicycle Racks

62%
60% 57%

50%
50%
46%

40%
36%
32%
30%
27%

20%

10%

0%
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Bicycle Integration with Public Transport. Figure 1

Trend in percentage of buses with exterior bicycle racks in the USA, 2001–2009 [18]
 Bicycle Integration with Public Transport 199

to operate, and do not take up space on the vehicles
themselves [15].
Another important form of bike-transit integration
is the permission to take bikes on board public trans-
port vehicles, since that enables cyclists to ride their
bikes to and from public transport stops at both ends of
their trips. Few public transport systems permit bikes
to be taken on board buses unless they are compact,
folding bikes. But most systems permit bikes on light
rail, metro, and suburban rail trains, except during
peak hour periods when crowding makes this infeasible
[15]. Moreover, an increasing number of public trans-
port systems are providing special accommodations for
bikes on trains, such as bike racks, bike hooks, special
bike holding areas near the doors, and even special bike
cars – although rarely [15, 33].
The last aspect of bike-transit integration is the
coordination of bike routes with public transport
stops. There are no national statistics available on the
extent of implementation, and it would be hard to
quantify at any rate. Nevertheless, the eight case studies
qualitatively assess this aspect of bike-transit integra-
tion in each of the cities.
Case Studies of Bike-Transit Integration
All eight of the case study cities are large, but they vary
considerably in metropolitan area population, ranging
from 2.2 million in Portland, Oregon, to 18.2 million
in New York City. The cities also vary widely in
their geographic locations, climate, and topography.
Most important for this study, they vary greatly in the
share of trips covered by bicycling and public transport,
as shown in Figs. 2 and 3.
In 2006, public transport shares of work trips for
central city residents ranged from only 13% in Portland
to 55% in New York (Fig. 2). Public transport trip
shares for metropolitan areas ranged from 5% in

60.0

54.8

50.0
City
MSA

40.0
Percent of Work Trips

35.7
34.4
31.9
30.4
30.0
26.7

22.2
21.0
20.0
16.5
14.4 14.4
12.6 13.4
11.3
10.0
6.4
4.8

0.0
Portland Minneapolis Vancouver Chicago San Francisco Toronto Washington, New York
D.C. City

Bicycle Integration with Public Transport. Figure 2

Public transportation share of work trips in US and Canadian cities and metropolitan areas, 2006/2008 [20, 34]
200 Bicycle Integration with Public Transport
7.0
6.0
5.0
City
Percent of Work Trips
MSA
4.0
3.0
2.0
1.0
1.0
0.6
0.5
0.4
0.0
New York Chicago
City
Bicycle Integration with Public Transport. Figure 3
Bicycling share of work trips in US and Canadian cities and metropolitan areas, 2006/2008 [20, 34]
Minneapolis to 30% in New York. The Canadian cities
of Vancouver and Toronto have relatively high public
transport use for their intermediate size. Indeed, their
metropolitan areas have higher public transport mode
shares than the Chicago area. Other studies confirm
that, controlling for population size, Canadian cities
have slightly more than twice as many public transport
trips per capita as American cities [35, 36]. Without
exception, public transport use is higher in all eight
central cities than in their suburbs. The difference is
far smaller in Toronto and Vancouver than for the
American cities, because of the higher density of
Canadian suburbs as well as the greater provision
of public transport services there [35, 36].
Levels of cycling also vary greatly among the eight
cities (Fig. 3). Portland (6.0%) and Minneapolis
(4.3%) had the highest bike mode shares of work
trips in 2006, but Vancouver (3.0%) and San Francisco
(2.7%) were not far behind. By comparison, cycling to
6.0
4.3
3.0
2.7
2.3
2.0
1.7 1.7
1.5
1.0 0.9
0.6
Toronto Washington, San Francisco Vancouver Minneapolis Portland
D.C.
work in New York (0.6%) and Chicago (1.0%) is rare.
Similar to levels of public transport use, bicycling is
much higher in central cities than in the suburbs. The
differences for Canadian cities, however, tend to be
smaller than for American cities, and the reason,
again, is the much higher density of Canadian suburbs
vis-à-vis American suburbs [28].
There are no comparable statistics on levels of bike-
and-ride in each of these cities, since the most recent
American and Canadian national travel surveys only
report the main mode of transport for the work trip.
Figures 2 and 3 provide useful background, however, by
portraying the overall levels of cycling and public trans-
port in the eight cities and their corresponding metro-
politan areas. On the basis of European experience, it
would seem that the higher the levels of both cycling
and public transport use, the greater the potential for
bike-transit coordination. On the other hand, where
there are large imbalances between the two modes, it

might suggest that the modes substitute for each other
rather than complement each other. For example, the
extremely low level of cycling in New York might be
partly due to a high level of public transport use.
Conversely, Portland and Minneapolis, the cities with
the highest bike mode shares, have the lowest public
transport mode shares, suggesting that some bike trips
might substitute for public transport trips. There are
many possible explanations for the different levels of
cycling and public transport use, but a detailed analysis
is beyond the scope of this entry, which is limited to the
more modest goal of examining the nature and extent
of bike-and-ride programs in these eight case-study
cities [1, 4, 5, 7].
Efforts to integrate cycling with public transport
vary greatly among the eight case studies. New York
City, for example, has done little to promote bike-and-
ride, while San Francisco, Vancouver, and Portland
have implemented the entire gamut of integration mea-
sures. The following section highlights the most impor-
tant aspects of bike-and-ride policies in each city,
noting in particular the strengths and weaknesses of
current policies.
Unless otherwise indicated, the information for
these case studies was obtained by the authors directly
from bicycling planners, public transport systems, met-
ropolitan planning organizations, city transport
departments, and nongovernmental cycling and sus-
tainable transport organizations in each metropolitan
area. The same panel of transit and cycling experts also
reviewed the case studies of their cities at several stages
to check for accuracy, consistency, and completeness.
San Francisco
The San Francisco Bay Area has been a leader in bike-
transit integration efforts in North America. Its
regional metro system Bay Area Rapid Transit
(BART) provides bike parking at almost all 43 stations,
with a total of 4,313 bike parking spaces in 2009,
including 1,010 in secure bike lockers. In order to
increase convenience and flexibility, BART has been
introducing electronic bike lockers (294 as of 2009),
which are available on a first-come, first-serve basis and
do not require advance subscriptions. Caltrain, the
suburban rail line from San Francisco south to Palo
Alto and San Jose, provides bike parking at all
Bicycle Integration with Public Transport 201
32 stations, with a total of 1,100 bike lockers and 400
bike racks.
The San Francisco Bay Area had five of the ten bike
stations in the USA in 2009: 226 bike parking spaces at
two Caltrain stations and 433 spaces at three BART
stations. The BART and SF Caltrain Bike Stations pro-
vide free, attended bicycle parking, while the Palo Alto
facility is an unattended, fee-based garage. Utilization
rates of the bike stations vary widely, from over 100% at
the Berkeley BART station to only 11% at the Palo Alto
Caltrain station. BART will soon triple the size of the
Berkeley bike station and move it above ground to
increase accessibility.
Cyclists prefer to take bikes on board, however.
A 2008 survey found that 72% of bike-and-ride pas-
sengers carried their bikes with them, compared to only
28% who parked them at BART stations. Bikes are
allowed on BART trains except during peak hours in
the peak direction. Although Caltrain has no time
restrictions, cyclists are often denied boarding on rush
hour trains because all bike spaces are already occupied.
Neither BART nor Caltrain charge a fee for bringing
bikes on board. Moreover, Caltrain’s lead cars provide
special accommodations for 16–32 bikes, depending on
time of day and direction of travel. Most of the numer-
ous ferry lines in the Bay Area also permit bikes on
board with no extra fee. Folding bikes are allowed on
BART and Caltrain at all times, but are not permitted
on board San Francisco’s MUNI buses, streetcars, cable
cars, and light rail vehicles. Complementing bike access
to BART and Caltrain services, virtually all buses of all
public transport systems in the San Francisco Bay Area
are equipped with bike racks, free of charge to cyclists.
There is limited coordination of bike routes with
public transport routes. To avoid bus-bike conflicts,
bike routes are not usually located on bus routes in
San Francisco, but are on parallel streets. Due to the
extensive and fine-grained network of bus and rail
routes in San Francisco, bike routes often lead to public
transport stops, even without any explicit coordina-
tion. The Regional Bicycle Plan as well as the Bike
Plans of BART and Caltrain explicitly encourage coor-
dination of bike routes and facilities with public trans-
port. Outside of San Francisco, where public transport
routes and stations are farther apart, many communi-
ties make an explicit effort to coordinate bike routes
with key stops.
202 Bicycle Integration with Public Transport
Overall, bike-transit integration efforts in the
San Francisco Bay Area have been successful.
The percentage of public transport trips combined
with cycling has more than tripled since 1990. Never-
theless, several problems remain. For example, it is
difficult for cyclists to get across the San Francisco
Bay during rush hours, since bikes are prohibited
from BART trains in peak directions and not permitted
on the Bay Bridge. Similarly, Caltrain has problems
accommodating bikes on board in the peak hour and
often denies boarding to cyclists. The exclusion of bikes
from Muni’s light rail lines is also a problem for cyclists.
Overall, however, the available evidence suggests that
bike-transit integration efforts have been successful.
Portland
As shown in Fig. 2, Portland has the highest bike share
of work trips of any large American city (6.0%). Thus,
one would expect a high degree of bike-transit coor-
dination. Bike-and-ride in Portland, however, is quite
different from that in San Francisco and mainly
involves bikes on transit vehicles. TriMet, Portland’s
public transport system, estimates that ten times
more bikes are taken on their LRT vehicles than parked
at LRT stations (2,100 vs. 200 per weekday). There are
no fees, no permit requirements, and no time of day or
directional restrictions for taking bikes on LRT vehi-
cles. Every train has a low-floor car especially designed
to facilitate bike access, with waiting areas and four bike
hooks located near the doors. But passengers without
bikes have priority to board crowded trains. All buses in
the Portland area have bike racks, another inducement
for cyclists to ride with their bikes instead of parking
them.
By comparison, Portland does not provide much
bike parking at train and bus stops. In 2009, there were
a total of 670 bike parking spaces at TriMet LRT stops
and transit centers (major transfer hubs for several bus
or LRT lines). Of those, almost half were bike lockers.
In addition, there were city-owned bike lockers at
15 locations in downtown Portland, most of which
are near bus or rail services. In sharp contrast to San
Francisco, Portland does not have any bike stations,
which is surprising given its high bike mode share
and wide range of other pro-bike policies and pro-
grams. Bicycling planners and public transport officials
in Portland plan to improve bike parking at transit
stops by installing 196 bike lockers and 168 bike racks
at LRT extensions.
Portland cyclists prefer taking their bikes on board
transit vehicles. A survey by TriMet indicated that 76%
of cyclists would not be willing to park their bikes at
a transit stop even if there were sheltered and secure
bike parking available. The advantage of taking bikes
on buses or rail vehicles is that bikes can then be used at
both ends of the transit trip. So the aversion to parking
bikes at bus and rail stops may be due more to conve-
nience than to concerns over theft or vandalism.
Portland carefully and explicitly coordinates its
bikeway network with its public transport network.
The city has set the goal of a seamless link between
cycling and transit. Bike routes are designed to facili-
tate access to public transport stops. Most transit
centers are served by multiple bikeways. Moreover,
city planners give special consideration to enhancing
bike access to transit stops in outlying areas too far
away from the city center for most people to cover by
bike alone. Nevertheless, as in most cities, there are
some public transport stops that are difficult or dan-
gerous to access by bike. As the largest of America’s
three platinum level Bicycling Friendly Communities,
Portland has already made impressive progress at inte-
grating cycling with public transport and seems sure to
continue on that path.
Vancouver
Similar to San Francisco and Portland, Vancouver has
vigorously promoted the integration of public transport
and cycling. The unique advantage in Metro Vancouver
is TransLink, the fully integrated, multimodal regional
transportation authority. Unlike the other case studies,
public transport, major roadways, and bicycling in
Vancouver are all handled within the same agency. The
coordination of cycling and public transport is obvious
and natural in such a multimodal agency – as reflected in
TransLink’s plans, funding, construction projects,
vehicle procurement, and operating procedures. Over
the past 10 years, TransLink has spent over $12 million
specifically on bike-transit integration.
As in the San Francisco and Portland areas, all buses
in Metro Vancouver are equipped with bike racks.
Similar to San Francisco’s BART, bikes are allowed on
Vancouver’s SkyTrain except during peak hours in the

peak direction due to problems of overcrowding. Until
recently, there were no special accommodations for
bikes on SkyTrains, but all future vehicles will provide
a special area for bikes in the rear of each car with
a leaning rail and fold-up seats. Bikes are allowed at
all times on West Coast Express trains for a $.50 charge.
SeaBus ferries permit bikes on board at all times with-
out charge. Almost all of TransLink’s rail and ferry
services are fully accessible – through elevators,
ramps, or level boarding – thus facilitating bike-
and-ride.
There are bike racks at all SkyTrain and West Coast
Express rail stations as well as park-and-ride lots and
transit nodes with interchanges of several bus or rail
lines. In 2008, Vancouver had a total of 1,060 parking
spaces at transit stops: 660 spaces in racks and 400
secure bike lockers. TransLink plans to increase the
overall supply of bike parking at transit stops in the
coming years, with a special focus on improving
the quality of bike parking, especially secure short-
term bike parking. Bike stations at the most important
transit hubs are also being considered.
Thanks to its multimodal orientation, TransLink
explicitly coordinates bike routes with public transport.
For example, the construction of the Millennium,
Expo, and Canada SkyTrain lines included traffic-
protected, parallel bike routes to foster bicyclist access
to public transport. Another aspect of TransLink’s
multimodalism is the focused promotion of cycling in
central corridors where bus and rail vehicles are the
most crowded, and where cycling has the potential to
divert some of the overload and thus reduce crowding.
That coordination of demand and supply between the
two modes is rare and emphasizes the advantages of
multi-modal agencies such as TransLink.
Minneapolis
Although Minneapolis is, by far, the coldest of the eight
cities, it has the second highest bike share of work trips
after Portland (4.3% vs. 6.0%). Public transport’s share
of work trips is slightly higher in Minneapolis than in
Portland (13.4% vs. 11.2%), but much lower than in
any of the other cities (Fig. 2).
Metro Transit had 497 bike parking spaces at its
light rail and bus stops in 2007: 271 spaces in bike
racks and 226 bike lockers. Minneapolis has a staffed
Bicycle Integration with Public Transport 203
bike station, the Midtown Bike Center, with 100 bike
parking spaces, repairs, rentals, and a café. It is only
a block from the Chicago and Lake Streets transit hub,
which serves two of the city’s busiest bus lines. Perhaps
the city’s most ingenious policy is the official designa-
tion of 35,000 traffic sign posts as bike parking with
70,000 bike parking spaces, assuming two bikes parked
at each post. The city also considers that a form of bike-
transit integration, since many traffic sign posts are
near bus stops, including the bus stop post itself.
All Metro Transit and suburban transit buses are
equipped with exterior bike racks, and the city has five
stationary bike racks for first-time users to practice
loading their bikes. Every light rail vehicle has interior
vertical racks that accommodate four bikes. Bike-and-
ride has become increasingly popular in Minneapolis.
Metro Transit surveys in spring 2007 and fall 2008
found a doubling in the number of bicycles transported
on bus racks and a 41% increase in bikes on light rail.
On an average weekday in 2008 Metro Transit buses
carried 870 bikes. Only 4% of cyclists had to wait for
another bus due to racks being filled.
There is no explicit policy of coordinating bike
routes and transit stops in Minneapolis, and city offi-
cials emphasize the need to improve cycling facilities
feeding into public transport stops. Some bike routes
already lead to transit stops or parallel transit lines. For
example, the Hiawatha LRT line runs parallel to an off-
street bike path for most of its length in Minneapolis,
thus providing an alternative for cyclists on that route.
Chicago
With the second largest transit system in the USA,
Chicago has made impressive efforts to integrate
cycling with public transport. Its special distinction
lies in the innovative provision of bike parking at rail
stations, tailoring the design of parking facilities to each
station’s particular situation. With 6,420 parking spaces
at its rail stations, Chicago has about the same amount
of bike-and-ride parking as the San Francisco Bay Area,
and far more than other cities in North America.
There are 2,153 bike parking spaces at 131 of the 143
CTA subway and elevated rail stations and 4,267 spaces
at 50 of the 76 Metra suburban rail stations. Moreover,
indoor or sheltered parking is available at 83 CTA sta-
tions, more than any other transit system in North
204 Bicycle Integration with Public Transport
America. The specific location of bike racks inside the
stations provides both weather protection and greater
security, since they are usually placed within easy sight of
station attendants and other passengers. Chicago is cur-
rently installing additional sheltered bike parking for 382
bikes at four CTA stations, and the city has funding to
install bike shelters for 250 more bikes in 2010.
The largest bike station in the USA is located in
Chicago’s Millennium Park, immediately above the
terminal station for two of Chicago’s suburban rail
lines. The bike station is easily accessible from down-
town Chicago and the 18-mile Lakefront Trail. It pro-
vides secure, indoor parking for 300 bikes as well as
convenient lockers, showers and towel service, bike
rentals, bike repairs, and guided bicycling tours.
The CTA regularly monitors bike parking needs by
conducting bi-annual bike rack inventories and mea-
suring usage rates. It then works with Chicago DOT to
install additional racks where needed.
As in most of the case study cities, all of Chicago’s
buses have bike racks – including CTA buses as well as
PACE suburban buses. Bikes are permitted on CTA and
Metra trains except during weekday rush hours. And
just as most of the other cities, no fees or permits are
required for the use of bike racks on buses or for
bringing bikes on trains.
The biggest challenge to bike-transit integration in
Chicago is the difficult access to train platforms. Because
most of the rail lines are so old, only 54% of CTA stations
and 68% of Metra stations are ADA accessible. Thus,
cyclists are often forced to carry their bikes up long
flights of stairs. Few stations have elevators, and cyclists
are not permitted to use escalators.
Chicago DOT, transit agencies, and the cycling
community are aware of these problems and have
made improvements in bike-transit integration a top
priority. Chicago’s Bike Plan 2015 sets goals of further
expanding and improving bike parking inside and out-
side of rail stations, remodeling stations to make them
more accessible to bikes, providing more park-and-ride
facilities, and establishing a second bike station with
better transit connections.
Toronto
The combination of high rates of transit use and rising
cycling levels has prompted a range of efforts to integrate
the two modes in the Toronto metropolitan area. The
main approach has been to provide ample bike parking.
With over 15,000 post-and-ring bike racks throughout
the city, Toronto has more bike parking than any other
city in North America. That includes bike parking at
almost all rail stations. In 2008, there were 1,192 short-
term spaces in bike racks at Toronto Transit Commission
(TTC) subway stations and 579 short-term spaces in
racks at GO Transit suburban rail stations.
Yet there is a severe shortage of secure bike parking,
with only 114 bike lockers in the entire transit network.
Consequently, Toronto plans on greatly expanding the
supply of secure parking in 2009 and 2010 through
installation of more bike lockers and completion of
a new bike station at Union Station, the main transit
hub in downtown Toronto, providing bus, streetcar,
subway, and suburban rail connections. The bike sta-
tion will provide secure, sheltered parking for 200
bikes. Construction of an even larger bike station at
City Hall is planned to begin in 2010. That facility will
be close to several bus and streetcar lines. The GO
Transit suburban rail system is improving its bike
parking by expanding sheltered parking to all stations
by the winter of 2009–2010.
As in most cities with high levels of rail transit use,
bikes cannot be taken on TTC subways and streetcars
during weekday peak hours. Even when permitted,
there are no special provisions for bikes on TTC subway
cars. Similarly, bikes are not allowed on any GO Transit
trains headed toward Union Station in the morning
peak (6:30–9:30) or departing Union Station in the
evening peak (3:30–6:30). Folding bikes are permitted
on all public transport vehicles at all times. Bike access to
rail transit is limited by the lack of elevators in most
subway stations. As in Chicago, cyclists in Toronto must
carry their bikes up and down long flights of stairs to reach
the train platforms. Only 41% of TTC subway stations are
wheelchair accessible, while 75% of GO Transit stations are
accessible, either through elevators or ramps.
Toronto is making rapid progress equipping its
buses with bike racks, which can be used at any time,
even during peak periods. In 2008 only 55% of TTC
buses had bike racks, but all new buses have racks, and
every month about 40 older buses are retrofitted with
racks. By the end of 2010, all TTC and GO Transit buses
will have bike racks, thus facilitating bike and ride
throughout the region.

There is almost no explicit coordination of bike
routes with transit routes and station stops. Instead,
the guiding principle of bike route planning is to put
every Torontonian within a 5-min bike ride of the
bikeway network. The many transit stations and fine-
grained street network in much of the central city
facilitates bike access to TTC stations. In suburban
areas, however, many streets are circuitous and do not
connect across arterials, making it difficult for cyclists
to avoid major arterials while en route to a transit
station.
Although bike-transit integration in Toronto faces
some serious challenges, much progress is being made,
partly thanks to the Province of Ontario’s Metrolinx
BikeLinx Program, which helps finance bike parking at
stations and bike racks on buses. With the two new bike
stations, additional bike lockers, expansion of sheltered
parking, and completion of bike rack installation on all
buses by 2010, Toronto will have made great strides
coordinating cycling and public transport.
Washington
Bike-transit integration in Washington is similar to the
situation in the San Francisco Bay Area. Both metro-
politan areas rely on regional metro systems started in
the late 1960s and early 1970s: BART in San Francisco
and Metrorail in Washington. There is bike parking at
almost all of Washington’s 86 Metrorail subway sta-
tions, with a total of 1,800 bike racks and 1,300 bike
lockers. The parking facilities are popular, with usage
rates at most stations ranging from 50% to 100%.
In October 2009, a new bike station with spaces for
150 bikes will open next to Union Station, providing
convenient connections to Metrorail as well as subur-
ban trains leaving from Union Station. The bike station
will also offer bike rentals, repairs, and accessories as
well as storage lockers and changing rooms.
In 2008 a new bike sharing program began in
Washington, similar in technology to the Velib system
in Paris, but on a much smaller scale: only 120 bikes
compared to over 20,000 bikes in Paris. It facilitates
bike-and-ride because eight of the ten bike sharing
docking stations are at Metrorail stops. The short-
term rental bikes can be used to get to and from
Metrorail stations, thus serving as feeders and distrib-
utors for transit.
Bicycle Integration with Public Transport 205
Bikes are allowed on Metrorail trains except during
morning and afternoon rush hours on weekdays: two
bikes per car on weekdays, four bikes per car on week-
ends. Unlike the much older subway systems in New
York and Chicago, all 86 Metrorail stations have eleva-
tors (271 in total) and are ADA accessible. That facili-
tates access to platforms for cyclists as well, who are, in
fact, required to use the elevators and are not permitted
on escalators. In contrast to the Metrorail, VRE and
MARC suburban trains do not allow bikes on board at
any time unless they are folding bikes. All 1,450
WMATA buses have bike racks, but some buses run
by suburban agencies do not.
In theory, bike plans for the Washington area estab-
lish the goal of coordinating bike routes with transit
routes, but in fact, nothing has really been accom-
plished in this area except by accident. Overall, how-
ever, bike-transit integration in the Washington area is
successful. The biggest gap is the exclusion of bikes
from the Metrorail system during peak hours. To
some extent, this is unavoidable due to the
overcrowding of metro cars, just as on the BART,
Toronto, and Chicago subway systems. But it reinforces
the need to expand bike parking at the many Metrorail
stations where current capacity is insufficient.
New York/New Jersey
With 55% of all work trips by public transport, New York
City has, by far, the highest transit mode share of any city
in North America. Thus, one might expect substantial
efforts to coordinate cycling with public transport. In
fact, New York City’s transit systems have done little to
promote bike-transit integration, far less than any other
city in this study. The Metropolitan Transportation
Authority (MTA) does not provide bike parking of any
kind at the city’s 467 subway stations, so the only option
for cyclists is to park on nearby sidewalks. The MTA’s
suburban railroads, the Long Island Railroad (LIRR) and
Metro-North Railroad (MNR), offer bike parking at
some of their stations – but MTA has no information
on the total number of spaces (MTA, 2009).
Compounding the problem of insufficient bike
parking along the many subway and suburban rail
lines of the MTA, there is no secure bike parking at
any public transport terminals in Manhattan. Train,
bus, and ferry terminals do not even offer short-term
206 Bicycle Integration with Public Transport
parking in bike racks. Thus, cyclists must seek out
the occasional bike rack on sidewalks within a few
blocks of the terminals or risk having their bikes
confiscated if parked at traffic signposts, which is illegal
in New York.
NYC subways are unique in permitting bikes on
board trains at all times, but it is difficult to get bikes
to the platforms. Only 16% of New York’s subway
stations are ADA accessible via elevators or ramps. At
the remaining 84% of stations, cyclists must carry their
bikes up and down long flights of stairs, as they are
prohibited from using escalators in stations where they
are available. Bikes are allowed on the MTA’s two sub-
urban railroads (MNR and LIRR) except during
peak hours in the peak direction, but cyclists must
register in advance and purchase $5 lifetime permits.
Folding bikes are allowed at all times. In 2008, 18%
of LIRR stations and 52% of MNR stations were not
ADA accessible – considerably less than the 84%
inaccessible MTA subway stations, but still a problem
for cyclists having to carry their bikes up and down
stairs.
Bike-bus integration is almost non-existent in New
York City. Not a single bus in the MTA’s fleet of 5,929
buses has a bike rack. That contrasts sharply with 100%
of buses equipped with bike racks in most of the other
case-study cities. Only since spring 2008 have folding
bikes been allowed on most MTA buses.
There is no explicit effort to coordinate bike routes
with transit routes in New York City. That is not
a severe problem in most of the city because the transit
network is so dense that most neighborhoods are
served by a nearby subway or bus line. In the suburbs
and the outermost portions of the city, however, the
complete lack of bike-transit route integration is
a serious shortcoming. Precisely in those lower density
areas where cycling would provide an ideal feeder mode
to more distant transit stops, bikeways are almost
exclusively recreational paths that do not connect to
practical destinations such as transit stations.
Not only does New York compare unfavorably in its
bike-transit integration to the seven other case study
cities, but it is surpassed by the New Jersey portion of
the Greater New York metropolitan area. In 2008, New
Jersey Transit (NJT) offered bike parking at 90% of its
suburban rail stations and 80% of its light rail stations,
with a total of 2,400 spaces, including 376 secure bike
lockers. Unlike the MTA in New York, NJT provides
bike racks at all three of its major terminals in Hoboken
and Newark (Penn Station and Broad Street). Bikes
are allowed on all suburban rail and light rail lines
except during rush hours in the peak direction, and
no permits or fees are required. Folding bikes are
allowed on all NJT vehicles at all times. Roughly half
of NJT’s 2,000 buses are equipped with bike racks, and
an additional 200 buses are outfitted with racks each
year. By 2014, 95% of NJT’s buses will have bike racks.
The main problem in New Jersey is the almost complete
lack of bike paths and lanes leading to NJT rail stations
and bus stops.
Summary of Case Studies
Most of the case study cities have greatly improved the
coordination of bicycling and public transport in
recent years. They have increased bike parking at transit
stops and better accommodated passengers wanting to
take their bikes with them on buses and rail vehicles.
Only a few transit systems have measured the actual
extent of bike and ride, but the available evidence is
encouraging. In Washington DC, for example, the
number of bicyclists riding on Metrorail increased by
60% between 2002 and 2007. At some stations, cyclists
accounted for up to 4% of all passenger boardings. In
Minneapolis, Metro Transit carries over 250,000 bicy-
cles annually and reports a doubling of bikes on buses
between spring 2007 and fall 2008. Roughly 4% of
Portland MAX light rail passengers carry their bikes
onto the vehicles with them. In the San Francisco Bay
Area, the share of passengers accessing BART stations
by bike rose from 2.5% in 1998 to 3.5% in 2008, with an
average of 10,920 bike-and-ride trips per day.
As shown in Table 1, there is considerable variation
among the eight case studies. The San Francisco Bay
Area, for example, provides the full gamut of bike-
integration measures and has been at the vanguard of
innovations to promote bike-and-ride. By comparison,
New York’s transit systems have made few provisions to
accommodate cyclists, lagging behind the other case
study cities in both the quantity and quality of bike-
integration measures. All eight of the cities have plans
to further improve bike-transit integration. Thus, it
seems certain that the promising trends of recent
years will continue.
 Bicycle Integration with Public Transport 207

Bicycle Integration with Public Transport. Table 1 Overview of bike-transit integration measures in eight large Amer-
ican and Canadian cities [37–54]
Bikes on transit
Bike parking at transit Bike routes and transit Highlights: strengths
stops and stations Bike racks on buses Bikes on trains stops and challenges
San Francisco (population city: 0.8 m, metro: 4.2 m)
BART provides 4,313 All buses have bike BART allows bikes on all Bike routes often parallel Strengths: Racks on all
bike parking spaces, racks (usage rates vary non-peak trains and MUNI bus routes and buses; extra space for
including 716 from 17% to 43%) during peak times in intersect with transit bikes on Caltrain cars;
conventional and 294 non-peak directions. stops. No explicit bike access to BART off-
electronic bike lockers BART expanded on- planning process peak; extensive bike
board space for bikes on coordinating transit and parking at BART and
select cars cycling routes in SF Caltrain stations; five
bike stations
Caltrain’s 32 suburban Caltrain: special lead cars The Regional Bicycle Challenges: No bike
rail stations offer 400 designated for bike use Plan as well as the Bike access on MUNI light rail;
bike racks and 1,100 bike all day, space for 16–32 Plans of BART and need more dedicated
lockers bikes Caltrain encourage space for bikes on
coordination of bike Caltrain; no coordination
Five bike stations with
routes and facilities with of bike routes with MUNI
659 bike parking spaces
public transport routes; no rush hour
at BART and Caltrain
access for bikes on BART
Stations
Portland (population city: 0.5 m, metro: 2.1 m)
Bike racks and/or lockers All buses have bike MAX light rail: bikes Explicit coordination of Strengths: Racks on all
at almost all TriMet LRT racks permitted if space is bike routes with transit buses; dedicated space
stops and transit hubs available on cars, but stops with the goal of on LRT for hanging bikes
providing 670 bike non-cyclists passengers establishing a seamless near doors; free access
parking spaces; have priority link between the two for bikes 24/7; easy
including 320 secured modes access to low-floor trains
spaces
15 downtown locations Challenges: Lack of good
with bike lockers; usually quality and quantity of
close to light rail or bus parking at most train
stops stations; poor design of
access to stations by
bike; cyclists are often
denied boarding on LRT
during peak
Vancouver (population city: 0.6 m, metro: 2.1 m)
660 bike parking spaces All buses have bike SkyTrain: bikes allowed Central Valley Greenway Strengths: Bike racks on
in bike racks at SkyTrain racks anytime except in peak and BC Parkway facilitate all buses; bikes allowed
stations and at transit periods in peak direction access to and from on SkyTrain except
transfer nodes transit stations during peak hours in
peak direction;
integration of bike
network and transit
stops; bike parking at all
major transit stops; 24/7
no cost bike access on
SeaBus
400 secure bike lockers SeaBus: bikes allowed at Translink’s $2.55 million Challenges: Need more
at most Sky Train and all all times annual cost sharing of and better bike parking;
West Coast Express cycling infrastructure need more capacity on
stations favors projects that SkyTrain to
facilitate access to transit accommodate cyclists
208 Bicycle Integration with Public Transport

Bicycle Integration with Public Transport. Table 1 (Continued)

Bikes on transit
Bike parking at transit Bike routes and transit Highlights: strengths
stops and stations Bike racks on buses Bikes on trains stops and challenges
Minneapolis (population city: 0.4 m, metro: 3.2 m)
271 bike parking All buses have bike 27 light rail vehicles with No explicit coordination Strengths: Bike racks on
spaces and 226 bike racks. Bike on bus on-board interior vertical of bike routes and buses and trains; bike
lockers next to Metro demonstration racks in racks that accommodate transit. But connection racks and lockers at most
Transit stops and at various locations 4 bikes per vehicle to transit lines is train stations; marketing
park-and-ride lots. Usage a prioritizing criterion for information on bike
rate of 28% of bike bike projects funded by transit integration
parking spaces at the MPO provided by Metro
Hiawatha LRT line Transit
City bicycle ordinance The Hiawatha LRT line Challenges: Bike racks
encourages parking parallels an off-street needed at four more LRT
bikes at sign posts, bike path for most of its stations; lack of bike
which are within a block length in Minneapolis paths and lanes leading
of most bus stops. Bike to train stations and bus
station 1/10 of a mile stops
from the Chicago and
Lake Street transit hub
Chicago (population city: 2.7 m, metro: 9.5 m)
2,153 bike parking All CTA and PACE buses Bikes permitted on CTA The routing of on-street Strengths: Ample bike
spaces at 131 of the 143 have bike racks and METRA trains during bikeways in Chicago parking at 131 CTA
CTA stations and 4,267 off-peak. Bikes not takes the location of stations, of which 83
bike parking spaces at 50 permitted on escalators transit stations into provide indoor parking.
of the 76 METRA in stations account. CTA system State-of-the-art
suburban rail stations. maps show bike parking Millennium bike station
Most (83) CTA stations possibilities at stations at downtown terminus
provide indoor or of suburban rail line; CTA
sheltered bike parking. regularly collects bike
Usage rates of 90% and rack usage data to add
50% for indoor and parking where needed
outdoor racks
Largest bike station in Challenges: The CTA
the USA with indoor subway is old with
parking for 300 bikes, limited ADA facilities;
right above the terminal flights of stairs make it
station for two of difficult to bring bikes on
Chicago’s suburban rail board
lines
Toronto (population city: 2.5 m, metro: 5.5 m)
1,771 short-term bike 55% of TTC buses and Bikes permitted on TTC The guiding principle of Strengths: Extensive
parking spaces and 114 all GO Transit buses subways, RT trains, bike facility planning is transit system with high
bike lockers at TCC have bike racks streetcars, and buses to put every resident mode share; bike racks
subway and GO Transit without bike racks with a 5 min bike ride of on most buses; bikes
stations except during peak the bike network allowed off-peak on
hours streetcars, subways, and
suburban rail; bike
parking at most transit
stops
 Bicycle Integration with Public Transport 209

Bicycle Integration with Public Transport. Table 1 (Continued)

Bikes on transit
Bike parking at transit Bike routes and transit Highlights: strengths
stops and stations Bike racks on buses Bikes on trains stops and challenges
New bike station at All TTC buses will be Bikes permitted on GO The Waterfront Trail Challenges: No bikes
Union Station (opening equipped with bike Transit trains except on which spans the Greater allowed on rail transit
spring 2009) with secure, racks by end of 2010 trains arriving and Toronto Area parallels during peak hours; lack
sheltered parking for 200 departing from Union the busy GO Rail of elevators in many
bikes. Second bike Station in peak direction Lakeshore corridor and is stations; more bike
station planned in close proximity to parking and secure
many stations lockers needed at
stations; no coordination
of bike routes with
transit stops
Washington, DC (population city: 0.6 m, metro: 5.3 m)
1,800 bike racks and All WMATA buses have Bikes allowed on Metro Drafting of DC bike plan Strengths: Bike racks on
1,300 bike lockers at bike racks trains during off-peak. took the location of all buses; bikes on Metro
Metro stations. No bike Cyclists must use metro stops into except during peak;
parking at bus stops, elevators between the account, particularly for lockers and racks at all
except major transit street and the station the bike route network stations; SmartBike
hubs with rail stations platforms and facility short-term rentals at
recommendations 8 Metro stations
Bike station for 150 bikes Bikes not allowed on Challenges: Restricted
to open in July 2009 at MARC trains. VRE allows bike access hours on
Union Station bicycles on some trains Metro; bikes restricted
on VRE and MARC
suburban rail
New York City (population city: 8.2 m, metro: 18.8 m)
No dedicated bike No buses with bike Bikes permitted on the No explicit coordination; Strengths: 24/7 bike
parking at rail stations or racks in NYC; Roughly NYC subway at all times, but the transit network access and no fees or
transit terminals. Some half of NJ Transit busses but only 16% of stations in the city is so dense permits on NYC subway
bike parking provided by have bike racks in 2010 ADA accessible, with that most of the bike system, most extensive
NYCDOT on sidewalks, (planned increase to elevators or ramps network is close to in North America; big
often near transit stops. 95% by 2014) transit increase in bike racks
No secure bike parking and bike parking on NJ
of any kind in NYC Transit
Bike parking at a third of All suburban trains allow No coordination of bike Challenges: No bike
MTA suburban train bikes on board during network with transit racks on MTA buses in
stations in NYC; bike off-peak hours, but limit stations and routes in NYC; limited parking at
lockers at 18 LIRR and 6 of 2–4 bikes per car on suburbs; most suburban MTA suburban rail
MNR stations; bike weekdays; NJ Transit LRT cycling facilities are stations and no bike
parking at 90% of NJ allows bikes on trains recreational and do not parking at two main
Transit train stations and during off-peak hours lead to transit stops train stations in
80% of LRT stations (a with a maximum of 4–6 Manhattan; no secure
total of 2,400 bike bikes bike parking; difficult to
parking spaces, with 376 access platforms: only
bike lockers) 16% of subway stations
with elevator access
210 Bicycle Integration with Public Transport
Future Directions
North American cities have been making impressive
progress integrating cycling with public transport.
Since 2000, the percentage of buses with bike racks
has almost tripled. Bike-rail integration has also
advanced. Most light rail, metro, and suburban rail
systems permit bikes on their rail vehicles except dur-
ing peak hours, and they increasingly provide special
accommodations for cyclists such as bike hooks, racks,
and rails in special areas of rail cars. Complementing
provisions for bikes on transit vehicles, bike parking at
transit stops has been vastly expanded over the past
10 years, with large increases in the number of racks as
well as improvements in the convenience, security, and
shelter of bike parking.
While cycling and public transport have consider-
able synergies, there are some inevitable conflicts. Sur-
veys in some cities indicated that cyclists prefer to take
their bikes with them on rail vehicles so they can use
them at both ends of the trip. That can cause problems
during peak hours, however, when all available capacity
is needed to accommodate passengers, and there is no
extra room for bikes. Taking bikes on buses is much less
of a problem since bike racks are external and do not
reduce passenger-carrying capacity. But even bike racks
can be filled to capacity during the peak, forcing cyclists
to wait for later buses.
Paradoxically, bike-and-ride can become problem-
atic where it is most successful. Capacity problems are
most likely to arise in cities with well-used public
transport and high levels of cycling. That is why the
European approach to bike-and-ride has favored the
provision of ample, sheltered, secure bike parking
at transit stops instead of accommodating bikes on
transit vehicles. Similarly, in North American cities
with overcrowding of rail vehicles during rush hours,
the focus should probably be on providing improved
bike parking at rail stations. Not only is more parking
needed, but it should be of higher quality, with more
sheltered and secure spaces. Major transit terminals
should include multi-service bike stations, such as
those in northern Europe. Similar to the concept of
“complete streets,” an appropriate goal of transit sys-
tems in North America should be to provide “complete
stations,” which fully accommodate the needs of
cyclists. That includes making rail platforms more
accessible to cyclists, which would also improve acces-
sibility for persons with disabilities.
Such bike-and-ride provisions cost money, but they
are much cheaper than park-and-ride facilities for
motorists [15]. Transit systems should shift their focus
from park-and-ride to bike-and-ride, which is more
cost-effective as well as more environmentally friendly.
To encourage that shift, federal, state, and local govern-
ment agencies should vastly expand funding for further
improvements in bike-and-ride measures.
Bibliography
Primary Literature
1. Hegger R (2007) Public transport and cycling: living apart or
together? Public Transp Int 2:38–41
2. Brons M, Givoni M, Rietvield P (2009) Access to railway stations
and its potential in increasing rail use. Transp Res A 43:136–149
3. Givoni M, Rietveld P (2007) The access journal to the railway
station and its role in passengers’ satisfaction with rail travel.
Transp Policy 14:357–365
4. Martens K (2004) The bicycle as a feedering mode: Experiences
from three European countries. Transp Res D 9:281–294
5. Martens K (2007) Promoting bike and ride: the Dutch experi-
ence. Transp Res A 41:326–338
6. McClintock H, Morris D (2003) Integration of cycling and light
rail transit. World Transp Policy Pract 9(3):9–14
7. Rietveld P (2000) The accessibility of railway stations: the role
of the bicycle in the Netherlands. Transp Res D 5:71–75
8. Tschopp J (1997) Integration of bicycles and public transport
in Basel, Switzerland. In: Tolley R (ed) The greening of urban
transport. Wiley, Chichester, pp 441–446
9. Pucher J, Buehler R (2007) At the frontiers of cycling: policy
innovations in the Netherlands, Denmark, and Germany.
World Transp Policy Pract 13(3):8–57
10. Litman T (2009) Bicycle parking, storage, and changing facili-
ties. VTPI, Victoria
11. Pucher J, Dill J, Handy SL (2010) Infrastructure, programs and
policies to increase bicycling: an international review. Preven-
tive Medicine 50(S1):S106–S125
12. Pucher J, Buehler R (2008) Making cycling irresistible: lessons
from the Netherlands, Denmark, and Germany. Transp Rev
28(1):495–528
13. Bassett D et al (2008) Walking, cycling, and obesity rates in
Europe, North America and Australia. J Phys Activ Health
5(6):795–814
14. USDOT (1998) Bicycles and transit: a partnership that works.
U.S. Department of Transportation, Federal Transit Adminis-
tration, Washington
15. TRB (2005) Integration of bicycles and transit – TCRP Synthesis
Report 62. Transportation Research Board, National Research
Council, Washington, DC

16. APTA (2009) 2009 public transportation fact book appendix a:
historical tables. American Public Transportation Association,
Washington
17. APTA (2009) Transit statistics. American Public Transportation
Association, Washington
18. APTA (2009) 2008 public transportation fact book. American
Public Transportation Association, Washington
19. U.S. Census Bureau (2009) American fact finder: 1990 decen-
nial census, journey to work. U.S. Department of Commerce,
Washington
20. U.S. Census Bureau (2009) American fact finder: 2008 Ameri-
can Community Survey, Journey to Work. U.S. Department of
Commerce, Washington
21. StatCan (2009) Place of work highlight tables: employed
labor force by mode of transportation. Statistics Canada,
Ottawa
22. Clarke A (2003) Green modes and US transport policy: TEA-21.
In: Tolley R (ed) Sustainable transport: planning for walking
and cycling in urban environments. Woodhead Publishing,
Cambridge
23. Thunderhead Alliance (2007) Bicycling and walking in
the U.S. – Benchmarking report 2007. Thunderhead Alliance,
Prescott
24. USDOT (2004) National bicycling and walking study: A ten-
year status report. U.S. Department of Transportation, Federal
Highway Administration
25. TAC (1998) A new vision for urban transportation. Transporta-
tion Association of Canada, Ottawa
26. Transport Canada (2003) Sustainable development strategy
2004–2006. Transport Canada, Ottawa
27. CUTA (2008) Canadian urban transit fact book. Canadian
Urban Transit Association, Toronto
28. Pucher J, Buehler R (2006) Why Canadians cycle more than
Americans: a comparative analysis of bicycling trends and
policies. Transp Policy 13(1):265–279
29. APTA (2006) Transit infrastructure database. American Public
Transportation Association, Washington, DC
30. APTA (2008) 2008 Transit infrastructure database. American
Public Transportation Association, Washington, DC
31. Dutch Bicycling Council (2006) Continuous and integral: the
cycling policies of groningen and other European cities.
Fietsberaad, Amsterdam
32. Buehler R, Pucher J (2010) Cycling to sustainability in
Amsterdam. Sustain: J Environ Sustain Issues 21:36–40
33. APTA (2008) 2008 Public transportation vehicle
database. American Public Transportation Association,
Washington, DC
Bicycle Integration with Public Transport 211
34. Statistics Canada (2009) Employed labour force by means of
transportation. Statistics Canada, Ottawa
35. Pucher J, Lefèvre C (1996) The urban transport crisis in Europe
and North America. Macmillan, Basingstoke
36. TRB (2001) Making transit work: insight from Western Europe,
Canada and the United States. Transportation Research
Board, National Research Council, National Academy Press,
Washington, DC
37. Alliance for Biking and Walking (2010) Bicycling
and walking in the United States: 2010 Bench-
marking report. Alliance for Biking and Walking,
Washington, DC
38. WMATA (2010) Bike and ride. Washington Metropolitan Area
Transit Authority, Washington, DC
39. WABA (2010) Bikes on transit. Washington Area Bicyclist
Association, Washington, DC
40. TransLink (2010) Cycling support services study: strategic
plan. Third Wave Cycling Group. Via Architecture and Halcrow
Consulting, Vancouver
41. SFMTA (2009) San Francisco bicycle plan. San Francisco
Municipal Transport Authority, San Francisco
42. NYCDOT (2010) Bicycling. New York City Department of
Transportation, New York City
43. NJTransit (2009) NJTransit – bike program. New Jersey Transit,
Newark
44. MetroTransit (2010) Resources for bicycling. MetroTransit,
Minneapolis
45. METRA (2009) Bikes on trains program. METRA,
Chicago
46. DCDOT (2006) Bicycling master plan. Department of Transpor-
tation, Washington, DC
47. Bike Walk Twin Cities (2008) Snapshot Minneapolis: bicycling
and walking. Bike Walk Twin Cities, St. Paul
48. City of Minneapolis (2010) Bicycles and transit. City of
Minneapolis, Minneapolis
49. City of Chicago (2009) Bike 2015 plan-chapter 4. City of
Chicago, Chicago
50. City of Portland (2010) Portland bicycle plan for 2030. Office of
Transportation, Portland
51. DCDOT (2010) Bicycling in Washington, D.C. Department of
Transportation, Washington, Washington, DC
52. City of Toronto (2001) Shifting gears: city of Toronto bike plan.
City of Toronto, Toronto
53. CTA (2009) Bike and ride. CTA, Chicago
54. McDonald’s Cycle Station (2006) History of the bike station.
Chicago Bike Station, Chicago
212 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
Bus Rapid Transit and Light Rail
Transit Systems: State of Discussion
CLIFF HENKE
Parsons Brinckerhoff, Newark, NJ, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
BRT Developments
Light Rail Developments
Current and Future Trends
Conclusion
Bibliography
Glossary
Bus rapid transit (BRT) An integrated bus-oriented
public transport service composed of six basic ele-
ments: higher-quality, more styled buses; a well-
designed service plan, including route structure is
faster and more frequent transit service than typi-
cally experienced with regular route local bus ser-
vice; upgraded, appealing passenger stations;
roadway infrastructure that gives BRT priority in
mixed traffic or separated from other traffic entirely
to help ensure faster travel times; intelligent trans-
portation technologies, not only for traffic priority
as mentioned above but also for real-time passenger
information and vehicle tracking to ensure more
ambitious service schedule; and clear, distinct
branding and marketing [1].
Bus rapid transit vehicle A bus used in BRT service,
typically with advanced features such as aerody-
namic styling, hybrid electric drive or other
advanced propulsion, and real-time tracking sys-
tems, similar to what are now used in rail transit
systems.
Buy America Provisions of federal transit law that
govern purchases of goods with federal surface
transportation assistance grants. Certain aspects of
this policy date back to protectionist sentiments of
the 1930s. In 1978, Congress stipulated that for
purchases of rolling stock, these would be consid-
ered to be US made if they were assembled with at
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
least 60% of components and materials by value
supplied by US sources. The rules have been subse-
quently refined and subject to controversy several
times since [3].
Capital assistance/expenditures Government subsi-
dies for expenses related to the purchase of equip-
ment and facilities, such as stations, buses, and
trains; maintenance facilities; and support systems.
Such equipment typically means property that has
a useful life of more than 1 year. Capital expenses do
not include operating expenses that are eligible to
use capital funds.
Commuter rail service Regional-distance, usually
locomotive-hauled passenger train service often
sharing tracks with freight railroads. Chicago’s
Metra service is an example.
Federal Transit Administration A sub-cabinet-level
government agency of the US Department of Trans-
portation that administers a variety of grant pro-
grams that help fund bus and rail transit projects in
US cities, Puerto Rico, and the US territories, as well
as rural areas and federal lands.
Fixed-route bus service Traditional form of transit
bus service, typically with frequent stops along
a defined route. The bus services of the Chicago
Transit Authority are an example.
Light rail transit (LRT) A form of urban rail public
transportation that generally has a lower capacity and
lower speed than metro (also known as heavy rail)
systems, but higher capacity and higher speed than
traditional streetcar systems. These rail systems are
typically electrically powered, usually by overhead
catenary. They ideally operate their own rights-of-
way separated from other traffic but if necessary can
also operate in mixed traffic on city streets [2].
Modal split Term of art in public transportation for
market share. That is, the percentage of journeys
taken with the various forms of public transporta-
tion, as compared with walking, bicycling, or the
private automobile.
Operating assistance/expenditures Government sub-
sidies for expenses associated with the operation of
the transit agency and its passenger services, includ-
ing vehicle operations (e.g., bus, train, “dial-a-ride,”
and vanpool services), vehicle and facility mainte-
nance, and general administration (e.g., marketing,
# 2012, DOI 10.1007/978-1-4419-0851-3
 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
top management, insurance, and finance). Operat-
ing expenses also include salaries, wages, and ben-
efits as well as outsourced services under contract.
Privatization In its strictest definition, it is the return
of publicly owned and operated public transport
services to the private sector. However, in the USA,
it is also defined as the contracting of publicly
owned services to private companies, private par-
ticipation in the capitalization or ownership of
these publicly owned services, or the franchising
of these services to private firms.
Public transport service Bus and rail services operated
either by public or private organizations on a scheduled
or on-demand basis. These services are distinct from
charters, which are hired for groups such as tours.
Public–private partnerships/models Synonymous
with most definitions of privatization except the
complete return of transit services to private own-
ership and operation with minimal oversight.
Purchased transportation services Bus and train ser-
vice provided by others through a contract; also
called contract service.
Rolling stock Industry term for buses and trains and
other vehicles used in public transport service.
Transit-oriented development Land use patterns
such as higher densities around BRT and rail sta-
tions that also contribute to greater transit usage on
these systems.
Transportation Equity Act for the 21st Century Also
known as TEA 21, the landmark surface transpor-
tation assistance law passed in 1998 that ended
operating assistance for agencies that served urban-
ized regions with fewer than 200,000 residents.
It also for the first time guaranteed a stream of
funding for capital assistance over the 6-year period
in which the law was effective, which helped public
transportation agencies better manage and plan
their capital expenditures.
Travel time savings Industry term for the average
estimated savings per trip on new transit systems
as compared with the average travel time from
a commute taken by an automobile or by the pre-
vious bus network serving the route.
Urban Mass Transportation Act The 1964 federal law
that began the federal government’s involvement in
the oversight and financial assistance of regional,
state, and local public transport services.
213
Vehicle maintenance Activities associated with bus,
railcar, and support vehicle (e.g., fueling and repair
trucks) maintenance (repairs and routine replace-
ment of parts) and servicing (cleaning, fueling,
etc.), also including inspection and oversight of
these activities. In addition, vehicle maintenance
includes repairs due to vandalism and accidents.
Vehicle operations Those activities associated with
providing scheduled and on-demand passenger ser-
vices, such as vehicle dispatching and scheduling,
driver and technician training, ticketing and fare
collection management, and system security.
Definition of the Subject and Its Importance
Increasingly, light rail transit (LRT) and bus rapid
transit (BRT) are the two solutions cities throughout
the world most often turn to in a broad variety of urban
public transport applications to address a wide range of
mobility challenges. While regional rail schemes (also
called commuter rail in the USA and Canada) are also
experiencing rapid growth, and many cities that have
metros (also known as heavy rail in the USA and
Canada) continue to expand their networks, LRT and
BRT are even faster growing modes.
Whether LRT, BRT, metros, or commuter railroads,
virtually all large cities in the developed world now
boast at the very least starter urban passenger rail
systems, in many instances restoring only partly the
networks these cities had a century or more ago.
Many of these have since supplemented the initial seg-
ments with a second wave of lines to build out their
networks. Meanwhile, a growing number of midsize
cities in the developed world, including a rapidly grow-
ing list of cities in developing countries, have also
added networks. Call these stages the first two waves
of the worldwide public transport renaissance. Now
public transport infrastructure investment is entering
a third wave of the renaissance, in which the growth of
public transport investment becomes an essential and
integral part of a nation’s or a region’s balanced trans-
port strategy.
BRT is perhaps the fastest growing public transport
mode in a century. Only 12 years ago, 17 US cities were
selected by the then Federal Transit Administrator
Gordon Linton to participate in his BRT Consortium.
Just 5 years later, according to congressional sources,
214 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
more than 50 and as many as 75 communities have
a BRT project in some stage of development. Today,
virtually every project seeking federal financial support,
regardless of mode, also studies a BRT alternative, thus
putting the mode on equal policy consideration with
other public transport strategies.
The simplest and most often used definition of BRT
is a public transport concept that “thinks like rail but
uses buses.” The interest in it is borne out of desire to
offer increasingly congested cities a flexible, relatively
inexpensive alternative to more highways or more
expensive and longer-to-implement transit alterna-
tives. Applications range from feeder systems to rail
lines (e.g., Los Angeles) to downtown circulators
(e.g., Orlando and Denver) to urban transport spines
in their own right that can change the region’s modal
split and stimulate transit-oriented investment (e.g.,
Bogotá, Curitiba and Eugene, Oregon).
Similarly, applications of LRT range from spines for
smaller cities that cannot justify the expense of metro
systems to feeder systems to metros or regional passen-
ger railways to circulators that often run in the street
with other traffic. These last systems are often called
trams or streetcars, as well. The flexibility of LRT also
results in systems that operate in their own rights-of-
way either underground, in elevated structures or at
grade, as well as combinations of all of the above.
Since 1975, 23 cities in North America alone have
fueled the rebirth of LRT development. Prior to that
year, only seven cities had such services, as most were
discontinued in favor of road and highway investments.
LRT services now operate in 24 US cities, as well as three
each in Canada and Mexico. Still more are under con-
struction or in various other stages of development [4].
Introduction
BRT and LRT systems are being planned, constructed,
and opened for service at a faster pace than previously,
because these projects today tend to be smaller
projects than metros, and as such are better suited
for smaller applications, such as network spines for
small cities, lines for suburb-to-suburb commutes –
long the overwhelming share of journeys to work in
the USA and a growing trend elsewhere as well – and
for smaller-scale feeders to previously established
systems.
In addition, these systems, particularly smaller,
often street-running LRT lines known as trams or
streetcars, but also BRT in a growing number of
instances as well, serve as outstanding reshapers of the
built environment and as economic catalysts.
They also help provide solutions to another world-
wide trend: the aging of the population in the devel-
oped world. It has already taken hold in Japan, Europe,
and Canada, and is becoming an increasing challenge
of the USA. Older persons generally prefer to drive less,
and they require smaller homes. Many also prefer to
live nearer to better health-care services, nearer educa-
tional, cultural, and arts institutions, and otherwise
“where the action is,” or where their extended families
are – that is, in cities.
This third wave is also driven by other powerful
demographic forces. These include the population
growth worldwide, as well as the concentration of that
population in cities. This urbanization includes not
simply the teeming populations of the developing
nations, but also in the USA and other developed
economies.
As a consequence, this current wave of public trans-
port investment could be the largest expansion yet.
However, as will be explained below, this third wave
will require a different approach to project develop-
ment and management.
BRT Developments
There are two major reasons why BRT has been so
attractive to city planners. First, BRT projects are
typically a fourth of the cost of comparable light rail
projects. In addition, BRT can be implemented in
roughly a third of the time – or even less – to design
and build [5].
For example, the Los Angeles County Metropolitan
Transportation Authority (MTA) built and opened its
Orange Line BRT in the West San Fernando Valley to
light rail standards (to enable a future conversion if it
so chose) for $330 million; it took 3 years to design and
construct, even with court-ordered stoppages to
address neighborhood lawsuits that were concerned
about noise, as well as delays caused by construction
interruptions from one of the rainiest winters on
record in the region. A roughly comparable LRT line
in the service area, the Gold Line, took more than
 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
a decade to develop and opened at a cost of approxi-
mately $1 billion. Moreover, the Orange Line carries
more weekday passengers than the Gold Line [6].
According to the US Senate’s Banking Committee
staff analyses, more than 50 American communities are
developing BRT projects; that number grew with the
addition of roughly 2 dozen BRT earmarks in
SAFETEA-LU. The accompanying table provides a list
of these projects.
Based on a BRT market study performed for the
FTA in fall 2002 and updated in 2007 by the advanced
transportation research consortium WestStart –
CALSTART (Pasadena, CA), as many as five new BRT
systems per year and/or ten new BRT corridors could
open each year between 2007 and 2010 [7].
Interest in BRT outside the USA is even stronger.
Based on success of the TransMilenio, which carries
nearly two million boardings per weekday in Bogotá,
the government of Colombia has launched a new policy
to develop BRT in six other cities. Nearby Ecuador is
also planning BRT lines in at least three of its cities.
Cities in Peru and Chile are also planning systems.
These successes in many ways emulate the achieve-
ments of Curitiba, Brazil, which many point to where
BRT began. Despite having the second highest auto-
mobile ownership rates in Brazil (only after the
country’s capital, Brasilia), approximately 70% of
weekday journeys are on public transport, mostly on
the city’s BRT network.
In Asia, Beijing is in the throes of building a six-line
network, which began as part of its preparations for the
2008 Summer Olympics; as many as a hundred more
cities in China are studying networks, as the govern-
ment there has included BRT as a transport solution for
corridor applications whose volumes warrant less than
a tram or light rail investment. The Korean Ministry of
Transport is not only involved in assisting several cities
in network planning and construction, it is working
with a manufacturing consortium to design and com-
mercialize a next-generation BRT vehicle.
Several African cities view BRT as a cost-effective
mobility solution in their plans to cope with rapid
population growth, most of which is urbanization.
Accra, Ghana as well as Johannesburg and Cape Town
in South Africa have opened new systems (the latter
two in preparation for hosting the FIFA World Cup this
past summer).
215
India’s Ministry of Urban Development has issued
a new report, “National Urban Transport Policy,”
which encourages its lower-density cities to develop
BRT. Accordingly, the ministry has approved BRT pro-
jects in Ahmedabad, Indore, Jaipur, and Pune.
Like South America, Europe and Australia were at
the forefront of BRT and continue to witness new pro-
jects. In the UK, projects are part of Quality Bus Cor-
ridor projects, many of which featuring guided bus
systems, including two in Leeds, and one each in
Edinburgh, Ipswich, and Bradford. The next to open
will be the world’s longest guided busway, the 26-km
Cambridgeshire Guided Busway built on an aban-
doned railway right-of-way between Cambridge and
St. Ives. Manchester and Luton have also schemes in
development.
The industry publication METRO has been track-
ing the growth of BRT in its annual “BRT 25” survey. Its
most recent survey [8] found that the average project
cost for BRT networks under development is $143.8
million for new projects.
Light Rail Developments
Though not as fast as BRT in recent years, growth in
LRT systems around the world has also been impres-
sive. In the USA, only 11 cities operated modern LRTor
their historic counterparts, streetcar services. Today,
more than 30 do, and many of those have opened
additional lines to build out their networks [9].
In other North American and South American
countries, the growth rate has not been as high as in
the USA, but most either continue to operate the net-
works they started or have expanded them. In Mexico,
LRT is available in Guadalajara, Monterrey, and Mexico
City. In Canada, five cities operate such systems,
including Toronto, Edmonton, Calgary, Ottawa, and
Vancouver. A sixth, Waterloo, Ontario, is planning
such a system.
A recent phenomenon has been the growth in plan-
ning, constructing, and opening streetcar lines (also
called trams in Europe and some parts of Asia). Interest
in streetcars represents a return to an old idea, mainly
because of their history in attracting and focusing
strong economic development adjacent to their lines.
Indeed, the Smithsonian Institution’s recent exhibit on
public transport history explained that interurban
216 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
streetcar lines were built throughout the USA to con-
nect new real estate developments [10]. Henry
Huntington became rich not just because of his Pacific
Electric rail empire in Southern California but also
because of similar, though smaller, enterprises in
other parts of the nation [11].
Today’s applications comprise either modern tech-
nology, such as the systems recently opened in
Portland, Oregon, Seattle and Tacoma, Washington,
and Nottingham, England, or the one opened in
the early 1990s in Strasbourg, France; or heritage
(or vintage) systems, such as the one in Kenosha,
WI. However, most streetcar lines in the USA and
elsewhere in the world employ modern vehicles and
other technologies.
Interest in these two modes in recent years has
been more intense than that surrounding light rail in
the late 1970s and early 1980s and commuter rail in the
1990s, the last emerging markets in public transport. In
fact, BRT and streetcars very likely will exceed the other
US rapid transit markets not only in project cost but
also in the number of projects, even at a time when
these other public transport modes are also growing
robustly.
In most of the developing nations, BRTwill likely be
the favored mode, while in many European nations
light rail-based systems will be favored. Rail will also
continue to be preferred in more densely populated
cities of Japan, Korea, and China, though in these
countries the mode of choice will just as likely be
metro systems as LRT, particularly for the “backbone”
lines of networks. However, BRT and tram systems will
increasingly become surface network supplements even
in these cities, creating stronger networks than benefit
all modes of public transport – what telecoms and IT
experts call “network effects.”
This trend will continue if not accelerate for four
fundamental reasons. First, throughout the world,
there is a growing recognition that greater investment
in public transport improvements is needed to cope
with worsening congestion and increasing urbaniza-
tion throughout the world. Shanghai, Bangkok,
Abidjan, Sao Paulo, Mexico City, Cairo, and even
Paris and London face ever-increasing metropolitan
growth and its attendant problems. Indeed, a recent
study found that every other person on the globe lives
in a city and that by 2030 60% of the population will
live in cities, 23 of them with more than 10 million in
population [12]. Only a handful could be characterized
as high density.
Second, BRT and streetcars are now considered
legitimate tools in urban transport planners’ toolkit.
Third, BRT, but particularly rail-based public transport
modes, are increasingly viewed as good catalysts for
economic development, which can also attract private
capital to build and operate these systems. Finally,
growing fiscal pressures make smaller-scale and faster-
to-implement projects more viable relative to more
expensive and longer-to-deploy alternatives.
Each of these modes shares some characteristics,
but each also has strengths and weaknesses relative to
the other modes. For example, BRT is more flexible and
typically faster-to-implement, more scalable, and less
risky from a project management perspective. On the
other hand, streetcars can attract private capital more
easily, because their return on investment tends to be
much higher – as much as 2–3 times higher – than BRT.
Current and Future Trends
Both rail-based and road-based public transport
investments present a much different set of opportuni-
ties for policy-makers and thus will continue to require
different strategies. These are briefly summarized
below and elaborated upon later:
First, BRT and LRT projects in the future will tend
to be smaller but many more will be developed. Each
project will be smaller than the transit projects in the
past, but there will be many more such investments
than traditional, larger rail projects, even though all
public transportation investments continue to grow
rapidly. Implementation and management of these
requires a more nimble and flexible oversight, one
that is able to respond to changing conditions more
quickly.
Second, such smaller systems will be implemented
with much more interdisciplinary interaction not only
within the project team but also with some different
stakeholders. Often because they are far more likely to
be built in smaller cities that have less technical exper-
tise than their larger counterparts, BRT and streetcar
systems will be implemented for project sponsors that
will tend to rely more heavily on turnkey and
concession-based project delivery methods, such as
 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
design-build, design-build-operate-maintain, and
build-operate-transfer techniques to transfer project
risk to private sector experts, accelerate or even help
finance projects than with larger urban rail transport
projects. This will require project sponsors and their
contractors and other consultants, as well as city offi-
cials, professional planners, the business community,
and other project stakeholders to analyze opportunities
in a more interdisciplinary, collaborative approach.
Third, smaller projects will shift financing of them
away from central government funding to more
decentralized sources, including state, regional, and
local governments and, increasingly, even private
sources. The progress of this trend will depend on
how each country’s government is structured. In the
USA, even though federal public transport assistance
programs have recently been modified to better assist
BRT and streetcar projects, they will likely rely less on
federal support and more on state, local, or even private
sources of funding, which will require different project
management, outreach, government affairs, and finan-
cial planning strategies than larger projects. This is
partly because the explosive growth in interest in such
projects has already outstripped available funding. This
resource competition combined with lower-average
project costs and faster implementation for BRT and
streetcars prompts many localities to bypass centralized
funding and approval processes altogether.
In many countries, this is already the norm. For
example, the central governments of Germany and
France contribute well less than 50% of project cost.
Moreover, private concessions are responsible for
designing, constructing, and operating most new
urban transport projects outside the USA [13].
Meanwhile, another interesting trend that has
emerged has been the consideration of BRT designs
that can be converted at a future date to LRT. Although
many cities have designed busways and bus rapid tran-
sit running ways that can be converted to future light
rail operation, few cities have done so. Seattle’s
Downtown Transit Tunnel is the most recent such
example.
Several cities, however, are beginning to look at
BRT as an incremental step to future LRT, whether as
political cover against criticism of rail advocates who
see BRT as a lower quality investment or as a way to lay
groundwork for an incremental strategy, when
217
ridership and political support warranted a higher-
capacity solution.
Various scenarios for designing BRT running ways
to accommodate future light rail have been advocated
and studied. An interesting recent experience occurred
in San Francisco, where city planners there examined
how its Geary Avenue Corridor BRT Project could be
designed to accommodate light rail transit (LRT) at
some point in the future. San Francisco County’s Prop-
osition K specifically envisioned that Geary, one of the
busiest corridors in the city, accommodate a “BRT
service with exclusive lanes and dedicated stations. . .
designed and built to rail-ready standards.” The San
Francisco County Transportation Authority (SFCTA)
staff concluded earlier that light rail was not financially
possible within the provisions of the referendum’s
30-year Expenditure Plan and thus drafted the plan in
a way that required all center-running BRT design
alternatives for Geary to be “rail-ready” [14].
Two center-running alignments with exclusive lanes
were selected for detailed “rail-readiness” evaluation
using three definitions of rail readiness. The first defi-
nition is a minimal package of design guidelines that
would produce an alignment that does not preclude
a future conversion to light rail; such an approach
would use LRT-standard horizontal and vertical clear-
ances, grades, adjacent tangents, and turning radii, as
well as stations sited at locations that can accommodate
a light rail platform (typically 180 ft vs. a typical 120 ft
BRT platform).
The second definition goes well beyond this
approach, to comprise a series of investments that
would lead to LRT conversion much more quickly,
with less incremental financial expenditure and fewer
impacts to nearby businesses and transit riders. This
definition includes the design criteria of the first defi-
nition, but would also provide additional design detail
for surface and subsurface infrastructure, including all
trackwork (rail, fasteners, and concrete supportive
slab); all electrical and communications ductbanks,
manholes, catenary pole and substation foundations
for traction power cables and train control wiring;
ductbanks and concrete boxes for stray-current protec-
tion against corrosion; any necessary drainage work;
and utilities relocation needed, including work to pre-
serve access by these utilities so as not to interrupt BRT
service. This definition considerably narrows the
218 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
differences between this version of BRT and the cost
and construction considerations for light rail. The third
definition used in San Francisco is between these
extremes in the range of investments, though closer to
the first definition.
Because of budget constraints as well as the fact that
light rail for this corridor was not likely in the near
future or even two decades after revenue opening of
BRT, the first definition of rail readiness was selected by
city planners. They were concerned that even though
the other definitions’ additional up-front investment
obviated some construction when any decision was
made to convert to LRT, the remaining construction
needed for LRT would still require a great deal of
development time as well as create disruptions to
neighborhood businesses and residences [15].
This analysis comports with the experiences of
other cities seeking to design BRT running ways for
a future LRT. For example, the staff of the Metropolitan
Transit Authority of Harris County, Texas (known
locally as Houston Metro) returned to a policy that
committed to LRT for all five of its future rapid transit
lines, after it initially considered “rail-ready” BRT
(which was even given a new term, “guided rapid
transit”) on four of these five lines under development.
Houston Metro staff had repeatedly made represen-
tations to the public that the lines can be converted
without disruption to service, to include the rail
embedded into the corridor and all systems except for
the electric traction system [16]. In the end, however, it
abandoned any “rail-readiness” strategy and opted to
eliminate the incremental step [17].
The difficulties of designing BRT for a future LRT
are underscored by the initial design of Seattle’s
Downtown Transit Tunnel. It was designed to accom-
modate LRT at some point in the future, and even
included rails embedded in the tunnel’s roadway.
Other important rail transit design elements were
incorporated at that time, including rail-oriented hor-
izontal and vertical geometry requirements, tunnel
clearances for LRVs, station platform lengths to accom-
modate four-car LRV trains, station widths, and sizing
of structural elements to support LRV loads.
However, LRT technology has changed significantly
since the original tunnel’s design in the late 1980s, in
many ways which were not anticipated when the tunnel
was designed roughly two decades ago. For example,
the conversion, which was completed in late summer
2007, included a new LRT traction power system (with
attendant retrofit of grounding and other corrosion
protection), track and platform modifications to
accommodate low-floor LRVs not available in North
America at the time the tunnel was designed, upgraded
fire/life safety systems to comply with toughened reg-
ulations, and improved train control and communica-
tions systems to support an integrated joint bus and rail
operation. Importantly, the tunnel bed was also
lowered to accommodate both low-floor LRVs and
buses [18].
In addition, the bus service that used the DSTT was
rerouted to the surface streets of the area during tunnel
reconstruction. To minimize these adverse impacts,
additional police officers were assigned at key intersec-
tions who supplement the police force that was also
reassigned to surface bus operations. A large public
information campaign was also launched to explain
the transition. The cost of these measures combined
with the retrofit work exceeded $100 million [19].
In all three scenarios of rail readiness studied in San
Francisco, the most minimal “does not preclude LRT”
approach would have cost $2.8 million more than the
baseline cost estimate for a two-lane BRT center-
running exclusive running way, or slightly less than
a 3% premium. For the second scenario, total project
costs would have approached nearly 100% more than
a comparable baseline center-running BRT running
way cost estimate. The third scenario of rail convert-
ibility studied for the Geary Corridor, which includes
more features than the first definition but not as many
as the second, would have been approximately 10%
more than the baseline design [20].
Importantly, some significant costs are excluded
from these comparisons, however. For example, vehi-
cles, commissioning and testing costs, and any neces-
sary retrofit of maintenance facilities to accommodate
future LRV designs, catenary and substation technolo-
gies are not included, because these technologies have
been undergoing the most rapid evolution and are thus
more difficult to project in the future [20].
Conclusion
When modern light rail systems were developed, the
mode represented the avant-garde of the public
 Bus Rapid Transit and Light Rail Transit Systems: State of Discussion
transport renaissance, as a growing number of cities
throughout the world planned and constructed LRT
systems. That wave of the renaissance gave way to
subsequent, more rapidly accelerating phases involving
BRT and streetcars, as cities increasingly looked upon
public transport and strategies that were central to
addressing urbanization, congestion, sustainable
mobility, and economic revitalization. Because those
challenges will be no less daunting tomorrow, the pub-
lic transport renaissance, particularly the part that
includes BRT and street-running light rail, continues
to be bright.
Bibliography
Primary Literature
1. Federal Transit Administration (2008) Bus rapid transit.
Government Printing Office, Washington, DC
2. American Public Transportation Association (2005–2010)
Fact book. American Public Transportation Association,
Washington, DC
3. Committee on Rules, United States House of Representatives,
U.S. Congress (2005) Safe, accountable, flexible and efficient
transportation equity act: a legacy for users: report of the
conference committee. U.S. G.P.O, Washington, DC
4. Schumann JW (2010) Status of North American light rail transit
systems: year 2009 update, joint international light rail confer-
ence: growth and renewal, 19–21 Apr 2009, Los Angeles,
California. Transportation Research Board, Transportation
Research Circular E-C145, July 2010, p 3
5. U.S. General Accounting Office (2001) Mass transit: bus rapid
transit shows promise: report to Congressional Requesters,
GAO report #GAO-01-984. GAO, Washington, DC
6. Metropolitan Transportation Authority (MTA) (2005) FY03-
FY07 metro strategic performance plan (Updated FY05). Met-
ropolitan Transportation Authority, Los Angeles
7. Kantor D, Peak M, Wnuk L (2008) Analysis of BRT vehicle
market and ridership trends. WestStart-CALSTART, Pasadena
8. Schlosser N (Apr 2010) New BRT projects gather momentum
across the U.S. Metro, p 22
9. Schumann, ibid, p 4
10. Smithsonian Institution (2005) American on the move:
streetcar city, Washington, DC: growth of the capital’s
suburbs. http://americanhistory.si.edu/onthemove/exhibition/
exhibition_4_7.html. Accessed 12 Sept 2005
11. Richmond J (2005) Transport of delight: the mythical concep-
tion of rail transit in Los Angeles. The University of Akron Press,
Akron
12. Union of International Public Transport (2003) Ticket to the
future: three stops to sustainable mobility. Union of Interna-
tional Public Transport, Brussels, March
13. Henke C (1996) The french answer. Metro, May–June
219
14. San Francisco County Transportation Authority (2006)
Geary corridor bus rapid transit design principles &
guidelines, approved by Geary Citizens Advisory Committee,
1 Dec 2006
15. Kirschbaum J, Chang T (2005) ACTION – adopt a motion to
approve the proposed approach to the concept of rail-ready
design in the Geary BRT study and the proposed Q&A lan-
guage, memorandum to the Geary Citizens Advisory Commit-
tee, 18 Feb 2005
16. (Aug 2005) Convertible transit. Railway Age
17. Sallee R (2005) Metro told using buses over rail hurts credibil-
ity. Houston Chronicle, 28 June 2005
18. Parsons Brinckerhoff Quade, Douglas Inc. (2005) Sound transit
long-range plan update, issue paper no.5: convertibility of BRT
to light rail, prepared for sound transit, Mar 2005
19. Wood E, Shelton DS, Shelden M (2005) Designing BRT for
LRT convertability: an introduction for planners and
decision-makers, presented at the annual meeting of the
transportation research board, Washington DC, final version
Nov 2005
20. Parsons Brinckerhoff Inc., Technical Memorandum to
San Francisco County Transportation Authority (Feb 2008)
Geary BRT rail readiness analysis
Books and Reviews
Bill Millar (president of APTA) (2004) Testimony before the sub-
committee on transportation, treasury & independent
agencies, committee on appropriations, U.S. house of repre-
sentatives, 19 Mar 2004
Congressional Budget Office (1981) The effect of OPEC oil pricing
on output, prices and exchange rates in the United States and
other industrialized countries. Government Printing Office,
Washington, DC, February
Cudahy BJ (1990) Cash, tokens and transfers: a history of urban
mass transit in North America. Forham University Press,
New York
Dittmar H, Ohland G (2004) The new transit town: best
practices in transit-oriented development. The Island Press,
Washington, DC
Jackson RJ, Kochtitzky C (2002) Creating a healthy environment:
the impact of the built environment on public health. Sprawl
Watch Clearinghouse, Washington, DC
Shapiro RJ, Hassett KA, Arnold FS (2002) Conserving energy
and preserving the environment: the role of public transpor-
tation. American Public Transportation Association,
Washington, DC
Shelby RC et al (2003) Bus rapid transit and other bus service
innovations: hearing before Senate banking, housing and
rban affairs committee. U.S. G.P.O, 24 June 2003
Small K (2005) Unnoticed lessons from London: road pricing and
public transit. Access 26(Spring):10–15
U.S. Department of the Treasury and Council of Economic Advisers
(2010) An economic analysis of infrastructure investment.
Government Printing Office, Washington, DC, October 11
220 Bus Rapid Transit, Institutional Issues Related to Implementation
Bus Rapid Transit, Institutional
Issues Related to Implementation
MARK A. MILLER
California Partners for Advanced Transit and Highways
(PATH) Program, Institute of Transportation Studies,
University of California, Berkeley, CA, USA
Institute of Transportation Studies, School of Public
Affairs, University of California, Los Angeles, CA, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Institutional Issues: A Broad Perspective
Institutional Issues: Stakeholder Experience
Future Directions
Acknowledgments
Bibliography
Glossary
Dual-mode buses Buses that operate under both
manual and partially-to-fully automated control.
Intelligent transportation systems A system or ser-
vice designed to make the movement of people or
goods more efficient, safer, more economical, and
less polluting. Such systems work by applying
advanced and emerging technologies in informa-
tion processing, communications, and electronics
to surface transportation needs. People can use
these systems or experience their influence at
home, at work, driving in his/her automobile,
waiting at a bus stop, crossing a street, riding on
a bus, or looking for a parking space. Examples of
ITS technologies with transit applicability either
already deployed in particular settings or at least
under investigation, include advanced vehicle iden-
tification systems, electronic fare payment systems,
interactive trip planning systems (kiosk, personal
computer), transit signal priority systems, safety
and security systems, intelligent vehicle systems
(e.g., collision warning systems (frontal, side, and
rear), precision docking, and lane-keeping assis-
tance systems) and operations management sys-
tems (e.g., computer-aided dispatch system,
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
automatic vehicle location systems, automated
scheduling and dispatch software, automatic pas-
senger counter systems, and vehicle component
monitoring systems).
Multimodal Involves more than one mode of trans-
port for passengers.
Quality of service The overall measured or perceived
performance of transit service from the passenger’s
point of view.
Stakeholders A person, group, or organization that
has direct or indirect stake in an organization or
enterprise because it can affect or be affected by that
organization’s or enterprise’s actions, objectives,
and policies. Key stakeholders in the implementa-
tion of a bus rapid transit system include public
transit agencies and municipal or state departments
of transportation.
Technology push/market pull A term customarily
used in a company’s business strategy for a new
product or innovation that implies that a new
invention is pushed through research and develop-
ment (R&D), production, and sales functions onto
the market without proper consideration of
whether or not it satisfies a user need. On the
contrary, an innovation based upon market pull
has been developed by the R&D function in
response to an identified market need.
Transit choice riders Riders who choose to use transit
for their trip making, particularly during peak travel
time periods for work-related trips, even though they
have other means of travel, especially a motor vehi-
cle, available to them. Such riders may choose transit
over other modes for a variety of reasons, including
saving money, avoiding driving in congested traffic,
being able to use travel time productively for other
activities, and helping the environment.
Transit-oriented development (TOD) A mixed-use
residential or commercial area designed to maxi-
mize access to public transport and often integrates
features to encourage transit ridership.
Definition of the Subject
Bus rapid transit (BRT) is an innovative, high-capacity,
lower-cost public transit travel mode that can improve
urban mobility to help make bus transit more attractive
# 2012, DOI 10.1007/978-1-4419-0851-3

by enhancing customer quality of service with an ulti-
mate goal of increasing ridership that contributes to
relieving traffic congestion. BRT systems can easily be
customized to community needs and incorporate state-
of-the-art, low-cost technologies as part of their flexi-
ble and incremental implementation approach.
For purposes of this entry, the following definition
of bus rapid transit taken directly from [1, 2] is used:
" Bus rapid transit is a flexible, rubber-tired form of rapid
transit that combines stations, vehicles, services, run-
ning ways, and intelligent transportation systems into
a fully integrated system with a strong image and
identity. Bus rapid transit applications are designed to
be appropriate to the market they serve and their
physical surroundings, and they can be incrementally
implemented in a variety of environments (from rights-
of-way totally dedicated to transit to streets and high-
ways where transit is mixed with traffic).
Running ways for BRT include mixed traffic lanes,
curbside bus lanes, and median busways on city streets;
reserved lanes on freeways; and bus-only roadways,
tunnels, and bridges. Most stations are located curbside
or on the outside of bus-only roadways and arterial
median busways. Similarly, BRT stations have low plat-
forms since many are already being served or are plan-
ning to be served by low-floor buses. Conventional
standard and articulated diesel or compressed natural
gas buses are in wide use for BRT operations, and there
is a continuing trend toward innovations in vehicle
design, including environmentally clean or green vehi-
cles; dual-mode operations in particularly appropriate
settings such as tunnels; low-floor buses; more and
wider doors; and use of distinctively branded bus
rapid transit vehicles and stations. Service innovations
include fare-collection procedures and station design.
Intelligent transportation systems range from existing
and more customary automatic vehicle locations sys-
tems, transit signal priority systems, and passenger
information systems to more advanced systems includ-
ing intelligent vehicle systems.
Introduction
Though there has been renewed interest in its use, bus
rapid transit is not a new concept. A brief discussion of
the beginnings of bus rapid transit is presented in this
Bus Rapid Transit, Institutional Issues Related to Implementation 221
section of the entry to provide an historical context of
BRT to better appreciate from where their institutional
issues derive (section “Institutional Issues: A Broad
Perspective”).
The Origins of the Bus Rapid Transit Concept
As early as the 1930s in the USA, bus rapid transit was
suggested in a transportation plan for Chicago, Illinois,
that called for converting rail rapid transit lines to
express bus operation on certain highways with on-
street distribution in the central business district
(CBD) [1]. In the late 1950s, transportation agencies
looked for ways to implement high-quality, low-cost
transit service; for example, in a 1957 study [3],
California investigated high-speed bus operations for
commuter travel in the San Francisco Bay Area. The
proposal called for using a freeway – then under con-
struction – for high-speed bus service utilizing what
was then considered modern improvements including
park-and-ride lots, pedestrian friendly designs, and
improved amenities. The report acknowledged that
BRT was the “most economic form of rapid transit
that can be operated in this area under present
conditions” [3].
Design studies for bus rapid transit within freeway
medians were developed in the late 1950s in the
Washington D.C. area, which recommended that “in
planning of future radial freeways a cross section . . . be
provided to afford maximum flexibility and reserve
capacity for vehicles as well as for the mass movement
of people” [1]. In 1959, the St. Louis, Missouri trans-
portation plan included an 86-mile bus rapid transit
system, of which 42 miles were to be special grade-
separated bus roadways. The focus of this proposal was
an elevated loop road circling downtown St. Louis [1].
In 1963, Crain [4] presented a BRT concept, calling
for transportation planners to devise new transit ser-
vices that replicate the high performance, door-to-door
service offered by automobiles, while remaining within
the economic reach of most cities. Such a system, Crain
states, “combine(s) the best features of rail rapid transit
and conventional bus operations by retaining the flex-
ibility of one while obtaining some of the speed and
capacity of the other” [4]. Crain also described the use of
exclusive lane use and preferential traffic controls for the
rapid bus concept. People-throughput capacity of such a
222 Bus Rapid Transit, Institutional Issues Related to Implementation
service was also emphasized in order that this concept be
effectively sold to politicians and the public [4].
Crain promoted most of the features that are being
incorporated into today’s BRT systems including exclu-
sive lanes, transit signal priority, rapid fare-collection
techniques, quick boarding and alighting, and enclosed
pre-boarding payment stations. His preliminary cost
estimates showed BRT producing high-capacity service
(10,000–20,000 passengers per hour) at a fraction of
the cost of rail. His estimates also showed that BRT
would cost only 2% of the cost of rail for at-grade BRT
service and only 10% the cost of rail for an elevated
BRT system [5]. He also discussed the importance of
providing up-to-the-minute schedule information to
passengers (a precursor to today’s automated vehicle
location systems and passenger information systems
technologies), the importance of overcoming the neg-
ative sentiments of transit (bolstering its image), and
the need to properly sell the service (marketing). All of
these concepts have been incorporated into what we
today know as bus rapid transit.
In 1970, Milwaukee, Wisconsin, proposed a
transitway plan that included 107 miles of express bus
routes over the freeway system and an 8-mile, east–west
transitway equipped with 39 stations (excluding down-
town) and 33,000 parking spaces [6].
Dual-Mode Systems
Several additional reports from the 1960s outlined
many of the same goals and operational characteristics
sought in today’s BRT but envisioned an even more
sophisticated system. Many of the BRT concepts that
were explored during the late 1960s and early 1970s
focused on providing BRT with, at least, partial system
automation. The most common concept involved the
use of dual-mode buses. The idea was to combine the
flexible collection and distribution capabilities of a bus
system with the high line haul performance of an
exclusive right-of-way rail system, all with a single
seat trip and utilizing automation technologies. The
most emphasized features of the systems during this
period tended to be full bus automation (for a portion
of the route) and exclusive right-of-way of the line haul
segment. The hope was that this automation compo-
nent would permit a reduction in operating costs by
eliminating the need for a driver for a portion of the
trip, while providing greater line haul capacity than
could be achieved through manual driving alone.
Meanwhile, exclusive right-of-way (busway or bus
lane) would remove the vehicle from the most heavily
congested portions of the trip, and permit higher oper-
ating speeds, making the transit trip more attractive.
Interest in the dual-mode concept at the time declined
due to (1) insufficient technological capabilities for
cost-effective automation of vehicles, (2) operational
savings based on the assumption that the vehicle would
be “driverless” for the trip’s line haul portion was
problematic, and (3) costs of grade separations
envisioned for busways greatly reduced the capital-
cost savings as compared to a rail system [6].
Early Busways
During the 1960s and early 1970s, bus rapid transit
projects were implemented in several regions within
the USA, though appeared to be diluted versions of
Crain’s original concept. Most bus services that were
implemented utilized exclusive bus lanes on freeways,
but none really utilized a mix of multiple strategies.
These projects, therefore, functioned merely as express
bus services, whose effectiveness depended primarily
on congestion free bus lanes to provide most of the
operational improvements. The most notable projects
during this period were the (1) Shirley Busway in
Washington D.C. [7, 8], (2) El Monte Busway in Los
Angeles, California, (3) Holland Tunnel bus lanes in
New Jersey and New York, and (4) bus lanes on various
bridges in the San Francisco Bay Area. Most of
these facilities were eventually converted into high
occupancy vehicle (HOV) facilities to improve their
utilization. Primary factors that contributed to these
systems not utilizing the entire BRT toolbox of strate-
gies include (1) insufficiently developed or cost-
effective technologies and (2) lack of political will to
champion their use. Domestically, during the following
two decades interest in BRT generally waned as the US
transit industry’s mode of choice for high traffic corri-
dors tended toward rail-based projects [6].
International Experience
During the 1960s, Curitiba, Brazil, experienced rapid
population growth of approximately 3% per year due

to large in-migration from rural surroundings resulting
in overwhelming traffic congestion. A 1965 regional plan
called for the establishment of developmental policies for
transportation whose goal was to encourage growth pat-
terns that would be sustainable and supportive of a viable
public transit system. The plan called for a mono-
centric city with four radial corridors that would bear
most of the burden of growth and development in the
region. Lacking adequate financial resources for a rapid
rail “Metro” system, an approach taken by many of its
larger neighboring Latin cities, Curitiba opted for a Bus
rapid transit system. Whereas, an underground Metro
was priced at approximately $60–$70 million dollars per
kilometer ($97–$113 million dollars per mile) (1,965
$US), the express bus roadways were only $200,000
per kilometer ($323,000 per mile) [9–11].
The busway opened for service in 1972, and incor-
porated many, if not most, of the features outlined in
Crain’s paper [4, 5]. The system relied on features such
as a high or raised boarding platform to match the
vehicle floor height [12], pre-boarding fare payment,
signal priority, onboard automated announcement
systems [12] and high-frequency service to ensure
high-quality Metro-like service. Over the last 30 years,
Curitiba has been able to incrementally expand and
upgrade the system as funding allowed and demand
warranted. Instituting strong land use controls, the
government has also been able to effectively guide
growth to encourage development patterns along
structural axes that reinforce and encourage use of the
bus system. This has kept Curitiba’s central city both
a vibrant and a pleasant urban environment.
With the success in Curitiba, several other develop-
ing Latin American cities such as Sao Paulo, Brazil,
Porto Alegre, Brazil, and Quito, Ecuador, have since
implemented systems based on the Curitiba public
transit model.
Similarly, Ottawa, Canada also turned to the Bus
rapid transit concept when it decided to upgrade its
transit system. Ottawa made the decision to use the bus
as the backbone of the city’s transit system, and started
its “Transitway” in 1973 with approximately 6.8 miles
of bus lanes. By 1983, the first of several busways were
opened providing buses with exclusive, grade-
separated right-of-way. By 2000, the system utilized
16 miles of exclusive right-of-way, approximately
7.8 miles of priority lanes, and 2 miles in mixed traffic
Bus Rapid Transit, Institutional Issues Related to Implementation 223
with nearly 200,000 trips per day capable of carrying
10,000 riders per hour in each direction [6, 13].
The Transitway has features that make it special. In
the outlying areas away from the urban core, the
Transitway’s right-of-way is an exclusive bus lane that
is at- and above-grade depending on location usually
with one lane per direction. At each station, the right-
of-way expands to two lanes per direction to allow for
buses to stop without blocking other buses. Moreover,
at some locations, the exclusive right-of-way is
directionally separated by barrier or a grassy median.
In a few locations, the bus uses the freeway for short
distances (1–2 miles) in an exclusive bus lane. This lane
is the freeway’s right-most lane that is an exit-only lane
for use only by Transitway buses. It is separated by
means of a virtual barrier with diamond symbols
painted on the bus lane. Some of the Transitway’s
stations have similarities to rapid rail stations, not in
the sense that they are underground (in fact, there is
only one tunnel station on the Transitway’s route), but
that they are infrastructurally extensive. For example,
some of these stations are adjacent to commercial office
development and are linked together via enclosed walk-
way bridges. At one location, the Transitway has a stop
immediately outside a major department store that is
part of a suburban mall. The extensively built stations
are fully enclosed to accommodate Ottawa’s winter
weather with multiple stops for entry onto and exit
from buses. Also, bus routes for non-Transitway lines
converge at several of the Transitway’s stations.
In addition to the Transitway’s connection with Otta-
wa’s International Airport, it also connects with Otta-
wa’s central train station via an enclosed walkway
bridge linking the two stations [6].
Ottawa is typically heralded as another shining
success story for bus rapid transit and several other
Canadian cities have since embraced this concept.
Though most do not use exclusive right-of-way, pro-
jects in cities such as Montreal, Quebec, and Vancouver,
British Columbia have turned toward advanced tech-
nologies, in lieu of exclusive facilities, to achieve higher
bus performance.
The Return of Bus Rapid Transit to the USA
During the late 1970s, BRT made a comeback to
the USA with the implementation in Pittsburgh,
224 Bus Rapid Transit, Institutional Issues Related to Implementation
Pennsylvania of the first of three major segments of its
busway system. In 1977, the 4.3-mile South busway
became operational followed in 1983 by the 6.6-mile
East Busway with a 2.3 mile extension in 2003; in 2000,
the West busway – 5 miles in length – officially opened.
This busway system, combined with several new Light
Rail Transit projects, has produced an integrated, mul-
timodal system for the greater metropolitan Pittsburgh
area [14]. Bus rapid transit success stories in cities
such as Curitiba, Ottawa, and Pittsburgh, coupled
with continuing congestion and air pollution in
urban areas have sparked a renewed interest in the
bus rapid transit concept. In 1998, the US Department
of Transportation’s Federal Transit Administration ini-
tiated support for ten BRT demonstration projects,
which helped generate further interest in bus rapid
transit [15–17].
Contents of This Entry
This entry begins with a discussion of institutional
issues associated with the implementation of bus
rapid transit systems from a macroscopic perspective.
Initially, issues were organized into categories includ-
ing, for example, intergovernmental and interorgani-
zational, political will, and land use. The subsequent
discussion of each category follows with a listing of
specific institutional issues within each category.
The entry then turns to a more focused look at
specific institutional issues that, based on experience,
stand out from other issues relative to the following
two criteria: level of importance and level of difficulty
to resolve. Specific examples are provided in this sec-
tion of the entry to illustrate issues.
Institutional Issues: A Broad Perspective
The implementation of bus rapid transit systems tra-
verses numerous stages consisting of system design,
development, testing, evaluation, and deployment cul-
minating in a completed and fully operational system.
Moreover, all these activities take place in a context
with organizational stakeholders participating at vari-
ous levels. In fact, an overriding theme running
through the BRT implementation process is the notion
of stakeholders, their agendas, and the relationships
among them. As each stage of BRT implementation
proceeds through its more technological, design, and
operational aspects, questions may arise concerning the
effects of actions taken or policy decisions made, which
are often of a nontechnical nature and usually referred
to as institutional issues. These less-technical and less-
operational issues, which are numerous and diverse,
need to be considered and effective arrangements made
that address them to successfully implement a bus
rapid transit system.
Bus rapid transit systems will not necessarily experi-
ence the same set of institutional issues because each BRT
deployment will be affected by local and regional factors.
Moreover, even when the same issues arise in different
settings, there will likely be local and regional site-specific
differences. The importance of identifying and working
out such issues should not be underestimated as they do
contribute to the overall success of implementing bus
rapid transit systems in terms of how transit operations
and quality of service for passengers are enhanced.
Institutional issues may be grouped into the follow-
ing general categories [6]:
● Intergovernmental and Interorganizational
● Intra-agency
● Political Will
● Public Relations and Marketing
● Funding and Finance
● Labor and Human Factors
● Safety and Liability
● Land Use and Planning
● Physical Environment
Intergovernmental and Interorganizational
It is rarely the case that a transit agency can develop
a bus rapid transit system without the coordination
and cooperation of multiple agencies and often
overlapping governmental jurisdictions. Even if this
were possible, there is benefit in seeking the coopera-
tion and support of other agencies. However, the
multi-jurisdictional and/or multi-stakeholder aspects
can make the process of decision-making and imple-
mentation more complex as each stakeholder usually
brings their own philosophies, priorities, and agendas
to the table. Achieving agreement among all affected
stakeholders – whether political jurisdictions or other
transportation organizations – often proves to be

a difficult task. To have a system that works effectively
requires the transit agency to achieve agreement with
localities and other agencies on infrastructure, opera-
tions, and responsibilities.
When planning for the deployment of bus rapid
transit systems, there are, at a minimum, two distinct
types of stakeholders with primary roles. One is the
local and/or regional transit agency whose interest lies
foremost in reducing its own costs while also enhancing
the quality of transportation services that it delivers to
its passengers. The other primary stakeholder is the
local and/or regional highway and traffic department
along the route the transit agency’s bus runs and this
latter stakeholder could include multiple operators
depending on whether the bus runs through multiple
political jurisdictions. Other stakeholders with a role
to play in the implementation of bus rapid transit
might also include the following agencies and/or
organizations [18]:
● Municipal environmental, health, and urban devel-
opment and public works departments
● Construction industry and other potential industry
supporters
● Regional and/or metropolitan planning
organizations
● State or provincial department of transportation
● Relevant national or federal transportation agen-
cies, for example, in the USA such agencies include
the Federal Transit Administration and Federal
Highway Administration
● Advocacy groups, for example, proponents of com-
peting or complimentary rail projects
● Motorists and their representative organizations
● Economic development agencies
● Business and merchant associations
● Neighborhood or corridor resident associations
● Traffic and transit law enforcement
● Public transportation experts and consultants
● Nongovernmental organizations
● Public transportation passengers, for example, the
Bus Riders Union in Los Angeles, California
● Nonmotorized transportation users
● General public
The significance of these stakeholders’ roles and
influence depends on local and regional conditions
Bus Rapid Transit, Institutional Issues Related to Implementation 225
encompassing the bus route/traffic corridor where the
bus rapid transit system is to be implemented.
Typically, a public transit agency will interface with
other government agencies during regular service dis-
cussions; however, these interfaces become more criti-
cal with the development of a bus rapid transit system;
it is typical for a public transit agency to need to
coordinate with an organization it had not worked
with prior to the bus rapid transit project. These agree-
ments should set out agency and staff responsibilities
giving particular attention to the clarification of roles.
Achieving consensus, let alone agreement, among all
affected stakeholders, whether political jurisdictions or
other transportation organizations may at times prove
to be a challenging and possibly difficult task. To have
a system that works effectively requires the transit
agency to achieve agreement with localities and other
agencies on infrastructure, operations, and assignment
of responsibilities. However, the primary objectives of
transit agencies – to provide high-level, high-quality
service for their customers at minimum cost – may
conflict with the objectives of highway and traffic agen-
cies whose performance is often judged more on
enhancing vehicle-moving than people-moving capac-
ity. These often-competing objectives can complicate
the implementation of bus rapid transit strategies and
may require significant coordination and cooperation
if multiple transportation and traffic agencies are
involved. An example of conflict of objectives occurs
when a BRT project is planned to operate in mixed
traffic on public running ways such as when working to
achieve the operational benefits of BRT service, prefer-
ential treatment on the running way is often required.
The selection and incorporation of new BRT infra-
structure may also prove difficult. For many projects,
public transit agencies are “tenants” of streets or high-
way departments, often dependent on the cooperation
of the roadway operator for the right to use their
infrastructure such as the roadway, signal poles and
boxes, etc. Many BRT strategies also utilize infrastruc-
ture, which not only needs to be incorporated into the
existing roadway facilities, but must be operated and
maintained as well. With new infrastructure come
additional financial responsibilities. Reaching agree-
ment on acceptable designs, and operational and main-
tenance responsibilities would likely involve
coordination among these agencies.
226 Bus Rapid Transit, Institutional Issues Related to Implementation
Bus rapid transit operations will be effective only if
laws and regulations regarding transit prioritization are
enacted, adhered to, and enforced. Effective enforce-
ment carries with it financial responsibilities that may
be thrust reluctantly upon local jurisdictions. Infra-
structure and enforcement are two factors vital to the
success of BRT, which often lie outside the control of
the transit agency. Enlisting the support of the affected
parties in these two areas may ultimately determine the
success and effectiveness of the BRT system.
Bus rapid transit prioritization may also occur at
the expense of other roadway users such as in the case
of queue jump lanes. This issue will require discussion
and agreement on where and how these preferences
should be given, while considering its impact on
existing roadway operations. Any impacts to existing
services will require the consent and support of high-
way agencies or street departments, and local
governments.
Finally, in attempting to make a seamless transit
system, BRTservices should be coordinated with neigh-
boring transit agencies. This may require them to revise
their operations and schedules to provide better feeder
services for the BRT system.
The number and complexity of the agreements will
depend upon the type of facility and the governmental
organization(s) in the area. There are generally
a number of elements of the system that are out of the
control of the transit agency. Cooperation among agen-
cies is critical to the successful introduction and oper-
ation of a BRT system. In these cases, it is necessary to
define and codify these responsibilities. Intergovern-
mental agreements will be required with a number of
different agencies covering items such as right of use
(how long, conditions for extension or termination of
agreement, state in which the facility is returned to the
appropriate agency, legal responsibility, watering main-
tenance of landscaping, maintenance of running ways,
trash collection, graffiti removal, advertising, enforce-
ment, signal timing, lighting). In some cases, local or
state laws may be enacted, repealed, or modified to
implement various BRT elements or practices, such as
the use of the BRT facility by emergency service
vehicles.
A specific example to illustrate this type of institu-
tional issue brings together technological aspects, oper-
ational plans, and institutional concerns of
implementing bus rapid transit. For Los Angeles’
Wilshire-Whittier Boulevard Metro Rapid service,
which opened in 2000, the Metropolitan Transporta-
tion Authority (MTA) implemented a number of bus
rapid transit features as elements of its Metro Rapid
service including transit signal priority along the
heavily traveled corridor, which traverses the cities of
Santa Monica, Beverly Hills, and Commerce in addi-
tion to the city of Los Angeles and each of these munic-
ipalities controls traffic signal operation within their
respective jurisdictions. Moreover, along the corridor,
the municipal boundaries are such that the city of
Los Angeles is interspersed among the other three
municipalities in a noncontiguous fashion. Thus, for
the Wilshire-Whittier corridor, MTA and the four traf-
fic signal operators, that is, the local municipalities, are
the primary stakeholders. Initially, transit signal prior-
ity was implemented only within the city of Los Angeles
as the other cities wanted demonstrative proof of tran-
sit signal priorities’ benefits before relinquishing con-
trol over the operation of traffic signals in their
jurisdictions. To date, transit signal priority still
remains implemented only in the city of Los Angeles
while negotiations between MTA and the other juris-
dictions continue.
The following list summarizes intergovernmental/
interorganizational institutional issues:
● Integration of multiple priorities, objectives, and
agendas
● Impacts of BRT on roadway operations
● Streets/highway departments “relinquishing” con-
trol of their infrastructure
● Agreement on performance measures
● Maintenance responsibilities for shared infrastruc-
ture and hardware/software
● Responsibility for enforcement on bus lanes/
busways
● Institutional fears of new technologies
● Coordination on selection and implementation of
technologies
● Coordinating other transit agencies’ services and
BRT operations
Intra-agency
Institutional issues may arise not only among transit
agencies, political jurisdictions, and highway traffic

agencies, but also internally within an individual tran-
sit agency. Concerns over preferences in funding and
scarce resources, the delegation of responsibilities, and
increased responsibilities for staff may result in internal
resistance and morale issues for a transit agency. Unless
there are additional funding sources available,
increased spending on one route will usually mean
decreased funding on others.
Bus rapid transit systems may require additional
resources to support the service offered. Additional
operations, new technologies, new vehicles, and new
infrastructure will require training and maintenance.
Achieving agreement on roles and responsibilities may
be difficult if employees are merely required to shoul-
der additional duties and responsibilities for BRT with-
out additional compensation or support.
Many transit agencies may still be unfamiliar with
at least a portion of available BRT strategies and
agencies may need time and resources to locate or
develop both design and operational standards for
many of these strategies. Initiating BRT service may
also require additional work to reschedule existing
services to support the BRT system. As many BRT
systems will operate as trunk lines, feeder services will
need to be coordinated to achieve the full benefit of the
system.
Even gaining internal agreement on what strategies
should be implemented, what fare structures should be
used, and what technologies are appropriate will again
require time and resources. Many agencies will need
additional time to identify and digest best industry
practices for these issues. Even then, identifying and
attempting to accommodate a transit agency’s depart-
ments’ needs may cause internal conflict. As new strat-
egies may affect the duties of department staff, it is vital
that they are consulted and strategies are selected with
their concerns in mind.
The following list summarizes intra-agency institu-
tional issues:
● Concerns (or perceptions) that BRT is given special
preference over other transit services
● Defining and agreeing on new roles, responsibil-
ities, and organizational structures to support
BRT
● Creation of design and operational guidelines
for BRT
Bus Rapid Transit, Institutional Issues Related to Implementation 227
● Determining an appropriate fare structure and
medium
● Internal coordination on selection of technology
● Coordinating schedules of other transit routes with
BRT operations
● Insufficient understanding of the “state of the art”
of technologies and how they can be used in BRT
operations
Political Will
The deployment of a bus rapid transit system is one of
many stages in the process of design, development,
testing, evaluation, and finally deployment of
a completed system. At each stage, decision-making
stakeholders are involved in a variety of ways
that impact the specific deployment path a particular
bus rapid transit system will take. The decision-makers
are by definition major players in the political
arena that govern the local jurisdictions in which the
bus rapid transit system would operate. The commit-
ment to bus rapid transit by such major players is
of crucial importance to the success of a bus rapid
transit system.
One concern in the political arena may be whether
a proposed bus rapid transit system is a solution in
search of a problem, whether appropriate technologies
are being used, and whether there is a market for these
enhanced bus transit services. Does it result from
a technology push or market pull?
To establish and sustain a high level of interest and
commitment to BRT, a political champion will likely be
required. Whether it is an individual or organizational
entity, a political champion would aid in coalition
building and sustaining interest in BRT, which could
wax and wane with the whims of the political process.
The strength and capability of the political champion
would help determine if the project could weather the
political storms of opposition arising from various
quarters, for example, the local business community,
the mass media, or local residents. However, gaining
such championing decision-makers often requires
proof of the operational and quality-of-service benefits
of BRT, but political support is required to perform the
testing that could result in the quantifiable benefits,
resulting in a chicken-or-the-egg dilemma requiring
resolution.
228 Bus Rapid Transit, Institutional Issues Related to Implementation
There may also be issues regarding legislative
restrictions on the procurement of new vehicles that
could delay or slow deployment of BRT. Legal issues
may also arise as a result of changes in service associ-
ated with BRT operation. For example, route changes,
the elimination of stops, or lengthening the distance
between consecutive stops could potentially present
legal challenges.
Many political officials may be reluctant to under-
take a bus rapid transit project due to the perceived risks,
especially in relation to upsetting powerful special inter-
est groups. For example, motorists and existing public
transportation operators may tend to resist such change.
Thus political officials may end up playing it safe by
avoiding any type of major public transportation initia-
tive that will risk alienating specific stakeholders. How-
ever, when officials take the perceived low-risk path of
inaction or very slow action, the ensuing political
rewards will certainly be reduced [18].
The following list summarizes politically related
institutional issues:
● Concerns of BRT being a top down solution
● Perceived or actual competition of BRT with rail
transit
● Lack of domestic BRT success stories
● Lack of empirical evidence of BRT’s operational
effectiveness
● Finding political champions to support BRT
● Concerns over long-term level of interest, potential
for waning
● Local and business community opposition to the
removal of, or restrictions on, parking spaces for
BRT use
● Local and community opposition to BRT
● Concerns over the distribution of the costs and
benefits of BRT
● Legal issues associated with service changes
● New vehicle procurement
Public Relations and Marketing
The success of implementing a bus rapid transit system,
as with nearly any new product, service, or system, largely
depends on how well it is sold to the public, which often
requires setting expectations. Setting high, yet realistic
expectations is crucial to gain support for BRT. Failure
to produce what was promised could lead to disappoint-
ment and a loss of public confidence and support. Bus
rapid transit may also require a significant public educa-
tion campaign on interacting with new transit agency
strategies, features, and technologies such as bus lanes,
signal priority systems, queue jump lanes, and new fare-
collection systems. Moreover, the transit agency needs to
consider public views about its current performance.
Before taking on the additional responsibilities of a BRT
system a transit agency must ensure it is performing
satisfactorily, or it could face political and public oppo-
sition as it embarks on new ventures [19].
The following list summarizes public relations and
marketing institutional issues:
● Educating the public on BRT, and managing per-
ceptions and expectations
● Concerns over transit agency’s existing perfor-
mance and reputation
● Concerns over effects of BRT on existing roadway
operations
● Educating pedestrians and motorists on interacting
with BRT
● Educating users on changes in and uses of multiple
fare structures
Funding and Finance
During the 1960s and 1970s, interest in BRT in the USA
waxed and waned. Though there is renewed interest,
the fear of history repeating itself in the USA may still
cause concern among transit agencies considering BRT
for their communities and lead to their reluctance to
embrace bus rapid transit. Though the up-front capital
costs for most BRT projects are relatively small com-
pared to other capital-intensive modal alternatives
such as rail, transit agencies will still be responsible
for the operations and maintenance of the new system.
With continued fiscal pressures facing transit agencies,
concerns may arise over the long-term financial com-
mitment to bus rapid transit. Similarly, bus rapid tran-
sit will also require additional financial commitments
for staff, training, and enforcement [19].
The following list summarizes funding- and
finance-related institutional issues:
● Concerns over long-term funding commitments to
BRT at the state and federal levels
 Bus Rapid Transit, Institutional Issues Related to Implementation 229

● Concerns about BRT redirecting funds away from ● Use of technology, for example, automated vehicle
existing service or other routes location systems for monitoring schedule
● Lack of understanding of funding mechanisms adherence
available for BRT ● Different responsibilities between bus rapid transit
● Agency reluctance to expand services due to current and non-BRT routes
fiscal constraints
● Ability to use existing buses or need for new fleet Safety and Liability
● Capital costs of BRT
Bus rapid transit may involve new procedures, technol-
● Cost of operating and maintaining (O&M) new
ogies, or personnel tasks. The potential thus exists for
technologies and infrastructure
system components not to function as anticipated,
● Cost of additional staff and/or training to
raising safety and liability issues. Stakeholders need to
support BRT
consider whether bus rapid transit changes the assign-
● Cost of additional facilities to support BRT
ment of risk and responsibility if technologies or strat-
● Cost of and responsibility for enforcement, for
egies do not function as intended. Safety issues
example, proof of payment
regarding pedestrians and motorists and their interac-
tion with bus rapid transit components, such as signal
Labor and Human Factors
prioritization and queue jump lanes, will also need to
Another important stakeholder group that must be be considered and addressed [19].
considered for bus rapid transit is the transit agency The following list summarizes safety- and liability-
staff, especially bus drivers. What will be the impact of related institutional issues:
bus rapid transit on bus drivers? With the implemen-
● Insurance industry–initiated changes in assignment
tation of bus rapid transit, there could be concerns over
of risk and responsibility for bus transport
additional work and responsibilities without assur-
● Potential changes in liability associated with tech-
ances of additional staff, needed resources, and/or
nological and/or operational malfunctions of BRT
pay. Bus drivers would have a direct and potentially
systems
the closest connection of all agency employees to any
● Safety issues arising from changing interaction of
new technologies implemented as part of a bus rapid
pedestrians and motorists with new technologies
transit system. How would such technologies affect bus
and/or strategies
drivers? Would it mean any change in the roles and
● Safety concerns of residents along BRT corridors
responsibilities of their job? Could the implementation
of a bus rapid transit system mean that some bus
drivers would lose their jobs or be replaced by individ- Land Use and Planning
uals with more familiarity with and experience in the
Large-scale public transportation projects often influ-
use of certain technologies? Will there be assurances to
ence travel patterns and surrounding land uses, over
retrain and reeducate bus drivers to use these new
both the short and the long term. Bus rapid transit
systems?
would be no different, as high-level transit services,
Another concern of bus drivers could be the use or
which BRT attempts to replicate, have often signifi-
perceived misuse of technology by transit agency man-
cantly altered surrounding land uses. However, bus
agement for driver performance monitoring. Drivers
rapid transit may raise concerns over how it fits into
may fear their employers using “Big Brother” privacy-
a region’s overall transportation planning environment
threatening tactics under the guise of improving transit
and how it will affect local land uses.
operations and service [19].
Anecdotally most transit planners believe that
The following list summarizes labor-related insti-
fixed guideway systems have a positive impact on
tutional issues:
land use, particularly around the station areas, whereas
● Lack of support from transit agency staff bus-based systems have at best a neutral impact.
● Changing role of drivers Bus-based systems have been shown to have
230 Bus Rapid Transit, Institutional Issues Related to Implementation

a beneficial impact at a community level; however, ● Need for strong community involvement in plan-
property adjacent to stops/stations is considered less ning phases builds support
desirable.
One of the biggest concerns of communities devel- Physical Environment
oping bus rapid transit systems is convincing them that
The physical imposition of a BRT system may also raise
BRT will provide the benefits that they associate with
political and institutional challenges. Many project
fixed guideway modes. The advent of BRT systems has
areas, especially in older city centers, may simply lack
challenged this premise. The new BRT systems that
the physical space to accommodate certain BRT
have chosen to incorporate a fixed guideway element
strategies. In other areas, transit agencies may encounter
have demonstrated that bus-based transit systems can
opposition if BRT competes with, or at least is viewed as
have a positive impact on urban form and land values.
competing with, other interests for high-value
Although the BRT concept is still relatively new in the
real estate. This may inflate costs or overly complicate
USA with only a handful of systems in operation, there
operational requirements. Though eminent domain is
is a growing body of evidence that suggests that BRT
an option, it usually is an undesirable, drawn-out pro-
systems can support existing land users and promote
cess. Ensuring there is adequate and obtainable physical
higher density residential, office, and commercial land
space could present problems for certain projects.
use, particularly around BRT stations. North American
Secondly, image is a strong marketing tool for BRT.
examples of this trend include Boston, Pittsburgh, and
Many design guidelines suggest making the system
Ottawa, where $1,250 million, $302 million, and $675
unique and easily identifiable. Many projects are
million of new or improved development, respectively,
included as part of their systems station area improve-
has occurred. The continuing development of more
ments. However, these improvements are usually being
BRT systems will provide further evidence of this effect;
inserted into existing urban design. Finding station
however, as land use improvements tend to lag transit
area designs that promote a strong image, while being
investment, examples of this trend may take a while to
acceptable to local interests, may be a challenge.
be realized [20].
The following list summarizes physical-environ-
Some BRT systems have benefited by initially
ment-related institutional issues:
developing a number of key stations where land use
development potential exists and linking them with ● Availability and acquisition of right-of-way or phys-
transit facilities which incrementally increase to fully ical space
exclusive busways. In this way, they could make strategic ● Reaching agreement or consensus on bus stop/sta-
infrastructural investments at specific locations without tion area enhancements
the need to improve the whole corridor to the same level. ● Need to educate and address public concerns
The following list summarizes land-use- and plan- regarding the potential effects of BRT on the phys-
ning-related institutional issues: ical environment

● Coordinating a BRT project with local planning
agencies’ land use policies
● Gaining community support for transit-oriented
development
● Possible reluctance of potential developers to invest
along BRT corridors due to a perceived lack of
permanence as compared to rail
● Integrating BRT projects into the metropolitan
planning process
● Planning requirements such as transportation
improvement programs that could delay imple-
mentation of a bus rapid transit system
Institutional Issues: Stakeholder Experience
Transit agency experience has shown that there are
specific institutional issues that are primary and stand
out from others with respect to the criteria of level of
importance and ease of resolution. Eight primary issues
are listed below (in no particular order) [21]:
1. Local and business community opposition to the
removal of/restrictions on parking spaces for
BRT use
2. Availability and acquisition of right-of-way or
physical space

3. Integration of multiple priorities, objectives, and
agendas
4. Concerns over long-term funding commitments
to BRT
5. Impacts of BRT on roadway operations
6. Finding political champions to support BRT
7. Gaining community support for transit-oriented
development
8. Educating the public on BRT and managing per-
ceptions and expectations
Integration of Multiple Priorities, Objectives,
and Agendas
The integration of multiple priorities, objectives, and
agendas often lies at the heart of institutional issues.
When several institutions come together to discuss
issues of common interest, each brings its own organi-
zational experiences, cultures, and goals. A “win-win”
strategy might not always be achievable, but BRT
project members need to acknowledge and be
thoughtful of other agencies’ issues and concerns.
Modal biases and agendas have historically infiltrated
transportation planning. However, in recent years, with
the recognition that multimodal transportation
systems tend to be the healthiest, there are greater
levels of cooperation. Many transportation organiza-
tions, however, still have responsibilities to their
respective agencies or jurisdictions, and are still
expected to protect their own interests. Though greater
cooperation nevertheless must also be accompanied by
continuous dialogue to discuss and better understand
stakeholders’ concerns and by attempts to address
them throughout the BRT development and imple-
mentation process.
For example, close working arrangements between
traffic engineers and transit planners are essential in
developing busway and bus lane designs, locations of
bus stops and turning lanes, and application of traffic
controls. A good program of traffic controls and sign-
age will help ensure safe vehicle and pedestrian cross-
ings of busways and bus lanes. Excessively long traffic
signal cycle lengths to accommodate exclusive bus
phases should be avoided.
Los Angeles’s successful Metro Rapid bus opera-
tions on its Wilshire-Whittier and Ventura Boulevard
lines are a direct result of cooperation between the
Bus Rapid Transit, Institutional Issues Related to Implementation 231
MTA and the city’s Department of Transportation
(DOT). These two agencies found that (1) a modest
“advance” or “extension” of the traffic signal green time
(or a delay of the red signal time) of up to 10 s per cycle
can reduce bus delays with negligible impacts on cross
street traffic, (2) bus headways should not be less than
2.5–3.0 min to enable major cross streets to “recover”
from the time lost, and (3) far-side stops are essential.
An at-grade busway has fewer traffic impacts on
intersecting roads than a typical arterial street. However,
relatively light bus volumes require traffic control strate-
gies that ensure safety at grade crossings. Positive protec-
tion, such as bus-actuated traffic signals, is essential.
However, if accidents persist, gating of bus crossings may
be appropriate. The busway should be treated as though it
were a high-speed light rail line on a private right-of-way.
If there is sufficient conflict among various modes (i.e.,
vehicles, transit vehicles, and pedestrians), it is important
to incorporate gates to reduce these conflicts [1].
Finding Political Champions to Support Bus Rapid
Transit
Though public support is critical in implementing a BRT
system, it is usually not attainable through transportation
agencies alone. Finding a political champion to support
a BRT initiative may be critical in gaining public support.
Politicians are typically the final decision-makers and
often have the clout to produce results. Gaining the ear
and voice of influential politicians is one of the most
often-cited means of achieving results [19].
Roadway-Related Issues
The following three issues may be covered under the
umbrella of roadway-related issues:
● Local/business community opposition to removal
of/restrictions on parking spaces for BRT use
● Availability and acquisition of right-of-way or physi-
cal space
● Impacts of BRT on roadway operations
Bus rapid transit intends to provide the high-quality
service associated with rail transit at a much lower cost. In
many BRT projects, this is accomplished by providing
buses with exclusive or nearly exclusive right-of-way, so
operations are not affected by urban street congestion.
However, obtaining the required right-of-way may be
232 Bus Rapid Transit, Institutional Issues Related to Implementation
difficult. Most BRT projects operate, for at least a portion
of their route, in developed urban areas where physical
space for transportation improvements is typically not in
abundance. In several projects, this space comes from
currently utilized roadway lanes or from existing parking
lanes. Operators of the roadway facilities (typically
municipal streets departments or state departments of
transportation) used for BRT projects will be interested
in how BRT operations would affect their facilities. In
cases where projects look to utilize roadway space that is
currently on-street parking, businesses and residents may
be opposed to the “loss” of parking, even if it is only for
peak-period directions and times. For example, at BRT
stops/stations the use of queue jump lanes and/or bus
bulbs may also be considered. The use of these design
attributes in order to improve the level of service may,
however, conflict with concerns of the local business
community over its opposition to the removal of or
restrictions placed on parking space availability that
may be necessary to accommodate such operational and
service plans for BRT. Therefore one of the major insti-
tutional considerations in assessing where BRT is
a realistic alternative for specific corridors or roadway
facilities is the availability of physical space to accommo-
date BRToperations. Proper consideration must be given
to identify whether there are competing interests for space
and how BRToperations may impact these facilities [19].
Concerns over Long-Term Funding Commitments
to BRT
Concerns over long-term funding commitments to
BRT include it merely being the popular favorite and
the implications for transit agencies should the BRT
concept fall out of favor. Some BRT projects will
require a great deal of capital investment, often requir-
ing transit agencies to shoulder the risk of having
greater capital to maintain without recovering suffi-
cient additional revenue to cover those costs. Whether
these are items that a transit agency can afford to
operate and maintain, may cause reluctance on the
part of transit agencies to embrace BRT [19].
Gaining Community Support for Transit-Oriented
Development
Allaying the fear of the unknown is often a responsibility
that must be borne when presenting an untested concept
to the public. Many BRT projects have incorporated land
use strategies to encourage and reinforce transit usage.
However, for most outside of the transportation and
planning communities, the concept of transit-oriented
development (TOD) is new. For many, higher density
and mixed use equals more crowding and greater conges-
tion. Attempting to obtain public support for TOD may
be a challenging undertaking, especially if there are not
many local examples to aid the public’s understanding.
Proactively educating the public on the underlying objec-
tives of TOD may assist a transit agency in avoiding
public opposition to it [19].
Educating the Public on Bus Rapid Transit, and
Managing Perceptions and Expectations
This issue may be critical in maintaining continued
support of and interest in bus rapid transit. Agencies
must perform a balancing act on this issue. For BRT to
be embraced by both the public and the decision-
makers, the concept must be “sold” to them but it
must be sold in the correct manner and amount. Set-
ting unrealistically high expectations can lead to disap-
pointment and a loss of support both from the public
and from the decision-makers. It is important that
balance be maintained between the “hype” and actual
results.
Future Directions
Throughout the world, bus rapid transit systems continue
to be implemented, with more than 135 systems in over
28 countries on 6 continents and 90 cities in North and
South America, Europe, Asia, Australia, and Africa in
both the developed and developing worlds. In the USA
alone, there are nearly 60 existing BRT systems with
significant progress made in the number of implemented
systems in the last 12 years alone, which marks a special
milestone year in the history of BRT in the USA because
1998 was the start of the most recent resurgence of
interest in the USA in bus rapid transit with the US
Department of Transportation’s formal initiation of and
support for a US BRT Demonstration Program.
BRT has shown that it is indeed a viable rapid
transit mode within the USA. The strength of the
mode lies in its diverse range of BRT strategies and its
flexible and incremental character of deployment that
allows transit agencies to customize their BRTsystem to

the local environment including institutional concerns,
available right-of-way, or traffic conditions.
For example, agencies with both budget and time
resource constraints and little, if any, additional right-
of-way capacity may consider a BRT system modeled
after Los Angeles’s Metro Rapid service. The Metro
Rapid uses lower-cost strategies including ITS technolo-
gies (transit signal priority systems), frequent headway-
based service, a branded and unique identity for the buses
and bus stops, low-floor buses, real-time passenger infor-
mation systems, and a simple route structure with fewer
stops to minimize total travel time and enhance travel
time reliability on mixed traffic urban arterials. If addi-
tional right-of-way capacity is available, agencies may
want to consider implementing services on dedicated
bus lanes within existing roadways, such as the EmX
Line along the Franklin Avenue corridor between Eugene
and Springfield, Oregon, where a traffic lane was
converted for BRTuse. Another example is the Healthline
BRT system along the Euclid Avenue corridor in Cleve-
land, Ohio, where one traffic lane per direction along its
4.5 mile long corridor was removed to build a bus lane
within the median. With still more available right-of-way,
on the high end of the roadway capacity continuum are
exclusive running ways, such as the Metro Orange Line in
Los Angeles built along a 14-mile former rail right-of-way
owned by MTA. Such systems offer the greatest potential
for achieving high speeds and high levels of travel time
reliability required for and desired by public transit choice
riders.
The next decade will continue to see more BRT
systems implemented on an international scale espe-
cially in the developing world. In the USA, BRT will
continue playing a major role in transit systems
throughout the country. Cain A, et al. [22] have
suggested that if current trends continue in pioneering
BRT cities such as Los Angeles, which has already
implemented a network of approximately 25 Metro
Rapid BRT Lines beginning with its initial deployment
of the Wilshire-Whittier and Ventura Boulevard Lines
in 2000, the next major challenge will be transitioning
from individual BRT corridors into integrated net-
works of BRT routes. Such expansion offers transpor-
tation planners and engineers the opportunity to
increase the benefits of BRT from the corridor level to
the city or regional level, enhancing public transit rid-
ership that helps improve mobility and air quality,
Bus Rapid Transit, Institutional Issues Related to Implementation 233
reduce congestion, and promote economic growth.
Careful integration of such networks with existing
BRT and conventional bus services as well as non-bus
transit modes will be crucial to achieving a successful
transition. Nonetheless, in the context of a BRT net-
work as opposed to an individual BRT corridor, the
institutional structure may experience changes and it is
possible that additional institutional issues heretofore
not identified may arise or already existing issues may
be exacerbated, which of course will need to be
addressed as part of any successful BRT network imple-
mentation process.
Acknowledgments
Portions of sections “Public Relations and Marketing”
to “Safety and Liability” and “Finding Political Cham-
pions to Support Bus Rapid Transit” to “Gaining Com-
munity Support for Transit-Oriented Development”
above use material from citation [19] and are the copy-
right of the National Academy of Sciences, Washington,
D.C., 2001; and are reproduced with permission of the
Transportation Research Board.
Bibliography
Primary Literature
1. Levinson HS et al (2003) Bus rapid transit volume 1:
case studies in bus rapid transit. TCRP Report 90, Transit
Cooperative Research Program. Transportation Research
Board, Washington, DC
2. Levinson HS et al (2003) Bus rapid transit volume 2:
implementation guidelines. TCRP Report 90, Transit Coopera-
tive Research Program. Transportation Research Board,
Washington, DC
3. California Public Utilities Commission, Transportation Division,
Traffic Engineering Section (1957) Report on bus rapid transit
between Concord and Oakland-San Francisco. California
Public Utilities Commission, San Francisco, p 6
4. Crain JL (1963) The rapid transit bus concept. Stanford
Research Institute, Menlo Park, p 1
5. Crain JL (1963) The rapid transit bus concept. Stanford
Research Institute, Menlo Park, p 16
6. Miller MA, Buckley SM (2000) Institutional aspects of bus rapid
transit: a macroscopic examination. California PATH Working
Paper, UCB-ITS-PWP-2000-07, California PATH Program,
University of California, Berkeley
7. National Capital Transportation Agency (1963) A study of bus
rapid transit operations of the national capital region. National
Capital Transportation Agency, Washington, DC
234 Bus Rapid Transit, Institutional Issues Related to Implementation
8. Metropolitan Washington Council of Governments (1970)
Feasibility study for bus rapid transit in the Shirley highway
corridor. Metropolitan Washington Council of Governments,
Washington, DC
9. Rabinovitch J, Hoehn J (May 1995) A sustainable urban trans-
portation system: the ‘surface metro’ in Curitiba, Brazil. Envi-
ronmental and Natural Resources Policy and Training Project,
Working Paper No. 19
10. Rabinovitch J (1996) Innovative land use and public transpor-
tation policy: the case of Curitiba, Brazil. Land Use Policy
13(1):51–67
11. Rabinovitch J, Leitman J (1996) Urban planning in Curitiba. Sci
Am 274:26–33
12. Diaz R, Schneck, D. (2000) Bus Rapid Transit Technologies in the
Americas: An Overview. Transport Res Rec 1731:3–9
13. Bonsall JA (1989) Transitways: the Ottawa-Carleton experi-
ence. Ottawa-Carleton Regional Transit Commission, Ottawa
14. Web site: http://pittsburgh.pahighways.com/busways/.
Accessed 21 May, 2010
15. Federal Transit Administration (1998) Bus rapid
transit demonstration program. United States Department
of Transportation, Federal Transit Administration,
Washington, DC
16. Federal Transit Administration (1999) Proceedings of the bus
rapid transit demonstration program. United States Depart-
ment of Transportation, Washington, DC
17. Federal Transit Administration (1999) Bus rapid transit dem-
onstration program kick off workshop. United States Depart-
ment of Transportation, Washington, DC
18. Wright L, Hook W (2007) Bus rapid transit planning guide,
3rd edn. Institute for Transportation & Development Policy,
New York, pp 189–190
19. Miller MA, Buckley SM (2001) Bus rapid transit institutional
issues: the route from research to experience. Transit: bus
transit and maintenance, paratransit, and new technology.
Transportation Research Record, No. 1760. Transportation
Research Board National Research Council, National Academy
Press, Washington, DC, pp 34–41
20. Miller MA, Carey G, McNamara I, Zimmerman S (2006)
Development of bus rapid transit information
clearinghouse. California PATH Research Report, UCB-ITS-
PRR-2006-7, California PATH Program, University of California,
Berkeley
21. Miller MA, Buckley SM (2001) Institutional aspects of bus
rapid transit operation. California PATH Research Report,
UCB-ITS-PRR-2001-09, California PATH Program, University of
California, Berkeley
22. Cain A, Darido G, Thole C, Flynn J (2007) BRT update: an
overview of bus rapid transit in the United States. Mass Transit
Magazine
Books and Reviews
Diaz RB, Hinebaugh D (2009) Characteristics of bus rapid transit for
decision-making (CBRT). Federal Transit Administration, FTA-
FL-26-7109-2009.1
Eccles KA et al (2007) Design, operation, and safety of at-grade
crossings of exclusive busways. Report 117, Transit Coopera-
tive Research Program. Transportation Research Board,
Washington, DC
ECONorthwest, Parsons Brinckerhoff Quade & Douglas, Inc., et al
(2002) Estimating the benefits and costs of public transit
projects: a guidebook for practitioners. In: Report 78, Transit
Cooperative Research Program. Transportation Research
Board – National Research Council, National Academy Press,
Washington, DC
General Accounting Office (2001) Mass transit: bus rapid transit
shows promise, GAO-01-984. U.S. General Accounting Office,
Washington, DC
Kittelson & Associates, Inc. et al (2003) Transit capacity and quality
of service manual, 2nd edn. Transit Cooperative Research
Program, Report 100. Transportation Research Board,
Washington, DC
Kittelson & Associates, Inc. et al (2007) Bus rapid transit
practitioner’s guide. Transit Cooperative Research
Program, Report 118. Transportation Research Board,
Washington, DC
Levinson HS et al (1973) Bus use of highways: state of the art.
National Highway Cooperative Research Program, Report 143.
Transportation Research Board, Washington, DC
Levinson HS et al (1975) Bus use of highways: planning and
design guidelines. National Highway Cooperative Research
Program, Report 155. Transportation Research Board,
Washington, DC
Miller MA et al (2006) Development of deployment strategy for an
integrated bus rapid transit system. California PATH Research
Report, UCB-ITS-PRR-2006-09, California PATH Program,
University of California, Berkeley
Peak M, Henke C, Wnuk L (2005) Bus rapid transit ridership analysis.
Federal Transit Administration, U.S. Department of Transpor-
tation, FTA-CA-26-7068-2004.1
St. Jacques K, Levinson HS (1997) Operational analysis of bus lanes
on arterials. Transit Cooperative Research Program, Report 26.
Transportation Research Board, Washington, DC
Thole C, Samus J (2009) Bus rapid transit and development:
policies and practices that affect development around transit.
Federal Transit Administration, U.S. Department of Transpor-
tation, FTA-FL-26-7109.2009.5
Wilbur S et al (1975) Bus rapid transit options in densely
developed areas. U.S. Department of Transportation,
Washington, DC
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
Bus Rapid Transit: Worldwide
History of Development, Key
Systems and Policy Issues
DARIO HIDALGO
EMBARQ, The World Resources Institute Center for
Sustainable Transport, Washington, DC, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
BRT Components, Characteristics, and Impacts
BRT Status Around the World
BRT in the United States and Canada
BRT in Latin America
BRT in Europe
BRT in Asia
BRT in Australia and New Zealand
BRT in Africa
Future Directions
Bibliography
Glossary
Alternatives analysis Evaluation of different transit
choices available, usually requiring cost-benefit
analysis, life cycle costing, and sensitivity analysis.
Branding Name, term, design, symbol, or any other
feature that identifies a service or product.
Bus corridor A street or highway featuring bus prior-
ity measures, such as bus lanes or busways or traffic
signal priority.
Bus lane Street or highway lane(s) with priority des-
ignation for bus public transport services.
Bus priority Infrastructure or traffic signal measures
to speed up bus public transport services, especially
at intersections or junctions.
Busway Street or highway lane(s) exclusively desig-
nated for bus public transport services.
Comfort The capacity to give physical ease and
well-being. In public transport, it is usually
related to the provision of adequate space, soft
ride, easy circulation, and adequate information
for passengers.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
235
Commercial speed Distance traveled divided by the
time, as perceived by the passengers of the public
transport system. It includes time spent at bus stops
or stations, intersections or junctions, and the time
traveling between them.
Externalities An economic effect that is not reflected
in market prices. In public transport, this term
usually refers to travel time, air pollution, road
fatalities and injuries, and other effects on people
and the environment.
Fare collection Collection of entrance fee in a public
transport service. It may include direct access to the
service or tickets or passes to travel.
Headway Time interval between two consecutive tran-
sit vehicles.
Intelligent transport system (ITS) Is an umbrella
term for a range of technologies including
processing, control, communication, and electron-
ics that are applied to a transportation system.
Paratransit Is an alternative mode of flexible passen-
ger transportation that does not follow fixed routes
or schedules. It includes informal services as well as
formal services provided for users with special
needs, such as elderly and disabled.
Passenger capacity Number of passengers that can be
accommodated in a vehicle, station, service, or sec-
tion of the transport system. Can be measured in
units of space (passengers per square meter), vehi-
cles (passengers per bus), or time (passengers per
hour per direction).
Peak section load Number of passengers riding inside
the transit vehicles in the heaviest section of the
public transport system and direction per unit of
time, usually 1 h.
Public transport Is a shared passenger transportation
service which is available for use by the general public.
Also known as public transportation or transit.
Public–private partnership Describes a government
service or private business venture which is funded
and operated through a partnership of government
and one or more private sector companies. These
schemes are sometimes referred to as PPP or P3.
In public transport, this term usually describes the
service or concession contract for transit provision,
which may include development of infrastructure
and vehicles.
# 2012, DOI 10.1007/978-1-4419-0851-3
236 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
Rapid transit A system of public transport in an urban
area, using a combination of infrastructure, vehi-
cles, operations, and information technologies to
provide fast and reliable service.
Reliability Capability of being relied on; dependabil-
ity. In public transport, this refers to the provision
of timely services, with low variance in arrival and
travel times.
Service operation plan Public transport routes,
schedules, and itineraries, preferably planned in
advance and centrally controlled to provide service
according to the user needs.
Traffic signal Signaling devices positioned at road
intersections, pedestrian crossings, and other loca-
tions to control competing flows of traffic. Also
known as traffic lights, stoplights, traffic lamps,
traffic signals, or semaphores.
Transit-oriented development (TOD) Urban area
designed to maximize access by public and
nonmotorized transportation to encourage transit
ridership. A typical TOD has a rail or bus station at
its center, surrounded by relatively high-density
and mixed used development. It usually has ade-
quate facilities and attractive street conditions for
walking and cycling, good connectivity, and uses
traffic calming and parking management measures.
Definition of the Subject and Its Importance
Bus rapid transit (BRT) can be defined as a flexible,
rubber-tired form of rapid transit that combines sta-
tions, vehicles, services, running ways, and information
technologies into an integrated system with strong
identity [1]. Complete BRT systems offer fast, comfort-
able, and low-cost urban mobility [2]. BRT is an evo-
lution of bus priority measures, such as designated
busways and bus lanes, which were proposed, and in
some cases implemented, as early as 1937 throughout
the world [3]. The expression BRT was initially used in
the United States in a 1966 study [4]. The concept
gained popularity in Latin America after the successful
upgrade of busways in Curitiba, Brazil, to full-featured
BRT in 1982 [5]. The high performance, low cost, and
rapid implementation of this system, and its adapta-
tions to Quito, Bogotá, Mexico City, Beijing, Jakarta,
Los Angeles, Beijing, Istanbul, and Guangzhou, among
other cities, made the idea attractive for urban
transport planners throughout the world. As of January
2011, there were about 118 cities with BRT or bus
corridors, with 97 of the cities entering into the list
in the last 10 years, and at least 80 cities building,
designing, or planning BRT systems [6]. BRT is an
attractive option for public transport delivery, applica-
ble to a wide variety of conditions – from very low to
very high passenger throughput, and should be consid-
ered in alternatives analysis in most transit corridors.
Critics of BRT indicate that these systems are not per-
manent, use precious surface space, and exhibit opera-
tional and cost issues as compared with rail [7]. BRT
implementation requires strong political leadership,
sound technical planning, and adequate funding
levels [2, 8].
Introduction
The concept of BRT is not new [3], but it has only being
extensively deployed around the world in the last
decade [6]. The growth may be attributed in part to
the success of Curitiba [5] and its adaptations in dif-
ferent parts of the world [2, 3]. In general, BRT systems
exhibit low cost, rapid implementation, and high per-
formance, with significant positive impacts [2, 9].
Interesting trends are emerging in the BRT industry
[6], such as the implementation of citywide integrated
bus systems based on BRT, improved processes for
private participation in operations, increased funding
from national governments, and growth of bus manu-
facturers and technology providers. Technological
developments in vehicles and information systems are
also improving the quality, performance, and impact of
BRT [9].
Despite the growth, there are some outstanding
issues: BRT does not have a single meaning and image
and it is often regarded as a “second best” as compared
to rail alternatives [2]. Its ability to foster urban devel-
opment and the use of space designated for cars are
often questioned, as well as its actual costs and impacts
[7]. In addition, several systems in the developing
world suffer problems resulting from poor planning,
implementation, and operation, due to financial, insti-
tutional, and regulatory constraints [8].
This entry discusses the concept, its history, and
expected trends. In the first section the concept of
BRT is developed. The following sections show the
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 237

history and current status of BRT in different
geographical regions of the world – the United States
and Canada, Latin America, Europe, Australia and
Oceania, and Asia. The final section indicates future
directions.
BRT Components, Characteristics, and Impacts
Components of BRT include [9], running ways, sta-
tions, vehicles, fare collection, intelligent transport sys-
tems (ITS), service operation plans, and branding
elements. There could be different degrees of complex-
ity of each one of these elements, resulting in varied
system performance characteristics and system
impacts. Performance characteristics include commer-
cial speed, passenger capacity, capital and operational
productivity, and costs. Impacts of BRT encompass
user perception, travel time, reliability, comfort, and
externalities – air pollution, road safety, physical activ-
ity, and urban development, among others.
Advanced BRT systems typically involve integration
of (see Fig. 1, list adapted from [1, 2, 9]):
● Median running ways for exclusive use of the
BRT system buses which are separated from the
rest of the traffic through raised curbs (right of
way A or B, according to the classification by
Vuchic [10]).
● Stations with off-board payment and platforms
with level access for boarding to the buses.
● Large buses with multiple doors, special design
features, and lower emission levels than conven-
tional buses.
● Use of advanced electronic ticketing systems, such
as contactless fare cards, integrated to other
applications.
● Several information technology applications for
centralized control – such as automatic vehicle loca-
tion (AVL) and dispatch systems, and improved
user information systems – variable message signs
in stations and buses to indicate next bus and next
station in real time, and provide public announce-
ments; online and personal data appliance provid-
ing routing and schedule information, among other
applications.
● Combined service plans according to the passenger
demand characteristics, including high frequency
trunk line services combined with feeder services,
as well as accelerated and express bus services.
● Distinctive image differentiating running ways, sta-
tions, buses and overall service from other bus
services and transit applications in the city.
Few systems encompass all these components. Each
component is adapted to the local conditions and the
service needs [1], as well as budgetary constraints.

Distinctive
Image
Centralized
Control

Stations
Electronic
Ticketing

BRT Buses

Median Bus Ways

Frequent and Differentiated
Services

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 1
BRT system characteristics (Guadalajara, México, Macrobus BRT system, CTS Mexico 2010)
238 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 1 Maximum values for
some performance indicators in selected BRT systems

Performance Value System, city

indicator Definition (year) (date) System features
Commercial Distance/time as 42 km/h Metrobüs, Fully segregated busway on expressway, stations
speed perceived by the user (2011) Istanbul, every 1.1 Km [11]
on board (km/h) Turkey
Peak section Passengers/hour/ 45,000 TransMilenio, Median busway, level access stations with five
load direction (pphpd) pphpd Bogotá, platforms, overtaking lanes and combined
(2011) Colombia services –local, express, seven standees per square
meter, dense urban area [12]
Infrastructure Passengers/kilometer of 35,800 Guangzhou Median busway, with long station, overtaking lanes,
productivity busway (2011) BRT, China open operation 40 routes, very dense urban area
[13]
Capital Passenger boardings/ 3,100 Macrobús, Median busway, overtaking lanes relatively dense,
productivity bus/day (2010) Guadalajara, mixed use urban area [8]
México
Operational Passenger boardings/ 13.2 Metrovı́a, Median busway, dense urban area, very low fare
productivity bus kilometers (2010) Guayaquil, (USD 0.25 per trip) [8]
Ecuador

As a result, performance characteristics are varied.
Table 1 presents the maximum values of some perfor-
mance indicators observed:
BRT capital costs are in the range of one tenth to
one third of comparable rail systems [2, 14]. BRT costs
depend on the selection of system components and
performance requirements. Figure 2 shows comparison
of capital costs for some systems in the developing
world; most systems have been built and equipped at
a cost between 1 and 5 million USD per kilometer. The
higher cost of the Bogotá BRTsystem reflects the invest-
ment to achieve very high peak load (45,000 pphpd)
and commercial speed (28 km/h) [8].
Regarding system impacts, most systems have
resulted in higher passenger demand than expected
[2, 9, 14]; user satisfaction is frequently high [15, 16];
travel times are usually reduced as a result of higher
commercial speeds than buses in mixed traffic [2, 8, 9],
reliability increased due to the supporting infrastruc-
ture and communication technologies [9], and there is
documentation on positive impacts for several systems
regarding reduction of crashes, pollutant emissions,
and improved urban environments [9, 14, 17, 18]
Concerning comfort, most systems in developing
countries use very high occupancy standards and may
not be considered comfortable [8]; this is a result of
financial restrictions that require most transit opera-
tions in developing cities to be self sustainable. As
a result, productivity levels need to be very high. Critics
of BRT often cite comfort issues when comparing bus
systems with rail [7].
BRT Status Around the World
BRT is currently a key component of an effective urban
transport policy to address congestion, pollution, and
road accidents. A survey of BRT around the world [6]
indicates that there are about 118 cities with BRTor bus
corridors. The existing BRT and bus corridors as of
January 2011 comprise about 280 corridors,
4,300 km, 6,700 stations, and use 30,000 buses, serving
about 28 million passengers per day [6]. As of January
2011, about 49 new cities were building BRT systems,
16 cities were expanding their corridors, and 31 cities
indicated they were at in initial planning stages [6].
What used to be unusual in the early 1990s became
common in 2010. Most cities have either developed
BRT or bus corridors, or have considered their inclu-
sion in their transportation improvement plans.
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 239

RIT, Curitiba $2.4

Metrobus-Q, Quito $3.6

TransMilenio, Bogotá $12.5

Sáo Paulo $3.5

SIT-Optibús, León $1.8

Transjakarta $1.4

Metrobús, México City $2.8

BRT 1, Beijing $4.8

Megabus, Pereira $5.7

Metrovía, Guayaquil $2.0

Transantiago

Macrobús, Guadalajara $3.8

Janmarg, Ahmedabad $2.4

$0 $2 $4 $6 $8 $10 $12 $14

Total cost (infrastructure + equipment in USD millions) per km

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 2
Capital costs per kilometer for selected BRT systems (2009) (Source: Hidalgo and Carrigan [8])

Table 2 presents a regional distribution of BRT and
bus corridors around the world, which shows its pres-
ence in all continents. Latin America has about one
fourth of the cities in the world, but concentrates two
thirds of the ridership. There are a very high number of
cities and kilometers in Asia, which is the fastest urban-
izing region in the world. Usage in Europe, the USA,
and Canada is comparatively low in relation with the
total kilometers reported. Only two cities in Africa have
introduced BRT: Johannesburg, South Africa, and
Lagos, Nigeria. The only intercontinental BRT is in
Istanbul, Turkey, crossing the Bosphorus Strait.
Figure 3 shows the number of cities introducing
BRT or bus corridors since 1970, as well as the total
cumulative number of cities. Most cities in the list (97),
introduced BRT or bus corridors in the first decade of
the twenty-first century. The highest number of cities
introducing BRTor bus corridors in the last decade is in
China, followed by Indonesia, and the Latin American
region.
Some trends are emerging in the BRT Industry:
● Implementation of integrated transit systems based
on BRT: The initial phase of implementing isolated
corridors is evolving toward citywide systems inte-
gration, often including other modes of transport
like Metro (Sao Paulo [19], Santiago [20], and
Guangzhou [13]); Light Rail Transit (Guadalajara
[21], Monterrey); and conventional buses (Sao
Paulo [19], Santiago [20], and Bogotá [22]).
● Improved processes for private participation in
operations: Most BRT systems in developing coun-
tries operate under public–private partnership
agreements, using performance-based concession
contracts which are awarded either through com-
petitive tendering – with some provisions to include
240 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 2 Regional distribution
of BRT and bus corridors as of January 2011

Cities Corridors Kilometers Stations Buses Passengers/day

Africa 2 2 47 63 363 290,000
Asia 33 85 1,306 1,658 6,590 6,289,531
Europe 25 32 291 609 781 629,369
Europe/Asia 1 2 43 33 300 700,000
Latin America and the Caribbean 32 90 1,330 2,687 18,656 17,591,945
Oceania 5 12 324 142 1,411 345,800
USA and Canada 20 57 993 1,485 1,993 1,013,901
Total 118 280 4,335 6,676 30,094 26,860,546
Source: EMBARQ BRT/Bus Corridors Database [6]

Cities with BRT/Bus Corridors

25 125
Guanghzhou, Hefei, Yancheng, Zaozhuang -China
Jaipur -India; Pelembang, Gorontalo, Surakata - Indonesia
20 Bangkok - Thailand; East London Transit - UK 100
João Pessoa - Brazil; Barranquilla, Bucaramanga - Colombia 17
Estado México - México; Lima - Perú; Brampton - Canada 16
15 75
13
Bogotá TransMilenio
Los Angeles Metrorapid
10 50
8 888
Curitiba 7
5 5
5 4 25
2 2 2 2
1 1 1 11 1 1 1 11 11
0
0 0
1970 1975 1980 1985 1990 1995 2000 2005 2010

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 3
Cities with BRT/bus corridors 1970–2010 (Source: EMBARQ BRT/Bus Corridors Database [6])

incumbent operators, or direct negotiations of new
concessions with existing private providers [8].
● Increased funding from national governments: BRT
systems are eligible for funding from the national
or federal governments in several countries, like
Mexico – through the national fund of infrastructure
FONADIN [23]; Brazil – through the Program for
Growth Acceleration PAC 2 Mobility in Large Cities
[24]; and India – through the Jawaharlal Nehru
National Urban Renewal Mission JnNURM [25]
and the National Urban Transport Policy [26],
among others.
● Growth of bus manufacturers and technology pro-
viders: Commercial vehicle manufacturers and
information technology providers are increasing
their activity, especially in emerging economies
like Brazil, China, India, and Indonesia.
● Technological developments in vehicles and infor-
mation systems are also improving the quality, per-
formance, and impact of BRT [9]. Advanced
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
technologies include cleaner fuels – Compressed
Natural Gas CNG, Ultra Low Sulfur Diesel ULSD,
Biofuels, Hydrogen; improved propulsion
technologies – hybrid drive trains, electric; emissions
posttreatment – catalytic converters, particulate fil-
ters; improved central dispatch and real-time control
systems – automatic vehicle location (AVL) and
operations control; enhanced fare collection systems
using electronic ticketing and contactless fare cards;
improved user information systems – variable mes-
sage signs, communication with personal wireless
devices for routing, information of next buses and
public announcements, among others.
Despite the growth, there are some outstanding issues:
● BRT does not have a single meaning and image –
a broad spectrum of applications, from improved
bus service on mixed traffic to totally segregated
systems, are considered BRT [1, 2, 9].
● It is often regarded as a “second best” as compared
to rail alternatives without a fair evaluation or alter-
natives analysis [2, 7, 10, 14].
● Its ability to foster urban development and the use
of space designated for cars are often questioned, as
well as its actual costs and impacts [7].
Several systems in the developing world suffer
problems like [8]
● Rushed implementation – several components
incomplete at the time of commissioning, but grad-
ual improvement over time has been observed
● Very tight financial planning – user fares are not
defined technically generating risks in some systems
● Very high occupancy levels – the adopted standard
of 160 passengers/vehicle for articulated buses is not
acceptable by the users
● Early deterioration of infrastructure – lack of road
surface reinforcement or problems in design and
construction result in maintenance issues
● Delayed Implementation of fare collection systems,
which often require longer timetables than initially
expected and need very tight supervision
● Insufficient user education for initial implementa-
tion and system changes
These issues are often the result of poor planning,
implementation, and operation, due to financial,
241
institutional, and regulatory constraints; they are not
intrinsic to the BRT concept [8].
BRT in the United States and Canada
According to Levinson et al. [3], the idea of using buses
to provide mass transit is not new. Plans for bus sys-
tems, displaying the current understanding of BRT,
were proposed as early as 1937, when a “plan called
for converting three west side rail rapid-transit lines to
express bus operation on super highways with on-street
distribution in central areas and downtown,” in Chi-
cago. Similar major initiatives for buses on reserved
lanes on freeways were proposed in a 1956–1959 Wash-
ington D.C. Plan, a 1959 St. Louis Plan, and a 1970
Milwaukee Plan [3].
Several research studies, indicating the potential of
BRT and suggesting guidelines for the design of bus
facilities were completed between 1966 and 1975.
Levinson et al. [3] indicate that “most of these planning
studies focused on the facility aspects of BRT, often as
an adjunct to urban freeways. Little or no attention was
given to the station, service, and image/identity aspects
of BRT.”
The planning emphasis changed in the late 1970s,
from bus use of streets and highways toward the pro-
vision of high-occupancy vehicle (HOV) lanes and
light rail transit (LRT) due to environmental and per-
formance considerations [3]. LRT tended to be consid-
ered more fully in alternative analyses, due to lack of
information on the potential benefits and costs of BRT
[3]. More recently, BRT recaptured interest of planners
and funding agencies [3]. The Federal Transit Admin-
istration (FTA) sponsored a BRT conference in 1998,
using Curitiba’s BRT system as a model, published
major documents highlighting BRT, established a BRT
Consortium with 17 supporting cities in 1999, and
launched a BRT “Demonstration Program” involving
15 cities [3].
The most important BRT systems and bus corridors
in the United States and Canada by January 2011 are
presented in Table 3, including date of initial implemen-
tation, some features, and estimated ridership. There are
20 cities, featuring 57 corridors and 993 km, serving
about 1,013,901 passengers per weekday.
Systems in the United States are not usually
displaying full BRT features. The most
242 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 3 Most relevant BRT
systems and bus corridors in the United States and Canada by January 2011

Initial Passengers/
City year Name Corridors Kilometers Stations Buses day (year) Source
Pittsburgh, 1977 South, MLK East, 3 31.3 50 103 51,700 (2009) [6, 9]
Pennsylvania West Busways
Ottawa, Ottawa 1983 Transitway 3 30.0 37 195 97,739 (2008) [6, 9]
Miami, Florida 1997 South Miami-Dade 1 34.0 30 46 23,000 (2009) [6, 9]
Busway
Orlando, Florida 1997 LYNX Lymmo 1 5.1 22 9 4,475 (2009) [6, 9]
Chicago, Illinois 1998 Express 3 62.4 154 109 54,742 (2009) [6, 9]
Los Angeles, 2000 Metrorapid 21 390.2 650 893 464,600 (2009) [6, 9]
California
Oakland, California 2003 San Pablo Rapid 1 23.8 52 26 13,000 (2009) [6, 9]
Albuquerque, 2004 Rapid Ride 1 23.5 30 25 12,430 (2009) [6, 9]
New Mexico
Boston, 2004 Silver Line 2 11.7 61 62 30,874 (2009) [6, 9]
Massachusetts
Las Vegas, Nevada 2004 MAX 1 12.75 22 20 10,000 (2009) [6, 9]
Halifax, Nova Scotia 2005 Metrolink 3 39.4 16 15 7,266 (2008) [6, 9]
York, Ontario 2005 VIVA 5 63.6 59 71 35,300 (2008) [6, 9,
30]
Los Angeles, 2007 Orange Line 1 24.65 28 125 62,597 (2009) [6, 9]
California
Eugene, Oregon 2008 EmX 1 6.8 18 13 6,200 (2009) [6, 9]
Phoenix, Arizona 2008 LINK 4 128.0 92 5 2,372 (2009) [6, 9]
Cleveland, Ohio 2009 HealthLine 1 12.1 36 21 10,591 (2009) [6, 9]
Kansas City, Kansas 2009 MAX 1 10.2 42 9 4,450 (2009) [6, 9]
New York, New York 2009 Bx Select Bus 2 26.4 52.7 220 110,187 (2010) [6, 31]
Snohomish County, 2009 Swift 1 28.9 16 10 4,878 (2010) [6, 32]
Washington
Brampton, Ontario 2010 Züm 1 28.5 17 15 7,500 (2010) [6, 33]

advanced applications are the Orange Line (Los
Angeles, California, 2007, Fig. 4), EmX (Eugene, Ore-
gon, 2008), and the HealthLine (Cleveland, Ohio,
2009). Evaluations of these and other applications in
the United States are available in the FTA document
“Characteristics of BRT for Decision Makers” [9] and
the National BRT Institute [27], a federal funded
program.
In Canada, the City of Ottawa, implemented a very
interesting concept, the Transitway, a fully segregated
corridor with services coming in and out, allowing for
direct trips without transfers [9, 28]. Plans to partially
replace the Transitway system with Light Rail are
underway [29]. The suburban communities of York
(2005) and Brampton (2010), in Ontario, also
improved transit using BRT.
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 243

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 4
Orange Line, Los Angeles, North Hollywood station (Source: Gary Leonard, http://www.metro.net/press/2005/10_october/
metro_160a.htm#North)

BRT in Latin America

Many Latin American cities have embraced bus rapid
transit as a key component of their transit systems.
Curitiba, Brazil, can be considered the cradle of mod-
ern BRT [5]. Since the 1970s, Curitiba’s administrators
have introduced constant innovations to the city’s bus-
based transit system. Originally, the bus system evolved
from conventional buses in mixed traffic to busways,
which were fitted with at-level boarding, prepayment,
and articulated buses, creating the first full BRT system
in the world in 1982 [5]. Later on, the city introduced
high capacity bi-articulated buses and the electronic
fare ticketing systems.
In 2009, the Curitiba integrated bus system was
upgraded, again, with the introduction of the Green
Line, its sixth BRT corridor which includes the opera-
tion of 100% bio-diesel articulated buses [5]. In 2010,
the city also introduced capacity enhancements for one
of its existing corridors (Fig. 5). Capacity expansions
are underway in all the original corridors.
The success story of Curitiba has been replicated in
several cities through the region, with adaptation to the
local conditions [34]. Quito replicated the Curitiba
concept in 1995, with the use of electric trolleybuses
to preserve the environment in its historic downtown
[34]. Bogotá, expanded the concept in 2000, using very
large stations with passing lanes, combined local
and express services, electronic ticketing system and cen-
tralized control, achieving extremely high peak loads – over
45,000 passengers per hour per direction (Fig. 6) [35].
The Curitiba, Quito, and Bogotá successes were
then replicated, with adaptations, to at least 29 cities
(Table 4). As of January 2011, there were at least 91 BRT
and bus corridors; 1,300 km; 2,700 stations; and 18,000
buses, serving more than 17 million passengers per day
[6]. Latin America has one fourth of the world cities
with BRT or bus corridors, serving two thirds of the
estimated global ridership [6].
Many other cities in the region, like Buenos Aires,
Medellin, Cartagena, Cúcuta, Arequipa, Asunción,
Chihuahua, Querétaro, San Juan, and Panamá City,
were building corridors or in final planning stages as
of January 2011 [6].
The highest concentration of BRT and bus corridors
is in Brazil with 16 cities and 828 km [6]. It is important
to indicate that most cities in Brazil, outside Curitiba and
244 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 5
Boqueirao busway in Curitiba integrated system after capacity expansion in 2010 (Source: Municipality of Curitiba 2010)

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 6
Bogotá TransMilenio BRT system peak section in Avenida Caracas – station with five platforms, prepayment, level boarding
and overtaking lanes for local and express services (Source: TRANSMILENIO S.A. 2001)
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 245

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 4 Most relevant BRT
systems and bus corridors in Latin America and the Caribbean by January 2011

Initial Passengers/
City year Name Corridors Kilometers Stations Buses day (year) Source
Curitiba, Brazil 1972 Rede Integrada de 6 72 376 2,200 2,260,000 [5, 6]
Transporte (2010)
Goiania, Brazil 1976 Rede Metropolitana de 3 35 70 200 200,000 [6, 36]
Transporte Coletivo (2010)
Porto Alegre, 1977 Prioridade Transporte 10 57.2 103 1,300 1,170,000 [6, 36]
Brazil Colectivo (2010)
Sao Paulo, 1979 Prioridade Transporte 10 301.3 602.6 7,604 6,843,664 [6, 36]
Brazil Colectivo (2010)
Juiz de Fora, 1982 Av. Visconde do Rio Branco 1 5 10 83 75,000 [6, 36]
Brazil (2010)
Campinas, 1988 Prioridade Transporte 3 13.4 9 178 160,000 [6, 36]
Brazil Colectivo (2010)
Mauá – 1988 Corredor Metropolitano 1 33 66 233 273,000 [6, 36]
Diadema, ABD (2010)
Brazil
Campo 1991 Sistema Integrado de 1 3.2 6.4 89 80,000 [6, 36]
Grande, Brazil Transportes (2010)
Fortaleza, 1994 Sistema Integrado de 1 3.6 7.2 86 77,000 [6, 36]
Brazil Transportes (2010)
Quito, Ecuador 1995 Metrobus-Q 3 42.2 82 531 440,000 [6, 8]
(2010)
Uberlândia, 1997 Sistema Integrado de 1 17 13 78 70,000 [6, 36]
Brazil Transportes de Uberlândia (2010)
Bogota, 2000 TransMilenio 7 84 144 1,769 1,700,000 [6, 37]
Colombia (2010)
Natal, Brazil 2001 Sistema Integrado de 2 5 19 200 180,000 [6, 36]
Transportes (2010)
Leon, 2003 Optibus 4 31 61 90 700,000 [6]
Guanajuato, (2010)
México
Monterrey, 2003 Metrobus 3 101 201 85 125,281 [6]
México (2010)
Belo 2004 Prioridade Transporte 2 23.7 47.4 483 435,000 [6, 36]
Horizonte, Colectivo (2010)
Brazil
Feira de 2005 Sistema Integrado de 2 3.5 7 83 75,000 [6, 36]
Santana, Brazil Transportes (2010)
Mexico City 2005 Metrobus 3 67 108 275 620,000 [6]
(2011)
Salvador, Brazil 2005 Sistema de Transporte 2 4 8 167 150,000 [6, 36]
Coletivo por Ônibus (2010)
246 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 4 (Continued)

Initial Passengers/
City year Name Corridors Kilometers Stations Buses day (year) Source
Guayaquil, 2006 Metrovia 2 63 60 215 300,000 [6, 38]
Ecuador (2010)
Guatemala 2007 Transmetro 2 35.0 70 400 400,000 [6]
City (2010)
Merida, 2007 Transmerida 2 30 34 8 8,000 (2010) [6, 39]
Venezuela
Pereira, 2007 Megabus 3 19.15 37 137 115,000 [6, 40]
Colombia (2010)
Recife, Brazil 2007 Sistema Estrutural Integrado 1 9 28 444 400,000 [6, 41]
(L-O) (2009)
Santiago, Chile 2008 Transantiago 3 26 52 130 130,000 [6]
(2010)
Cali, Colombia 2009 MIO 6 179.0 358 470 215,000 [6, 40]
(2010)
Guadalajara, 2009 Macrobus 1 16 27 144 130,000 [6]
Mexico (2010)
Barranquilla, 2010 Transmetro 1 13.4 15 92 32,000 [6, 40]
Colombia (2010)
Bucaramanga, 2010 Metrolinea 1 8.9 24 131 75,000 [6, 40]
Colombia (2010)
Estado Mexico 2010 Mexibus 1 16 32 63 63,000 [6, 42]
(2010)
João Pessoa, 2010 Corredor de Ônibus de João 1 2.5 5 111 100,000 [6, 43]
Brazil Pessoa (2010)
Lima, Perú 2010 Metropolitano 2 27 35 627 160,000 [6, 44]
(2010)

Goiania, keep the general model of busways, rather than
full BRT systems. There was a major upgrade of the
integrated system in Sao Paulo in 2005 [19], which
included the introduction of an elevated fully segre-
gated busway – Expresso Tiradentes (Fig. 7). Similarly,
Santiago, Chile introduced a major reform in bus services
in 2008, which includes several busways, and prepayment
in selected stations, but keeps most fare collection off
board and has few segregated busways [20].
BRT in Europe
European cities have preserved and enhanced their
transit systems in high capacity corridors using metros
and tramways mainly, with buses only considered in
medium capacity corridors. Heddebaut et al. [45] con-
sider that high capacity BRT configurations do not suit
the European context (lack of available space, undesir-
able urban cuttings, and low demand). These authors
prefer the term “Bus with a High Level of Service
(BHLS),” rather than BRT, to refer to the European
applications introduced in several cities (Table 5).
Most applications have bus lanes in congested areas
(downtown), sharing the road with taxis and bicycles,
and in few cases lanes in expressway (High-occupancy
vehicle lanes in Madrid and Grenoble, Metrobüs in
Istanbul).
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 247

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 7
Expresso Tiradentes, Sao Paulo, Brazil (Source: Sao Paulo Transportes SPTrans 2008)

Case studies compiled by the European Union pro- BRT in Asia

ject COST [46] indicate that BHLS corridors are well
BRT is evolving very rapidly in several Asian countries,
adapted to new urban zones, small towns, and
such as China [49], Indonesia [50], India [51], and Iran
medium-sized conurbations [45]. They also allow for
[6]. Applications are very varied, from very basic bus
a variety of configurations (feeder-trunk and direct
corridors (e.g., Delhi and Pune) to very complete sys-
services), and permit transformation into tramway sys-
tems (e.g., Ahmedabad and Teheran). Table 6 lists the
tems once there is sufficient demand. Some applica-
most relevant corridors as of January 2010.
tions used advanced vehicles, including optical or
There are at least 33 cities, with 85 corridors,
magnetic or physical guidance (e.g., Castellón de la
1,306 km, 1,720 stations, 6,590 buses, serving 6.3 mil-
Plana, Eindhoven, Leeds).
lion passengers per day. Forty-five percent of the cities
Outside the European Union, the most widely used
started operation in 2009 and 2010 and many other
application is the Istanbul Metrobüs, which connects
cities are building and planning BRT and bus corridors
Europe and Asia in the only intercontinental BRT sys-
across the region.
tem (Fig. 8). This system has central busways on
One important advance in BRT development is the
expressway (fully segregated BRT), very long platforms,
application in Guangzhou, China (Fig. 9), which uses
low floor buses (articulated and bi-articulated), and
direct services, with large stations, overtaking lanes and
achieves very high commercial speed (42 km/h) and
the use of advanced technologies for control and user
peak throughput (30,000 passengers/hour/direction)
information [49]. This system features very high
[11]. It carries 700,000 passengers per day in a 45 km
throughput (30,000 passengers per hour per direction),
corridor, under expansion in 2011.
with small number of transfers.
Table 5 shows the most relevant corridors in
Europe, including Turkey. There are at least 27 cities
BRT in Australia and New Zealand
in Europe with bus corridors, 35 corridors, 335 km,
and 1,083 buses, serving 1.3 million passengers per day Australia has a long tradition on BRT, as one of the first
(53% in Istanbul) [6]. world systems, the Adelaide North East Busway started
248 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 5 Most relevant bus
corridors in Europe as of January 2010, including Istanbul, which extends to Asia

Initial Passengers/day
City year Name Corridors Kilometers Stations Buses (year) Source
Paris, France 1993 TVM 1 19.5 29 33 65,000 (2009) [6, 47]
Ipswich, UK 1994 Ipswich Rapid 1 0.5 1 1 739 (2010) [6]
Transit
Leeds, UK 1995 Superbus 2 4.2 8.4 62 31,000 (2005) [6, 48]
Dublin, Ireland 1997 QBC 1 8.4 17 68 34,000 (2009) [6, 46]
Rouen, France 2001 TEOR 3 13 52 64 45,000 (2009) [6, 47]
Utrecht, The 2001 TVM 2 8.2 16 42 33,500 (2005) [6]
Netherlands
Caen, France 2002 Twisto TVR 1 15.7 34 24 48,000 (2009) [6, 47]
Nancy, France 2002 TVR (GLT) 1 8 30 18 37,000 (2009) [6, 47]
Amsterdam, The 2002 Zuidtangent 1 33 66 80 40,000 (2009) [6, 46]
Netherlands BRT
Bradford, UK 2002 Manchester 1 3.7 7.4 7 5,467 (2005) [6, 48]
Road
Eindhoven, The 2003 Phileas 1 15 30 12 28,500 (2009) [6, 9]
Netherlands
Crawley, UK 2003 Fastway 2 24 48 23 6,000 (2005) [6, 46]
Prato, Italy 2004 LAM 1 15 30 120 60,000 (2009) [6, 46]
Edinburgh, Scotland 2004 Fastlink 2 5 10 55 44,000 (2005) [6, 48]
Lyon, France 2006 C-Lines 1 4 10 5 4,700 (2009) [6, 47]
Nantes, France 2006 BusWay (Line 4) 1 7 15 20 21,000 (2009) [6, 47]
Kent, UK 2006 Fastrack 1 15 30 14 5,726 (2005) [6, 48]
Luton, UK 2006 FTR train to 1 5 2 4 30,000 (2010) [6]
plane
York, UK 2006 FTR First York 1 5 10 11 5,500 (2010) [6]
Lorient Triskell, 2007 BHLS 1 5 15 13 15,000 (2009) [6, 47]
France
Maubeuge, France 2008 BHLS - Viavil 1 7.5 14 20 5,000 (2009) [6, 47]
Toulouse, France 2008 Toulouse BSP 2 11 17 11 13,300 (2009) [6, 47]
Istanbul, Turkey 2008 Metrobus 2 43 33 300 700,000 (2010) [6, 11]
Castellon de la Plana, 2009 TVRCAS 1 1 4 2 4,000 (2010) [6]
Spain
Cambridge, UK 2009 Guided Busway 1 25 50 46 36,938 (2005) [6, 48]
Swansea, UK 2009 FTMetro 1 13.5 27 10 5,000 (2010) [6]
London, UK 2010 East London 1 20 40 18 9,000 (2010) [6]
Transit
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 249

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 8
Istanbul Metrobüs, intercontinental BRT system (Source: Dario Hidalgo, EMBARQ, March 2010)

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 6 BRT and bus corridors
in Asia as of January 2010

Initial Passengers/day
City year Name Corridors Kilometers Stations Buses (year) Source
Taipei, China 1996 Busways 10 30.3 143 1,680 1,680,000 (2008) [6, 52]
Taiwan
Kunming, China 1999 Busways 5 46.7 63 156 156,000 (2010) [6, 53]
Nagoya, Japan 2001 Yutorito Line 1 6.5 9 52 26,000 (2010) [6]
Seoul, North 2003 Median bus lanes 5 43 73 500 400,000 (2009) [6, 53]
Korea
Jakarta, 2004 TransJakarta 10 172.2 241 520 276,643 (2010) [6, 50]
Indonesia
Beijing, China 2005 Beijing BRT 3 34.5 60 200 200,000 (2008) [6, 53]
Batam, Indonesia 2005 Bus Pilot Project 2 36 20 22 1,472 (2010) [6, 50]
Hangzhou, China 2006 Hangzhou BRT 2 18.8 50 104 153,000 (2010) [6, 53]
Pune, India 2006 Pune BRTS 2 16.5 33 54 54,000 (2010) [6]
Bogor, Indonesia 2007 Trans Pakuan 2 60 56 30 2,978 (2010) [6, 50]
Tehran, Iran 2007 Tehran BRT 5 91 114 1,125 1,440,000 [6]
250 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 6 (Continued)

Initial Passengers/day
City year Name Corridors Kilometers Stations Buses (year) Source
Changhzou, 2008 Changhzou BRT 2 41 51 240 240,000 (2010) [6, 53]
China
Chongqing, 2008 Chongqing BRT 1 6 9 10 1,500 (2010) [6, 53]
China
Dalian, China 2008 Dailan BRT 1 9 13 87 87,000 (2010) [6, 53]
Jinan, China 2008 Jinan BRT 4 34.4 46 165 162,750 (2010) [6, 53]
Xiamen, China 2008 Xiamen BRT 3 48.9 40 177 177,750 (2010) [6, 53]
New Delhi, India 2008 Delhi BRTS 1 5.8 9 96 95,565 (2010) [6]
Yogyakarta, 2008 Trans Jogja 3 90 76 75 12,702 (2010) [6, 50]
Indonesia
Zhengzhou, 2009 Zhengzhou BRT 1 26.6 38 170 84,000 (2010) [6, 53]
China
Ahmedabad, 2009 Janmarg 3 39 61 60 102,013 (2010) [6, 54]
India
Bandung, 2009 Trans Metro 1 16 15 10 589 (2010) [6, 50]
Indonesia Bandung
Manado, 2009 Trans Kawanua 2 51 39 27 108 (2010) [6, 50]
Indonesia
Pekanbaru, 2009 Trans Metro 2 74 58 20 5,691 (2010) [6, 50]
Indonesia Pekanbaru
Semarang, 2009 Trans Semarang 1 30 53 20 2,560 (2010) [6, 50]
Indonesia
Guangzhou, 2010 Guangzhou BRT 1 22.5 26 800 800,000 (2010) [6, 53]
China
Hefei, China 2010 Hefei BRT 2 12.7 14 65 65,250 (2010) [6, 53]
Yancheng, China 2010 Yancheng BRT 1 8 21 20 20,000 (2010) [6, 53]
Zaozhuang, 2010 Zaozhuang BRT 1 33 24 20 20,000 (2010) [6, 53]
China
Jaipur, India 2010 Jaipur bus 1 7.1 10 20 6,200 (2010) [6]
Gorontalo, 2010 Trans 3 90 84 15 1,920 (2010) [6, 50]
Indonesia Hulonthanlangi
Palembang, 2010 Trans Musi 2 60 106 15 1,920 (2010) [6, 50]
Indonesia
Surakarta, 2010 Batik Solo Trans 1 30 53 15 1,920 (2010) [6, 50]
Indonesia
Bangkok, 2010 Bangkok BRT 1 15.9 12 20 10,000 (2010) [6, 53]
Thailand
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 251

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 9
Guangzhou, China, BRT corridor (Source: Karl Fjelstrom [49] http://www.transportphoto.net/photo.aspx?
id=7599&c=Guangzhou&l=en)

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Table 7 BRT and bus corridors
in Australia and Oceania as of January 2010
Initial Passengers/
City year Name Corridors Kilometers Stations Buses day Source
Adelaide, Australia 1986 O-Bahn 1 12 3 138 28,000 (2009) [6, 55]
Brisbane, Australia 2001 Brisbane 3 23.8 21 1,002 242,000 (2009) [6, 55]
Busway
Melbourne, Australia 2003 SmartBus 4 233 56 122 36,200 (2009) [6, 55]
Sydney, Australia 2003 Busways 3 49.1 58 140 32,400 (2009) [6, 55]
Auckland, New 2005 Northern 1 5.9 4 9 7,200 (2009) [6, 55]
Zealand Busway

operation in 1986 [55]. Brisbane, Melbourne, Sydney,
and Auckland introduced bus corridors between 2001
and 2005 (Table 7). Bus corridors in Australia and New
Zealand have very diverse design features, from guided
busways (Adelaide) and fully grade-separated bus-only
roads (Brisbane, sections of Sydney), to on-street
busways (Auckland) and bus lanes (Melbourne) [55].
According to Currie and Delbosc [55], the BRT
concept continues to be very attractive in the region
with rapid increase in kilometers and ridership, espe-
cially on established systems. The authors identify risks
in the provision of vehicles and accommodating
high patronage growth, but highlight the relative cost-
effectiveness. BRT development in Australia and
252 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 10
Brisbane BRT (Source: Karl Fjelstrom December 2004, http://www.transportphoto.net/photo.aspx?
id=1303&c=Brisbane&l=en)

Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues. Figure. 11
Rea Vaya, Johannesburg’s BRT, South Africa (Source: Rea Vaya (2010) http://www.reavaya.org.za/photo-gallery/
state-of-the-art/category/1)
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues 253

New Zealand has exceeded rail in the last decade; nev- other modes of transport (metro, regional rail,
ertheless, rail has received greater attention lately [55]. and standard buses). Pedestrian and bicycle access
Brisbane (Fig. 10), exhibits very advanced design will be improved, and bike- and car-sharing ser-
features, and recently completed a downtown tunnel. vices will be incorporated around BRT stations.
2. Automatic guidance systems, using video, mag-
BRT in Africa netic or physical devices, will evolve to ensure
optimal docking at stations. In some cases, prior-
Infrastructure and service development in African cit-
ity lanes will be intermittent; only a few meters
ies is beginning to pick up attention by the local
ahead and behind the bus will be reserved using
authorities, as urbanization and GDP increase. Gov-
advanced signal and communications technolo-
ernments are beginning to recognize the need for orga-
gies, at other times the lanes may be used for
nized transit, to replace low-quality informal
general traffic [59]. This will eliminate the percep-
paratransit services [56].
tion of “empty lane” in systems with headways
Two cities, Lagos, Nigeria, and Johannesburg,
beyond 2 min.
South Africa, started operations of BRT systems in
3. The buses will have several advanced technology
2009. Lagos implemented a corridor on the parallel
components: hybrid buses, plug-in hybrids, or
roads of an existing expressway with a length of
fully electric with high autonomy or extra-fast
22 km, 26 stations, and 220 buses, carrying 220,000
battery charging systems at each station. Alterna-
passengers per day [57]. However, it is not considered
tive technologies will evolve, such as solar
a full BRT, as the bus lanes are on the kerb side, with
recharging hydrogen fuel cell, biofuels from sus-
some sections on mixed traffic and access to the buses
tainable feedstock, among others.
requires stairs [56].
4. Programming and operations will improve due to
Johannesburg launched a full BRT system in prep-
the availability of intelligent transport systems
aration to the Football World Cup in 2009 (Fig. 11).
(ITS), including the integration into the signal
The system is 25 km long (out of 122 km planned),
networks (advanced transit signal priority
with 33 stations and 143 buses. It connects the high-
[TSP]). Improved control (Global Positioning
density community of Soweto with the central business
System [GPS], Automatic Vehicle Location
district and carries 70,000 passengers per day [58].
[AVL], and Operations Support Systems) will
Cape Town (South Africa) started operations of its
allow optimal regulation and dynamic equilib-
BRT in 2011. There are systems in intermediate plan-
rium of the supply (buses) and the travel demand
ning stages in Port Elizabeth and Pretoria (South
(passengers). Very advanced routing systems will
Africa); Dar es Salaam (Tanzania) and Accra (Ghana)
be possible, including super express, express, and
[56]. Other cities considering BRT include Kampala
local, services, as well as direct services, to reduce
(Uganda), Nairobi (Kenya), and Bloemfontein –
transfers.
Ethekwini, East London – Buffalo City, Ekurhuleni,
5. Fare collection systems will be integrated into cell
Polokwane, and Rustenburg (South Africa) [56].
phones and turnstiles will no longer be required.
The main challenges for African cities are the crea-
Tickets will be deducted automatically from elec-
tion of local capacity to oversight and operate systems,
tronic wallets, which will have other multiple
transforming the paratransit informal services
applications, just as debit cards. User fares will be
and funding the capital costs (infrastructure and
differentiated by time of day, day of week, and
buses) [56].
length according to the supply and demand con-
Future Directions ditions of the system, increasing efficiency and
service quality.
BRT will evolve in several directions, namely:
6. User information systems will be dynamic, with
1. BRTwill gradually become an integral part of most electronic panels that indicate the next bus and
transit systems, as the cities grow their corridors public announcements, helping users to navigate
into networks and integrate BRT services with the system in advanced screens or cell phones.
254 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
7. Land use along the corridors will include good
accessibility, high-density, and mixed uses (Tran-
sit-Oriented Development [TOD]). BRT will sup-
port urban development and vice versa.
8. Contracts with private operators will be based on
performance, with dynamic adjustment formulas
of their revenue according to changes in input costs.
9. Infrastructure, technology, training, education,
and control will ensure safety, with a vision of
zero tolerance of severe injuries and fatalities.
10. BRT will facilitate physical activity, low or zero
emissions, and universal accessibility (disabled
people).
Many of these advances are already applied in sev-
eral systems around the globe or under research and
development. Integration into future BRT systems will
happen in the next few years.
Bibliography
1. Levinson H, Zimmerman S, Clinger J, Gast J, Rutherford S,
Bruhn E (2003) Bus rapid transit, vol 2: implementation
guidelines. Transit cooperative research program – report 90,
vol II. Transportation research board, National Academies,
Washington, DC
2. Wright L, Hook W (eds) (2007) Bus rapid transit planning guide,
3rd edn. Institute for Transportation and Development Policy,
New York
3. Levinson H, Zimmerman S, Clinger J, Rutherford S, Smith RL,
Cranknell J, Soberman R (2003) Bus rapid transit, vol 1: case
studies in bus rapid transit. Transit cooperative research
program – report 90, vol I. Transportation research board,
National Academies, Washington, DC
4. Wilbur Smith and Associates (1966) Under commission from
American Automobile Manufacturers Association (Levinson
HS, Principal Investigator). Transportation and parking for
tomorrow’s cities, New Haven, Cited by 3
5. Lindau LA, Hidalgo D, Facchini D (2010) Curitiba, the cradle of
bus rapid transit. Built Environ 36(3):274–282, http://www.
atypon-link.com/ALEX/toc/benv/36/3
6. Hidalgo D (2011) The global bus rapid transit (BRT) industry,
EMBARQ, The world resources institute center for sustainable
transport, Washington, DC, Jan 2011. http://www.embarq.
org/en/the-global-bus-rapid-transit-brt-industry. Accessed 10
May 2011
7. Light Rail Now (2011) Bus rapid transit analyses. http://
www.lightrailnow.org/facts/fa_brt.htm. Accessed 10 May
2011
8. Hidalgo D, Carrigan A (2010) Modernizing public transporta-
tion, Lessons learned from major bus improvements in Latin
America and Asia, World Resources Institute, Washington, DC.
http://www.embarq.org/en/modernizing-public-transportation.
Accessed 10 May 2011
9. Diaz RB (ed), Hinebaugh D (2009) Characteristics of BRT for
decision makers (CBRT), 2nd edn. Publication FTA-FL-26-
7109.2009.1, Federal Transit Administration, United States
Department of Transportation, Washington, DC
10. Vuchic VR (2007) Urban transit systems and technology. Wiley,
Hoboken
11. IETT (2011) Istanbul public transport authority, Istanbul Met-
ropolitan Administration, Turkey. http://www.iett.gov.tr/en/
section.php?sid=58. Accessed 11 May 2011
12. TRANSMILENIO SA (2011) Empresa de Transporte del Tercer
Milenio S.A., Alcaldı́a Mayor de Bogotá, D.C., Colombia. http://
www.transmilenio.gov.co/WebSite/Default.aspx. Accessed 11
May 2011
13. ChinaBRT (2011) Guangzhou BRT. http://www.chinabrt.org/
en/cities/guangzhou.aspx. Accessed 11 May 2011
14. Hensher D (1999) A bus-based transitway or light rail? Con-
tinuing the saga on choice versus blind commitment. Road
Transp Res 8(3):3–21
15. CTS Mexico (2011) Usuarios aprueban Macrobús, identifican
temas por resolver, Centro de Transporte Sustentable de Méx-
ico, México. http://www.ctsmexico.org/node/426. Accessed
11 May 2011
16. CTS Mexico (2011) V Encuesta de Opinión al Usuario de
Metrobús, Centro de Transporte Sustentable de México,
México. http://www.ctsmexico.org/node/160. Accessed 11
May 2011
17. INE (2008) The benefits and costs of a bus rapid transit system
in Mexico City, Instituto Nacional de Ecologı́a, México. http://
www.epa.gov/ies/pdf/mexico/Metrobus%20Mexico%20Phase
%20III.pdf. Accessed 11 May 2011
18. Hidalgo D, Pereira L, Estupiñan N, Jiménez PL
(2010) TransMilenio de Bogotá, Un Sistema de Alto
Desempeño e Impacto Positivo – Principales Resultados de
Evaluación Ex-Post de las Fases I y II. XVI Congreso
Latinoamericano de Transporte Público y Urbano –CLATPU,
México, 6–8 Octubre 2010
19. Hidalgo D (2009) Citywide transit integration in a large city:
the Interligado system of São Paulo, Brazil. Transp Res Rec
J Transp Res Board 2114:19–27 (Developing countries)
20. Muñoz JC, de Ortúzar JD, Gschwender A (2009) Transantiago:
the fall and rise of a radical public transport intervention. In:
Saleh W, Sammer G (eds) Travel demand management and
road user pricing: success, failure and feasibility. Ashgate,
Farnham, pp 151–172 (Chapter 9)
21. Hidalgo D, Voukas Y, Freiberg G, Lopez A, Alveano S (2010) The
Macrobus system of Guadalajara, Mexico: an evolved concept
in BRT planning and implementation for medium capacity
corridors, XVI Pan-American conference of traffic and
transportation engineering and logistics – Lisbon, 16–18 July
2010
22. TRANSMILENIO SA (2010) Sistema Integrado de Transporte
Público SITP. http://www.transmilenio.gov.co/WebSite/
Contenido.aspx?ID=AcercaDelSITP. Accessed 11 May 2011
 Bus Rapid Transit: Worldwide History of Development, Key Systems and Policy Issues
23. Mier y Terán C (2009) PROTRAM federal mass transit program,
World Bank transport forum 2009, 31 March 2009. http://
siteresources.worldbank.org/INTENERGY/Resources/335544-
1232567547944/5755469-1239633250635/Carlos_Mier_Y_
Teran.pdf. Accessed 11 May 2011
24. Ministerio das Cidades (2010) PAC 2 Mobilidade Grandes
Cidades, Governo do Brasil. http://www.cidades.gov.br/
index.php?option=com_content&view=article&id=776:pac-2-
mobilidade-grandes-cidades&catid=68&Itemid=103. Accessed
11 May 2011
25. Ministry of Urban Development, Government of India
(2010) Jawaharlal Nehru National Urban Renewal Mission
JnNURM. http://jnnurm.nic.in/. Accessed 11 May 2011
26. Ministry of Urban Development, Government of India
(2007) National Urban Transport Policy. http://www.
urbanindia.nic.in/policies/TransportPolicy.pdf. Accessed 11
May 2011
27. National BRT Institute (2011) http://www.nbrti.org/. Accessed
12 May 2011
28. Hofman A (2011) Advanced network planning for bus rapid
transit: the “Quickway” model as a modal alternative to “Light
Rail Lite”, Federal transit administration, FTA-FL-26-
7104.2007.4. http://www.nbrti.org/docs/pdf/BRT%20Network%
20Planning%20Study%20-%20Final%20Report.pdf. Accessed
12 May 2011
29. OC Transport (2011) Ottawa’s light rail future. http://www.
octranspo1.com/routes/ottawas_light_rail_futurem. Accessed
12 May 2011
30. http://www.yrt.ca/. Accessed 10 Jan 2011
31. http://www.nyc.gov/html/dot/downloads/pdf/ssi09_projects/
ssi09_fordham_sbs.pdf. Accessed 10 Jan 2011
32. http://www.commtrans.org/News/New.cfm?id=1386. Accessed
10 Jan 2011
33. http://en.wikipedia.org/wiki/Z%C3%BCm. Accessed 10 Jan 2011
34. Hidalgo D, Carrigan A (2010) BRT in Latin America – high
capacity and performance, rapid implementation and low
cost. Built Environ 36(3):283–297
35. Hidalgo D, Lleras G, Hernández E (2010) Capacidad en
Sistemas de Transporte Masivo en Buses - Desarrollo de
Fórmulas y Aplicación al Caso de TransMilenio de Bogotá,
XVI Congreso Latinoamericano de Transporte Público y
Urbano –CLATPU, México, 6–8 Octubre 2010
36. ANTP (2011) http://portal1.antp.net/site/simob/Lists/
MrsCdds09/Mcidades.aspx. Accessed 10 Jan 2011
37. TRANSMILENIO SA (2010) http://www.transmilenio.gov.co/
WebSite/Contenido.aspx?ID=TransmilenioSA_TransmilenioEnCi-
fras_EstadisticasGenerales. Accessed 10 Jan 2011
38. Fundación Metrovı́a (2010) www.metrovia-gye.com/start.htm.
Accessed 10 Jan 2011
39. Trolebús Mérida (2010) http://www.venezuelavanza.es.tl/
TROLEBUS–g-MERIDA-g-.htm. Accessed 10 Jan 2011
255
40. Ministerio de Transporte (2010) Sistemas de Transporte
Masivo en Colombia. Unidad Especial de Transporte Masivo,
Bogotá
41. Municipalidad de Recife (2010) http://www.recife.pe.gov.br/
pr/secplanejamento/caxanga.html. Accessed 10 Jan 2011
42. http://es.wikipedia.org/wiki/Mexib%C3%Bas. Accessed 10
Jan 2011
43. http://www.agenciat1.com.br/4101-onibus-ganham-faixa-
exclusiva-em-joao-pessoa/. Accessed 10 Jan 2011
44. Menckhoff G, Ochoa C, Ardila A (2010) El Metropolitano de
Lima: Implementación y Primeros Resultados de un Nuevo
Sistema BRT. Congreso Latino Americano de
Transporte Público y Urbano, CLATPU, Ciudad de México,
Octubre 2010
45. Heddebaut O, Finn B, Raubel S, Rambaud F (2010) The Euro-
pean bus with high level of service (BHLS): concept and prac-
tice. Built Environ 26(3):307–316
46. COST BHLS (2010) www.bhls.eu. Accessed 10 Jan 2011
47. Rabuel S (2010) Buses with high level of service: choosing and
implementing the right system, centre for the study of urban
planning, transport and public facilities, CERTU. http://www.
nbrti.org/docs/pdf/buseshighlevelofservice_bhns_2009_english.
pdf. Accessed 10 Jan 2011
48. http://www.ashford.gov.uk/pdf/Planning_Pentland_6c_WS_11.
pdf. Accessed 10 Jan 2011
49. Fjelstrom K (2010) Bus rapid transit in China. Built Environ
26(3):363–374
50. Ernst JP, Sutomo H (2010) BRT´s influence on public transport
improvements in Indonesian cities. Built Environ
26(3):344–352
51. Tiwari G, Jain D (2010) Bus rapid transit projects in Indian cities:
a status report. Built Environ 26(3):353–362
52. Chang M (2008) BRT in Taiwan. http://ce11012.ce.ntu.edu.tw/
paper%5CBeijing%20skchang.pdf. Accessed 10 Jan 2011
53. ITDP (2011) BRT in China and Asia. http://www.chinabrt.org/.
Accessed 10 Jan 2011
54. CEPT University (2010) Janmarg 14th month report, CEPT
University Ahmadabad, 4 Dec 2010
55. Currie G, Delbosc A (2010) Bus rapid transit in Australasia: an
update on progress. Built Environ 26(3):328–343
56. Gauthier A, Weinstock A (2010) Africa: transforming paratran-
sit into BRT. Built Environ 26(3):317–327
57. Moberola D (2010) The Lagos BRT system, LAMATA. http://
siteresources.worldbank.org/EXTAFRSUBSAHTRA/Resources/
DP09-Lagos-BRT.pdf. Accessed 10 Jan 2011
58. Mcaul C (2010) Rea Vaya, the Johannesburg BRT. http://www.
sutp.org/documents/Armin-Nairobi-event/BRT%20Joburg%
20Presentation_McCaul.pdf. Accessed 10 Jan 2011
59. Viegas J, Baichuan L (2004) The intermittent bus lane signals
setting within an area. Transp Res C Emerg Technol
12(6):453–469
256 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Bus Rapid Versus Light Rail Transit:
Service Quality, Economic,
Environmental and Planning Aspects
VUKAN R. VUCHIC1, RICHARD M. STANGER2,
ERIC C. BRUUN1
1 loop configuration, typically serving internal circu-
Department of Electrical & Systems Engineering,
University of Pennsylvania, Philadelphia, PA, USA
2 Vehicles may be either steel wheel or rubber tired.
Los Angeles, CA, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Process and Controversies in Selecting Transit Modes
Negative Trends in the Application of Urban Transit
Modes Between 1945 and 1975
Inherent Differences Between Industrialized and
Developing Countries
Light Rail Transit
Upgrading Bus Services into Bus Transit Systems and
Bus Rapid Transit
Comparison of LRT and BRT Modes
Trends and Expected Future Directions
Bibliography
Glossary
Access time The component of door-to-door travel
time spent leaving the point of trip origin to
a transit stop.
Annualized operating cost The direct cost of oper-
ating and maintaining a transit route or set of
routes over a whole year; an average value.
Articulated bus A bus vehicle with three (exceptionally
four) axles and two (exceptionally three) body sec-
tions connected by body joints that provide a contin-
uous interior for passengers. Body articulations allow
the bus to make turns occupying a moderately wider
body profile than a regular, two-axle bus.
At-grade BRT Bus rapid transit that is predominantly
on ROW B and has most of the following
elements: signal priority at intersections and other
preferential treatments elsewhere; well-spaced stops
like stations, with good passenger amenities;
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
regular, reliable services throughout the day; off-
board fare collection; real-time arrival information;
large, distinctive buses allowing rapid passenger
exchange; and a clear, branded image.
Automated guided transit (AGT) Automated vehi-
cles, operated on ROW A, in either a shuttle or
lation needs of large institutions or downtowns.
Average operating speed A performance indicator
computed by dividing the length of a route or line
by the travel time required.
Bogie or truck (United States) Frame containing
axle, motor, and brake assemblies and pivots
under the body/chassis of a rail vehicle.
Build-operate-transfer (BOT) A project implementa-
tion method where one single entity is responsible
for building transit infrastructure, procuring all
needed equipment and managing the operations
for an initial period time before the eventual hand-
over to public ownership.
Bus lane Traffic lane for exclusive (or dominant) use
by buses. It may be:
Contraflow – Operating in the opposite direc-
tion from other traffic.
Exclusive – Physically separated lanes (usually
two) for bus use only.
High Occupancy Vehicle (HOV) Lane – A traffic
lane open only to vehicles with more than
a minimum number of persons onboard.
Regular – Lane on urban street or motorway
reserved only for buses, separated by markings,
signs, or cones but not physically separated.
Bus rapid transit (BRT) or bus transit with high
level of service (BHLS) Bus transit system
with mostly ROW B, long spacings between
stops, signal priorities, vehicles with distinct
image, and other rail-like components for high
performance.
Bus transit system (BTS) A bus service designed as
a coordinated system with average stop spacings of
at least 300 m, some bus lanes, passenger-friendly
vehicle design, operations control, passenger infor-
mation, etc., for higher speed, reliability, and
efficiency.
Consist A set of rail vehicles coupled into a train.
# 2012, DOI 10.1007/978-1-4419-0851-3
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Cross-subsidization The concept of using excess rev-
enues (profits) from one line or route and using
them to support another line that earns revenues
less than its costs.
Deadheading Moving a transit unit to a line from
a depot or storage location or repositioning
a vehicle to another line without passengers.
Design-build model An approach to major projects
making the same firm/consortium responsible for
both the design and construction phases.
Design-build-operate-maintain (DBOM) model An
approach to major projects in which the same firm/
consortium is responsible for the design, construc-
tion, operation, and maintenance phases.
Diametrical line (or route) A transit line that has
a radial alignment from a suburb to center city,
crosses it, and continues into another suburb.
Economy of scale The unit cost of output decreases
with the quantity of output.
Elasticity of demand The percentage change in rider-
ship over the percentage change in some other
quantity (e.g., fare or frequency)
Frequency The rate at which transit units pass a fixed
point, usually expressed per hour; it is the inverse of
headway but usually expressed in minutes.
Global positioning system (GPS) The currently dom-
inant vehicle location system, based on taking posi-
tion fixes from multiple satellites.
Guideway A travel way (rail track, guide beam, and
other guiding surfaces) that physically guide vehi-
cles. Guideways for nonrail vehicles always require
ROW A.
Headway The time interval between transit units pass-
ing a fixed point, usually expressed in minutes. It is
the inverse of frequency, which is usually expressed
as departures per hour.
High occupancy vehicle (HOV) lane A traffic lane
open only to vehicles with more than a minimum
number of persons onboard.
Hybrid vehicle A bus or rail transit vehicle that com-
bines two different forms of propulsion like an
internal combustion engine and an electric motor,
which operate in optimum combination for each
regime, such as in acceleration, low or high speed,
etc.
Intelligent transportation system (ITS) A package of
hardware and software specifically designed for
257
improving transportation operations, information
to the public, and/or information for service
planning.
Interlining The practice of having a vehicle continue
onto another line or route rather than reversing at
a terminus.
Level of service (LOS) The combined service charac-
teristics experienced by the user.
Light rail transit (LRT) A transit mode utilizing pre-
dominantly ROW B, sometimes A or C, on different
network sections. Its electrically powered vehicles
operate in one- to four-car transit units. The mode
has a wide range of LOS and performance
characteristics.
Light rail vehicle (LRV) An electric rail transit vehi-
cle, powered or unpowered, with up to seven body
sections and overall length of 20–50 m capable of
operating on ROW A, B, and C.
Line capacity The maximum throughput of transit
units (TUs) or spaces measured at one point on
a line on a per-unit time basis; it has a wide range of
values depending upon safety regime and assumed
crowding standards.
Linked trip A trip that includes all segments a user
makes on public transport vehicles.
Load factor or coefficient of utilization Ratio of
passengers in a vehicle/transit unit and its total
capacity in spaces.
Locally preferred alternative The project alternative
selected locally that is then submitted to the federal
government for further review and a funding deci-
sion (specific to the United States).
Marginal cost The cost to carry one more vehicle on
a roadway, one more person in a transit vehicle, or,
in general, the cost of providing one more unit of
service output.
Maximum load section (MLS) The segment of a line
or route on which the maximum number of pas-
sengers is carried.
Metropolitan planning organization (MPO) A gov-
ernmental body legally required to do transporta-
tion forecasting and to propose a transportation
improvement/investment plan for a metropolitan
region (specific to the United States).
Modal split The fraction of trips by a particular mode
along a route or line, between two points, or across
an entire region.
258 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Monorail An urban transit technology that operates
exclusively on right-of-way A using in most cases
tires for both lateral and vertical support on a beam
with a special cross section, typically a proprietary
design.
National Transit Database (NTD) A source of oper-
ating statistics and cost information for each transit
agency receiving federal funds that is broken into
categories used for cost modeling and estimating
(United States only).
No-build scenario A project alternative used to com-
pare what may happen if no action is taken. The no-
build alternative may sometimes include modest
Transportation Systems Management investments.
O-Bahn Bus transit technology in which buses have
two horizontal wheels and roadway has two lateral
surfaces; the buses can operate on regular streets or
enter sections where guiding wheels stretch out and
provide guidance against the lateral surfaces so that
the driver does not steer.
Operating speed Average speed on a line or network
including times for stops, but not terminal times
and deadheading.
Opportunity cost The forgone possibilities when
resources are committed to a particular project
alternative.
Pantograph A spring-loaded mechanism mounted
atop electric vehicles that glides along and collects
current from a suspended high-voltage catenary.
Paratransit In the narrow definition, services pro-
vided as a public service on a demand-responsive
basis; in the broader definition any for-hire services
that are not fixed route (e.g., taxis, jitneys, van-
pools, airport shuttles, etc.).
People mover The informal name for Automated
Guided Transit (AGT) system.
Public–private partnership (PPP) A method of
project development in which responsibilities
and finances are divided amongst various
parties in recognition of the benefits each can
receive, hopefully to the mutual benefit of all
parties.
Radial line (or route) A line or route that begins in
CBD or at a center of activity and goes to suburbs,
in the direction of lower demand density.
Real-time passenger information (RTPI) The esti-
mated arrival time for a vehicle at a particular
stop based on tracking of actual location versus
scheduled location; it can be delivered with signs
in the field over the Internet, over Personal Digital
Assistants (PDAs), cell phones, etc.
Regular Lane on urban street or motorway reserved
only for buses, separated by markings, signs, or
cones but not physically separated.
Right-of-way (ROW) Broadly speaking, any path or
way on which a transit vehicle travels. Transit rights-
of-way (ROW) are classified in three categories:
Category A Right-of-Way (ROW A) – Fully con-
trolled ROW without (or with fully protected)
grade crossings or any legal access by other vehicles
or persons; also called “grade-separated” or “exclu-
sive” ROW. It can be a tunnel, aerial, or at-grade
level.
Category B Right-of-Way (ROW B) – ROW hor-
izontally separated from other traffic (by curbs,
barriers, grade-separations, etc.), but with grade
crossings for vehicles and pedestrians usually at
regular street intersections.
Category C Right-of-Way (ROW C) – Surface
streets with mixed traffic. Transit may have prefer-
ential treatment, such as reserved lane but not
physically separated lanes, or it may travel in gen-
eral traffic lanes.
Short turning Transit line on which some transit units
turn back at a station closer than the outlying
terminal.
Shuttle A route that consists of only two stops, one at
each terminus.
Space-averaged load factor A performance indicator
used to measure vehicle space consumption effi-
ciency over the length of a route, computed as the
ratio of total space-distance consumed over total
space-distance offered.
Streetcar A rail transit mode consisting of electrically
powered rail vehicles operating in one- to three-car
transit units, usually on ROW B or C.
Sustainable Development The concept that economic
development should take place so that
nonrenewable resources are not depleted and natu-
ral systems of restoration and pollution mitigation
are not overloaded.
Tangential route (or Line) A route or line that does
not enter into a central area, but instead con-
nects peripheral locations, generally having the
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
property of more even loading because demand
does not increase continually as with radial
routes.
Timed-transfer system (TTS) A network consisting
of transit lines and one or several transit centers or
focal points at which transit units from intersecting
lines arrive simultaneously, allowing easy passenger
transfer in all directions.
Tramway See Streetcar
Transit signal priority (TSP) The control of traffic
signal timing, and possibly street layout, designed
to favor transit vehicles over private vehicles.
Truck See Bogie
Unlinked trip A trip counted separately, even if it may
actually be a transfer between two transit vehicles.
Zonal service Transit service in which trains operate
express or local along different sections of the route,
with connections between zones only at a few joint
stations.
Definition of the Subject
The term “urban public transit” encompasses a family
of modes each serving a particular transport market.
These modes can be arrayed by their costs and the
quality of service they provide. Toward one end of the
spectrum is the typical urban bus mode. At the other
end are the regional rail systems with their high speeds
and high capacities. In between, urban transit modes
have ranges of cost and performance that overlap, and
choosing between them may be difficult. Over the last
several decades, the modes most often seen as compet-
ing for selection are bus rapid transit (BRT) and light
rail transit (LRT).
This debate between the BRTand LRTadvocates has
too often been destructive, pitting two important
urban transit modes against each other when the real
“enemy” is the overuse and overfunding of the private
car. Policymakers are often left more confused as they
listen to the different opinions of their transit advisors.
Sometimes they are led to believe that they will be able
to save money with BRT yet get the same level of transit
service and capacity as with LRT; in other cases they
realize that an LRT solution is the better one even if it
requires higher investment because of its stronger pas-
senger attraction, permanence, and stronger positive
impacts on the city.
259
What would be useful is to understand why and
how the BRT and LRT modes evolved and what roles
they play within a public transit network. Important in
this context is to see BRT as several distinct services,
each important, but few providing the speed and
capacity of light rail transit.
Introduction
There has long been an ongoing debate in urban trans-
portation on the role of bus and rail modes in urban
transportation. That debate is actually a “subproblem”
of the overall urban transportation problem – the rela-
tionship between public transit, private cars, pedes-
trians, and other modes. This problem occurs in all
countries when auto ownership increases rapidly. The
debate eventually defines the main issue: medium and
large cities face the “collision of cities and cars.” The
excessive traffic congestion caused by cars, which is
wasteful, environmentally damaging and degrading to
a city’s livability, can be solved either by extensive
construction of highways and parking facilities or by
controlling the use of private cars and offering com-
petitive transit systems.
The policies of “adjusting cities to cars” were par-
ticularly strong in the United States. From 1945 to
1970, there was massive financial support for highway
construction with negligible support for transit and the
total neglect of pedestrians.
This wave of “auto mobilization” of cities also
encompassed transit planning. A campaign swept the
country against rail modes and for the replacement of
streetcars by “flexible buses” in mixed traffic. This cam-
paign, fueled by highway, car manufacturers, and the
oil industry, also actively worked against rail rapid
transit systems. A major consulting firm developed an
all-bus plan for Washington, D.C., a city that obviously
needed rapid transit. The plan showed buses in huge
underground stations with insufficient capacity, lack of
signalization, and unacceptable exhaust gases. More-
over, the system would have required higher investment
and operating costs than the rail. There was even
a campaign alleging that “flexible” paratransit was
more efficient for car-based cities than “fixed” rail
transit.
Yet, even as the construction of metro and light rail
transit lines intensified in dozens of cities in North
260 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

America and around the world over the last 4 decades,
extreme anti-rail transit views continue in some coun-
tries. In the United States, through the 1950–1970
period, the only urban transit modes to choose from
were low-investment buses and trams (streetcars), and
very high-investment rail rapid transit (metros or sub-
ways). The evolution of the tram into light rail transit
(LRT), first introduced in Europe in the 1960s and then
in the United States by 1980, gave urban planners a rail
mode with some of the attributes of rapid transit that
was more cost-effective in medium-density corridors
and medium-size cities. This evolution was made pos-
sible by relocating trams from city streets and mixed
traffic onto their own, separated (generally) at-grade
rights-of-way.
Over the last 2 decades, a similar evolution has
occurred with buses. Early on freeway bus lanes were
introduced. Later, select bus routes on urban streets
were upgraded through a series of smaller steps includ-
ing longer bus stop spacing and preferential lane and
signal treatments. As the ultimate step in this evolution,
several cities have built largely grade-separated busways
(e.g., Ottawa and Bogotá). This wide range of bus
improvements was given the single label bus rapid
transit (BRT) (Fig. 1).
However, the introduction of the BRT concept
has again ignited a campaign that BRT can provide
the same service as rail transit but at a much lower
cost. As further discussion will show, BRT offers
a different set of features than rail transit and it
differs considerably from LRT, so that services of
the two modes are by no means fully comparable.
Nor are their permanence and impact on cities
comparable.
The debate between BRT and LRT advocates is
important because policymakers who want to
improve their transit systems often do not understand
what each mode is able to deliver. The ongoing
arguments among their advisors on the merits of
BRT versus LRT are destructive for two reasons:
they waste time and too often further delay already
long-delayed decisions, and they play into the hands
of those who are against any improvements to public
transit.

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 1
Two BRT vehicles in Bogota, Colombia can carry all the people in the cars shown in six highway lanes
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
This section reviews the evolution of both LRT and
BRT and explains how each might meet the transit
needs of cities. It defines the subcategories of bus ser-
vices that fall under the umbrella term “BRT” in an
effort to separate bus systems which have many LRT
features (primarily separated ROW B, signal priority,
and long lengths between stops) from buses which have
only a few operational improvements.
Process and Controversies in Selecting Transit
Modes
Civic leaders, transportation planners, and the public
in general are increasingly recognizing that urban
transportation policies, specifically the selection of
transit modes, have a major impact on the type and
quality of transportation, as well as on the entire city’s
character and quality of life.
Selection of transit modes is therefore a major step
in urban transportation planning. Many factors that
should be considered in comparing and selecting tran-
sit modes can be grouped into several categories:
– System performance: transporting capacity, speed,
reliability, safety, comfort, and other elements.
– Economic aspects: investment and operating
costs estimated for the projected quality of
service, volume of attracted passengers, and applied
fares.
– Direct and indirect impacts: The effects of the
planned transit system on the area it serves and
the entire city.
– Role of transit on the livability, sustainability,
and global impacts on weather change, energy con-
sumption, and other broad aspects: The planned
transit line or network must analyze these long-
term aspects, which are gaining increasing
importance.
– Quality and permanence: Numerous other factors,
such as the ability to protect the transit right-of-way
from use by other vehicles.
– Local preference: The preference of the population
for different transit modes.
– Influences by different lobbies: Groups (e.g., labor
unions, business interests), often driven by financial
interests rather than public interest, often wish to
influence major transit decisions.
261
To compare transit modes systematically, it is very
helpful to consider the elements that define modes.
They are:
– ROW category: There are three types of ROW:
A (fully grade separated), B (separated from other
traffic), and C (within mixed traffic).
– Technology: Its main components are support (road
or rail), guidance (steered or guided), propulsion
(diesel or electric), and method of driving and
control (manual or signalized).
– Type of service: This characteristic includes line
length and functional role (short-haul, regular city
service or regional).
While planning and attention to mode selection
must be performed carefully, excessive controversies
can be very harmful because they often cause long
delays and lead to incorrect decisions, such as selection
of a mode without sufficient capacity, with an inferior
ability to attract passengers, or a mode that is later
degraded by a failure to protect its ROW.
The “family of transit modes” can be formed by
listing the modes from the lowest capacity minibuses
on ROW C through medium-capacity modes on ROW
B, including BRT and LRT, and AGT, metros and
regional rail systems with long trains on ROW
A providing the highest capacity and performance.
When the average investment of these modes per kilo-
meter of line is plotted on the ordinate, the diagram of
the family of modes in Fig. 2 shows three areas with
their increasing performance and investment costs.
Negative Trends in the Application of Urban
Transit Modes Between 1945 and 1975
This entry discusses many of the factors that led to the
creation of the distinct BRT and LRT modes.
Impact of Auto-highway Expansion on Selection of
Transit Modes
Public transport companies, facing heavily subsidized
highway construction for private vehicles and virtually
no public assistance for transit investments and
operations, were forced to convert their systems from
electric streetcars and trolleybuses to motorbuses.
Part of this conversion was logical when diesel
262 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

Passenger volume
Investment cost per pair pair of lanes –K ($/km) x 106
120

in corridor
100
RGR

80
Auto/fwy
RRT
60

40 Rapid transit
LRT
BRT

Semirapid transit
20
Street SCR
transit RB Level of service

400 800 1200 1600 2000

Productive capacity – Pc (space-km/h2) x 103

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 2
Relationship between system performance (productive capacities) and investment cost of different generic classes of
transit modes

buses were upgraded so that they became more eco- Change from rail to bus vehicles which are smaller,
nomical and required less investment than streetcars on less comfortable, have fewer doors, and cannot be
lightly traveled lines. However, conversion of heavily coupled in trains (lower line capacity)
traveled streetcar lines, particularly those which had Change from electric to diesel traction, which
separated rights-of-way (category B), was made with downgrades performance, generates exhaust gases
explanations that streetcars were obsolete, buses were and noise, and prevents operation in tunnels
“more flexible” to mix with general traffic, their lines
could be easily changed, etc. Such conversions were Downgrading of Transit Services
mistakes that led to downgrading of transit services in
There were secondary impacts of the conversion from
general.
rail to bus transit. In many cities conversion to buses
If urban transportation is analyzed as a multimodal
was followed by the belief that “flexibility” of buses
system, the nearly complete conversion of streetcars/
should be used to replace a single heavily traveled line
tramways that was done in the United States, United
with high frequency of service by a large number of bus
Kingdom, France, and a number of other countries,
routes that provide better area coverage but have longer
had the following results:
headways, lower speed, and reliability of service. Such
+ Lower investment cost for line infrastructure (tracks extensive networks, which still exist in many cities,
and overhead cables) require less transferring, but offer services that have
Loss of transit ROW B and the operation of “flexi- long headways and irregular alignments, resulting in
ble” transit vehicles in general traffic, which resulted low recognition and image. The belief was that copying
in reduced speed, reliability, and loss of distinct services private cars offer would increase transit rider-
image of transit services ship. However, the opposite happened. Extensive
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
networks of bus services with long and irregular head-
ways attracted significantly fewer passengers than
a smaller number of streetcar lines with frequent
services.
Replacement of Streetcars by Buses Led to Major
Losses of Passengers
The substitution of networks of rail lines having fre-
quent, fast, and reliable services with extensive net-
works of infrequent bus services was not only less
attractive, but in the long run its lower image and
permanence diminished the role of transit and made
it supplementary to private cars, particularly in small-
and medium-sized cities. This was demonstrated by the
fact that the number of passengers carried by buses in
US cities consistently decreased even during the period
between 1950 and 1970, when buses were introduced in
many cities to replace streetcars.
Decreasing transit ridership, increasing traffic con-
gestion and deterioration of cities led many cities to
search for higher quality of transit services. The expe-
rience with “flexible” buses led to realization that mak-
ing transit vehicles independent of general traffic is the
basic element in achieving transit competitive with
private cars.
Many cities tried to introduce reserved bus lanes,
but found little support from traffic engineers, and
great difficulties getting the police to enforce lane sep-
aration. The concept of LRT with tracks physically
separated from other traffic by curbs was a logical
solution to this problem, and since the late 1970s,
about 20 cities in North America have introduced
LRT systems with separated and protected ROW on
their main lines. With the introduction of rail vehicles
with greater comfort, more space, and higher speed
than buses, many cities were very successful in reversing
downward trend of transit ridership.
Two Major Trends in Transit Role Definition and
Mode Selection: “Flexible Buses” or Upgrading
Streetcars into LRT
Between the 1950s and the 1970s, policies on the selec-
tion of transit modes varied among cities depending on
local conditions, methods of decision-making, financ-
ing sources, and the expertise of transportation
263
planners and transit agencies. Nevertheless, a review
of trends during this period of rapid car ownership
increase in North America and Europe showed two
distinctly different major groups of cities.
Cities with Policies That Encouraged Automobile Use:
These policies facilitated unlimited private car use.
They were followed in most US, UK, French, and Span-
ish cities, and they resulted in massive replacements of
streetcars/tramways by buses in mixed traffic as well as
extensive construction of freeways in urbanized areas.
Metro was considered the only alternative for high-
performance transit competitive with private cars.
Cities with Policies That Fostered Balanced Inter-
modal Transportation: These policies implemented par-
allel improvements of streets/highways and transit
systems to achieve a balanced urban transportation
system. They were the basic transportation policies in
most cities in Germany, Switzerland, Austria, Benelux,
and most Scandinavian countries as the only solution
leading to livable cities. The main element in achieving
transit competitive with private cars was recognized to
be physical separation of transit, that is, the provision
of ROW B and A. Selection of these separated ROWs
made LRT the logical choice, because rail offers signif-
icantly higher performance, comfort, and image in the
city, as well as greater permanence of transit ROW.
These advantages were given more weight as, since the
1970s, most busways in the United States were
downgraded into HOV facilities, greatly diminishing
the priority of buses over other types of vehicles
(Fig. 3).
The results of the two sets of urban transportation
policies were drastically different with respect to mode
selection as well as livability of cities. The cities which
followed balanced transportation policies systemati-
cally upgraded tramways and buses on streets into
semi-rapid transit modes to counter the increasing
attraction of private cars, and to make transit indepen-
dent of traffic congestion. The cities attempting to
accommodate the increasing volumes of auto traffic
found themselves with low-quality transit buses in
ROW C facilities and with an inability to build exten-
sive metro networks due to its high investment cost.
Eventually, many of these cities realized that they
needed transit modes that offer much higher perfor-
mance than do buses on streets, but at significantly
lower investment costs than metro systems. The LRT
264 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 3
Initially an exclusive bus roadway, the El Monte Busway in Los Angeles, USA has been degraded into an HOV facility, even
allowing single driver hybrid cars

mode met this requirement, and it has been built in
dozens of cities in the United States, France, United
Kingdom, Spain, Canada, and others. Later develop-
ment of the BRT mode provided another opportunity
for these cities to build medium-capacity modes,
narrowing the “gap” between buses on streets and
metro systems.
Inherent Differences Between Industrialized and
Developing Countries
What follows is a generalization. There are, of course,
wide differences between particular nations, and even
cities within the same nation. Moreover, some econo-
mies have such rapid development that the context for
planning transport is continually changing. While
rapid change increases the need for dramatic action, it
also increases the risk of unforeseen consequences from
implementing a project. The following are conditions
that make the selection of modes in developing coun-
tries different than their selection in industrialized
countries.
Demographics: Richer countries, by definition, have
a population with more travel mode and destination
choices. Thus, most new public transportation enhance-
ments in such countries must pay heed to the needs and
desires of so-called “choice” riders. While the fare levels
are not unimportant, they are less important than cre-
ating attractiveness for persons who can afford the high
investment and operating costs of auto ownership.
By contrast, in developing countries a large percent-
age of the population is “captive” riders. They will be
either walking or using public transportation. They are
much more sensitive to the fare level. If the fare level
can be kept very low by allowing greater crowding, this
option is often employed even at the expense of losing
many choice riders.
In rapidly developing countries there is a need for
public transport investments that address both choice
and captive riders. On the one hand, suppressing auto
use by affluent persons becomes an important goal. On
the other hand, an increasing population of lower and
medium-income persons and higher development
densities demands a goal of providing high capacities
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
with moderate fares. The question then arises whether
one mode can address both goals simultaneously or
whether different systems must be built for different
passenger segments.
Wage levels: Public transport systems in the devel-
oped world typically use 70–80% of their operating
budget for salaries, wages, and fringe benefits. Thus, it
becomes of vital importance to identify and employ
technologies that increase worker productivity. Some
well-capitalized public transport systems have seen
continuing increases in service output with little
increase in staff size. The ability to increase consist
size without additional drivers is one of the main rea-
sons for upgrading high-demand bus corridors to rail
modes. Furthermore, state-of-the-art self-service fare
collection like pay-by-phone, computer-aided
dispatching for field supervision, Internet-based itin-
erary planning services, and other advanced technolo-
gies can pay for themselves. The ultimate productivity
gain is conversion of rail lines to full automation to
remove the driver.
By comparison, well-organized transit agencies in
developing countries often pay a small fraction of the
hourly labor rate paid in richer countries, and with
fewer benefits. At the same time, funds to purchase
advanced technologies are likely to be scarce. In such
an environment, the case to go to rail technologies, or
even to larger buses, is not as compelling.
Employment policies and labor regulation: In many
developing countries, governments view public trans-
port as a significant source of employment. In some
cases, publicly owned services are overstaffed. In many
more cases, the vast majority of transport services are
provided by owner-operators or persons renting or
leasing vehicles from owners of bus fleets. In such
situations, attempts to build public transport networks
on high-capacity corridors using very large vehicles
often meet strong opposition. A recent and dramatic
case is the entirety of South Africa. Efforts to design
BRT networks in Gauteng (Johannesburg-Pretoria),
Cape Town, and Nelson Mandela Bay all met intense,
and even violent, opposition from minibus syndicates.
While it may be logical to suggest that the mobility
of millions of persons is more important than the
employment of a few thousand persons, it is a fact
that politicians can be very wary of labor opposition,
particularly in the context of high unemployment.
265
In such cases, the higher labor productivity of rail
or large BRT buses is not a particularly strong
selling point. (Actually, rail systems do have a lot of
other employment in track, station, and vehicle
maintenance.)
Financing capacity: The ability to finance major
construction projects can be a problem for many cities
even in the richer countries. For many poorer cities and
countries, it means increasing taxes or depending upon
international investment banks that attach conditions
to their grants and loans. Either option can be unat-
tractive to politicians.
One of the consequences of the difficulty to finance
transit projects is the controversy between building
BRT and metros. This is true even for very large
cities that need the capacity of a metro due to
large populations, long travel distances, and severe
street congestion. The choice in such cases is often
between the lower investment costs per kilometer
of lower-performance BRT, or the higher investment
for a higher performance and more permanent rail
system.
Organizational capacity and effectiveness: In many
developing countries, the existing institutions at the
city, regional, and/or national level do not have the
ability to design, build, or manage modern high-
capacity public transport systems. This lack of capabil-
ity acts as a deterrent to even consider many projects.
One case is instructive. A need for a metro network in
Delhi was recognized in 1997, but neither the national
government nor prospective international lenders
believed that the existing public agency could design,
build, or manage such a system given its record of gross
overstaffing and inefficiency. Thus a new corporation
to be headed by a proven manager was created. In the
end, the corporation not only built an excellent metro
rail system, but it was well poised to compete against
international firms for future projects in other
Indian cities.
Public participation: Most of the industrialized
countries have public participation laws designed to
allow affected parties as well as the average citizens to
influence the design process. These laws are
a democratic response to cases where citizens
receive a new transportation project that they would
not have selected had there been more input. This
process slows down the design effort, but it also
266 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
presumably results in a more beneficial and just project
in the long run.
Many developing countries, even with well-
intentional planners, do not have formal public partic-
ipation processes. Thus, planners must design systems
with less knowledge of the needs and desires of poten-
tial users. The technical expertise of planners and engi-
neers is therefore of primary importance.
On the other hand, in some developing coun-
tries governments have greater powers than in dem-
ocratic industrialized countries. This can make the
design and build steps far easier, quicker, and more
efficient.
Right-of-way acquisition: Obtaining reserved ROW
B is the basic element determining the feasibility of
semi-rapid transit, BRT or LRT. The possibility of get-
ting a protected ROW depends on many factors, often
different in developing and industrialized countries.
Advocates of BRT claim that 8–10 m strips of land
can be obtained easily in many developing countries
because auto traffic volumes are lower than in indus-
trialized countries. However, in numerous cases the
expected acquisition of such rights-of-way has proven
to be unrealistic.
The reasons for this discrepancy of theoretical
claims and real-world implementation are many.
Although car ownership in those countries is still low,
street traffic congestion is often even more acute than
in industrialized countries because their streets are even
more congested by trucks, minibuses, rickshaws, taxis,
bicycles, pedestrians, and other participants. The diver-
sity of modes makes traffic conditions even more cha-
otic and difficult to regulate. This makes acquisition of
protected traffic lanes extremely difficult. For example,
in several cities, fleets of buses that were bought for
planned BRT systems had to be stored in expectation
that the BRT infrastructure will be built later. In
Jakarta, there were legal challenges of exclusion of
other traffic from bus lanes, while in New Delhi regu-
lations allowing and prohibiting taxis from BRT lanes
was changed several times.
In industrialized countries there is also a major
difference between BRT systems in theory and in prac-
tice, but for different reasons. A number of successful
reserved busways with most BRT elements were
downgraded by their conversion to high occupancy
vehicle (HOV) lanes. The decisions for these changes
were made by judges in courts (discontinuance of HOV
lanes on the Santa Monica Freeway in Los Angeles), by
city councils (discontinuance of a bus mall in Philadel-
phia), by state legislatures (conversion of the El Monte
Busway in Los Angeles to HOV lanes), and by other
bodies which have no professional expertise in urban
transportation.
Decisions to retain the exclusive use of busways
by buses or to allow other vehicles vary among cities
and countries. In the United States, many exclusive
busways have been downgraded to HOV facilities, in
many cities taxis are allowed to share busways, and
most recently hybrid passenger cars are given this
privilege. In Madrid, motorcycles are permitted in
busways.
Modal integration: In the more developed countries,
it is common for one agency to either provide all
services, regardless of mode, or else manage them
under one coordinated fare structure. The operator of
a particular vehicle in no way depends upon the fares
collected (or not) from boarding passengers. In such an
environment, it is possible to institute significant or
even total network reorganization with the advent of
a new mode. Thus the lower income person is not
unduly penalized for needing to transfer between
services.
By contrast, most developing countries have cash-
based operators that depend on retaining all or part of
the money collected. At the same time, riders are forced
to pay a full fare each time they switch between vehicles
or modes. Thus, both the operators and riders have
reason to oppose the imposition of a new higher-
capacity mode within their existing network. The for-
mer sees a revenue loss and the latter sees paying
additional fares to use the new service.
Traffic enforcement: Police control and enforcement
of exclusive transit rights-of-way is very different for
the two semi-rapid transit systems. LRT requires very
little police control, mostly limited to intersections.
A BRT lane, when it is immediately adjacent to other
traffic lanes, is always attractive to road vehicles travel-
ing in other lanes, so that police enforcement must
always exist at intersections and bus lanes between
intersections.
Police enforcement depends on local police effec-
tiveness, which varies among cities. For example, cities
like Singapore and New York have stricter controls than
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Boston and Moscow. Similar variations exist among
cities in developing countries, but in many of their
cities the practice of bribing policemen degrades the
traffic discipline even more.
Light Rail Transit
Cities which adopted policies of implementing bal-
anced transportation concentrated on upgrading tran-
sit by developing rail transit networks on ROW
categories B and A. This has been the most important
measure making transit independent of street traffic
congestion and therefore competitive with the private
automobile. Transit on exclusive ROW logically led to
the development of advanced articulated rail vehicles,
increased operating speeds, reliability, comfort, and
safety.
Major changes in medium-capacity rail transit,
most of which took place in Germany, The Nether-
lands, Belgium, and several other European countries,
created a transit mode that is actually more similar in
its performance and efficiency to metro/rapid transit
than to conventional streetcars/tramways, so that it was
given a new name: light rail transit, LRT, or Stadtbahn
in German (Figs. 4–6).
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 4
Tramway on ROW category C in Milan, Italy
267
Development of LRT Mode
Following several decades of system design, rights-of-
way development, and technical and operational inno-
vations, LRT has become a very diversified mode. The
most significant innovations that led to the creation of
LRT and its introduction in dozens of cities are listed in
the box below.
It is interesting to note that during the 1970s and
1980s many cities built tunnels for center-city sections
of LRT to make their operations as close to the metro as
possible. This was the case, for example, in Frankfurt,
Stuttgart, Edmonton, and Hannover. Since 1990, how-
ever, an increasing number of cities have designed LRT
systems that run through pedestrian zones with
decreased or eliminated general traffic. For example,
in cities like Strasbourg, Calgary, Denver, Barcelona,
and Istanbul, LRT trains travel through central cities
at lower speeds than tunnels would allow, but that often
does not result in longer passenger trips because it
eliminates for passengers the time and effort of walking
two levels into and out of underground stations
(Fig. 7). Heavily traveled lines with longer trips and
a need for higher speed are, however, still being placed
in tunnels, on aerial structures or other types of
268 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 5
LRT on ROW category B in Paris, France

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 6
LRT on aerial structure in Los Angeles, USA
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 7
LRT within a pedestrian zone in Karlsruhe, Germany
exclusive ROW, as is the case in Dallas (Fig. 8),
Cologne, Copenhagen, St. Louis and Rouen.
Major Steps in the Development of LRT and Its Role
as the Leading Medium-Capacity Transit Mode
● Consolidation of extensive tramway networks into
fewer, high-performance LRT lines.
● Systematic replacement of ROW C (street running)
by ROW B and A, that is, separated ROW.
● Use of street medians, tunnels, or aerial alignments
on the same lines.
● Intermodal integration with bus and metro lines
through construction of joint transfer stations.
● Development of high-quality rails and switches that
provide quiet and virtually perfect riding comfort,
far superior to buses and rubber-tired AGT systems.
● Articulated vehicles offering spacious comfort,
operated as one- to four-car trains up to 90 m long
with capacities as high as 750 spaces.
● Change from unidirectional tramway-type to bidi-
rectional metro-type rolling stock.
● Self-service fare collection allowing rapid boarding/
alighting of passengers on all doors and use of any
type of fare.
269
● Low-floor (0.25–0.35 m above top of rails) vehicles
facilitate boarding/alighting of all passengers and
meet the needs of the disabled.
● Because of increased TU capacity, one-person
crews, and increased operating speeds, the produc-
tivity of operating personnel (expressed as passen-
ger-km/h/driver) has increased about 20 times from
tramways. Productivity is also about five times
greater than on the highest capacity BRT (750 vs
150 spaces per driver, not considering typically
higher operating speeds of LRT than BRT).
● Operation of LRT in central cities is performed effec-
tively in tunnels, or, in many cities, by running
directly in pedestrian malls and zones (with lower
operating speeds on such sections).
● Integration (track sharing) of LRT with regional rail
lines for services to suburbs and nearby cities.
Major Types of LRT Applications
Successful construction of LRT in many European cit-
ies led to numerous professional conferences and the
adoption of LRT as the leading medium-capacity tran-
sit mode in many countries which had eliminated
streetcars/tramways during the 1950–1970 era: United
270 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 8
High-capacity high-speed LRT train in Dallas, USA
States, Canada, France, United Kingdom, Spain, and
a number of others. During this new wave of building
LRT since the 1970s, about 100 cities have either
expanded or built new lines and networks. In many cities
one or several new LRT lines have become the main
transit system carriers with significant increases in tran-
sit ridership. This was the case, for example, in San
Diego, Calgary, Sacramento, Portland, Strasbourg, Bor-
deaux, Sevilla, Birmingham, Denver, and many others.
The successful introduction of LRT in so many cities
of different sizes and characters is largely due to the
extreme diversity of characteristics of the LRT mode.
An analysis of the present broad family of tramway/
LRT modes has defined ten categories of this mode by
its physical, operational features and role it plays in the
city’s transportation. They are listed in the box below.
Ten Categories of LRT and Its “Neighbors”: Tram-
ways and Mini-metro Modes
1. Conventional tramway mostly with ROW B and C:
Melbourne, Moscow, St. Petersburg, Toronto
2. Modernized tramways with signal priority,
improved stops, prepaid fares, etc.: Amsterdam,
Oslo, Prague, Vienna, Zurich
3. New tramways built in redesigned pedestrian-
oriented city centers (some also using historic
vehicles as center-city shuttles): Grenoble, New
Orleans, Nice, Portland Streetcar
4. LRT developed from upgraded tramways: Berlin,
Cologne, Rotterdam, San Francisco, Stuttgart
5. LRT with metro network: Boston, Brussels,
Philadelphia
6. LRT systems in suburbs of megacities: Hong Kong
(Tuen Mun), London (Croydon), New York (Hud-
son-Bergen), Paris (lines T1, T2, and T3)
7. New high-performance LRT systems: Birmingham,
Calgary, Dallas, Denver, Nantes, Portland, San Diego
8. Light Rail Rapid Transit (LRRT): LRT on ROW A only,
also referred to as a “mini-metro”: Kuala Lumpur,
Manila, Philadelphia (Norristown Line)
9. Automated LRT: same as #8, but fully automated
and without driver: Copenhagen, Kuala Lumpur,
London (Docklands), New York (JFK Airport), Van-
couver (Skytrain)
10. LRT/Regional Rail integrated systems, popularly
(although incorrectly) known as “Tram-Train”:
Bern, Karlsruhe, Kassel, Manchester, Saarbrucken
Review of LRT Characteristics
An analysis of the development and broad deployment
of LRT in many cities shows that this transit mode
has many characteristics that make it an optimal
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
solution for cities that need medium-capacity transit
systems, that is, those that perform much better
than regular buses on streets, but require a
substantially lower investment cost than metro sys-
tems. Main characteristics of LRT are summarized in
the following box.
LRT has several other physical characteristics that
distinguish it from road vehicles. First, being a guided
system, LRT uses considerably wider and longer vehi-
cles than buses and it can operate in trains. With the
high capacity of transit units, LRT can provide a higher
line capacity per track than road vehicles can provide
per lane. Second, electric propulsion gives LRT systems
better performance (acceleration, grade-climbing abil-
ity, braking with regeneration of energy), no exhaust or
noise along the line, and the ability to operate in tun-
nels, which vehicles with internal combustion engines
do not have. Finally, rail technology makes signaliza-
tion, automatic train protection (ATO), and other
automated safety features possible, so that LRT can
operate at higher speeds and at a higher degree of
safety than road vehicles.
Positive and Negative Characteristics of Light Rail
Transit Compared to Its Peer Medium-Capacity
Modes
+ Rail guidance technology has a simple basic mecha-
nism: four to eight wheels per vehicle running on
two steel rails.
+ The simplest, fastest, fail-safe switching of all guid-
ance technologies.
+ Steel-wheel-on-steel-rail contact produces extremely
low rolling resistance, so that rail modes require the
lowest energy consumption per ton of weight of all
presently operational transit modes for any given
dynamic performance.
+ Rail is the only guidance technology that allows
not only at-grade crossings, but also on-street
running.
+ Its simplicity and ruggedness give rail technology
low maintenance requirements and high durability.
+ It copes better with adverse weather conditions (rain,
snow, and ice) than rubber-tired technologies.
+ Modern rail vehicles provide virtually the ultimate in
stable, smooth riding comfort.
271
– Investment cost of rail lines is higher than that for
buses, but lower than the cost of other guidance
technologies, such as Rubber-Tired Rapid Transit
and Maglev.
– The lower adhesion coefficient of steel-on-steel than
rubber on dry concrete surfaces gives LRT and other
rail modes two relative disadvantages.
– Rail systems cannot negotiate as steep gradients as
can rubber-tired systems.
– They must be operated with higher degree of safety
because of their longer stopping distance.
– Although modern rail vehicles run extremely quietly
on straight sections and mild curves, when negotiat-
ing sharp curves, rail vehicles produce more noise
and vibration than do rubber-tired vehicles.
Roles and Impacts of Light Rail Transit
The interaction of transit systems with land use plan-
ning, physical form, and quality of life in cities is
mainly dependent on the infrastructure of the transit
system. While transit modes with flexible routings,
such as buses on streets, do not have much impact on
city planning and form, permanent infrastructure of
guided systems provides the certainty that investment
decisions require, so that LRT can be used to influence
the form and character of a city.
Mayors of several cities which have recently built
LRT systems, including Portland, Denver, and Salt Lake
City, have often pointed out in their speeches that the
introduction of LRT did not only improve transporta-
tion services, but it changed the character and
increased the livability of their cities as well. Similarly,
many cities in France, Spain, Ireland, and other coun-
tries have used the construction of new LRT systems as
the pivotal project for the reconstruction of entire
corridors or central urban areas to make them more
pedestrian-oriented rather than car-dominated. Good
examples of this are Strasbourg, Lyon, Toulouse, Nice,
Dublin, and Valencia (Spain).
Upgrading Bus Services into Bus Transit Systems
and Bus Rapid Transit
Local bus routes are the workhorses of the transit
industry, and must operate under challenging condi-
tions. They operate in mixed traffic like all other
272 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
vehicles. They have no signal priority, and only rarely
any reserved bus lanes. They serve many bus stops,
often 150–200 m apart, which vary from a small sign
on the side of the road to shelters and benches. The
driver does fare payment and control at the front door.
And local bus routes necessarily serve both short, local
trips and long, commuter trips.
Over time, a number of measures that incremen-
tally improve local bus services have been developed
and gained increasing acceptance. Taken together, they
provide a level of service that has become popularly
known as bus rapid transit. Unfortunately, BRT is
a term that has been too loosely used to describe too
wide a range of bus service enhancements. This has
somewhat tarnished the term.
This chapter will first review efforts to make buses
act more like light rail within their own guided rights-
of-way and within their own tunnels. Concurrently,
bus planners began to focus on taking the bus mode
and greatly improving its physical and operating envi-
ronment. The next section will discuss those measures
that have given buses preferential treatment in its ongo-
ing struggle with all other vehicles using the streets.
Guided Buses and Buses in Tunnels
In the late 1970s and early 1980s, there were a number
of guided busways proposed and a few were built.
These efforts were attempts to provide the bus mode
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 9
Sole major O-Bahn system in Adelaide, Australia
some of the attributes of rail transit, or at least make
them appear to have the attributes of rail transit: speed,
safety, and exclusive, narrow rights-of-way. Other
applications used various mechanical arms or wheel
assemblies to allow buses to track a raised, guiding
curb. Anything raised, of course, made it difficult to
move across intersections, and interrupting the contact
with the raised curb proved too unsafe. Other designs
used electronic or optic sensors to guide the vehicle.
The most lasting of these guided buses was the
German O-Bahn concept that was initially built in
Essen, Germany, but that test track did not get
extended and has been superseded by light rail.
An O-Bahn was also built in Adelaide, Australia, and
it is still operating there. This concept uses concrete
beams with guiding curbs within its own, fully grade-
separated right-of-way. This design proved successful
to install, if expensive. It has not been replicated else-
where, except for a few short sections of bus lines
(Fig. 9).
Several bus tunnel projects were built to allow buses
to avoid surface congestion. The first such project in
America was Pittsburgh’s original South Busway opened
in 1977. It made use of an old tramway tunnel that was
converted to serve both buses and light rail trains. There
are, however, no stops or stations in the tunnel. The
2.1-km Seattle Bus Tunnel began construction in 1987
and cost $450 M to complete. Operations began in 1990
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
using dual-mode diesel-electric trolleybuses. It became
a joint bus and light rail tunnel in 2007. The older buses,
which cannot operate with the light rail overhead cate-
nary system, have been replaced with new diesel-electric
battery hybrid buses. Thus, tunnels for BRT have cost in
some cases more than tunnels for rail modes, including
LRT and metro systems.
By far the longest bus tunnel built in North America
has been a section of the Silverline near South Station in
Boston. The project was poorly conceived and represents
an extremely expensive facility with very poor results. This
bus tunnel cost about $500 million per mile ($350 million
per km) to build. To solve the problem of diesel motor
exhaust, dual-mode (diesel/electric) buses have been intro-
duced. The tunnel has curves with limited sight distances,
but has no signal protection, so that speed is limited on
different sections to 10–20 km/h. At the exit of the tunnel,
there is a traffic signal that further delays trolleybuses
1–2 min. The line then continues to the airport, where
buses stop along a frequently congested four-lane roadway.
Picking up passengers with luggage who have to pay fares
on board also slows the service down.
The Silverline in Boston demonstrates that claims
that BRT is always much cheaper to build than LRT is
not correct. Actually, in many cases (Seattle, Boston)
BRT tunnels have involved much higher investment
cost and provided much lower system performance
than rail tunnels.
The Development of Bus Preferential Treatments
For many decades buses were treated no differently
than any other street vehicle. The result was slow ser-
vice and poor reliability. Gradually, planners began to
realize that providing preferential treatment for buses
increased transit speeds, attracted riders, and lowered
operating costs. These have increasingly been incorpo-
rated into street designs, but there is still a reluctance to
provide buses with the priority that their higher pas-
senger loads warrant.
Intersections are where a number of improvements
can be made. One basic improvement is to allow buses
to go through an intersection from a right-hand turn
lane where there is often a bus stop. Providing a lead
“queue jumping” green signal from this lane or
a “bypass” lane through the intersection also reduces
travel time.
273
The separation of buses onto their own lanes pro-
vides much more benefit, but this if often quite difficult
in a congested city and is done, even now, only rarely.
Reserving the curb lanes for buses is the most common
approach because bus stops can still be on sidewalks.
Bus lanes can also be located in the center of streets.
This is usually an expensive proposition, however,
because streets may need to be further widened for
bus stops, turn lane storage, etc. Contraflow bus lanes
take advantage of wide streets or street pairs to carve
out the necessary travel way for buses; in the proper
place they can be useful.
Finally, buses have begun to be given (by traffic
engineers) various levels of traffic signal priority
(TSP). TSP for bus transit usually comes in the form
of a “green wave”, or “green extension/red truncation”
which is limited, for example, to less than 10% of the
cycle length. While traffic engineers may give signal
priority to buses at many intersections along the
route, they are rarely given at all intersections.
A higher level of transit signal priority is traffic
signal preemption. Light rail trains are almost always
given traffic signal preemption when operating faster
than 35 mph and preemption often comes with cross-
ing gate protection. These gates are justified because
the weight, speed, and long stopping distances of trains
require stronger protection and allow higher transit
speeds than buses.
Express Bus Services
The first step often taken by transit agencies to enhance
local bus services is to supplement the route with lim-
ited-stop service. Fewer bus stops means shorter travel
times for passengers going longer distances. This is
especially appealing during peak commuting hours,
which is when many limited-stop bus services operate.
Reducing bus stops is a prerequisite for all higher levels
of bus operations.
Other than fewer stops, however, many of the fea-
tures of local bus services remain: no preferential treat-
ment, no dedicated bus lanes, onboard fare payment and
control, and no special amenities or service branding.
Enhanced Bus Transit Service
These services add further travel time savings to lim-
ited-stop bus services. They retain and even reduce the
274 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
number of bus stops of limited-stop bus services. They
included bus stops with improved shelters and seating,
arrival-time message boards, and if possible extended-
curb bus stops that remove the time buses lose moving
out of, and then back into, the flow of traffic. Special
bypass lanes at intersections, lead “queue jumping”
green signals, and special turning lanes can help give
buses an advantage at busy intersections. Priority signal
treatment that gives approaching buses an extended
green signal can be a significant travel time improve-
ment. The ultimate improvement is to give the
buses their own bus lanes separated from
other traffic. However, as noted above, rarely are des-
ignated street lanes of any length given over to bus
operations.
Bus services with enough of these enhancements are
often branded to reflect their special status and service
quality. Often newer, articulated, cleaner-energy buses
with special paint schemes and catchy names are used
to differentiate these services. For example, this net-
work is called Metro Rapid Bus in Los Angeles, Special
Bus Services (SBS) in New York, and the Silverline in
Boston.
Although these types of services are often given the
label BRT, they usually do not include enough enhance-
ments to warrant the label. Enhanced bus services typ-
ically lack three key timesaving features: fully separated
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 10
The Orange Line BRT in Los Angeles on ROW category B uses articulated buses to increase line capacity
bus roadways, off-bus fare collection, and traffic signal
preemption.
At-Grade Busways
There are at least two major at-grade busways in the
United States: the South Miami-Dade Busway and the
Los Angeles Orange Line. The Eugene, Oregon’s Emer-
ald Express busway may be another, although it does
not have the same level of treatment the others do.
(Seattle has an approximately 2 km long busway seg-
ment south of the bus tunnel.) The first two facilities
appear to have all of the ingredients needed to provide
service comparable to light rail. They operate over
separate bus roadways; they have off-vehicle fare col-
lection; they have traffic signal priority at most inter-
sections they cross; and they have only limited service
over city streets. To make up for the inability to train-
line buses, these busways use longer, articulated buses
to provide more line capacity (Fig. 10).
That being said, the service provided is not compa-
rable to light rail operations. The main reason is that
neither the Orange Line nor the South Miami Busway
has crossing gates and therefore both must slow to
below 32 km/h (20 mph) at each intersection, even
with signal priority [1]. The use of crossing gates with
at-grade busways has not been allowed by traffic
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
engineers because their operation takes too much green
time away from cross traffic [2]. At some intersections
the buses on the Orange and South Miami-Dade
Busways have no signal priority at all.
Crossing gates are mandatory for LRT above
a certain speed typically 55 km/h (35 mph). Because
of them, ROW B for LRT can act like a ROW A by
providing full protection for the light rail train.
Bus Transit System: BTS
As mentioned above, regular bus services in many
cities, including most US cities, are operated very inef-
ficiently: they have stop spacings of 150–300 m (at
every intersection), passengers board and alight at any
door, colliding with those who are boarding and trying
to pay through a single channel door; many buses do
not stop close to the curb; there are few bus lanes and
nearly no signal priorities at any intersections. Recent
efforts to improve such operations have been rather
liberally given the term “bus rapid transit.” This trend
has resulted in downgrading the BRT concept and
caused confusion.
This text uses three definitions of bus systems.
Regular, conventional (and often inefficient) bus ser-
vices are referred to as “Regular Bus – RB.” Bus
services which are improved in station spacings, oper-
ations, etc., similar to many bus services in European
countries and those in US cities which have been
upgraded, are designated here as “Bus Transit System –
BTS.” Finally, “bus rapid transit – BRT” are the
systems which have distinctly higher performance
with elements defined rather precisely in the following
section.
Bus Rapid Transit
Definition of Bus Rapid Transit There is a consensus
[3] that bus rapid transit is a bus system that has the
following elements superior to regular bus systems:
1. Reserved (physically separated) lanes or roadway –
ROW B or A not shared with other vehicle catego-
ries (taxis, HOVs, and others)
2. Distinctive lines with frequent, reliable service and
regular headways during all daily hours
3. Distinctive stops with passenger information and
protection spaced at least 300–600 m
275
4. Distinctively designed bus vehicles with large
door-to-capacity ratios, low floor, or high
platforms
5. Bus preferential treatment at all signalized
intersections
6. Use of ITS technology for monitoring vehicle
movements, passenger information, and fare
collection
However, this definition is rather “idealistic,”
because there is virtually no bus system in the
world that has all these elements. Therefore, it is con-
sidered that BRT is a system that has most of these
elements.
Several bus systems in the world have achieved an
exceptionally high quality and fall in this definition of
BRT. Often cited are Bogotá’s TransMilenio (Fig. 11),
Brisbane’s Busways, and Ottawa’s Busways. Pittsburgh’s
Busways could be among them if its image and other
elements (station amenities, off-board fare collection)
were improved.
The most important requirement for a busway to be
comparable to an LRT line is to give it a similar right-
of-way and, therefore, a similar operating speed.
Because at-grade busway buses will not get the same
signal preemption as do light rail trains, greater empha-
sis – and investment – must be made to give busways
more grade-separation at intersections. The need to
provide a high level of priority is particularly true in
the central business district when a 90 km/h (55 mph)
approach on a busway can degenerate into an 15 km/h
(9 mph) slog through congested downtown streets.
Successful BRT systems try to make this investment
(Fig. 12).
BRT stops are also well spaced and more like sta-
tions with good weather protection, seating, a sense of
security, and real-time bus arrival information. These
elements are common with LRT stations, but are too
often short-changed on busway lines.
BRT systems use large buses with wide doors for
easy and rapid passenger exchange. Articulated buses
are common and even double-articulated buses are
used. To help market the better BRT service(s), its
buses have special bus colors and logos.
BRT systems try to provide BRT buses any
preferential treatment necessary through any at-grade
intersections or bottlenecks off the ROW A. A number
276 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 11
Bogotá TransMilenio BRT line

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 12
Bogotá’s BRT distribution through central city pedestrian area

of such measures were discussed in the preceding
section. It does little good to have great service
only on that portion of the bus route that is fully
grade-separated if the overall experience of the passen-
ger is mediocre.
BRT systems monitor all aspects of their operation
to make sure buses are on time and safe, to provide the
traveling public with real-time arrival information and
respond to their questions, and to process quickly and
efficiently the collection of fares.
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
BRT systems have robust bus operations with sev-
eral types of services available. While there is typically
the busway-only service equivalent to a light rail line,
there may also be express bus services and services
typical of a freeway busway with suburban bus routes
using the common busway trunk segment. Two things
make such a complex operation possible: the design of
the busway that allows buses traveling in the same
direction to pass each other and the full grade-
separation of the busway. This mix of bus operations
is beyond the abilities of almost all light rail systems.
The extra tracks needed for trains to pass other trains
traveling in the same direction are rarely built. Express
trains may skip-stop or be scheduled to just catch up
with a local train ahead, but do not pass the local train.
True BRT is not “cheap LRT” or “LRT Lite” [4].
Building a substantially grade-separated busway with
well-designed stations can be very expensive, particu-
larly within the congested central city. But with a robust
and well-conceived service plan using several types of
bus routes the resulting service can have some advan-
tages over LRT with respect to easier introduction of
line connections, branches, and express/local services if
stops have four lanes (see Fig. 11). It is, however,
doubtful that a BRT line can achieve on any given
ROW width (two tracks/lanes, or four tracks/lanes)
the capacity of a light rail line capable of coupling
multiple cars into a single unit.
BRT Evaluation: Successes and Failures At this time
(2011), the broad experiences of recent decades
achieved by dozens of cities – and demonstrated by
cities recognized as leaders in efficient transportation
that contributes to their livability and sustainability –
can be summarized in the following major points:
● Balanced intermodal transportation is greatly supe-
rior to transportation systems based on car depen-
dency, meaning excessive restrictions on certain
major modes such as transit, walking, or bicycling,
or impracticalities on their use.
● Balanced transportation must incorporate and
favor public transit systems which consist, particu-
larly in medium and large cities, of different inte-
grated modes, such as buses, BRT, trolleybuses,
LRT, metro, and regional rail.
● Large cities, with more than one million people,
usually have rail modes as their main carriers, but
277
rail systems cannot achieve their full potential if
they are not complemented by and integrated with
buses.
There have always been promoters of individual
modes claiming that they are superior to other
modes. The strongest such pressures have been for
private cars, and their political successes have created
car-dependable cities that are inferior to intermodal
cities with respect to quality of life, economic vitality,
and sustainability. Single-mode promoters have also
caused “waves” of discussions that the most efficient
modes are paratransit, monorails, various automated
guided systems (AGT), including the totally infeasible
personal rapid transit (PRT, now reappearing under the
name “Pod cars”). These extremist views that a single
mode is the dominant or only solution for all cities
have eventually been discredited. Intermodal transit
systems utilizing the diverse family of transit modes,
from historic streetcars to BRT and regional rail sys-
tems, are obviously the trend for the future.
To be fair, promoters of single modes have fostered
some useful innovations. Paratransit has become more
diversified since the 1970s. Some 2 dozen monorails
have been built in various cities, and AGT systems are
used in an increasing number of airports as well as for
transit lines in some cities, particularly as downtown
circulators.
It is unfortunate that BRT, which has brought very
significant progress to many cities, is promoted by
some organizations, government bodies, and individ-
uals as the single mode superior to all other modes.
This fringe of extremist BRT promoters is actually
causing considerable harm by weakening the concept
of intermodal transit and by actually strengthening
the already dominant pro-automobile lobbies.
Finally, these “BRT evangelists” harm the cause of
BRT by promising that this mode can solve all trans-
port problems, so that its problems or failures in
different cities lead to questioning the value of the
entire BRT concept.
Many elements of BRT were introduced during the
1970s and 1980s in several cities, although the term
“BRT” had not yet been coined. São Paulo developed
several high-density corridors with high-frequency ser-
vice by buses and trolleybuses utilizing four-lane sta-
tions with overtaking of buses, platooning of buses
278 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
from different branches for increased capacity on the
joint trunk, etc. Pittsburgh, Washington (Shirley High-
way), and Los Angeles (El Monte) introduced busways,
although more for commuter services than for regular
transit. Ottawa built a network of busways and made
more integration of their services with land use plan-
ning around their stations. (The limitation of this sys-
tem has been that its busways lead into downtown
streets, which represent a capacity bottleneck for the
entire network.)
To clarify the strengths and limitations of the BRT
mode, several leading BRT projects are evaluated here.
Curitiba: Following innovations in bus system
design in various cities, Curitiba in Brazil built
a network of BRT lines which is still considered the
best planned one because it was designed as an inte-
grated part of the city’s land use plan. The BRT lines
represent the spines of major corridors with high-
density buildings throughout the city.
In addition to the planning aspects, Curitiba
became the leader in designing high-level platforms
and high-floor buses for level boarding/alighting
through wide doors on articulated and double-
articulated buses. Fare collection at the entrance to
specially designed stations in the median of the busway
was another element that increased line speed and
capacity. Curitiba is considered as a leader in the devel-
opment of the BRT concept.
Bogotá’s TransMilenio BRT System: This BRT system
has drawn great publicity and is called by some the
“Gold Standard of BRT.” It is therefore useful to exam-
ine its strengths and achievements, as well as its limi-
tations and shortcomings.
The city of Bogotá, capital of Colombia, suffered
from chronic congestion that was getting steadily worse
with its growth of population and car ownership. It was
served by remarkably primitive transit services
consisting mostly of 8- and 10-m long private bus
lines without any coordination, integration, or joint
fares among hundreds of companies and individuals
who owned and operated them. The city faced two
problems: how to organize transit services into
a coordinated system, and then how to build
a network of high-capacity semi-rapid (mostly inde-
pendent) or rapid (fully separated) transit lines.
Rail rapid transit was planned for a long time, but
problems in coordinating, organizing, and financing
the project stalled progress. Then, under the leadership
of Mayor Enrique Penalosa, a BRT system was built for
a major corridor, opening in December 2000. This was
followed by other lines, which now represent a network
of high-capacity lines.
The achievements of TransMilenio are quite
remarkable. The system represents a huge leap forward
over the chaotic, low-quality and low-capacity
disintegrated bus lines. These small-bus services still
carry the majority of trips in Bogotá and still represent
a serious problem because of their poor service quality
and inability to attract choice riders. The TransMilenio
system has successfully attracted a substantial portion
of trips and provided them with much faster and more
reliable services. Now, finally, the city with a population
of about seven million persons and growing has an
increasing network of fast lines largely independent
from street congestion, which also reduce the traffic
loads and congestion on existing streets.
Along with these significant achievements,
TransMilenio has several serious limitations, which
should be carefully considered in further transit plan-
ning for the city. They are:
● The success in attracting very large passenger vol-
umes is overwhelming the system. TransMilenio
operates at and beyond its capacity for long periods
of the day. That results in extremely low passenger
comfort and lower service reliability than the sys-
tem could provide with below-capacity passenger
volumes.
● TransMilenio busways are located in the median of
corridors with 8–12 traffic lanes that can be crossed
by pedestrians only via long overpasses provided at
distances of several hundred meters. This design
provides only a single access to every bus station,
so that walking distances from the closest buildings
and streets to a bus station entrance amount in
many cases to several hundred meters and involves
crossing ramps and bridges unprotected from the
weather. This design is very unattractive to users,
reduces the competitiveness of transit services for
car owners, and represents a major divide between
streets and buildings on its opposite sides. This is
the opposite impact on a city’s livability from the
impacts that LRT services have when they go
through pedestrian zones.
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects 279

● While electrically powered rail transits (LRT or attractive alternative. Thus, in the choice between
metro) provide quiet vehicles without exhaust and BRT and LRTor metro, buses represent a major step
long-term sustainability, buses with diesel or other backward with respect to a city’s livability and sus-
internal combustion engines represent a much less tainability (Figs. 13–16).

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 13
Back-up of buses at TransMilenio station reduces service reliability

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 14
Long overpasses are only access to many TransMilenio stations
280 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Bus Rapid Versus Light Rail Transit: Service Quality,
Economic, Environmental and Planning Aspects.
Figure 15
TransMilenio corridor is a major barrier to pedestrians in
Bogota
These limitations explain why there is so much
interest in building a metro in Bogotá. Low- and
medium-capacity BRT lines represent the optimal,
most economic solution when matched with similar
passenger demand. However, a BRT line operating
beyond its capacity requires a major investment in
a line that provides a higher quality of service, attracts
more choice riders, and enhances the city’s livability.
Bogotá is a city with a steadily increasing population
that brilliantly converted uncoordinated midi-bus
services into a coordinated network of regular buses
and BRT lines. It has used taken advantage of corridors
that allowed the four-lane busway, but now must
invest in rail transit to further improve its quality of
transit service, economic vitality, and sustainability.
The question is actually not whether Bogotá’s higher-
capacity transit system should be BRT or LRT/metro:
the city obviously needs both, rail for its most heavily
loaded lines, BRT serving a network of medium-
capacity lines.
Mexico City’s Insurgentes BRT: Similar to Bogotá,
the Insurgentes corridor was served by uncoordinated
minibuses with erratic, unreliable service. The first
20 km and 36 stations of the Insurgentes BRT facility
opened in 2005; in 2008 an additional 9 km opened. As
with the TransMilenio BRT system this line operates in
the median of a major arterial road requiring passen-
gers to access stops only through one entrance after
long walks. Buses stop one at a time and cross many
signalized intersections one at a time, so that the max-
imum capacity obtained using one articulated bus per
signal cycle amounts to 5,000–6,000 persons/h.
The BRT line was so much better than the mini-
buses that the line attracted 280,000 persons/day, which
exceeded its capacity and resulted in extreme, some-
times dangerous, overcrowding.
Also similar to TransMilenio, the Insurgentes line
was built faster and at lower investment cost than an
LRT line would have required. However, while its cur-
rent operation can be deemed a success based on its
daily ridership, its capacity limit was reached at the
time the line opened. An upgrade to rail technology
could allow double or triple the capacity still
maintaining only one transit unit (train) per signal
cycle through intersections. Only the street block
length would limit the potential train length. Safety
would also be increased with the decrease in the num-
ber of bus braking and accelerating cycles and with the
added signal protection on any segment where train
speed exceeds 70 km/h. LRT would have provided so
much greater capacity, resulting in correspondingly
greater riding comfort. Given the present ridership
levels, one could argue that conversion to a metro line
would even have merit.
The Bogotá and Mexico City examples show fairly
typical decisions many cities face as they ponder how to
improve their transit services. If there is an avenue in
which ROW B can be obtained, BRT usually requires
a lower investment and simpler/shorter implementa-
tion. However, LRT would provide a much higher
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 16
Exhaust pollution from TransMilenio operations
capacity, better riding comfort and passenger attrac-
tion over the long term. Moreover, with respect to the
city’s livability and sustainability LRT is clearly supe-
rior to diesel (or even CNG) buses, particularly in
sensitive inner city areas.
Istanbul Freeway Median BRT: Istanbul is by any
definition a megacity spanning a very large region. Due
to the long travel distances and extremely heavy con-
gestion, the traveling public wastes much time. Even
the limited access highways can be chaotic with thou-
sands of buses stopping along the roadside, thus resem-
bling ROW C much more than ROW A in character. In
2007, the first 18 km and 14 stations within one such
freeway opened as a BRT system (Phase I). Typically, its
stops and dedicated lanes were located in the middle of
the limited access highway. In 2008, an additional
29 km and 25 stations were added (Phase II). At that
time it was projected that an additional 11 km would be
built for a total of 50 km at a relatively modest total cost
of $125 M (Fig. 17).
Phases I and II already carry 858,000 passengers per
day, and many passengers are saving over 1 h per day in
commuting times. Thus, it would seem to be a major
success. However, this very success indicates that the
conditions are strong for an upgrade to rail technology.
There are close to 2,500 bus trips made per day on this
281
busway with headways often down to 20 or 30 s of
separation. Consequently, there is a significant amount
of air pollution and fuel consumption from the enor-
mous volume of buses. Furthermore, bus speeds and
passenger comfort are limited and safety is
compromised due to hundreds of thousands of accel-
erating and braking cycles per day at highway speeds.
The ready availability of the right-of-way and the
long station spacings due to their location at overpasses
would allow for the running of large high-capacity
trains that accelerate faster and obtain higher speeds
than do buses. The high demand and ROW A clearly
justifies rapid transit technology, not LRT technology.
Headways would still be short while time savings and
passenger comfort would be greatly increased. Safety
would be enhanced by automatic train protection.
Total energy consumption and both air and noise pol-
lution would be reduced. Operating cost savings would
be substantial and could be used to bolster other parts
of Istanbul’s network including more feeder services.
It is true that far fewer bus drivers would be needed,
but this decrease in employment would be far offset by
the increase in economic vitality and job opportunities
opened with the corridor’s faster travel times. (There
would also be some offsetting employment created to
maintain and repair the guideway and vehicles.)
282 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 17
BRT with ROW category A in Istanbul, Turkey
Comparison of LRT and BRT Modes
Light rail transit and bus rapid transit (but not Bus
Transit Systems) are functionally similar modes. As
semi-rapid transit (mostly independent of street traffic),
both provide much higher service performance and
have more positive impacts than the basic street transit
mode – regular (conventional) buses (RB). Both require
investments that are significantly higher than those for
RB, but much lower than for metro systems. However,
LRTand BRTare drastically different in their technology,
operations, quality of service, and their short- and long-
term impacts on the city. In most cases, they typically
provide different “investment cost/performance pack-
ages” so that their comparison is very complex.
Applications of the two modes vary between two
extremes. In medium-sized cities or corridors BRT can
fully meet the capacity and performance needs at lower
cost than LRT. At the other extreme, in many medium-
or large-size cities LRT offers far superior performance,
operates efficiently in pedestrian-oriented areas or in
short tunnels under city centers, and has many more
positive impacts for a city’s livability than does BRT.
Between these two poles, where BRT and LRT are,
respectively, obviously the superior choices, are many
applications with different trade-offs so that either one
of the two systems may be the preferred mode.
Review of Modal Characteristics
Table 1 presents a comparison of the basic characteris-
tics of RB, as the most widely used street transit mode,
BRT and LRT. The characteristics are grouped in three
categories: system components, lines and operational
components, and system characteristics.
As Table 1 shows, the elements in which BRT differs
from RB are ROW category, larger vehicle capacity if
double-articulated buses are used, different line designs
and stop spacings, higher investment costs, and its
greater passenger attraction.
Focusing on the BRT and LRT modes, their com-
parison can be summarized in a number of character-
istics, as presented in the following box [5].
Some of these comparisons are factual and obvious,
proven in real-world situations. Several of the major
differences will be discussed here.
Advantages (+) and Disadvantages ( ) of LRT
over BRT
+ Separate ROW (B or A) for LRT is easier to achieve
because LRT uses rail tracks instead of roadway lanes,
and due to its different technology requires no phys-
ical protection and police enforcement, as do
busways (Fig. 18).
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects 283

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Table 1
Comparison of regular bus, bus rapid transit, and light rail transit characteristics

Mode►
Characteristic▼ Regular bus(RB) Bus rapid transit (BRT) Light rail transit (LRT)
System components:
ROW C B (C, A) B, (A, C)
Support Road Road Rail
Guidance Steered Steered Guided
Propulsion ICE ICE Electric
TU control Driver/visual Driver/visual Driver/signal
Vehicle capacity (spaces) 80–120 80–180 100–250
Max TU size Single vehicle Single vehicle 1–4 car trains
Max TU capacity 120 180 4  180 = 720
Lines/operational elements:
Lines Many Few Few
Headways Long/medium/short Long/medium/short Medium/short
Stop spacings (meters) 80–250 200–400 250–600
Transfers Few Some/many Many
System characteristics:
Investment cost/km Low High Very high
Operating cost/sp Medium Medium Low
System image Variable Good Excellent
Pass. attraction Limited Good Strong
Impacts on land use and city Least Moderate Strongest
livability

+ LRT has better vehicle performance than BRT
because of its electric traction.
+ LRT produces no exhaust along the line and much
lower noise than BRT (except that in a few cities,
trains are required to use horns at grade crossings).
+ LRT is often designed to serve as the central element
for access and image of pedestrian areas in central
cities; a busway with high-frequency bus services is
much less compatible with “pedestrianized” areas.
+ LRT can use tunnels, BRT cannot.
+ LRT vehicles are more spacious and comfortable and
have better riding quality than buses.
+ LRT has a stronger image; it is more popular and
attracts more riders.
+ LRT has a stronger positive impact on urban devel-
opment than BRT.
Investment costs for LRT are higher than those for
BRT.
For the first LRT line in a city, introduction of a new
technology requires more extensive construction of
infrastructure as well as new equipment, and it
involves longer implementation.
Diesel Bus Technology Versus Electric Rail
Technology
The basic benefits and drawbacks of diesel and electric
traction, and rubber-tired and rail technology are
284 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 18
Recently opened LRT line in Dublin, Ireland on a ROW category A section

inherent in the BRT and LRT modes. All BRT systems
use the internal combustion engine for propulsion. In
the past the fuel was diesel. Recently, new BRT systems
in the industrialized world use less polluting liquid
natural gas (LNG) or compressed gas (CNG) as fuel
sources, but engine performance does not fundamen-
tally change.
More detailed analysis of BRT and LRT character-
istics caused by their different types of traction and
road/rail technology are given in the following two
boxes, respectively [6].
Several relevant points from the above comparisons
deserve some discussion here.
● LRT in most cases costs more to construct
initially. Although rail equipment and facilities
last longer, the required higher initial invest-
ment is often a major problem. An exception
to LRT costing more than BRT might be where
ridership demand requires a category A BRT to
avoid at-grade crossings, but LRT can provide
that capacity with a less expensive category
B design.
● LRT has much more capacity because rail technol-
ogy allows the coupling of railcars into trains; cou-
pling cannot be done with buses. This is particularly
important for transit projects using category
B rights-of-way. Traffic engineers limit the amount
of green time given to transit. If the expected rider-
ship demand on the line is high, the transit unit
used must carry the most passengers. In this regard,
LRT has the added advantage of requiring crossing
gates for safety.
● BRT has the potential to offer more service options,
not just the end-to-end operation typical for LRT.
A BRT line can offer end-to-end service, but can
supplement it with express bus services or multiple
route-to-trunk line services typical of freeway bus
operations.
● If capacity and service quality require that ROWA is
used, that is usually more complicated for buses
than rail vehicles and results in bus services which
are much inferior to rail. The best example is the
Silverline BRT in Boston, where a 1-mile long tun-
nel required an investment cost of about $500 mil-
lion, and buses now operate in it at speeds limited
to 10–25 km/h. Capacity of buses operating in sin-
gle-lane tunnels is only a fraction of the capacity
LRT would provide.
● Road and rail technologies also have a direct impact
on operations of their rights-of-way. Separation of
ROW for LRT needs no enforcement because in
most designs road vehicles cannot use tracks
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
because they are not paved. BRT’s advantage, that it
uses the same road surface as other street lanes, is
actually a serious disadvantage with respect to con-
trol of their use. Many bus lanes and busways have
been first downgraded by allowing other vehicles to
share them (first taxis and vehicles making turns at
intersections, then HOVs, any cars who pay tolls
in such lanes, and then even hybrid private cars).
In many cases these steps led to eventual losing of
ROW B for buses and discontinuation of the BRT
system (see Fig. 3).
Electric Rail Traction Compared with the Internal
Combustion Engines of Buses
+ Produces higher, smoother acceleration (limited by
passenger comfort)
+ Produces far less noise and vibration
+ Produces no local air pollution
+ Can easily use tunnels due to absence of pollution
and good dynamic performance
+ Allows for regenerative braking, thus reducing
energy consumption
+ Is much cleaner to operate and maintain
Requires additional investment for electrification of
lines and construction of substations
Overhead power lines are sometimes criticized
for aesthetic reasons, although many passengers
like the distinct identity the catenary gives
the line
Is susceptible to power failure that stops all vehicles
on a section of line
Rail Technology Compared with Rubber-Tired Buses
+ Makes coupling of cars into trains possible, drastically
increasing line capacity
+ Rail guidance allows use of wider and longer cars and
requires narrower free profile
+ Has much higher safety because of automatic signal
control, although at additional investment and main-
tenance costs
+ Provides a far more stable, smoother ride quality
285
+ Is far more energy-efficient due to steel wheels on
steel rails
+ Can better cope with adverse rain, ice, and snow
conditions
Requires substantially higher investment, although
the vehicles and guideways are more durable and
require lower ongoing maintenance
Cannot negotiate as steep a gradient, nor as sharp
curves as buses
Requires a higher degree of safety because of longer
stopping distances
Can be much louder when going through sharp
curves
Is normally slower around any given radius of curva-
ture and may therefore have a slower travel time
between stations than bus technology
Line Capacities
An extensive debate has been going on with respect to
the line capacity of different transit modes. To check
the validity of various claims of modal capacities, two
values will be quoted for each mode: the first, for the
bus mode, assumes regular buses with 90-space capac-
ity (6 persons/m2) and with single stops with fast
boarding/alighting, allowing average headways as
short as 1 min. Such a system provides a line capacity
of 5,400 spaces/h.
The second value for bus line capacity assumes that
each stop has four lanes, allowing overtaking and “leap-
frogging” of buses. In this way, capacity can be
increased significantly. This is seldom possible in
urban areas, but some examples exist. Portland, Ore-
gon, organized a pair of bus streets with two lanes for
buses and overtaking of buses in stop areas, which
achieved capacity of 180 buses/h or average headways
of only 20 s. With an average capacity between regular
and articulated buses of 110 spaces, this results in a line
capacity of 19,800 spaces/h.
The Comonor project, a high-capacity bus corridor
in São Paulo, Brazil, claimed to have achieved capacity
of 24,000 spaces/h by operating buses in platoons over-
taking each other at stops along the line. However, that
description turned out to be theoretical, and actual
286 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
capacity provided on São Paulo streets was consider-
ably lower.
New claims of much higher capacities for new BRT
systems came particularly with the opening of the
TransMilenio system in Bogotá, Colombia. That sys-
tem was systematically designed for high capacity: most
buses are articulated and double-articulated, all fares
are collected at station entrances, and the system has
two lanes on some sections, but four lanes in station
areas allowing bypassing and overtaking of buses. This
is particularly the case on the main line built in a wide
median of a very wide freeway right-of-way. However,
the line has a number of signal-controlled intersec-
tions, which limit the green time for the BRT line.
Even if it is assumed that all buses are double-
articulated and occupied with an extremely high aver-
age density of 6 persons/m2, and that it is possible to
have an average headway of buses crossing signalized
intersections of 20 s, such an operation would amount
to about 28,000 spaces/h. Although these assumptions
are higher than in real world (for one, not all buses are
double-articulated), advocates of BRT systems have
claimed capacities exceeding 30,000, later 35,000 and
even as high as 53,000 persons/h. The latest revisions to
a couple of prominent BRT design manuals written for
applications to developing countries appear to be
correcting these overblown numbers downward,
toward lower, although still unrealistic values of over
30,000 persons/h.
For comparison, the capacity of LRT lines operated
with trains consisting of two articulated cars (such as in
Manila) with realistically achievable 2-min headways is
about 15,000 spaces/h. With more optimistic assump-
tions, corresponding to those used by BRT promoters,
LRT lines using four-articulated car trains as operated
in several cities and 1.5 min headways would achieve an
offered capacity of 32,000 spaces/h. With four tracks
for bypassing in stations, this capacity could be
increased further, but the critical locations would be
at intersections. As the measurements of actual opera-
tions of the Orange Line BRT (Los Angeles) show, LRT
can provide substantially higher capacity at street
crossings than buses because of the higher capacity of
LRT trains compared to individual buses, even double-
articulated ones.
Under similar assumptions, ten-car trains of
a metro system operating at 1.5-min headways would
provide a capacity in the order of 60,000 spaces/h,
while the four-track metro line in New York carried
89,000 persons/h in the 1940s (presently, passenger
numbers are smaller, because trains in most of the
industrialized countries are operated with higher
comfort levels which result in 4–5 persons/m2).
Another deceptive feature of these claims of BRT
capacities is that they represent the obsolete, simplistic
measure of line capacity as a single number, without
regard to service quality and level of service. With
respect to highway capacity, it was already in 1950 real-
ized that capacity in terms of vehicles per hour should be
related to the level of service (LOS), which was defined as
LOS A through F. Capacity of transit lines is now also
related to performance and LOS. For example, if one
transit line carries 14,000 persons/h with an operating
speed of 16 km/h and another line carries 14,000 with
a speed of 24 km/h, their performance is by no means
the same. Lehner [7] introduced the concept of pro-
ductive capacity – product of transporting capacity and
speed – to include this characteristic. It can be insight-
ful to plot the productive capacity against investment
cost. Vuchic [8] specified that evaluation of transit
capacity should include the following five factors:
● Offered line capacity in spaces/h
● Operating speed in km/h
● Load factor or capacity in persons/space, where
spaces include both seated and standing spaces
assuming a certain degree of crowding
● Comfort standard in square meters/person (or
degree of crowding in persons/m2)
● Reliability of service in percent of departures on
time
Consequently, capacity of transit modes cannot be
evaluated only by hourly throughput of persons; it
must include level of service.
Comparisons of capacities of different modes must
also be made for comparable conditions and assump-
tions. For example, comparing capacities of BRT and
LRT must assume that both modes have either two
lanes/tracks, or four lanes/tracks in stations. Thus, the
Insurgentes BRT line in Mexico City should be com-
pared with a two-track LRT line, such as in Cologne or
Manila. The TransMilenio BRT line in Bogotá should
be compared with a four-track LRT line, which would
have capacity well over 40,000 spaces/h.
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects 287

Compared modes must also have the same load
factors and comfort standards. Reliability and operat-
ing speed would then have to be measured or predicted
on the basis of characteristics of each mode.
Recent Studies Comparing Different Modes of Real-
World Transit Systems
Studies comparing transit modes generally fall into two
categories. The first compares modes that are being
planned or hypothesized, that is, not in operation.
These types of studies are needed, of course, when
doing major planning work, but they must use pro-
jections of ridership and costs rather than actual rider-
ship and costs. The alignments and operational
assumptions made for each alternative should be as
similar as the characteristics of different modes allow.
The second category of studies compares modes
operating in one or more cities. A major problem
with these types of studies is that each system operates
in a different cultural, economic, and operational con-
text, but their great advantage is that their features and
numerical values are realistic, not hypothetical and
subjective.
One recent study that looked at one city with sev-
eral bus and rail modes operated by a single agency is
described below.
Case Study: Comparison of Transit Modes in Los
Angeles Over the past 25 years Los Angeles’ metro
transit network has evolved from an all-bus system to
a network of many transit services and modes. As of
spring 2008, this network included a rapid transit line,
three light rail lines, three busways, 19 enhanced bus
transit routes, and 95 local bus routes. For the first time
the opportunity exists to compare a variety of transit
modes all operated by the same agency within the same
urbanized area eliminating differences in labor costs
between urban areas, in operating and management
practices, in efficiencies of scale, and in time and
methods of data collection [9].
Various performance and cost measures were cal-
culated for each mode operated by the Los
Angeles metro. The results of this analysis are shown
in Table 2 for performance measures, in Table 3 for cost
measures.
The Orange (BRT) Line is much slower than the
average of the LRT lines. This is because it does not

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Table 2
Performance measures of Los Angeles metro transit modes (2008)
Performance measure Local bus Rapid bus Orange Line BRT Light rail Rapid transit
Ave. peak hour speed (km/h) 20.6 24.0 29.3 41.4 51.8
Ave. trip length (km) 5.7 7.3 9.4 11.3 8.0
Daily riders: (experienced) (possible max) 30,000 30,000 25,000 84,000 140,000
30,000 30,000 35,000 100,000 300,000

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Table 3 Cost
measures of Los Angeles metro transit modes (2008$ unless noted)
Cost measures Local bus Rapid bus Orange Line Light rail Rapid transit
Capital cost ($M/km) N/A $ 0.15 (2000) $17.3 (2005) $ 38.9 (2003) $130 (2000)
Operating cost per pass-km $0.60 $0.39 $0.30 $0.31 $0.27
Operating subsidy per pass-km $0.47 $0.29 $0.22 $0.25 $0.20
288 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects

have full signal priority and catches red lights at a third Summary Comparison of BRT and LRT Modes
of its intersections. Its buses must also slow below
Table 4 gives a summary comparison between the BTS,
32 km/h (20 mph) at all crossings. It also has nowhere
BRT, and LRT modes. Overall, the two bus modes have
near the carrying capacity of the light rail lines and is
advantages with respect to investment cost, construc-
nearing capacity with about 25,000 daily riders. One
tion time, and the complexity of system implementa-
LRT Line (Blue) has had a weekday ridership above
tion. However, LRT demonstrates superiority in most
80,000 for a number of years.
measures of service efficiency and level of service:
On the other hand, the Orange Line was built for
speed, comfort, image, and passenger attraction. Even
less than half of the capital cost per kilometer of the
more important may be that LRT is superior in its
Gold (LRT) Line during the same period. Time will tell
positive impacts on land use planning, urban form,
whether that investment was prudent given the much
livability, and sustainability. Focusing on the BRT-LRT
slower speed and the looming capacity limit of the
comparison, BRT tends to be a faster and usually (but
busway. The Orange Line may be an example of
not always) lower investment solution, but LRT pro-
a BRT line that should have been built fully grade-
vides higher service quality, stronger passenger attrac-
separated for higher speed and capacity, but if it had
tion, and a more powerful step in increasing the city’s
been, its cost might have exceeded the cost of an equiv-
quality of life and sustainability.
alent light rail line operating on the existing ROW B.

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Table 4
Summary comparisons of the BTS, BRT, and LRT modes
Superior
Mode characteristic Bus transit system BTS Bus rapid transit BRT Light rail transit LRT mode
Investment cost Medium High Very high BTS
Implementation Complexity Short Medium Long BTS
and time
Operating cost Lower for low Lower for low Lower for high Depends
passenger volumes passenger volumes passenger volumes
Operating speed Medium High High Depends
Ability to accommodate Low Some with four-lane Low except with four- BRT
service options stops track stops
Capacity Low Medium High LRT
Type of energy and traction Internal comb. engine Internal comb. engine Electric LRT
Vehicle performance Good Good Excellent LRT
Air pollution and noise Poor Poor No local pollution, low LRT
noise
System image and passenger Fair Good Excellent LRT
attraction
Potential to influence land Limited Fair Very good LRT
development
Contribution to livable urban Some Limited Excellent LRT
environment
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects 289

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 19
Five-section low-floor LRT vehicle provides high capacity and can make sharp turns on ROW category B in urban streets in
Helsinki, Finland

Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects. Figure 20
Buses and light rail trains share ROW category B in Helsinki

Trends and Expected Future Directions the inevitable, debilitating congestion that these new
cars bring is to invest in attractive urban transit systems
City populations in many developing countries are
capable of meeting future demands.
growing rapidly. Those countries that also have grow-
For large cities with many million inhabitants, high
ing economies and rising income levels are seeing omi-
population densities, and heavily traveled corridors,
nous growth in auto ownership. One defense against
290 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
rail rapid transit (metro) is clearly the most efficient
transit mode for the basic network. Megacities like
Tokyo, Beijing, and São Paulo will have to extend
their metros as the population continues to grow.
The trend of developing medium-capacity modes,
which started with the growth of LRT and later BRT,
will continue. These modes are being introduced to
supplement metros in suburban areas and as transi-
tional solutions when metros cannot be built fast
enough for financial or physical reasons.
The two modes most capable of meeting the need in
many cities in developing countries are LRT and true
BRT. Both, however, are costly investments. One future
direction is therefore to convey to transit policymakers
that the investment decision they face is between these
two modes, not the false promise of inexpensive, inad-
equate lower-cost bus modes.
In developed countries, a future direction for study
is to appraise honestly the cost and performance of true
BRT systems and their more comprehensive service
options. In some applications the LRT model of
a high-capacity rail line fed through transfers from
connecting bus routes may not provide the most attrac-
tive, convenient service. In other cases, if a BRT would
be loaded to capacity as soon as it is opened (which is
the case with some lines in Bogotá and Insurgentes in
Mexico), a rail system – LRT or metro – would obvi-
ously be a more efficient and permanent solution.
Understanding where each of these options works
best will be quite useful.
Finally, the damaging argument within the transit
community between BRT and LRT must subside if
transit is to compete most aggressively with the private
car. In a world that needs to have far fewer private cars
depleting our oil reserves, strangling our cities, and
fouling our air, the focus must be to provide the most
attractive, cost-effective transit, bus, or rail for a given
circumstance. It should not be to have one mode to the
exclusion of another. A good example for this is
Bogotá, where the debate should not be BRT versus
LRT, but BRT and a rail system (LRT or metro). This is
particularly the case with lines where volumes are very
high and environmental aspects are very important. In
such cases the transit system must enhance the livabil-
ity of the city, minimize its air and noise pollution,
and join rather than divide its neighborhoods
(Figs. 19 and 20).
Bibliography
Primary Literature
1. Eccles KA (2007) Design, operation, and safety of at-grade
crossings of exclusive busways. TCRP Report 117. TRB,
National Research Council, Washington, DC, pp 19–20
2. Stanger R (2007) An evaluation of Los Angeles’s orange line
busway. J Public Transp 10(1):103–119 (Center for Urban
Transportation Research, Tampa)
3. Kittelson & Associates (2007) Bus rapid transit practitioner’s
guide. TCRP Report 118. TRB, National Research Council,
Washington, DC
4. Hoffman A (2008) Advanced network planning for bus rapid
transit: the “Quickway Model” as a modal alternative to “Light
Rail Lite”. Department of Transportation, Federal Transit
Administration, Washington, DC
5. Vuchic VR (2007) Urban transit systems and technologies.
Wiley, Hoboken, p 541
6. Vuchic VR (2007) Urban transit systems and technologies.
Wiley, Hoboken, p 116
7. Lehner F (1978) Der maximale Wirkungsgrad des
Personaleinsatzes; (Maximum efficiency of personnel assign-
ment). Alba Buchverlag, Dusseldorf
8. Vuchic VR (2005) Urban transit operations, planning, and eco-
nomics. Wiley, Hoboken, p 88
9. Stanger R (2010) Comparative performance of Los Angeles’
transit modes. Presented at the 11th joint light rail conference,
9 Apr 2009. Transportation Research Board and American
Public Transportation Association, Los Angeles, Transporta-
tion Research Board Circular 143
Books and Reviews
Bruun E (2005) BRT versus LRT operating costs: comparisons using
a parametric cost model. Transportation Research Record No.
1927. Transportation Research Board of the National Acade-
mies, Washington, DC, pp 10–20
Grava S (2003) Urban transportation systems: choices for commu-
nities. McGraw Hill, New York
Hess D, Taylor B, Yoh A (2005) Light rail lite or cost-effective
improvements to bus service?: evaluating costs of
implementing bus rapid transit. In: Transportation Research
Record: Journal of the Transportation Research Board, No.
1927. Transportation Research Board of the National Acade-
mies, Washington, DC, pp 22–30
Kittelson & Associates (2007) Bus rapid transit practitioner’s
guide. TCRP Report 118. TRB, National Research Council,
Washington, DC
Koonce P, Ringert J, Urbanik T, Rotich W, Kloos B (2002) Detector
range setting methodology for signal priority. J Public Transp
5(2):115–135 (Special edition on BRT, Center for Urban Trans-
portation Research, Tampa)
Kühn F (2002) Godard & Fatorzoun (eds) Urban mobility for all – La
Mobilité Urbane pour Tous. Swets & Zeitlinger, Lisse, pp 357–365
 Bus Rapid Versus Light Rail Transit: Service Quality, Economic, Environmental and Planning Aspects
Levinson H, Zimmerman S, Clinger J, Rutherford S, Smith R,
Cracknell J, Soberman R (2003) Bus rapid transit, volume 1:
case studies in bus rapid transit. TCRP Report 90. TRB, National
Research Council, Washington, DC
Levinson H, Zimmerman S, Clinger J, Gast J, Rutherford S, Bruun
E (2003) Bus rapid transit, volume 2: implementation
guidelines. TCRP Report 90. TRB, National Research Council,
Washington, DC
Newman P, Kenworthy J (1999) Sustainability and cities. Island
Press, Washington, DC
Sislak KG (2000) Bus rapid transit as a substitute for light rail transit:
a tale of two cities (CD-ROM). Presented at the eighth joint
conference on light rail transit, Dallas
Stutsman J (2002) Bus rapid transit or light rail transit – how to
decide? Los Angeles Case Study. In: Transportation Research
291
Record: Journal of the Transportation Research Board, No.
1973. Transportation Research Board of the National Acade-
mies, Washington, DC, pp 55–61
TCRP (Transit Cooperative Research Program) (2005) Innovations
in bus, rail and specialized transit. Research Results Digest 70.
Transit Cooperative Research Program, National Academy
Press, Washington, DC
Thompson GL, Matoff TG (2003) Keeping up with the Joneses:
planning for transit in decentralizing regions. J Am Plann
Assoc 69(3):296–312
Vuchic Vukan (2005) Planning and selection of medium and high
performance transit modes. In: Urban transit: operations, plan-
ning and economics. Wiley, Hoboken (Chap. 12)
292 Bus Versus Rail Implications for Transit-Oriented Development
Bus Versus Rail Implications for
Transit-Oriented Development
MING ZHANG
Graduate Program in Community and Regional
Planning, School of Architecture, The University of
Texas at Austin, Austin, TX, USA
Article Outline
Glossary
Definition and Importance
Introduction
Transit Technologies
Cost Characteristics of Bus Versus Rail Mass Transit
Capacity of Bus Versus Rail Transit
Transit Impacts on Property Values
Density for Bus and Rail Transit
Best Practice of Bus- Versus Rail-Based TOD
Conclusions
Future Directions
Bibliography
Glossary
BRT Bus rapid transit (BRT) refers to a variety of
mass transit systems that utilize buses to provide
faster, more efficient services than a conventional
bus line. BRT runs on existing roadways or
dedicated rights-of-way and offers such operational
features as high-capacity, low-floor bus vehicles,
traffic signal prioritization, and real-time
information services.
LRT Light rail transit (LRT) refers to electrically
propelled rail vehicles operating singly or in trains.
It obtains power from overhead cables and runs on
reserved but not necessarily grade-separated
rights-of-way. LRT provides service capacities and
speeds typically lower than metro rail.
Metro rail Metro rail typically consists of steel-
wheeled, electric-powered vehicles operating in
trains of two or more cars on a fully grade-separated
right-of-way. Alternative names for metro rail
include subway and heavy rail. It offers higher
capacities and speeds than light rail.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
New urbanism New urbanism refers to an urban
design movement led by a group of architects and
urban planners in the USA since the early
1980s. New urbanism promotes walkable neighbor-
hoods and attacks suburban sprawl through
nonconventional physical planning and design
approaches. Traditional neighborhood
design (TND) and transit-oriented
development (TOD) are two examples of new
urbanist practice.
TOD Transit-oriented development (TOD) refers to
mixed-use, medium- to high-density development
around the transit station with pedestrian- and
cyclist-friendly environmental design. The term
was coined by Architect Peter Calthrope while
other terms, for example, Transit-Focused Design
and Transit Villages, have also been used to describe
similar development concepts.
Transit ridership The quantity of passengers riding
on the transit over a given period of time, for
instance, a day or a year.
Transit value capture Transit value capture refers to
the process by which all or a portion of increments
in land value attributed to public transit
investments are recaptured by the public sector.
The idea can be traced back to the thinking of
Henry George and his followers. The rationale
states that transportation projects improve accessi-
bility to the adjacent land. This improved
accessibility is capitalized in property values,
generating a windfall for private landowners. Public
agencies can capture a portion of that windfall by
utilizing a variety of methods. The captured value
can then be used to finance transportation or other
public infrastructure.
VMT Vehicle miles traveled (VMT) measures the
distance in miles that vehicles are driven over a given
period of time, for instance, a day or a year.
Definition and Importance
Transit-oriented development (TOD) refers to mixed-
use, medium- to high-density development around the
transit station with pedestrian- and cyclist-friendly
environmental design oriented to transit services.
Integrating transit with land use has been a common
# 2012, DOI 10.1007/978-1-4419-0851-3

practice in most European, Asian, and Latin American
cities. It is not a brand-new idea in the USA either.
In the streetcar era, for example, developments largely
concentrated along transit corridors in Boston,
Chicago, San Francisco, and many other streetcar
communities. TOD began to regain popularity in the
early 1990s when the US Architect Peter Calthorpe
coined the term in his book The Next American
Metropolis: Ecology, Community, and the American
Dream [1]. Other terms have been used by transporta-
tion professionals, urban planners, and the real estate
industry to express similar development concepts,
for example, “transit-sensitive land use” [2], “transit-
oriented design” [3], “transit-focused development”
[4], and “transit villages” [5]. TOD represents these
multi-professional interests in developing around
transit. Amid a worldwide trend of rising motorization
and growing vehicle miles traveled (VMT), TOD is
being widely promoted as a strategy to achieve sustain-
able transportation. Currently, there are over 4,000
transit sites in the USA offering significant potential
to practice TOD [6].
At the center of TOD is transit. There are a variety of
transit technologies; broadly they can be grouped as
bus-based or rail-based. Bus- versus rail-based transit
presents distinctive operational and service character-
istics and involves quite different levels of public invest-
ments. Accordingly, bus-TOD versus rail-TOD entails
different planning and design considerations. Whether
to develop bus- or rail-based mass transit has long been
a topic of policy debate. Recent developments in bus
rapid transit (BRT) and light rail transit (LRT)
throughout the world add to existing heated
discussions on the advantages of bus- versus rail-based
transit systems. This entry describes the comparative
characteristics of bus- versus rail-based transit and
discusses their implications for TOD.
Introduction
Transit-oriented development (TOD) is a strategy to
integrate transportation with land use by focusing
development on transit. A number of factors have
motivated many communities in the USA to pursue
TOD. Fighting sprawl and sprawl-induced problems is
among the top. Community development in the USA
after World War II can be largely characterized as
Bus Versus Rail Implications for Transit-Oriented Development 293
low-density, strictly separated land uses, and
car-oriented design. In a sprawling built environment,
transit becomes operationally unviable and driving
becomes a necessity. The consequences are ever-
growing vehicle miles of travel (VMT) and increasing
roadway congestion, energy consumption, and vehicle
emissions. TOD offers potentials to curb sprawl and to
help reduce the driving-related problems. Many state
and local governments are encouraged to implement
TOD by the recent steady increase in federal funding
for transit. For example, the three transportation bills
in the less than 20 years, namely, ISTEA to TEA21 to
SAFETEA-LU, authorized increase in transit funding
from $25 to $36 to $52.6 billions, respectively.
TOD is being widely advocated in Europe, Asia, and
Latin America as well although their cities have tradi-
tionally developed in relatively high densities and rich
mixed uses. The TOD interest grows out of the concern
that new developments in many of these cities tend to
depart from the traditional urban form. Amid rising
income and associated growth in private motorization,
neighborhoods and workplaces are becoming
increasingly car oriented and pedestrian/cyclist hostile.
Densities in these cities are high relative to the
North American cities; yet the high density may
not function well if it is not fully integrated with the
transit. Dysfunctional density likely leads to what the
New Urbanist Andres Duany called “high-density
sprawl” [7]. Careful TOD planning, design, and
implementation are essential for transit and its
surrounding built environment in order to benefit
from density.
A TOD typically has five attributes: transit access,
a district of walking distance from the transit, mixture
of different compatible urban functions, medium to
high development density, and pedestrian- and
cyclist-friendly environmental design in the TOD
district. These attributes are interrelated and their
composition for a specific TOD will vary by the spatial,
socioeconomic, and technological context in which the
TOD is proposed. TOD planners have created TOD
typologies, specifying variations in design attributes
for TODs in different spatial context, for instance,
downtowns, urban neighborhoods, employment
centers, and other regional locations [8]. As transit is
at the center of TOD, TOD attributes will also vary by
transit technologies.
294 Bus Versus Rail Implications for Transit-Oriented Development
Transit Technologies
Transit technologies evolve along with advances in
vehicle technologies, information and communica-
tions technologies, as well as power and energy
technologies. Table 1 shows the most commonly seen
urban transit modes grouped into two categories:
bus-based and rail-based. A comprehensive list and
characterization of urban transit can be found in
Vuchic [9].
TOD attributes and the ways in which these
attributes interact will differ depending on the specific
transit that the TOD is proposed for.
● A particular type of transit technology is associated
with certain levels of cost. In general, the cost
consists of two parts: capital and operation. Capital
costs include vehicle cost and infrastructure cost
(land acquisition and clearance and system
complexity). Operating cost includes labor,
insurance, fuel, administration, and others.
● Different transit technologies and operating
plans deliver various levels of service
capacities. Main factors affecting service capacities
include vehicle size, service frequency, and road
conditions.
● Carrying capacity, along with other operational and
contextual factors, determines the level of services
(LOS). Existing empirical studies have shown that
the extent of land-use impacts of transit
development is positively associated with LOS. In
other words, higher capacity systems such as metro
likely have stronger positive impacts on the value of
land near the station.
● More expensive land leads to higher development
intensity, which will result in higher population or
employment density.
● Higher population and job density provides a larger
pool of transit ridership. If there are more people
riding on the transit, farebox revenue will increase.
Higher density also provides a larger basis for
economic activities and a larger tax base for the
local community. The farebox and tax income help
reduce cost burden of building and operating transit
systems. Ridership increases, however, may
trigger a greater need for expanded services, which
can lead to increases in operating and potentially
capital costs.
Cost Characteristics of Bus Versus Rail Mass
Transit
The main topic of debate on bus versus rail transit
development is cost. Recent discussions largely focus
on cost comparisons between BRT and LRT due to
a number of performance similarities between them.
On a per user basis, rail transit systems generally cost
more to build than bus transit but less to operate owing
to relatively higher capacities associated with rail. Still,
within each type of transit, there are large variations in
technical specifics and operational characteristics.
Capital Cost Comparison Between Bus and Rail
Transit
Table 2 shows the average capital cost per route mile for
BRT, LRT, and MRT based on selected systems
reviewed by USDOT [10], GAO [11], and BAH [12].
On average, BRT costs $10.24 millions (in 1990 dollars)
per mile to build, less than half of LRT ($26.4 million)
and less than one-tenth of MRT ($128.2 million).
Nevertheless, average cost–based comparisons between
bus and rail will likely generate controversies [13].
Table 2 also shows the ranges of capital costs of the
transit systems studied. Notably, within same type of
transit technology, capital costs vary considerably.
For instance, the BRT in Miami, FL costs $5.6 million
per mile to build, whereas the West Busway BRT in
Pittsburgh, PA costs $41.7 per mile, higher than the
average capital cost of LRT. The most expensive LRT
among the 21 projects reviewed by USDOT [2] is in
Buffalo, NY, which costs $90.19 per mile, over 40%
higher than the per-mile capital cost of the metro in
Miami.
Many factors contribute to the large variations
in the capital costs of transit systems. Costs of BRT
projects, for example, include vehicles (regular or
articulated), the roadway cost (busways or bus lanes),
station structures, park-and-ride facilities, communi-
cations, and traffic signal systems. For rail transit
systems, guideway elements may take a large share of
capital costs, depending on types of structure (e.g.,
tunnel or elevated). Different ways of accounting the
costs by including (or excluding) various capital cost
elements will change the average cost figures quite
significantly. Hence, when comparing different types
of transit systems with respect to capital costs, it is
 Bus Versus Rail Implications for Transit-Oriented Development 295

Bus Versus Rail Implications for Transit-Oriented Development. Table 1 Transit technologies and descriptions
Typical
System technology and Service Average Station Typical power
description Example systems geography speed spacing headway Guideway source
Bus-based
Regular bus: A road Cities with transit Urban 15–30 km/h 0.2–1.0 km 8–20 On street Gasoline
vehicle designed to carry services
multiple passengers.
Buses vary in capacity
from a dozen to several
hundred passengers.
Trolley bus New Orleans Urban 15–30 km/h 0.2–1.0 km 8–15 On street Electric
A passenger bus
operating on tires and
having an electric motor
that draws power from
overhead wires.
Bus rapid transit Pittsburg, Curitiba, Urban, 25–50 0.4–1.5 10–20 Shared or Gasoline
BRT is a relatively new Bogota, Beijing Regional exclusive
umbrella term for urban ROW
mass transportation
services utilizing buses to
perform premium
services on existing
roadways or dedicated
rights-of-way.
Rail-based
Streetcar Portland Urban 15–25 0.4 8–15 On street Electric
Bus on rails typically
operating on city streets
Light rail Dallas: DART; Urban, 30–60 1–1.5 5–30 Exclusive Electric
With an overhead power Denver: Regional or shared
supply and LRT utilizes ROW
predominantly reserved
but not necessarily
grade-separated rights-
of-way. Electrically
propelled rail vehicles
operate singly or in trains.
LRT provides a wide
range of passenger
capabilities and
performance
characteristics at
moderate costs.
Heavy rail Washington, D.C. Urban, 80–130 km/h 1–3 3–10 Grade- Electric
Metro; San Regional separated,
Francisco: BART; Exclusive
New York City: ROW
MTA Boston: MBTA
Chicago: CTA
296 Bus Versus Rail Implications for Transit-Oriented Development
Bus Versus Rail Implications for Transit-Oriented Development. Table 1 (Continued)
System technology and Service Average
description Example systems geography speed
Commuter rail Boston Purple Regional 50–120 km/h
Lines: MBTA; New
York Long Island
Railroad San Jose:
CalTrain
Bus Versus Rail Implications for Transit-Oriented Development. Table 2 Capital cost comparisons of bus versus rail
(in 1990 US$)
BRT
Average cost per route mile (millions) $10.2
Cost range (millions) $5.6  $41.7
Number of projects reviewed 9 21
Sources: Calculated from [10–12]
important to maintain consistency in cost reporting
and to examine/compare specific cost elements
whenever data is available. Such itemized comparisons
allow clear identification of specific factors driving up
the system development cost.
Table 3 shows the breakdown of capital cost by
subsystems for the three types of rapid transit technol-
ogies. Data for LRT and MRT come from a USDOT
study of nine completed projects [2], whereas cost data
for BRT are for two proposed systems in Atlanta,
GA and Abuquerque, NM. Note that, although BRT
costs less on average than LRT ($16.6 and $19.5 million
per mile, respectively), for several items BRT costs
exceed those of LRT. For instance, the average land
acquisition cost (right-of-way) for the BRT projects
($3.018 million per mile) is almost twice as much as
that of LRT project ($1.52 million per mile). This is
partly because BRT requires a wider right-of-way than
LRT. BRT guideway cost ($6.495 million per mile) is
much higher than that of LRT ($4.289 million per
mile) due to construction of dedicated busways.
Station cost for BRT is also slightly higher than that
of LRT. In most cases, a BRT station is located in
highway median areas and requires elevated structure
Typical
Station Typical power
spacing headway Guideway source
3–10 km 15–30 min. Exclusive Diesel or
ROW hybrid
LRT MRT
$26.4 $128.2
$9.4  $90.2 $63.9  $169.6
4
for passengers to access the station from both sides of
the road. Capital cost for conventional bus is not
reported here because it runs mostly on existing road-
ways free-of-charge and does not require guideways.
Operating Cost Comparison Between Bus and Rail
Transit
Table 4 below compares operating costs between the
bus and the rail transit modes. Cost per vehicle revenue
mile measures the average cost of operating individual
vehicles. It is the annual cost of operating a vehicle
divided by the total annual number of miles traveled
while the vehicle is in revenue service (e.g., excluding
the miles traveled to/from the maintenance yard).
Among the four transit modes listed in the table, LRT
exhibits the highest cost at $9.3 per vehicle revenue
mile. MRT has a relatively lower cost than LRT mainly
because it can operate with as long as ten or more cars
per train. Standard bus and BRT have the lowest
vehicle-mile costs. Operating cost per vehicle hour is
another way of measuring the cost of transit operation
regardless of the number of passengers carried.
MRT, which typically operates at a higher speed than
other transit modes, displays the highest vehicle-hour
 Bus Versus Rail Implications for Transit-Oriented Development 297

Bus Versus Rail Implications for Transit-Oriented Development. Table 3 Per mile capital cost breakdown of bus versus
rail (in 1990 US$1,000s)

BRT LRT MRT

$ % $ % $ %
Land (right-of-way) 3,018 15.0 1,520 7.8 7,436 5.8
Guideway 6,495 33.5 4,289 22.0 33,333 26.0
Trackwork or special conditions 983 8.5 1,686 8.7 5,385 4.2
Stations 1,387 9.1 1,094 5.6 33,333 26.0
Power and control 113 2.7 2,047 10.5 10,513 8.2
Facilities (yards and shops) 192 1.9 974 5.0 2,820 2.2
Eng./Mgt./Test (soft costs) 2,948 20.3 5,581 28.6 19,230 15.0
Vehicles 1,483 9.0 2,295 11.8 16,154 12.6
Total 16,620 100 19,486 100 128,203 100
Sources: Calculated from [10–12]

Bus Versus Rail Implications for Transit-Oriented Development. Table 4 Operating costs (1990$)
BUS BRT LRT MRT
$ per vehicle revenue mile 3.1 3.6 9.3 6.5
$ per vehicle revenue hour 45.0 78.8 125.0 152.0
$ per 1,000 place mile 60.0 72.3 96.0 49.2
$ per 1,000 passenger mile 616.4 496.9 578.0 282.0
Sources: Calculated from [10, 11]

cost. BUS and BRT once again rank the lowest in
operating cost measured on the per vehicle revenue
hour basis.
The other two measures of operating costs – cost
per thousand place mile (where places account for total
number of seats and permitted standings) and cost per
thousand passenger mile – take into consideration the
differences in vehicle capacity and system usage among
the transit modes. Notably, MRT has the lowest cost
at $49.2 per thousand place mile, whereas LRT’s
place-mile cost is the highest at $96. If usage is consid-
ered, BUS has the highest cost at $616.4 per thousand
passenger mile. Again, MRT is the least expensive at
$282 per thousand passenger mile. It should be noted
that figures for BRT in the table above are derived from
data in the GAO study [3]. The study does not report
cost per passenger mile. The figure of BRT’s $496.9 per
thousand passenger mile was estimated based on the
assumption that average trip distance of BRT
passengers was half of the BRT route length.
Overall, figures in the table above indicate that BRT
outperforms LRT in all of four operating cost catego-
ries. Nevertheless, cautions should be exercised when
interpreting these averages. Tennyson [13] has chal-
lenged that GAO’s study was unevenly biased against
LRT – the bus costs were reported too low due to
possible inconsistency and errors in the reported data.
As with the capital costs reviewed earlier, a wide range
of figures exist in each of the four operating costs. The
use of alternative measures may give different results on
298 Bus Versus Rail Implications for Transit-Oriented Development

(Constanr 2000 Dollars)

$8.00 7.5 7.6
7.2 7.2 7.2
6.9
$7.00 6.7 6.5 6.7
6.4
$6.00

$5.00

$4.00

$3.00 2.4 2.5 2.5 2.5 2.6

2.1 2.2 2.2 2.3 2.2 2.3 2.3 2.3 2.2 2.3
2.1
1.9 1.9 1.9 2.0

1.5
$2.00

1.4
1.4

1.4

1.4

1.4

1.4

1.4
1.3

1.3

$1.00
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Bus Commuter Rail Heavy Rail Light Rail

Bus Versus Rail Implications for Transit-Oriented Development. Figure 1

Trends of operating cost per unlinked passenger trip for bus and rail modes 1998–2007 [16]

Bus Versus Rail Implications for Transit-Oriented Development. Table 5 Maximum line capacities of transit modes

Transit unit capacity Maximum

Vehicle dimensions (Seat + standing Minimum frequency (Transit
Mode (L  W m) spaces) headway (s) units per hour) Line capacity
Standard bus 12.00  2.50 75 70–50 51–72 3,800–5,400
Articulated 18.00  2.5 120 80–60 45–60 5,400–7,200
bus
High-capacity 22.00  2.50 160 30–12 120–300 9,000–30,000
bus (BRT)
LRT (partially 24.00  2.65 3  170 = 510 or 2  150–75 24–48 12,200–26,900
separated 280 = 560
ROW)
MRT 21.00  3.15 10  240 = 2,400 150–120 24–30 67,200–72,000
Source: [17]

Capacity of Bus Versus Rail Transit

the cost performance of bus versus rail. Operating costs
for light rail systems vary widely depending upon the
system operating environment [14]. For example, oper-
ating costs in Buffalo were $67 million dollars a year for
an on-street system and $87 million dollar/year for an
in-tunnel system [5]. The cost per average weekday
boarding on Los Angele’s Orange LRT line is $16,743
and $45,762 for the Gold Line Extension [15].
The US National Transit Database reports that,
between 1998 and 2007, the operating cost per unlinked
trip has increased for all but MRT (Fig. 1) [16].
Bus and rail transit technologies differ in their design
capacities. Under different system operational plans,
their trunk line capacities may vary significantly. Table 5
provides a reference to the line capacities of standard
bus, BRT, LRT, and MRT [17]. The basic capacity of
a standard bus serves as a benchmark for line capacity
comparisons among these transit modes.
Standard buses can operate in short headways with
relatively high frequencies. Unit capacity may vary
depending on the bus seating configuration and

comfort standard. On average, standard buses can
achieve a line capacity of 3,800–5,400 places per hour.
Articulated buses have vehicle capacities approxi-
mately 50% greater than standard buses. Operating in
slightly longer headways and lower frequencies than
standard buses, articulated buses can offer a line
capacity between 5,400 and 7,200 places per hour.
High-capacity buses are used by many BRT systems in
the world. Examples include Sao Paulo and Curitiba,
Brazil, and Transmillennio in Bogota, Colombia.
They can achieve a line capacity level more than four
times the basic bus capacity. Sao Paulo, Curitiba, and
Bogota operate high-capacity buses along well-planned
bus corridors. Special operating features such as bus
convoys and off-board fare collection allow a total
throughput of about 300 buses per hour. At such
a high frequency, line capacities were reported to
achieve 24,000 and 30,000 places per hour in Sao
Paulo and Bogota, respectively.
LRT line capacities are close to those attained by
high-capacity BRT at 4.2 times that of basic bus
capacities. An LRT with fully separated ROW will
achieve line capacities higher than figures provided
above. MRT systems have a great diversity of features
that determine MRT line capacities. For instance,
dimensions of metro cars in Paris (15.00  2.5 m)
are much smaller than in Toronto and Hong Kong
(21.00  3.23 m). Maximum train lengths also vary,
from four cars in Boston to ten or more cars in New
York, Tokyo, and San Francisco. In these cities with
large-profile systems, line capacities often exceed
70,000 places per hour.
Table 5 provides a reference of transit line capacities
under design conditions. In reality, the capacities of
these transit systems vary dramatically due to such
Bus Versus Rail Implications for Transit-Oriented Development. Table 6 Capacity comparison among transit modes
BUS
Typical peak-hour capacity 1,000–3,000
Ratio to BRT capacity 0.35
Highest observed in the USA or Canada 3,000
Source: [18]
Note: (1) The BRT mode shown here combines those mixed with street traffic and those with dedicated busways. (2) BRT is used for ratio
calculation is for consistency with cost comparison shown in previous section.
Bus Versus Rail Implications for Transit-Oriented Development 299
factors as operational techniques, facility constraints,
transit demand, and road conditions. Current debate
on BRT-LRT capacity comparisons centers around
methods where these constraints are taken into consid-
eration. For example, the communication-based train
technology (CBTC) creates very small virtual blocks to
allow trains to approach one another in closer intervals.
CBTC can increase capacity by 10–15%. Automatic
train operation is an operating technology that allows
the train to optimize speed, which can increase line
capacity by 2–4%. Another factor differentiating the
line capacities of bus from train systems is the number
of lanes. Typically, BRT travels in a single lane per
each direction, whereas rail can follow multiple track
lanes. For example, New York has several three- and
four-track trunk lines.
Like cost variations, bus and rail transit each offers
a wide range of service capacities. Generally speaking,
rail transit offers larger capacities than bus transit. In
real-world applications, however, it is common to see
an overlapping range of capacity offerings by bus and
rail transit systems. For instance, BRT in Curitiba
achieves far greater higher capacities than the LRT in
Buffalo. Vehicle technologies, operation/management
efficiencies, and station/street conditions all have
significant effects on achievable capacities. Table 6
shows the range of peak-hour capacities suggested by
the literature for four aggregated transit modes [18].
Transit Impacts on Property Values
Analyzing rail transit’s effects on land use/land
development can be built upon the urban location
theories, which suggest, in a most succinct form, that
transportation defines the urban spatial access pattern
BRT LRT MRT
2,000–10,000 3,000–18,000 13,000–41,000
1.0 1.65 5.3
11,000 10,000 50,000
300 Bus Versus Rail Implications for Transit-Oriented Development
and accessibility determines the worth of land. In
making location decisions, households (or firms) bid
for locations by trading housing (and/or land)
consumption against commuting costs to the urban
center (e.g., the CBD). The closer to the center, the
better the access, and the higher the bid rent. Therefore,
we would expect a downward sloping accessibility
curve from the urban center. Parallel to the accessibility
curve is a downward gradient bid-rent curve.
Measuring price or rent levels associated with the
proximity to rail transit will provide quantified
information on the magnitude of rail transit’s influence
on land value and land use.
Research on property/land value impacts of transit
has accumulated a large volume of literature. A number
of meta-studies have been published [19, 20]. RICS
[19] reviewed about 150 references in the topical area.
Most of those reviewed are studies from the USA,
Canada, the UK, and Europe. Residential properties
have been the focus of the existing studies, while
commercial, retail, and office properties were also
examined.
Table 7 compiles findings of the reviewed studies of
transit impacts on property values. Due to differences
in methods used in the studies, direct comparisons of
the study findings are quite challenging. This entry
groups the studies by property types (residential,
commercial, or others) and by transit technologies.
The focus is on the magnitude of transit proximity
effects measured in monetary terms. Due to limited
information provided by some of the studies, only
part of the reported proximity effects could be
converted to 1990 US dollars in comparable units.
Mixed results have been reported on transit
proximity premiums associated with apartment rents
[21–24]. Results vary mainly depending on the types of
transit technologies. Most studies have found positive
transit proximity premiums for single-family homes,
with a few reporting negative impacts [25–33]. For
instance, Armstrong [33] found that single-family
home prices were 6.7% higher in communities with
a commuter rail station than in communities without.
However, a statistically significant property value loss
of roughly 20% was found for properties located within
122 m (400 ft) of the right-of-way. Weinberger [34]
argues that the presence of transit is a nuisance and may
drive down adjacent property values. They cite the
noise generated by transit and believe that transit
access allows undesirable people easy access to their
neighborhoods. Chen et al. [28] tested the nuisance
hypothesis in their study of light rail in Portland,
Oregon, but found that the benefits to property values
attained from proximity to light rail outweighed the
negative effects. As shown in Table 7, most studies have
identified positive effects of rail transit on nearby
properties although some reported negative effects.
LRT tends to have mixed impacts on commercial
properties [22, 23, 26, 34], contrasting to mostly
positive effects associated with MRT [26, 35–37].
Nelson’s [35] study of the Atlanta MARTA heavy rail
line showed a negative gradient of $75 reduced property
value per meter from a transit station, indicating a positive
benefit associated with access to transit. Weinberger’s study
suggest positive impacts on commercial property values
resulting from proximity to light rail stations, though no
specific capitalization rate is provided. The study by
Cervero and Duncan shows that in Santa Clara County,
commuter rail had a greater impact on commercial
property values than the light rail line. Weinstein and
Clower’s [38] study of the DART light rail system also
suggests appreciable effect on commercial properties in
the Dallas area from proximity to light rail.
Data on the effects of BRT on property values are
quite limited. Rodriguez and Targa [21] estimated that,
in Bogota, Columbia, every additional 5 min of walking
time to a BRT station reduced rental price by 6.8–9.3%,
which translates to a premium of US$0.05 per m2 for
every meter closer to the station after controlling for
the distance to (i.e., the nuisance effect of) the system
right-of-way. Also studying Bogota’s BRT, Munoz-
Raskin [25] confirmed that, in most cases, property
values were lower within 5-min distance of the system
than those beyond. Cervero and Duncan’s [22] study of
BRT in Los Angeles found that residential properties
near BRT stops sold for less while commercial
properties for more. The reason for this negative price
effect, according to them, is that the majority of BRT
stops are located in redevelopment areas of Los
Angeles. In another study looking at how income serves
as a determinant in communities’ valuing transit,
Nelson [10] found that the elevated MARTA heavy
rail in Atlanta had positive effects on home values
in low-income neighborhoods but negative effects in
high-income neighborhoods.
 Bus Versus Rail Implications for Transit-Oriented Development 301

Bus Versus Rail Implications for Transit-Oriented Development. Table 7 Transit impacts on property values
Standardized Standardized
proximity proximity
Transit premium in premium in
Study system Findings present US$ 1990 US$
Apartment (monthly rent in dollars per square meters for every meter closer to the station)
Rodriguez and BRT: Bogota 6.8–9.3% increase for every 5 min walk closer to 0.07 0.05
Targa (2004) [21] the station; or in 2002 US$439–653/0.1 km
Cervero and LRT: LA No significant effects found 0 0
Duncan (2002) [22] County
Cervero and LRT: San East line: within 1/2 mile; 2000 $104,827 more 0.27 0.21
Duncan (2002) [23] Diego than other locations
Benjamin and MRT: Rents decrease by 2.4–2.6% for each one-tenth (1.24) (1.15)
Sirmans (1996) [24] Washington, mile increase of distance from a metro station.
D.C. 1992 mean month rent $797
Single-family homes (dollars per home for every meter closer to the station)
Munoz-Raskin BRT: Bogota In most cases, negative impacts are reported for
(2006) [25] properties location within 5-min distance to the
system
Landis et al. LRT: San The typical home sold for 1990 $272 more for 2.72 2.72
(1995) [26] Diego Trolley every 100 m closer to a light rail station.
Landis et al. LRT: No effects found 0 0
(1995) [26] Sacramento
Landis et al. LRT: San Jose The typical house was worth $197 less for every (1.97) (1.97)
(1995) [26] 100 m it was closer to light rail.
Al-Mosaind et al. LRT: Portland The typical house sold for $663 more for every 21.74 24.02
(1993) [27] MAX, Eastside 100 ft nearer a light rail station.
Chen et al. LRT: Portland Beginning at a distance of 100 m from the 32.20 30
(1998) [28] MAX, Eastside station, each additional meter away from
line decreases average house price by 1992 $32.20.
Dueker and Bianco LRT: Portland Median house values increase at increasing rates 37.70 37.70
(1999) [29] MAX, Eastside as move toward an LRT station. The largest price
line difference (1990 $2,300) occurs between the
station and 200 ft away.
Lewis -Workman LRT: Portland On average, property values increase by $75 for 2.25 1.83
and Brod MAX, Eastside every 100 ft closer to the station (within the
(1997) [30] line 2,500 ft–5,280 ft radius).
Knaap et al. LRT: Portland The values of parcels located within ½-mile of the
(1996) [31] MAX, line rise with distance from the lines, but fall with
Westside line distance from the stations.
Landis et al. MRT: BART 1990 single-family home prices decline by $1–2 1.50 1.50
(1995) [26] per meter of distance from a BART station in
Alameda and Contra Costa Counties.
Lewis -Workman MRT: BART Average home prices decline by about $1,578 for 51.74 43.11
and Brod every 100 ft further from station.
(1997) [30]
302 Bus Versus Rail Implications for Transit-Oriented Development

Bus Versus Rail Implications for Transit-Oriented Development. Table 7 (Continued)

Standardized Standardized
proximity proximity
Transit premium in premium in
Study system Findings present US$ 1990 US$
Lewis -Workman MRT: New Average home prices decline by about $2,300 for 75.41 62.84
and Brod York City MTA every 100 ft further from the station areas.
(1997) [30]
Voith (1993) [32] MRT: Finds a premium for single-family homes with 7.75%
Philadelphia access to rail stations of 7.5–8.0% over the
SEPTA average home value.
Cervero and CRT: San Price increases by 17% in ½ mile of non- 84 63.75
Duncan (2002) [23] Diego downtown Coaster station
Landis et al. CRT: CalTrain Did not find a significant impact on house values (170) (170)
(1995) [26] from proximity to a rail station. Houses within
300 m of a CalTrain right-of-way sold at a $51,000
discount.
Armstrong CRT: Boston Single-family residences located in communities 6.70%
(1994) [33] MBTA, that have a rail station have a market value
Fitchburg approximately 6.7% greater than those that do
Line not. Also found a property value loss of about
20% for properties located within 400 ft of
a commuter or freight rail right-of-way.
Commercial property (dollars per square meter for every meter closer to the station)
Landis et al. LRT: San No effect found for commercial impacts 0 0
(1995) [26] Diego Trolley
Weinberger LRT: Santa Commercial space within a ¼-mile of a station 0.36 0.27
(2001) [34] Clara, County received an average of 2.3¢ to 5.0¢ more per
Guadalupe square feet than space located more than 3⁄4 -mile
line from a station. Office space sold within a ¼-mile
of a station received an average of $4.87 per
square feet more per gross building square feet
compared to space located more than 3⁄4 -mile
from a station.
Landis et al. MRT: BART Found no effect for commercial property. 0 0
(1995) [26]
Nelson (1998) [35] MRT: Atlanta Price per square meter falls by $75 for each meter 75.00 61.07
MARTA away from transit stations. Price rises by $443 for
location within special public interest districts.
FTA (2000) [36] MRT: Price per square foot decreases by about $2.30 0.07 0.05
Washington, for every 1,000 ft further from station.
D.C.
Fejarang et al. MRT: Los Commercial space within ½-mile of the rail 0.39 0.34
(1994) [37] Angeles corridor had an additional $31 increase in mean
sale price per square feet over the mean sales
price of a comparable control group outside of
the rail corridor, between 1980 and 1990.
 Bus Versus Rail Implications for Transit-Oriented Development 303

One factor that may partly explain the varying
empirical findings above is system maturity [12, 14].
Studies conducted on newly constructed transit
systems may not capture the positive valuation of
transit on property values because housing markets
have not had time to react to transit’s presence. Transit
ridership may serve as an indication of the magnitude
that transit can add to property values, since low
ridership indicates that transit is not an amenity to be
used [14]. Transit technology also matters, with
reliability and speed adding value to the system and
in turn being capitalized in property values. Unwanted
adjacent uses such as freight or industry will also play
an important role by reducing neighborhood quality of
life [13].
Density for Bus and Rail Transit
A question of general interest is what land-use densities
are needed in order to support specific transit
operations. The work by Pushkarev and Zupan [39]
remains one of the most influential publications on
density and transit uses. They plotted with the
logarithmic values of population and trip densities on
the horizontal and vertical axis, respectively.
The log-log graph exhibits virtually a straight line.
Specifically, in the New York Region, trips per square
mile by MRT increase from 12 at a residential density of
0.8 dwelling units per acre (or du/ac) to more than
60,000 at a density of 200 du/ac. Table 8 shows a typical
range of transit supporting densities [39, 40].
Recent studies have confirmed the pattern that
Pushkarev and Zupan found 3 decades ago: the closer
people live to transit and other services such as shops,
the more likely it is that they will utilize the transit and
drive less [41, 42]. The study by Frank and Pivo [43]
compared travel mode choice among different census
tracts with differing levels of employment and
population density. They found that transit usage and
walking increase as density and land-use mix increase,
whereas travel by single-occupancy vehicle (SOV)
declines. The relationship between density (employ-
ment and population) and mode choice for SOV,
transit, and walking displays a nonlinear pattern for
both work and shopping trips. Significant modal shifts
from SOV to transit and walking occur when employ-
ment density increases from 20 to 75 employees per
acre or reaches 125 or more employees per acre.

Bus Versus Rail Implications for Transit-Oriented Development. Table 8 Transit supportive densities

Minimum necessary
residential density
Mode Service (dwelling units per acre) Remarks
Dial-a-bus Many origins to many 3.5–5 Lower figure if labor costs twice those of
destinations taxis; higher if thrice those of taxis
Local bus Minimum to frequent services; 4–15 Average, varies as a function of downtown
1/2 mile route spacing; 20–120 size and distance from residential area to
buses per day downtown
Express bus Five buses during 2 h peak 15 Average density over From 10 to 15 miles away to largest
(BRT) – walk period 2 mile2 tributary area downtowns only
access
Express bus Five to ten buses during 3 Average density over From 10 to 20 miles away to downtowns
(BRT) –drive 2 h peak period 20 mile2 tributary area larger than 20 million square feet of
access nonresidential floorspace
Light rail Five-minute headways or 9 Average density for To downtowns of 20–50 million square
(LRT) better during peak hour. a corridor of 25–100 mile2 feet of nonresidential floorspace
Metro rail Five-minute headways or 12 Average density for To downtowns larger than 50 million
(MRT) better during peak hour. a corridor of 100–150 mile2 square feet of nonresidential floorspace
Source: From [39, 40]
304 Bus Versus Rail Implications for Transit-Oriented Development
Population density needs to exceed approximately
13 persons per acre in order to achieve major modal
shifts from SOV to non-driving modes for shopping
travel.
The Institute of Transportation Engineers in its
1989 publication suggests that, to support bus service
with headways of an hour, 4–6 du/ac and 5–8 million
square feet of commercial/office space in the corridor
is needed; for bus service with 30 min headways,
7–8 du/ac and 8–20 million square feet of
commercial/office space is needed; and for light rail
and feeder bus service density of 9 du/ac and
35–50 million square feet of commercial/office space
is desired [44]. Newman and Kenworthy found that
densities of at least 12–16 persons/acre are needed to
support urban transit [45]. Similarly, another study
found that transit use increases sharply when residen-
tial density moves from 7 to 16 du/ac [46]. Levinson
[47] found that a relationship between mode choice
and density was significant only in densities greater
than 10,000 persons per square mile. As much as 93%
of the transit demand in Portland, OR can be explained
by employment and housing densities [48]. Residential
densities of more than 80 du/ac were considered highly
supportive, those from 46–80 du/ac medium high,
those from 20–45 du/ac medium, 6–19 du/ac medium
low, and 0–5 du/ac as low in support of transit [49].
Standard bus service can be maintained at 7 du/ac, and
significant transit use increases can occur at 10 du/ac
and higher [46, 50]. Lawton has also shown that the
higher the density found in an area, the more likely
people will walk or take transit and produce less vehicle
miles traveled [51].
Some have argued that certain densities of office
and other industries can support transit without
high-density development [52–54]. Cervero [55]
investigated suburban employment densities and
found that ride sharing increased by 3.5% for every
5,000 jobs added. Retail and services located near places
of employment reduce VMT and prompt workers to
utilize transit [56]. Barnes and Davis [57] maintain that
commercial land use exerts much more influence on
travel choices than do residential densities. Additional
work by Cervero has found that suburban employment
centers, especially when mixed use is incorporated,
prompt the employees to commute via different
modes. Furthermore, those living near transit stations
are much more likely than others to commute to work
by transit [58–60].
Others have emphasized the link between tran-
sit usage, car ownership, and centrality instead of
transit usage to density [61]. It has been shown
that a 10% increase in population centrality
(corresponding with higher densities) lowers the
chance that a worker drives to work by a margin
of 2.1% [62]. Johnson concluded that not only
density but density and its spatial relation to the
transit stop are both important when evaluating
ridership potential [63]. Schimek [64] has argued
that, while density matters, its affects are minimal –
only providing a 0.7% reduction in auto travel with
a 10% rise in density. Another study noted that while
density seemed to explain bus usage significantly, it was
difficult to properly identify densities, and that
destination as well as departure points should be
taken into account [65].
Specific densities needed for BRT operation have
not been sufficiently identified, although figures for
the standard bus-based operations can be borrowed
and extrapolated. BRT may perform better in
lower-density areas than LRT [66]. In corridors with
several medium-density cores but with an overall
low-density landscape, BRT’s flexibility may make it
more suitable than LRT [67].
Transit analysis conducted by the Florida
Department of Transportation has ranked densities
as to how they may support public transit (Table 9).
Residential densities at 80 dwelling units or more
per acre are considered highly supportive, whereas
Bus Versus Rail Implications for Transit-Oriented Devel-
opment. Table 9 Florida DOT’s assessment of transit
supportive densities
Dwelling units per acre Support level
>80 High
46–80 Medium high
20–45 Medium
6–19 Medium low
0–5 Low
Source: [68]
 Bus Versus Rail Implications for Transit-Oriented Development 305

Bus Versus Rail Implications for Transit-Oriented Development. Figure 2

BRT-based TOD in Curitiba, Brazil

Bus Versus Rail Implications for Transit-Oriented Development. Figure 3

LRT-based TOD node, Mockingbird Station, Texas, USA
306 Bus Versus Rail Implications for Transit-Oriented Development

less than five dwelling units per acre are not transit of pedestrian paths and public as well as private vehicle
supportive [68]. movement. The success of the Mockingbird Station
TOD has proven that a properly conceived TOD can
succeed and flourish by serving the adjacent
Best Practice of Bus- Versus Rail-Based TOD
communities, increasing transit use while acting as
Curitiba is the capital city of the Brazilian state of a catalyst for place making.
Paraná in Brazil. It presents one of the most successful Hong Kong’s Mass Transit Railway Corporation
corridor-type TODs on BRT in the world. Curitiba’s (MTRC) is one of few successful cases in the world
BRT system is featured with exclusive rights-of-way, that the rail transit agency is able to fully self-finance
biarticulated buses, tube stations, and land-use zoning system construction and operation by practicing
coordinated with high-capacity bus corridors (Fig. 2). a Hong Kong model of TOD, namely, “Rail + Property”
The system plan is known as the Trinary Model. The or “R + P” (Fig. 4). Its central idea is to integrate
model has a two-lane street at the center where the property development with rail operation and
express buses have their exclusive lane. Adjacent to construction such that the profit gained from property
the express lane are two one-way streets moving in development pays the operational and capital costs of
the opposite directions. Lots on both sides of the the rail system. The main working procedures are as
central streets are zoned to develop at high densities. follows. First, MTRC studies development potentials in
One block from the one-way street on each side of the
corridor is a regular two-way street that carries a mix of
private vehicle flow and conventional bus services. Five
of these roads form a star that converges to the city
center. Land farther from the roads is zoned for
lower-density developments.
The success of Curitiba’s bus-based corridor TOD is
shown by a number of key statistics: as of 2008, there
were 340 bus lines, 58 km of busways, 1,100 km bus
routes, and 26 terminals. Daily the system serves
380,000 passengers by the corridors and 2.4 million
by the network. The BRT operates at a less than 1 min
headway in the peak hour, peak directions, with an
average speed of 30 km/h for the trunk line and
20 km/h for other routes. More than 70% of
commuters ride the bus to work.
The Mockingbird Station TOD represents the best
practice of TODs implemented around a single node of
LRT (Fig. 3). Located four miles north of downtown
Dallas, the Mockingbird Station is one of the largest on
the Dallas Area Rapid Transit (DART) rail line.
Developer Ken Hughes of UC Urban of Dallas initiated
and developed the 4-ha project in 2000–2001. Main
functions of the TOD site contains 16,536 m2 of retail,
restaurant, and cinema space; 12,727 m2 of office space,
211 loft apartments; and parking for 1,580 vehicles.
The main features of the Mockingbird Station TOD Bus Versus Rail Implications for Transit-Oriented Devel-
include its combination of adaptive use of historical opment. Figure 4
warehouses with new construction, fine design of Tsingyi station: example of Hong Kong’s “Rail + Property”
streetscape and public arts, and careful organization TOD model
 Bus Versus Rail Implications for Transit-Oriented Development 307

Bus Versus Rail Implications for Transit-Oriented Development. Table 10 Comparison among major transit modes
BUS BRT LRT MRT
Capacity: typical peak-hour passengers 1,000–3,000 2,000–10,000 3,000–18,000 13,000–41,000
Capital cost (1990$ per route mile, millions) $10.24 $26.4 $128.2
Operating cost ($ per vehicle revenue mile) $3.1 $3.6 $9.3 $6.5
Operating cost ($ per unlinked passenger trip) $1.51 $2.03 $1.23
Property impacts: apartment (1990$ per square meter for $0.05 $0.3 $1.15
every meter away from transit)
Property impacts: single-family home (1990$ per home $0–38 $43–62
for every meter away from transit)
Minimum density (dwelling units/acre) 4 9 9 12
Ratio to BRT
Capacity 0.35 1 1.65 5.30
Capital cost 1 2.58 12.52
Operating cost (per vehicle revenue mile) 0.86 1 2.58 1.81
Operating cost (per unlinked trip) 1 1.34 0.82
Minimum density 0.44 1.00 1 1.33

the area of approximately 500-m distance to the
proposed stations. It then obtains land development
right from the government for the (re)developable
properties. The land cost to MTRC (i.e., land premium
income to the government) is estimated based on the
no-rail scenario. Next, MTRC specifies development
plan for the properties in terms of use composition
(residential, retail, green space, transportation, etc.),
density (i.e., FAR), and other needed facilities. After
the government’s approval of the plan, MTRC develops
the rail system and the station area in partnership with
developers. Because of rail transit availability, property
value increases. MTRC then sells properties to
individuals or firms, sharing the profits with developers
and the government. The MTRC’s earnings will be used
for maintenance and expansion of rail services as well
as other expenses or investments.
MTRC’s R + P model proves to offer an all-win
solution, at least in the scope of MTRC’s services.
MTRC benefits from guaranteed funds for service
delivery and expansion. To the community, ordinary
citizens enjoy high-quality rail services while the share
of household expenditure in transportation remains
low (9%) [69]. To the developer, MTRC plays
a coordinator role, helping reduce start-up costs. The
government is a big winner as well. Since 2000 when
MTRC went public, the R + P practice has enabled
MTRC to operate high-quality rail services without
asking subsidy from the government. Furthermore,
the Hong Kong government has received a profit (net
initial equity contribution) of over HK$103 billion
from its 76% MTRC shares.
Conclusions
Table 10 summarizes modal characteristics of bus-
and rail-based transit systems. Main categories of
features for comparison include capacity, cost (cap-
ital and operating), land-use impacts, and transit
supportive densities. For easy reference, their attri-
butes are normalized by using BRT as the main
reference.
The comparisons depicted above indicate that, on
average, BRT has the potential to outperform LRT in
providing a moderate to high level of service capacity at
a moderate level of capital and operating costs in
neighborhoods with moderate population and job
densities. MRT, although most expensive to build, can
308 Bus Versus Rail Implications for Transit-Oriented Development
achieve over five times the capacity of BRT or LRT and
has demonstrated the largest positive impacts on
property values in the vicinity of rail stations.
The reviews provided in this entry on the compar-
ative costs, capacities, and land-use impacts of various
rapid transit technologies suggest that no simple
answer exists to the question. Perhaps no single form
of transit will fit all market needs of a community.
Each transit technology is efficient when it is in the
right place serving the right market. It is more likely the
case that all forms of transit systems coexist, each
serving a particular market niche. The key is to propose
context-sensitive TODs for integrated development
transit with land use.
Future Directions
Current TOD practices and studies have largely focused
on nodal TODs, that is, the stand-alone TODs around
individual bus stops or train stations. However, many
community objectives, for instance, affordable housing
provision, job-housing balance, and congestion relief,
cannot be well addressed at the individual TOD basis.
From a traveler’s perspective, whether she/he uses
transit depends not only on what the origin can offer
but also on what amenities the destination provides.
Hence, an important direction for future study is
Corridor TOD, which formulates development and
implementation strategies at the corridor scale along
bus or rail transit routes.
Another future direction is TOD-based value
capture for transit financing. Existing TOD initiatives
are mostly concerned with physical planning and
design of TODs as well as legal and political barriers
to TOD implementation. There are successful examples
as mentioned above on TOD-based value capturing
through integrated transit-land development.
How the successful experience can be transferred to
cities and countries with different legal, institutional,
and financial settings remains a challenge and warrants
future research.
Furthermore, the current urban TOD practice can
be expanded to other modes of transportation for
intercity travel, for example, air, ferry, and high-speed
rail. Stations and terminus for intercity transportation
typically have more complex traffic operations than
those of intracity transit. A new trend is to develop
the station or terminal areas as multifunctional
places rather than traditionally as single-function
transportation hubs. TOD principles can be applied
to these cross-region modes, while their technological
complexities require specific consideration.
Bibliography
1. Calthorpe P (1993) The next American metropolis: ecology,
community and the American dream. Princeton Architectural
Press, New York
2. Beimborn E, Rabinowitz H, Gugliotta P, Mrotek C, Yan S (1991)
Guidelines for transit-sensitive suburban land use design.
Report No. DOT-T-91-13. U.S. Department of Transportation,
Urban Mass Transportation Administration, University
Research and Training Program, Washington, DC
3. Urban Land Institute (1994) Transit-oriented design. Working
Paper 635, June 1994, The Urban Land Institute, Washington, DC
4. Porter DR (1998) Transit-focused development: a progress
report. J Am Plann Assoc 64(4):475–488
5. Bernick M, Cervero R (1997) Transit villages in the 21st century.
McGraw-Hill, New York
6. Reconnecting America, 2006. http://www.reconnecting
america.org
7. Gordon D (2003) New urbanism and smart growth: twins sep-
arated at birth? Places - Forum Environ 15(2). http://www.
escholarship.org/uc/item/8wp02013. Accessed 10 March 2010
8. Dittmar H, Ohland G (2004) The new transit town: best
practices in transit oriented development. Island Press,
Washington, DC
9. Vuchic V (2007) Urban transit-systems and technology. John
Wiley, Hoboken, NJ
10. USDOT (1993) Characteristics of urban transportation
systems, revised edn. U.S. Department of Transportation,
Washington, DC
11. GAO (2001) Mass transit: bus rapid transit shows promise. U.S.
General Accounting Office, Washington, DC
12. BAH (Booz-Allen & Hamilton, Inc.) (2003) Light rail capital
cost study update. Federal Transit Administration,
Washington, DC
13. Tennyson E (2008) Analysis of the general accounting office
report "Bus Rapid Transit Shows Promise." http://www.
lightrailnow.org/myths/m_brt002.htm. Accessed 30 July 2008
14. Pickrell D (1992) A desire named streetcar. J Am Plann Assoc
58(2):158–177
15. Vincent W, Callaghan L (2007) Evaluating the Los Angeles
orange line BRT. In: Presented at 86th annual meeting of the
transportation research board, Washington, DC
16. U.S. Federal Transit Administration (2008) National transit
database. At URL: http://www.ntdprogram.gov/ntdprogram/
pubs/NTST/2007/HTML/Operating_Costs_and_Performance_
Measures.htm. Accessed 10 March 2010
17. Vuchic V (2005) Urban transit: operations, planning and eco-
nomics. John Wiley & Sons, Hoboken, NJ

18. Kittelson & Associates (2003) TCRP report 100: transit capacity
and quality of service manual, 2nd edn. Transportation
Research Board, Washington, DC
19. RICS (2007) Land value and public transport. RICS policy unit.
http://www.rics.org/NR/rdonlyres/567DF930-261E-4B2A-89E4-
D466FEB84436/0/land_report4.pdf. Accessed December, 2007
20. Debrezion G, Pels E, Rietveld P (2007) The impact of railway
station on residential and commercial property value: a meta-
analysis. J Real Estate Finance Econ 35:161–180
21. Rodriguez DA, Targa F (2004) Value of accessibility to Bogota’s
bus rapid transit system. Transp Rev 24(5):587–610
22. Cervero R, Duncan M (2002) Land value impacts of rail transit
services in Los Angeles County. Report prepared for National
Association of Realtors, Urban Land Institute, Washington, DC
23. Cervero R, Duncan M (2002) Land value impacts of rail transit
services in San Diego County. Report prepared for National
Association of Realtors, Urban Land Institute, Washington, DC
24. Benjamin JD, Stacy Sirmans G (1996) Mass transportation,
apartment rent and property values. J Real Estate Res
12(1):1–8
25. Munoz-Raskin R (2007) Walking accessibility to bus rapid tran-
sit in Latin America: Does it affect property values? The case of
Bogota, Columbia. In: Presented at 86th Annual Meeting of
the Transportation Research Board, Washington, DC
26. Landis J, Guhathakurta S, Huang W, Zhang M (1995) Rail transit
investments, real estate values, and land use change:
a comparative analysis of five California rail transit systems.
UCTC Working Paper No. 285. University of California Trans-
portation Center, Berkeley
27. Al-Masaind MA, Duecker KJ, Strathman JG (1993) Light-rail
transit stations and property values: a hedonic price approach.
In: Transportation research record: Journal of Transportation
Research Board, No. 1400, TRB, National Research Council,
Washington, DC, pp 90–94
28. Chen H, Rufolo A, Dueker KJ (1997) Measuring the impact of
light rail systems on single family home values: a hedonic
approach with GIS application. Discussion paper 97–93,
Center for Urban Studies, Portland State University, Portland
29. Dueker KJ, Bianco MJ (1999) Light rail transit impacts in
Portland: the first ten years. In: Presented at the 78th
annual meeting of the transportation research board,
Washington, DC
30. Lewis-Workman S, Brod D (1997) Measuring the neighbor-
hood benefits of rail transit accessibility. In: Transportation
research record: Journal of Transportation Research Board,
No. 1576, TRB, National Research Council, Washington, DC,
pp 147–153
31. Knaap GJ, Ding C, Hopkins L (2001) Do plans matter?: the
effects of light rail plans on land values in station areas.
J Plann Educ Res 21(1):32–39
32. Voith R (1993) Changing capitalization of CBD-oriented trans-
portation systems: evidence from Philadelphia, 1970–1988.
J Urban Economics 33(3):361–376
33. Armtrong RJ (1993) Impacts of commuter rail service
as reflected in single-family residential property values.
Bus Versus Rail Implications for Transit-Oriented Development 309
In: Transportation Research Record: Journal of Transportation
Research Board, No. 1466, TRB, National Research Council,
Washington, DC, pp 88–98
34. Weinberger R (2001) Light rail proximity: benefit or detriment in
the case of Santa Clara County, California? In: Transportation
Research Record: Journal of Transportation Research Board, No.
1747, TRB, National Research Council, Washington, DC, pp104–113
35. Nelson AC (1998) Transit stations and commercial property
values: a cast study with policy and land-use implications.
J Public Transp 2(3):77–95
36. FTA (Federal Transit Administration) (2000) Transit benefits 2000
working papers: a public choice policy analysis. Federal Transit
Administration, Office of Policy Development, Washington, DC
37. Fejarang R (1994) Impact on property values: a study of the
Los Angeles metro rail. In: Presented at the 73rd annual meet-
ing of the transportation research board, Washington, DC
38. Weinstein BL, Clower TL (2002) An assessment of the DART LRT
on taxable property valuations and transit-oriented develop-
ment. Center for Economic Development and Research,
University of North Texas, TX
39. Pushkarev B, Zupan J (1977) Public transportation and land
use policy. Indiana University Press, Bloomington, IN
40. Calthorpe P (1993) The next American metropolis: ecology,
community, and the American dream. Princeton Architectural
Press, New York
41. Handy S, Cao X, Buehler TJ, Mokhtarian PL (2005) Link
between the built environment and travel behavior: correla-
tion or causality. In: Presented at 84th Annual meeting of the
transportation research board, Washington, DC
42. Handy S, Mokhtarian PL (2005) Which comes first: the neigh-
borhood or the walking? ACCESS 26:16–21
43. Frank LD, Pivo G (1995) Impacts of mixed use and density on
utiliation of three modes of travel: Single-occupant vehicle,
transit and walking. In: Transportation Research Record:
Journal of Transportation Research Board, No. 1466, TRB,
National Research Council, Washington, DC, pp 44–52
44. Holtzclaw J (2008) Using residential patterns and transit to
decrease auto dependence and costs, Natural Resources
Defense Council. http://www.smartgrowth.org/library/
cheers.html. Accessed October 30, 2008
45. Newman P, Kenworthy J (1989) Cities and automobile depen-
dence: an international sourcebook. Gower, Aldershot
46. Smith W (1984) Mass transit for high-rise high-density living.
J Transp Eng 110(6):521–535
47. Levinson HS (1996) Cities, transportation and change.
Transp Q 50(4):67–85
48. Nelson/Nyggard Consulting Associates (1995) Land use and
transit demand: the transit orientation index. Primary Transit
Network Study, Tri-Met, Portland, OR
49. Florida DOT (2008) South Florida East Coast corridor transit
analysis study alternatives development. http://www.
sfeccstudy.com/. Accessed 30 July 2008
50. Dittmar H, Ohland G (2004) The new transit town: best
practices in transit oriented development. Island Press,
Washington, DC
310 Bus Versus Rail Implications for Transit-Oriented Development
51. Lawton K (2008) The urban structure and personal travel: an
analysis of Portland, Oregon Data and some national and
international data, E-Vision 2000 conference. www.rand.org/
scitech/stpi/Evision/Supplement/lawton.pdf. Accessed 30 July
2008
52. Ewing R (1998) Transportation and land use innovations. APA
Press, Chicago
53. Loukaitou-Sideris A, Banerjee T (2000) The blue line blues: why
the vision of transit village may not materialize despite impres-
sive growth in ridership. J Urban Des 5(2):101–125
54. Kim S, Ulfarsson G, Gudmundur T (2007) Analysis of light rail
rider travel behavior: impacts of individual, built environment,
and crime characteristics on transit access. Transp Res A Policy
Pract 41(6):511–522
55. Cervero R (1989) America’s suburban centers: the land-use–
transportation link. Unwin Hyman Inc, Winchester, MA
56. Chatman DG (2003) How density and mixed uses at the work-
place affect personal commercial travel and commute mode
choice. In: Transportation Research Record: Journal of Trans-
portation Research Board, No. 1831, TRB, National Research
Council, Washington, DC, pp 193–201
57. Barnes G, Davis G (2001) Land use and travel choices in the
twin cities. Center for Transportation Studies, University of
Minnesota (www1.umn.edu/cts), Report #6 in the Series:
Transportation and Regional Growth Study, Minneapolis
58. Cervero R (1994) Rail-oriented office development in
California: how successful? Transp Q 48(1):33–44
59. Cervero R (1998) The transit metropolis: a global inquiry. Island
Press, Washington, DC
60. Cervero R (2006) Office development, rail transit, and com-
muting choices. J Public Transp 9(5):41–55
61. Miller E, Ibrahim A (1998) Urban form and vehicle usage. In:
Transportation Research Record: Journal of Transportation
Research Board, No. 1617, TRB, National Research Council,
Washington, DC
62. Bento A, Cropper ML, Mobarak AM, Vinha K (2008) The impact of
urban spatial structure on travel demand in the United States,
World Bank group working paper 2007. http://econ.worldbank.
org/files/24989_wps3007.pdf. Accessed 30 July 2008
63. Johnson A (2003) Bus transit and land use: illuminating the
interaction. J Public Transportation 6(4):21–39
64. Schimek P (1997) Understanding the relatively greater use of
public transit in Canada compared to the USA. Unpublished
Ph.D. dissertation. Massachusetts Institute of Technology.
Cambridge, MA
65. Messenger T, Ewing R (1996) Transit-oriented development in
the sunbelt. In: Transportation Research Record: Journal of
Transportation Research Board, No. 1552, TRB, National
Research Council, Washington, DC, pp 145–152
66. Currie G (2006) Bus rapid transit in Australasia. J Public Trans-
port (BRT special edition), 1–22
67. Falbel S, Rodriguez P, Levinson H, Younger K, Misiewicz S
(2006) Bus rapid transit plans in New York’s capital district.
J Public Transport (BRT special edition), 23–49
68. Florida DOT (2009) South Florida East coast corridor
study. URL http://www.sfeccstudy.com/index.html. Accessed
10 March 2010
69. Tang BS, Chiang YH, Baldwin AN, Yeung CW (2004) Study
of the integrated rail-property development model in
Hong Kong, Research Center for Construction & Real
Estate Economics, the Hong Kong Polytechnic University,
Hong Kong
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of 311

Cooperative Group of Vehicles and Driver assistance system Technical system supporting
human drivers in performing certain driving tasks.
Dangerous Situations, Recognition of Driver assistance systems can be classified into comfort
KYM WATSON1, CHRISTIAN FRESE2, THOMAS BATZ1, systems (not subject of this entry) and safety systems.
JÜRGEN BEYERER2,3 Motion planning Algorithmic planning of feasible,
1
Department Information Management, Fraunhofer collision-free vehicle trajectories.
Institute of Optronics, System Technologies and Image Prediction Projection of system states into the near
Exploitation IOSB, Karlsruhe, Germany future. Specifically, in this context, model-based
2
Institute for Anthropomatics, Vision and Fusion prognosis of vehicle position.
Laboratory, Karlsruhe Institute of Technology, Wireless vehicular communication Networking of
Karlsruhe, Germany multiple vehicles using wireless local area network
3
Fraunhofer Institute of Optronics, System or cellular communication technologies.
Technologies and Image Exploitation IOSB, Karlsruhe,
Germany Definition of the Subject
The subject of this entry is cooperation among cogni-
Article Outline tive vehicles with the aim of increasing traffic safety.
Glossary This includes the formation of cooperative groups, the
Definition of the Subject recognition of dangerous situations using stochastic
Introduction prediction methods, and the planning of cooperative
Cooperative Groups maneuvers to avoid collisions (see Fig. 1).
Recognition of Dangerous Situations
Introduction
Cooperative Decision Making
Future Directions Intelligent vehicle technologies have been a topic of exten-
Acknowledgments sive research for the last decades. Important research
Bibliography areas include perception of roads, obstacles and environ-
ment, situation recognition, vehicle control, and behavior
Glossary
planning. Most of the developed technologies can be used
Cognitive vehicle A road vehicle equipped with sen- both for driver assistance systems and for fully autono-
sors like video, radar, or lidar and information mous vehicles. Regarding driver assistance, some comfort
processing resources for environment perception. and safety systems have been released to market, e.g.,
Furthermore, cognitive vehicles may have human– adaptive cruise control, antilock braking, and emergency
machine interfaces, actuators for automatic brake assist systems [1, 2]. State-of-the-art autonomous
intervention or fully autonomous operation, and road vehicles have been demonstrated at the DARPA
wireless communication devices. Urban Challenge 2007, where several automated vehi-
Common relevant picture Data structure containing cles completed driving missions lasting for hours in
the information necessary for situation analysis and a restricted traffic environment [3].
decision making within a cooperative group. The Cooperative systems offer even more potential
information originates from the sensors of the vehi- for increasing safety and improving traffic flow.
cles, is enhanced by fusion algorithms, and is com- In recent years, there has been a lot of research on
municated to all vehicles of the group. wireless vehicle-to-vehicle communication, see [4]
Cooperative group (CG) A set of cognitive vehicles for an overview. Applications of vehicular communi-
which cooperate in order to mitigate dangerous cation technology include wireless local danger warn-
traffic situations. ing [5, 6], distributed sensor fusion [7–9], and
Dangerous situation Traffic situation with immediate improvement of traffic efficiency (e.g., platooning
threat of an accident. [10–12], intersection control [13, 14]). However,

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
312 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

Cognitive Formation of Cooperative

vehicle cooperative groups group (CG)

CG 2
Situation Common
recognition/ relevant
assessment picture
CG 1

Dangerous
situations

Utility function Behavior

planning

Instructions for the Optimal

individual vehicles decision

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 1

Architecture of a cooperative collision warning and avoidance system

this entry focuses on the safety applications of
recognizing dangerous situations and avoiding or
mitigating collisions by automated cooperative
maneuvers.
The benefits of cooperative behavior compared to
selfish actions of each driver have been shown for other
problems [15], and it is worthwhile to exploit the
potential of cooperation also for the improvement of
traffic safety.
Cooperative Groups
Before methods for cooperative collision warning and
avoidance systems can be designed, the set of vehicles
which can cooperate reasonably has to be determined. A
predefined, but not static, group structure enables an
efficient and unambiguous decision making in case of a
dangerous situation. Limiting the size of cooperative
groups is important because cooperative reasoning entails
significant computation and communication complexity.
A suitable distance measure for group formation
has to take into account the road topology [16]. Based
on this notion of spatiotemporal distance, a criterion
for the distributed formation of cooperative groups is
developed [17].
These concepts will be presented in detail, and
related work will be reviewed.
Motivation and Definition
In dangerous situations, fast and unambiguous cooper-
ative decision making is essential. Therefore, it is bene-
ficial to know the possible cooperation partners in
advance and to establish a decision structure. This is
achieved by the concept of dynamic cooperative groups.
A cooperative group consists of cognitive vehicles which
might encounter a dangerous situation requiring coop-
erative actions in the near future. Vehicles are
partitioned into groups, i.e., each vehicle belongs to
exactly one group at any point in time. The set of
possible cooperation partners is restricted to the vehicles
within the group. Each group has a designated coordi-
nator vehicle which handles the necessary communica-
tion (see section “Distributed Negotiation Protocol”
below). A cooperative decision may only be selected
and communicated by the coordinator. Otherwise,
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of 313

Cooperative group i

Cooperative group j

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 2

Even though all six cars in this highway example are close enough to communicate, only the two indicated sets of
vehicles constitute suitable cooperative groups because of the barrier between the lanes

ambiguous decisions might arise if two or more vehi-
cles recognize a dangerous situation at the same time.
Computation and communication efforts increase
rapidly with the number of vehicles involved. To enable
real-time cooperative reasoning, cooperative groups are
limited in size. This requires a reliable grouping crite-
rion which assigns all vehicles within one dangerous
traffic situation to the same group. This cannot always
be achieved with a limited group size, but the results will
never become worse than without cooperation.
Distance
Intuitively, it is clear that vehicles “near” to each other
should be in one group. However, it is not so clear what
a suitable notion of distance for vehicles within a road
network should be. For example, cars driving in differ-
ent directions on a dual carriageway are not able to
perform cooperative maneuvers when they are sepa-
rated by a guard rail or concrete barriers. This means
that sometimes cars which are very near to each other
in terms of Euclidean distance should nevertheless
belong to different cooperative groups (Fig. 2). There-
fore, an appropriate distance measure has to incorpo-
rate knowledge of the road network.
For navigation applications, it is common to repre-
sent the road network as a graph [18]. For the purpose
of defining a distance measure, a graph R is created by
partitioning the road area into polygons and identify-
ing each polygon with a vertex v of the graph (see
Fig. 3). Edges between neighboring polygons are
inserted according to the allowed driving directions
and lane change possibilities. The weight w(e) of an
edge e = (v1,v2) models the minimal time required for
driving from vertex v1 to the v2,
d2 ðv1 ; v2 Þ
wððv1 ; v2 ÞÞ :¼ ; ð1Þ
velmax ðv1 ; v2 Þ
where d2(v1, v2) is the Euclidean distance between the
centroids of the polygons associated with v1 and v2,
respectively, and velmax(v1, v2) is the maximal velocity
possible when driving from v1 to v2, determined by
speed limits and road curvature. Some simplifying
assumptions are made in order to obtain a conceptually
elegant formalism: maximum speeds varying with the
vehicle type are not considered in favor of a conserva-
tive approximation. Distances within one road polygon
are neglected. This does not cause any problems if the
discretization is fine enough.
The shortest-path distance pR ðv1 ; v2 Þ between arbi-
trary vertices v1, v2 of R corresponds to the minimal
time required for driving from v1 to v2. However, this
does not exactly capture the distance between two cars c1,
c2 positioned at v1 and v2, respectively, as the cars could
drive toward each other and meet much earlier at
a vertex vm in the middle of v1 and v2 (see Fig. 4). In
this case, a time span of maxfpR ðv1 ; vm Þ; pR ðv2 ; vm Þg
elapses until both cars have reached the meeting point
vm. Using this notion of virtual meeting points, the
314 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

a b

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 3

(a) Example of a road network, in which the partition into disjoint vertices is indicated with dotted lines. (b) The graph
R corresponding to (a)

chosen for processing. Once the two search trees reach

c2 nm a common vertex, this must be the nearest meeting
point vm, and the distance can be computed as
c1 dðc1 ; c2 Þ ¼ maxfpR ðvðc1 Þ; vm Þ; pR ðvðc2 Þ; vm Þg. The
pseudocode of the method is shown in Algorithm 1.

Algorithm 1 Computation of the distance d(c1, c2)

Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 4 1: procedure COMPUTEDISTANCE (R; c1 ; c2 )
The cars c1 and c2 can meet each other at vertex vm 2: INSERT (heap, c1, v(c1),0)
3: INSERT (heap, c1, v(c1),0)
4: for all vertices v of R do ▻ Initialization
5: for c = c1,c2 do
distance between two vehicles c1, c2 can be defined as 6: pR ðc; vÞ 1
7: end for
follows: 8: end for
9: while not ISEMPTY(heap) do
dðc1 ; c2 Þ :¼ min maxfpR ðvðc1 Þ; vm Þ; pR ðvðc2 Þ; vm Þg 10: (c,v,w) ← REMOVEMINIMUM(heap)
vm
11: pR ðc; vÞ w
ð2Þ 12: if pR ðc1 ; vÞ < 1 ^ pR ðc2 ; vÞ < 1 then
13: return maxfpR ðc1 ; vÞ; pR ðc2 ; vÞg
This distance corresponds to the minimal time 14: end if
within which both vehicles can reach a common meet- 15: for all edges e = (v,v0 ) of R do
ing point vm (see Fig. 5). It thus incorporates both 16: if pR ðc; vÞ ¼ 1 then
spatial and temporal aspects. The function v(c) deter- 17: INSERT(heap,c, v0 , w + w(e))
18: else if pR ðc; v0 Þ > w þ wðeÞ then
mines the unique vertex v of R so that the polygon
19: DECREASEKEY(heap,c,v0 ,w+w(e))
associated to v contains the geometric centroid of the 20: end if
vehicle c. The vehicles’ positions of course depend on 21: end for
time; however, this is not expressed explicitly here for 22: end while
the sake of notational brevity. Obviously, d(,) is sym- 23: return 1 ▻ No meeting point of c1
metric and nonnegative. and c2 is reachable within R.
24: end procedure
The distance measure d can be computed efficiently
by an appropriate modification of Dijkstra’s shortest- The computational complexity of the algorithm is
path algorithm [19], Chap. 24.3. Search frontiers are in Oðn log nÞ, where n is the number of explored edges.
expanded simultaneously starting from both v(c1) and This number can be restricted in two ways. First,
v(c2). In each step, the vertex v having the minimal a graph R is chosen that is restricted to the surround-
value of minfpR ðvðc1 Þ; vÞ; pR ðvðc2 Þ; vÞg among the ing of the vehicle (electronic horizon method). Second,
vertices not yet visited by the respective search tree is the algorithm can be stopped when the searched
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
nm c2
c1
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 5
Example of a situation in which the path length pR is not a
suitable distance. Assuming unit edge weights, we have pR
ðc1 ; c2 Þ ¼ 11 and pR ðc2 ; c1 Þ ¼ 10, while both cars can meet
each other at vm within three time steps: pR ðc1 ; vm Þ ¼ 3,
pR ðc2 ; vm Þ ¼ 2. Accordingly, d(c1, c2) = d(c2, c1) = 3
distance exceeds a threshold beyond which no suitable
cooperation is possible. Altogether, the real-time com-
putation of the distance measure is possible without
any difficulty using the described algorithm.
Objective Function
Based on the distance measure d, an objective function
rating a cooperative group G ¼ fc1 ; c2 ; . . . ; cm g of m
cognitive vehicles can be defined. Several terms have been
designed in order to capture certain criteria for the for-
mation of cooperative groups. Besides distance, these are
group size, relative velocity, and temporal consistency.
The minimum spanning tree has proven to be
a good indicator for the inter-vehicle distances within
a group. A group can be represented as a complete
graph with vertices ci and edge weights d(ci, cj). The
weights of the edges in the minimum spanning tree
MST ðGÞ of the group graph G are summed up to yield
the distance term sD of the objective function:
m X
sD ðGÞ :¼ dðci ; cj Þ ðm > 1Þ
m  1 ðc ;c Þ2MST ðGÞ
i j
ð3Þ
m
The prefactor m1 accounts for the fact that a graph
with m vertices has a minimum spanning tree
consisting of m  1 edges.
315
a b
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 6
Motivation for the velocity term. (a) Vehicles approaching
each other are more likely to encounter a common
dangerous situation. (b) Cooperative groups of vehicles
driving away from each other are less desirable
As mentioned above, cooperative groups should be
limited to a reasonable number of vehicles. On the other
hand, groups should not be too small either because this
would narrow the possibilities of cooperation. Espe-
cially, groups consisting of a single vehicle have to be
penalized because the distance function cannot be
defined for these groups. Accordingly, a desired group
size of m0 vehicles is chosen, and the term sS of the
objective function rates the deviation from this value:
sS ðGÞ :¼ ðm  m0 Þ2 ð4Þ
These two terms already enable the formation of
reasonable cooperative groups. However, two addi-
tional terms will be defined in order to select among
alternative group structures having almost equal dis-
tance and size values. Obviously, a dangerous situation
is more likely to occur within a cooperative group if its
vehicles are driving toward each other, i.e., if the dis-
tances are decreasing. By contrast, vehicles moving
away from each other are less desirable to be assigned
to the same group (see Fig. 6). These criteria are cap-
tured by the change rate of the distance measure d. The
relative velocity term sV of the objective function can
therefore be formulated using the derivative of d with
respect to time:
X
m
1 X@
sV ðGÞ :¼ dðci ; cj Þ ðm > 1Þ ð5Þ
i¼1
m  1 j6¼i @t
From the perspective of the cooperative group, sV
can be interpreted as the rate of contraction or
316 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

expansion, respectively, of the group. Contracting To enable real-time computation, strict optimality is
groups are more desirable and hence have a lower (pos- abandoned. A suboptimal anytime algorithm is
sibly negative) value of sV. designed starting from the greedy clustering method.
Finally, a temporal consistency term sT is defined in The greedy algorithm is initialized with each vehicle
order to avoid frequent changes of the group structure, being in its own single-vehicle group. In each stage of
especially oscillations of vehicles between two neighbor- the algorithm, every possible merging of two groups is
ing groups. Such oscillations could be caused by small evaluated by the objective function. The merging hav-
fluctuations of the distances, or by discretization artifacts. ing the largest improvement is carried out before the
A countermeasure is to improve (i.e., decrease) the objec- next stage starts. The greedy algorithm halts if no
tive value of a vehicle that has just joined a new group, merging remains which yields an improvement of the
Xm  total objective value.
ti if 0 < ti < tT
sT ðGÞ :¼ ; ð6Þ An anytime algorithm is obtained by inserting every
tT if ti ¼ 0 or ti  tT
i¼1 evaluated merging step into a priority queue and
where ti is the time elapsed since vehicle ci has joined processing the merging with the totally maximum
group G, and tT is a threshold value. The temporal improvement in each stage as long as execution time
consistency term penalizes vehicles changing groups is left (Algorithm 2). Experiments have shown that
(ti = 0) but rewards newly established groups (0 < ti the group structure resulting from this algorithm
< tT) in order to increase their stability. is identical to the optimal partition with a probability
The different terms introduced above can be com- of more than 99%, and that the increase in the
bined in a weighted sum to yield the objective function objective function is less than 1% in the remaining cases.
(
maxf0; lD sD ðGÞ þ lS sS ðGÞ þ lV sV ðGÞ þ lT sT ðGÞg m > 1
sðGÞ :¼
lS sS ðGÞ þ lT s T ðGÞ m¼1 Algorithm 2 Any-time group partitioning algorithm
ð7Þ 1: procedure COMPUTEPARTITION (C = {c1, . . ., cn})
rating a cooperative group G of m vehicles, where 2: P0 ← {{c1}, . . ., {cn}}
3: INSERT(heap,P0,s(P0))
lk  0 are weighting factors. The objective function 4: P∗ ← P0
s can be used to compare an existing cooperative group 5: while time left do
with hypothetical alternative groups. 6: (P,s) ← REMOVEMINIMUM(heap)
7: for all Gi 2 P do
8: for all Gj 2 P, Gj 6¼ Gi do
Computing Optimal Partitions 9: P PnfGi ; Gj g [ fGi [ Gj g
The objective function sðGÞ can be extended to a set 10: if P0 2
= closed then
11: INSERT(heap,P0 ,s(P0 ))
P ¼ fG1 ; . . . ; Gk g of mutually disjoint cooperative
12: INSERT(closed,P0 )
groups as follows: 13: if s(P0 ) < s(P∗) then
X 14: P∗ ← P0
sðPÞ :¼ sðGÞ ð8Þ 15: end if
G2P
16: end if
Given a set C = {c1, . . ., cn} of cognitive vehicles, the 17: end for
18: end for
optimal group structure P∗ according to the objective
19: end while
function is obtained as 20: return P∗
21: end procedure
P  ¼ arg min sðPÞ; ð9Þ
P2PðCÞ

where PðCÞ is the set of all partitions of C. The exact

Distributed Negotiation Protocol
minimization of (9) is possible using, e.g., depth-first
branch and bound search. However, the set of parti- To realize the group formation in a distributed vehicu-
tions grows rapidly as the number of vehicles increases. lar environment, the cognitive vehicles have to
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
negotiate the group structure without any centralized
instance. This implies that a mechanism is required
which performs reliably with incomplete and local
knowledge. The negotiation method chosen for group
formation is based on coordinator vehicles communi-
cating with neighboring groups.
The notion of neighborhood can be formalized
by means of the distance measure d introduced
above. Two groups G1 and G2 are neighbored,
N ðG1 ; G2 Þ, if
8G 8c 2 G 9c1 2 G1 9c2 2 G2
ð10Þ
maxfdðc; c1 Þ; dðc; c2 Þg > dðc1 ; c2 Þ:
Informally, this means that there does not exist any
group G in between G1 and G2 . The resulting set of
vehicles from neighboring groups is
[ Related Work
CðG0 Þ :¼ G0 [ Gj ð11Þ
N ðG0 ;Gj Þ
The optimal group structure for the local neighbor-
hood of G0 is obtained by evaluating (9) for CðG0 Þ.
The coordinator of group G0 can perform this compu-
tation approximately using the anytime algorithm
presented above.
With this notion of neighborhood, a negotiation
protocol can be designed as follows (see Fig. 7). Each
vehicle periodically sends its position information to its
group coordinator. Group coordinators exchange their
information with neighboring coordinators and com-
pute the optimal partition. If it differs from the current
group assignment, a change of group structure is initi-
ated. The change of group assignment is performed
after the involved coordinators have agreed.
Whenever possible, a change of the group coordina-
tor is avoided in order to minimize communication
and initialization costs. However, it is sometimes
inevitable to choose a new coordinator, e.g., when the
previous coordinator has left the group, or when a new
group is established. In these cases, group coordinators
may be selected arbitrary, e.g., based on the identifica-
tion numbers of the vehicles, or according to
certain criteria. For example, it may be preferable
to choose the central vehicle of the group, given by
X @
 e.g., all vehicles near to an intersection [13, 35, 36] or
arg min lD dðc; c 0 Þ þ lV dðc; c 0 Þ : ð12Þ
c2G c 0 2G
@t
317
Results
The described group formation method has been
implemented and integrated with a traffic simulator
[20]. Various highway, country road, and inner city
scenarios have been simulated successfully with real-
time performance. Figure 8 shows an example of the
results.
The approach has been evaluated with regard to the
application of cooperative collision avoidance. The
traffic simulator has been modified in order to provoke
accidents. In all considered scenarios, the vehicles
involved in the accidents have been assigned to the
same group in more than 90% of the cases, implying
that the accidents can potentially be prevented by coop-
erative maneuvers.
Related vehicle grouping approaches can be classified
according to the purpose of the groups. Often, groups
are used to structure the communication network
[21–26]. However, communication groups have to be
distinguished from application-specific groups [27].
The method described above is designed to be applied
to cooperative collision avoidance. Other conceivable
applications include cooperative local danger warning,
cooperative perception, intersection management, and
platooning. Different, but possibly similar group forma-
tion criteria may be needed for these applications. For
example, warning messages may be propagated by
oncoming traffic on a dual carriageway, while cooperative
maneuvering is not possible across the central barrier.
Many researchers neglect the issue of grouping and
assume that all vehicles within communication range
can cooperate [28]. Thus, the simplest approach is to
consider groups identical to the connected components
of the communication network [29]. Formation of
groups based on a threshold for the Euclidean distance
between vehicles has been evaluated in robotics [30], in
road traffic applications [31], and in air traffic control
[32, 33]. A more elaborated method incorporating
road topology information is described in [34].
For traffic efficiency applications, groups consisting
of all vehicles within a fixed area have been considered,
on a specific highway section [37–39]. Regarding colli-
sion avoidance, it is also possible to group all vehicles
318 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

car0:Vehicle car1:Vehicle car2:Vehicle car3:Vehicle car4:Vehicle

Vehicle position Vehicle position

Vehicle position

Group announcement

Group announcement Group announcement

Group position

Group assignment
Assignment response
Group change Group change

Vehicle position
Vehicle position

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 7

Sequence diagram of the proposed negotiation protocol. At the beginning of this example, G1 ¼ fcar0; car1g and
G2 ¼ fcar2; car3; car4g with car0, car2 being the group coordinators. Then car0 initiates a new group assignment that is
agreed by car2, resulting in G1 ¼ fcar0; car1; car2g and G2 ¼ fcar3; car4g

whose probability of mutual collision exceeds
a predefined threshold [40–42]. However, this requires
to compute the collision probabilities for all vehicles
within communication range.
Platoons consist of vehicles having common goals
[10, 43]. Therefore, criteria evaluating destinations
and, e.g., desired speed of vehicles are appropriate for
platoon formation [44, 45]. By contrast, cooperative
groups encompass possibly conflicting vehicles, inde-
pendent of their destinations.
Recognition of Dangerous Situations
An important target for a cooperative group (CG)
G ¼ fc1 ; . . . ; cm g of vehicles ci is the automatic
detection of short-term dangerous situations [46].
Short-term means within a forecasting horizon Dt
that is shorter than the reaction time of human drivers.
Such situations require an automatic reaction of the
cooperative group as they cannot be handled safely by
the drivers themselves. A typical value for the short-
term prediction horizon Dt is about 1 s for moderate
speeds of 20 m/s in curves.
With this objective, the likely path of each vehicle ci
in the time interval [t0, t0 + Dt] starting at the current
time t0 is predicted with an appropriate dynamic
model. Then, for each vehicle pair {ci, cj}, i 6¼ j the
progression of its mutual Euclidean distance ei,j(t) is
computed for t 2 [t0, t0 + Dt]. Here the Euclidean
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 8
Example of resulting cooperative groups. Cars belonging to
one group are depicted in the same color. Note that
vehicles physically separated – by the central barrier of the
highway or by the bridge – are assigned to different groups,
while the vehicles approaching the intersection form one
group, shown in green
distance between two cars is the smallest distance ||pi 
pj|| between points pi and pj of vehicles ci and cj at the
given time t, respectively. Thus this distance concept
differs from the definition of distance d for the forma-
tion of cooperative groups. The distance e depends on
the vehicle geometry and will be further refined below
to take the uncertainty of the predicted vehicle position
into account.
If at least one of the ei,j(t) falls below a distance
threshold e with t 2 [t0, t0 + Dt], then a dangerous
situation for the CG G arises, i.e., when
min  min  
 cle heading angle (measured positive anticlockwise
ci ; cj 2 G ei;j ðt þ tÞ < e ) Danger
i 6¼ j t 2 ½0; Dt 
ð13Þ
Figure 9 illustrates the detection strategy within the
time distance plane.
Vehicle Model and State Description
For an overview of vehicle models, associated sensors,
and state equations, see for example, the results of Tan
and Huang in [47], where preprocessing of the GPS
319
Mutual vehicle distances
a
c
e
b
t0 t 0 + Δt t
Time (s)
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 9
(a) Predicted mutual distances within a forecast horizon of
Dt are above a critical level e. (b) Distances too small and
time too short to manually master the situation ) danger.
(c) Enough time ahead for the drivers to manage the
situation safely
signals and measurement of yaw are also discussed. The
authors propose a system design for a vehicle to vehicle
cooperative warning system in [48]. The vehicle state
model and equations for vehicle tracking are as follows:
xðtÞ ¼ ½ xðtÞ yðtÞ fðtÞ ob ðtÞ T ð14Þ
2 3
vðtÞ cos fðtÞ
: 6 vðtÞ sin fðtÞ 7
xðtÞ ¼ 6 7
4 oz ðtÞ  ob ðtÞ 5 ð15Þ
nwb ðtÞ
where x and y are the east and north distances in the
earth-fixed coordinate frame respectively, f is the vehi-
from the x axis), v is the longitudinal velocity of the
vehicle, oz is the vehicle’s yaw rate, ob is the bias in
the yaw rate measurement, and nwb is a Gaussian
white noise.
The nonlinear differential equation (15) is solved
approximately using an Extended Kalman Filter (EKF
[49]) to track the vehicle position. The input values to
the EKF are v from a wheel speed sensor, oz(t) from
a yaw gyro and the Gaussian noise nwb. The observation
vector [x(t) y(t) f(t)]T comprises preprocessed
DGPS (Differential Global Positioning System)
320 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

measurements. The measurement and observation The lateral velocity is ignored as it is normally small.
values are transferred by the group members to the The model does not use any vehicle specific parameters
group coordinator in order to update the information in contrast to the more detailed bicycle model, cf. [47].
in the Common Relevant Picture. The group coordina- The state-space equations are nonlinear differential
tor executes the EKF algorithm to track the vehicle equations. An EKF can be used to solve them approx-
positions. imately, for example, as in [48]. Better solutions are
achieved with the Unscented Kalman Filter as described
Path Prediction in [50] and also applied in [46].
The propagation of sigma particles representing
Using the information of the Common Relevant
the initial state distribution is calculated using numer-
Picture, the group coordinator predicts the paths of
ical integration. The path of each sigma particle is
the vehicles in its neighborhood based on their state
weighted with an outcome probability. Following
information.
[50], the sigma particles of the unscented Kalman filter
The reference position of the vehicle is taken to be at
may be selected as
the center of the rear axle. The vehicle shape is modeled
as a rectangle with the parameters shown in Table 1.
xðt0 ; 0Þ ¼ xðt0 ; 0Þ ð18Þ
These parameters depend on the vehicle type. The
rectangle at vehicle position (0, 0) and heading angle xðt0 ; iÞ ¼ xðt0 ; 0Þ þ Qi for i ¼ 1; . . . ; L ð19Þ
0 is called the Reference Rectangle of the vehicle and has
xðt0 ;iÞ ¼ xðt0 ;0Þ  QiL for i ¼ L þ 1;... ;2L ð20Þ
vertices r1 ¼ ðRd ; 0:5Rw ÞT , r2 ¼ ðRl  Rd ; 0:5Rw ÞT ,
r3 ¼ ðRl  Rd ; 0:5Rw ÞT and r4 ¼ ðRd ; 0:5Rw ÞT . where x is the mean of x, L is the dimension of x, i.e., 4,
Assuming that the current time is t0, the vehicle and Qi is the ith row of the matrix square root (upper
paths are to be predicted for a time interval [t0,t0 + triangular Cholesky factorization):
Dt]. It is assumed that there are no changes of the driver pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
actions in this interval and that vehicle paths are inde- Q ¼ ðL þ lÞPx ð21Þ
pendent of each other. This assumption can be justified Px is the covariance matrix of x at the current
by taking a prediction horizon shorter than the time time t0. The scaling parameter l = a2(L + k)  L with
necessary for a specific human reaction. The path of a = 0.001 and k = 0.
a given vehicle is described with the equations Let x(t,i) be the propagation of x(t0,i) through the
differential equation (17) till time t for i = 0, . . ., 2 L.
xðtÞ ¼ ½ xðtÞ yðtÞ fðtÞ vðtÞ T ð16Þ The differential equation can be solved in MATLAB
2 3
vðtÞ cos fðtÞ with the ode45 function for non-stiff ODEs.
: 6 vðtÞ sin fðtÞ 7 The mean of the predicted x(t) is the weighted
xðtÞ ¼ 64 oz ðt0 Þ  ob ðt0 Þ 5
7 ð17Þ
mean of the x(t,i):
ax ðt0 Þ  ab ðt0 Þ
X
2L
ðmÞ
where it is assumed that the unbiased yaw rate oz(t0)  xðtÞ ¼ Wi xðt; iÞ ð22Þ
i¼0
ob(t0) and the unbiased acceleration ax(t0)  ab(t0) are
constant in the time interval. x, y, f and v are as above. with the weights defined as follows:
ðmÞ
W0 ¼ l=ðL þ lÞ ð23Þ
Cooperative Group of Vehicles and Dangerous Wi
ðmÞ
¼ 1=ð2ðL þ lÞÞ for i ¼ 1; . . . ; 2L ð24Þ
Situations, Recognition of. Table 1 Vehicle parameters
with values as used in the example The covariance matrix of x(t) is the weighted sum
Length Rl 5.0 m X
2L
ðcÞ
^ x ðtÞ ¼
P Wi fxðt; iÞ  xðtÞgfxðt; iÞ  xðtÞgT
Width Rw 1.8 m
i¼0
Distance of rear axle from back Rd 1.0 m ð25Þ

with the weights defined as follows (with b = 2):
ðcÞ
W0 ¼ l=ðL þ lÞ þ 1  a2 þ b
ðcÞ ðmÞ
Wi ¼ Wi for i ¼ 1; . . . ; 2L
For t 2 [t0, t0 + Dt], let A(t) be the point set of the
likely area occupied by the vehicle at time t given its
state as described by the mean xðtÞ and covariance
matrix P ^ x ðtÞ. A(t) is estimated by firstly using the
covariance ellipse as an estimate of the uncertainty of
the vehicle reference position:
 
EðtÞ ¼ v 2 R2 : v T QðtÞ1 v  g ð28Þ
where QðtÞij ¼ P ^ x ðtÞ for i, j = 1,2 and parameter
ij
g = 1, and secondly by considering the rotational uncer-
tainty of the vehicle about its reference position (head-
ing angle):
qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
sf ¼ P ^ x ðtÞ33 ð29Þ
If ¼ ½xðtÞ3  bsf ; xðtÞ3 þ bsf 
Figure 10 shows the scenario under discussion. Two
FðtÞ ¼ H fRðc; ri Þ : c 2 If ; i 2 f1;    ; 4gg
where sf is the standard deviation of the heading angle,
parameter b = 1, R(c, u) is the counterclockwise rota-
tion of vector u by c, ri are the vertices of the Reference
Rectangle of the vehicle, and H is the convex hull set
operator.
Finally, we set A(t) as
AðtÞ ¼ EðtÞ  FðtÞ  ðxðtÞ1 ; xðtÞ2 ÞT
where the  denotes the vector (Minkowski) sum of the
sets.
Now consider a group of m vehicles ci for
i = 1, . . ., m. An obstacle on the road can be modeled
as a stationary vehicle. The path prediction is performed
as above for each vehicle separately, assuming that there
are no dependencies in the interval [t0,t0 + Dt].
Let Ai(t) denote the likely area occupied by ci at time
t for i = 1, . . ., m.
For the assessment of collision danger, we compute
the Euclidean distance between the vehicles:
eij ðtÞ ¼ distðAi ðtÞ; Aj ðtÞÞ
 
¼ min k pi  pj k: pi 2 Ai ðtÞ; pj 2 Aj ðtÞ
Cooperative Group of Vehicles and Dangerous Situations, Recognition of 321
and consider the minimum in the prediction interval
[t0, t0 + D t]:
ð26Þ
eijm ¼ minfeij ðtÞ : t 2 ½t0 ; t0 þ Dtg ð35Þ
ð27Þ
A dangerous state is detected if eijm < e for at least
one vehicle pair ci, cj, where e is a distance threshold (in
the example set to 1 m).
The geometric computations in (28), (32), and (33)
may be executed efficiently with packages for polytope
calculus such as the Multi-Parametric Toolbox (MPT,
[51]) with the optimization toolbox YALMIP [52].
Distances below 5 m as considered here require an
accurate polytope approximation of the vehicle and
covariance ellipse. For vehicles whose predicted trajec-
tories are sufficiently distant, say at least 30 m, only the
positions of the Reference Points need be computed.
For intermediate distances above 5 m, the polytopes
can be replaced by bounding rectangles.
Example
ð30Þ
ð31Þ vehicles c1 and c2 (vehicle sizes as in Table 1) are
approaching each other in a 45 curve on a country
road. The center line of the road is composed of
two clothoids, each of arc length l = 100 m. The cur-
vature k of the center line as a function of the arc
length from the bottom left is

s=a2 for 0  s  l
kðsÞ ¼
ð2l  s Þ=a2 for l  s  2l
ð32Þ p
where a ¼ 2l= p. The smallest radius of curvature is
k(l)1 = 4 l/p ’ 127 m. We assume that each lane of
the road is 4 m wide and that the vehicles travel in the
center of their lane. The longitudinal velocity of
both vehicles is a constant 20 m/s = 72 km/h. At time
t = 0, c1 and c2 are at the top right and bottom left
corners, respectively.
In addition, a stationary, broken down vehicle c3 of
the same size is located at 110 m from the bottom left in
the middle of the right lane. So c2 will hit c3 in about 5.2 s
if no action is taken and we, therefore, consider the pre-
diction time interval [3.2 s, 5.2 s] in detail in this example.
The state transition equations (17) assume that the
ð33Þ unbiased yaw rate is constant in the time prediction
interval. This is only an approximation in the curve
ð34Þ where the yaw rate must change continually to follow
322 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
100
c2
80
60
y (m)
40
20
0
0 20 40 60
Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 10
Initial positions of vehicles ci for i = 1, . . ., 3 and the planned trajectories of c1 and c2
the curve precisely. An improved prediction must
either use information about the change of yaw rate
derived from a presumed vehicle-driver behavior or the
period without measurement feedback must be
reduced. This assumption is reasonable if we restrict
the prediction horizon to be below human reactivity.
The future trajectories of the vehicles are predicted
for the next 2 s given the uncertain information on their
positions, heading angle, and velocity (units: m, radian,
m/s). The covariance matrix of the state information is
assumed to be the same for all vehicles:
diag ð0:04; 0:04; 0:001; 0:1Þ
The standard deviations of the state are, therefore,
for the x or y position: 0.2 m, heading angle: 0.0316 rad
and velocity: 0.316 m/s.
Figure 11 shows the pairwise distances between the
three vehicles as calculated from (33) for the prediction
time interval [3.2 s, 5.2 s]. c1 passes the stationary c3
without danger, but the distance between c1 and c2 is
predicted to enter the danger zone at 4.8 s. As shown in
Fig. 12, this dangerous situation is due to the predicted
understeering of c1 and c2 in the curve. As expected, this
problem becomes more serious when the velocities or
curvature increase. In addition, the predicted distance
between c2 and c3 becomes dangerous at 5.1 s.
The next step for the cooperative group would be to
calculate and evaluate possible joint reactions and new
trajectories. In the example, c1 and c2 should move
c3
c1
80 100 120 140 160 180
x (m)
closer to the road borders, and c2 should decrease its
velocity to avoid both dangers. However, depending on
the parameter constellations, additional reactions are
feasible, such as c1 changing its velocity to allow c2 to
overtake c3. The development of cooperative maneu-
vers will be explored in more detail in the next section.
Cooperative Decision Making
Once a dangerous situation has been detected, the
system can either issue a warning and leave the conflict
resolution to the driver or it can intervene autono-
ð36Þ
mously by executing a cooperative maneuver.
Warning Systems
The following two questions are of particular impor-
tance for the design of collision warning systems:
1. Which is the criterion to issue a warning? Typically,
thresholds in terms of collision probability or
time to collision are defined [28, 53, 54]. It is
important to minimize false alarms because drivers
tend to ignore warnings if there are frequent false
alarms.
2. How is the warning presented to the driver? Possi-
bilities include optic, acoustic, and haptic warnings,
or combinations of these [55–57]. Ideally, the warn-
ing should induce the driver to perform an ade-
quate collision avoidance maneuver. For example,
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of 323

10
No danger Driver
9 response

8 e12 e23 e13

Vehicle distance (m)

7

ε 1
Danger
0
4 4.5 5 5.5 6
Time (s)

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 11

Predicted distances eij between the vehicles considering the uncertainty of their state. If uncertainty is not considered,
then min(e12) and min(e13) are both approx. 2.14 m

state at 4.8s
88

86

84
c1
82
c2
y (m)

80

78

76
c3

74

72
92 94 96 98 100 102 104 106 108 110 112
x (m)

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 12

Likely positions of vehicles at 4.8 s with solid rectangles for the actual positions and surrounding polytopes for the
uncertainty area. Prediction calculated at 3.2 s
324 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
vibrations of the steering wheel could indicate
a necessary steering maneuver to the driver. This
technique is used in lane departure warning systems
[58, 59].
Monitoring driver behavior is also of interest. For
example, a driver who has already initiated a collision
avoidance maneuver should not be distracted by an
alarm [53, 60]. Some of the systems specifically address
certain traffic scenarios, e.g., intersections, left-turning,
lane changes, or rear-end collisions [61, 62].
Cooperative Driving
Various approaches to realize cooperative driving are
conceivable, many of them tailored to specific traffic
situations or types of accidents. For example,
a cooperative variant of adaptive cruise control has
been developed, which increases the foresight horizon
in order to reduce the amount of necessary accelera-
tions and decelerations [11, 63]. A similar system is an
emergency braking assistant preventing rear-end colli-
sions based on cooperative information exchange [64].
Both systems only affect the longitudinal control of
the vehicle. Platooning systems additionally address the
task of lane keeping and the execution of some maneu-
vers like merging into a platoon [10, 65]. Lane assign-
ment for platoons with the aim of optimizing traffic
flow is another topic of research [12]. Collision avoid-
ance with obstacles or vehicles outside the platoon is
usually not covered by these approaches. A method
specialized on merging maneuvers is described in [38].
Cooperative intersection control systems can have
two different goals: the focus is either on collision
warning, as described in the previous section, or
on optimizing traffic flow. Systems of the latter
type usually assume fixed paths of the vehicles and
optimize order of precedence and velocity profiles
[13, 14, 66], see also section “Decoupled Planning”
below.
Cooperative control approaches keep formations of
aerial, underwater, or land vehicles, usually in off-road
exploration scenarios [67]. Few of these works address
collision avoidance among cooperative vehicles [68].
Other systems specialize on obstacle avoidance from
a single-vehicle perspective. Typically, prespecified
evasive trajectories are parameterized to fit the encoun-
tered situation [69, 70]. A system that can handle more
general situations, such as multiple obstacles, has to use
motion planning techniques [71]. Another approach
applies differential game theory to avoid the collision of
a single vehicle with a driver going the wrong way [72].
Altogether, motion planning techniques seem to be
the most general approach to cooperative driving in
dangerous situations as they can handle arbitrary
obstacles and road geometries. Therefore, motion plan-
ning is covered in more detail in the following section.
Multi-vehicle Motion Planning
In this section, methods are described which compute
a cooperative motion plan for given initial positions of
vehicles and obstacles. The plan must be feasible with
regard to the kinematic and dynamic capabilities of the
involved vehicles. Therefore, a level of abstraction
appropriate for cooperative motion planning has to
represent vehicle dynamics and geometry. Once
a feasible cooperative plan has been computed by the
group coordinator, it can be communicated to the
vehicles of the cooperative group. Within the tolerances
of the cooperative plan, each vehicle can perform a finer
planning of its own trajectory, e.g., optimizing comfort
criteria, before executing the maneuver (cf. Fig. 1).
This section starts with a formalization of the motion
planning problem, especially introducing the concept
of configuration spaces. Afterward, some motion-
planning algorithms are described. The methods can be
distinguished by the criterion if they do or do not make
decoupling assumptions in order to reduce the compu-
tational complexity.
Configuration Space Path planning and motion
planning originate from robotics, and car-like vehicles
may be regarded to be specializations of robots. An
important concept from robot motion planning is the
configuration space.
The configuration of a robot is a parametric speci-
fication of the geometric position of every point of the
robot [73]. The manifold of all possible configurations
is called the configuration space Q of the robot [74, 75].
A path is a parametric curve in the configuration space,
i.e., a continuous mapping ½0; 1 ! Q.
The relevant configuration parameters of a car-like
vehicle ci are its position xi, yi in the road plane and its
orientation (yaw angle) fi. It is common to assume
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
a sufficiently planar road so that vertical position, roll,
and pitch angle can be neglected. This results in the
three-dimensional configuration space Qi ¼ 2
,
where  is the unit circle parameterized by [0,2p).
Note that the angles 0 and 2p describe the same
orientation, which causes Qi to be a cylindrical mani-
fold [75].
The motions of a car are restricted by the orienta-
tion of its wheels, it cannot move in arbitrary direc-
tions. This is expressed by the following constraint on
incremental position changes dxi, dyi:
dxi sin fi ¼ dyi cos fi ð37Þ
Despite this constraint, all configurations in Qi
can be reached if no obstacles are present. Such
constraints which do not reduce the dimensionality
of the configuration space are called nonholonomic
[75]. The nonholomic properties of a car are of
special importance when parking maneuvers are con-
sidered. It is possible to plan a path without consider-
ing the nonholonomic constraint and to subsequently
transform it to enforce conformity with the
constraint [76].
The motion of a car is not only restricted by its
nonholonomic kinematics, but also by its dynamics.
Instantaneous accelerations or decelerations are
impossible due to inertia. A state space model can be
used to describe the vehicle kinematics and dynamics.
A simple state space model for a car-like vehicle is given
as follows,
0 1 0 1 0 1 sampling-based approaches [75], especially rapidly
x_ i vi cos fi 0 exploring random trees [77, 78], and elastic-band
B y_ i C B vi sin fi C B 0 C
B C¼B C þ B tan a C; ð38Þ
@ f_ A @ 0 A @ vi i A
i li
v_ i 0 ai
where vi denotes the longitudinal velocity, ai the accel-
eration, ai the steering angle, and li the wheelbase of the
car (cf. Fig. 13). The first summand on the right-hand
side of the equation models the state evolution, while
the second one characterizes the actuator behavior.
Acceleration and steering angle are the control inputs
of the system. They are bounded by the capabilities of
the vehicle, which depend on brakes, engine power,
steering geometry, tires, and the friction of the road
surface. A simplified model for the dynamic limits is
Kamm’s circle stating that longitudinal and lateral
acceleration ai,lon and ai,lat are limited by
325
u
a
( , )
l
f
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 13
Notation used in the vehicle model (38)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
ai;lon þ ai;lat
2  mi g; ð39Þ
where g  9.8 m/s2 is Earth’s gravitational acceleration
and mi  1 is the road-tire friction coefficient [71].
Additional bounds may be imposed on acceleration
and steering angle, e.g., ai,lon,min  ai,lon  ai,lon,max
and |ai|  ai,max.
When vehicle dynamics have to be considered, it is
usually necessary to solve a motion planning problem,
i.e., to compute a continuous mapping ½0; tmax  ! Qi .
This mapping is sometimes called a trajectory.
Unlike a path, it is explicitly parameterized by time.
Methods for vehicle motion planning include
planning [79]. Vehicles with different configuration
spaces, e.g., trucks pulling trailers, have also been con-
sidered [76].
Obstacles divide the configuration space into free
space and obstacle regions. A configuration qi 2 Qi
belongs to free space if every point of the robot in
configuration qi is collision-free with every obstacle.
For collision checking, both robot and obstacle
geometry have to be considered. The explicit construc-
tion of the free space is avoided by most practical
planning algorithms due to its algorithmic
complexity [75].
Path planning for multiple vehicles can be
performed in the composite configuration space,
which is the Cartesian product of the individual
326 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
configuration spaces [80]: Q ¼ Q1
. . .
Qm . The can perform a refinement of its individual trajectory
possibility of motion planning in the composite con-
figuration space is well known from robotics textbooks
[74, 75]. However, the computational complexity of
motion planning algorithms usually increases expo-
nentially with the dimensionality of the configuration
space [80]. If each individual configuration space Qi is
of dimension di , the composite configuration space has
P
dimension d ¼ m i¼1 di . Therefore, cooperative
motion planning in the composite configuration
space has rarely been used in practice. See [81, 82] for
some robotics examples. More common are
decoupling approaches which make simplifying
assumptions in order to lower the computational com-
plexity. However, these assumptions also restrict the
space of solutions, which might be particularly critical
in dangerous situations. Therefore, both decoupling
and cooperative planning methods are presented in
the following sections.
Altogether, the motion planning task for m vehicles
can be formalized as follows: compute a continuous
trajectory ½0; tmax  ! Q ¼ Q1
. . .
Qm which tence of the other vehicles. In a second stage of the
1. Starts at the initial states of the vehicles
2. Avoids collisions among the m cooperative vehicles,
considering their geometric models
3. Avoids collisions with both static and moving
obstacles
4. Keeps the vehicles on the road area
5. Respects the kinematic and dynamic constraints of
the vehicles
Additionally, a goal region is specified usually. This
has been omitted here because collision avoidance is
the main objective in dangerous situations.
When moving obstacles are considered, the obsta-
cle-free regions of the configuration space become
time-dependent. This can be incorporated by adding
a time axis, resulting in the configuration space-time
[75, 80]. The trajectories of moving obstacles may be
derived from prediction methods as described above in
section “Path Prediction.” The road constraint may be
dealt with by declaring all off-road areas as obstacles.
The resulting trajectory in composite configuration
space is communicated to the vehicles of the coopera-
tive group (cf. Fig. 1). Before execution, each vehicle
within the tolerances of the cooperative plan according
to more detailed models and specific criteria. In fact,
the main purpose of cooperative planning is not to find
an optimal trajectory, but to construct and select
a homotopy class of solutions in the composite config-
uration space [75]. Optimizations within the tolerances
of the correct homotopy class, such as smoothing, can
be performed by the individual vehicles.
Decoupled Planning Decoupling methods reduce
the computational complexity of multi-vehicle motion
planning by considering lower-dimensional subspaces
of the composite configuration space. Usually,
a subspace corresponding to a single vehicle is chosen.
The path-velocity decomposition is a well-known
method which decouples the composite problem into
m path planning problems and a velocity coordination
problem [83]. It uses path planning algorithms to find
individual paths in Qi for each vehicle. Each path
circumvents the static obstacles but ignores the exis-
algorithm, the relative velocities along the individual
paths are coordinated in order to avoid inter-vehicle
collisions [84]. This corresponds to a one-dimensional
reparameterization of each path. It is a path planning
problem in the m-dimensional coordination space.
Alternatively, paths may be coordinated incrementally,
resulting in m  1 two-dimensional coordination
problems. In the last stage of the algorithm, the time
scaling of the path is performed to obtain a motion
respecting the dynamic constraints. It is also possible to
perform the time scaling during the coordination stage,
which requires increasing the dimensionality of the
coordination space by one.
The path-velocity decomposition is an adequate
approximation in many robotic scenarios [85–87] and
has also been applied to road traffic problems [40]. It
shows certain parallels to resource allocation methods
which consider regions such as intersections to be
mutual exclusive resources and apply scheduling mech-
anisms to resolve conflicts [66, 88]. Other methods
weaken the restriction to a single path and allow the
coordination procedure to select a path within
a probabilistic roadmap [89, 90].
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of 327

a b

c d

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 14

Comparison of cooperative and decoupled motion planning. (a) Oncoming traffic scenario, (b) path-velocity decoupling,
(c) prioritized motion planning, (d) cooperative motion planning in the composite configuration space

The second approach to decouple multi-robot

motion planning assigns priorities to the robots
[91–95]. The robots are processed in the order of pri-
orities, which means that lower-priority robots already
know the planning results of the higher-priority ones.
The ith robot plans a motion in its configuration space-
time which considers the planned motions of the
higher-priority robots 1, . . ., i  1 as moving
obstacles but completely ignores the lower-priority
robots i + 1, . . ., m. This replaces the d-dimensional
planning problem in the composite configuration
space with m planning steps in configuration space-
times of dimension di + 1. Of course, the results depend
on the selected order of priority. Therefore, methods
for constructing somewhat optimal priorities are of
interest [96, 97]. However, prioritized planning may
be suboptimal even with the best priorities. The
approach has been applied to manipulators, mobile
robots, and vehicle-parking scenarios.
In dangerous situations involving multiple vehicles,
decoupled planners may fail to find an existing colli-
sion-free solution. This can already happen in very
simple scenarios (Fig. 14). Additionally, it is difficult
to consider vehicle dynamics in the path-velocity
decomposition and in the roadmap coordination
approach. Therefore, three recent methods for cooper-
ative planning in the composite configuration space are
presented in the following sections (Fig. 15).
Tree Search The current action of a vehicle can be
characterized by its acceleration and its steering angle,
see (38). The continuum of actions can be discretized by
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 15
Space-time visualization of a cooperative merging
maneuver. In this 3D illustration, x and y directions of the
road plane are depicted together with the time axis
pointing upward vertically. Motion sequences of three
vehicles can be seen, and a static obstacle appears as
a pillar
choosing a finite set of representatives. When an action is
applied for a certain interval of time Dt before the next
action is selected, a tree structure of decisions results.
The discrete action set should contain extreme actions
like maximum braking and maximum steering admis-
sible by the dynamics of the vehicle, given by (39). These
extreme actions are known to be optimal for a single
vehicle evading an obstacle [69, 71]. In the automobile
domain, time discretization is justified because actuator
limits prohibit a quick change of actions.
Each node ni of the tree T i corresponds to a state
xi = (xi,yi,fi,vi)T of the ith vehicle. Applying an action
328 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

acti = (ai, ai)T results in a successor state which can be t0 r

computed by integrating (38) (cf. [80]):
act10 act 10
vi ðt þ DtÞ ¼ vi ðtÞ þ ai ðtÞDt
tan ai ðtÞ
fi ðt þ DtÞ ¼ fi ðtÞ þ ðvi ðtÞþ12ai ðtÞDtÞ  Dt
li
8 act20 act20
> xi ðtÞ þ ðvi ðtÞ þ 12 ai ðtÞDtÞ  Dt  cos fi ðtÞ
>
>
< if ai ðtÞ ¼ 0 t1 n n
xi ðt þ DtÞ ¼
>
> x ðtÞ þ li
ðsinf ðt þ DtÞ  sin fi ðtÞÞ
>
:
i tan ai ðt Þ i

if ai ðtÞ 6¼ 0 act11 act11

8
> y ðtÞ þ ðv ðtÞ þ 1
a ðtÞDtÞ  Dt  sin fi ðtÞ
>
>
i i 2 i
< if ai ðtÞ ¼ 0
yi ðt þ DtÞ ¼
>
> y ðtÞ þ li
ðcosf ðt þ DtÞ  cos fi ðtÞÞ
>
:
i tan ai ðtÞ i act21 act21
if ai ðtÞ 6¼ 0 t2
ð40Þ
Cooperative Group of Vehicles and Dangerous
The tree T i consists of T + 1 levels, resulting in Situations, Recognition of. Figure 16
a planning horizon of tmax = T  Dt. It has ATi leaf Tree T of cooperative actions for m = 2 vehicles
nodes, where Ai denotes the number of actions. With
each action acti leading to tree node ni, a loss L(ni) is
incurred for the violation of constraints listed in sec- in the single vehicle trees T i and look them up if they
tion “Configuration Space.” Loss is a notion used in are needed several times. This can avoid many redun-
utility theory to determine preferences between alter- dant computations.
native actions [98]. The loss is composed additively of The single vehicle trees can also be precomputed
a term Lroad(ni) penalizing road departure, Lobst(ni) for completely as they are much smaller than T . This allows
collisions with obstacles, Lcoll(ni) for collisions among the establishment of lower bounds for the loss within
cooperative vehicles, and Lcontrol(ni) rating control a subtree. A lower bound h(ni) for a subtree of T i
effort with the aim of avoiding unnecessary braking rooted at ni is obtained by recursively taking the min-
and steering. The optimal action sequence corresponds imum values of the children of ni :
to a path from the root ri to a leaf node ni having
8
minimum accumulated loss Li . <0 if ni is a leaf
For cooperative motion planning, every action of hðni Þ :¼ min fLðni Þ þ hðni Þg otherwise
: n0 : child
one vehicle can be combined with every action of all i
of n i

other vehicles, resulting in a set of A = A1  . . .  Am ð41Þ

composite actions and a tree T having AT leaf nodes
(Fig. 16). Even for moderate values of A and T, explicit A lower bound h(n) for a node n of the cooperative
construction of this tree exceeds memory and compu- tree T is given by the sum of the values from
tational resources. Therefore, pruning techniques are the corresponding nodes ni(n) of the single-vehicle
required which allow the optimal solution to be found trees T i :
without searching the whole tree.
An important observation is that many computa- X
m
hðnÞ :¼ hðni ðnÞÞ ð42Þ
tions during traversal of T depend only on the action i¼1
sequences of a single vehicle. This holds for vehicle
states and for some loss components, namely Lobst, These lower bounds can be used for an efficient
Lcontrol, and Lroad. As a consequence, this information branch and bound search in T . It explores the tree in
occurs redundantly in T , for example, at the edges depth-first order, remembering the best solution found
highlighted in Fig. 16. It is possible to store these values so far and its loss value L∗. Subtrees are excluded from
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of 329

the search if it can be proven that they cannot contain ⎧ ⎫

⎪
⎪ ⎪
⎬
any part of the optimal solution. To this end, the sum ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
= L(r, n)
g(n): = L(r,n) + h(n) is computed for the root node n of ⎪
⎨ ⎭
the subtree, where L(r,n) is the loss accumulated on the L∗ = n ⎫
⎪
⎪ ⎪
⎬
path from the root r of T to n. If g(n) > L∗, the total ⎪
⎪
⎪
⎪
⎪ ≥ h(n)
loss of any path passing through the subtree must be ⎪
⎪ ⎪
⎭
⎩ n∗
greater than L∗. In this case, the subtree can be
excluded from the search without losing the guarantee
to find the optimum (Fig. 17). The resulting branch
Cooperative Group of Vehicles and Dangerous
and bound search method is shown in Algorithm 3.
Situations, Recognition of. Figure 17
Alternatively, it is possible to search T using the A∗
Criterion to prune a subtree
algorithm, which also relies on the lower bounds h(n)
to improve efficiency [99, 100].

Algorithm 3 General branch and bound method for

An important advantage of the tree search formula-
cooperative motion planning tion is the use of explicit vehicle, loss, and action models,
which can be chosen in the level of detail appropriate for
1: procedure SEARCH (t, n, L) the task under consideration. Precomputation of lower
2: for all act 2 A do
3: Compute child node nc, x(nc) and L(nc)
bounds significantly improves computational perfor-
4: if L + L(nc) + h(nc) < L∗ then mance of the method. By shifting most of the compu-
5: if t + Dt < tmax then tational effort into the precomputation stage, running
6: SEARCH (t + Dt, nc, L + L(nc)) times become more predictable, increasing the real-time
7: else capability of the algorithm [101]. Figures 18 and 19
8: L∗ ← L + L(nc)
show collision avoidance maneuvers planned with the
9: n ∗ ← nc
10: end if tree search method.
11: end if
12: end for Mixed-Integer Linear Programming The linear pro-
13: end procedure gramming framework allows large systems of linear
14: procedure PLAN(xstart) inequalities to be optimized efficiently. A linear pro-
15: Perform precomputation
16: L∗ ← 1
gram consists of an objective function, equality and
17: x(r) ← xstart inequality constraints, all of them being linear in the
18: SEARCH (0, r, 0) real-valued variables. An extension is mixed-integer
19: return n∗ linear programming (MILP) which constrains some
20: end procedure of the variables to be integer, or – as a special case –
binary. This makes the problem much harder.
The only remaining loss component is Lcoll, which As efficient standard software exists for solving
does not depend on the full cooperative action MILPs, cooperative motion planning has been formu-
sequence either but only on the actions of pairs of lated in this framework in recent years, with focus on
vehicles [90]. Therefore, similar techniques can be the air traffic domain [102–104]. Variables xik, yik
applied in order to avoid redundant computations denote the position of the ith vehicle at time t = k  Dt
and to obtain lower bounds. The collision information within a planning horizon k = 0, . . ., T. Further vari-
can be stored in two-vehicle trees T ij for every pair of ables are introduced for velocities and control inputs,
vehicles. There is, however, a trade-off between with appropriate inequality constraints based on vehi-
precomputation effort and savings during branch and cle dynamics. They are coupled by equations describing
bound search, and experiments show that a linearized version of a vehicle model like (38). Binary
precomputation of collision loss does not improve variables bijkl are required for the collision avoidance
overall performance in most cases [101]. constraints:
330 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

3 2
t=0 1

3 2 1
t = 1.3s

Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 18

Example of a cooperative merging maneuver planned by the tree search algorithm. While vehicle 1 changes to the left lane
due to the obstacle, vehicles 2 and 3 brake in order to allow the merging. The full planning horizon has been 2.5 s

trivially satisfied with bijkl = 1. The binary variables thus

Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 19
A cooperative collision avoidance maneuver in an
intersection scenario
xik  xjk þ bijk1 M > rij
yik  yjk þ bijk2 M > rij
xik þ xjk þ bijk3 M > rij
yik þ yjk þ bijk4 M > rij
X 4
bijkl  3
l¼1
bijkl 2 f0; 1g k ¼ 1; . . . ; T
Therein, rij is the minimum distance to be enforced
between the vehicle reference points (xik,yik)T and (xjk,
describe on which side the vehicles pass each other. It
may be necessary to discretize the space of directions
into more than the four paraxial directions of (43).
This can be accomplished by additional separating
half-planes [106]. Obstacle avoidance is modeled in
the same way, the only difference being that the
position of the second object is no longer variable.
Complex objects may be represented by a covering of
several geometric primitives.
Paralleling the tree search approach of the previous
section, MILP solvers often employ branch and bound
methods. However, in the tree search formulation, the
branching occurs over the vehicle actions and collisions
are tested depending on the selected actions, whereas
in the MILP formulation, the binary decision variables
describe collision avoidance constraints and the vehicle
actions are optimized dependent on the collision avoid-
ance decisions. As a consequence, time discretization is
tightly coupled with passing distances and cannot be
coarsened without impairing collision avoidance. By
contrast, in the tree search method, the number of
decision points can be reduced without any difficulty
ð43Þ in order to improve computational efficiency. This may
be an explanation for the observed running times of the
MILP approach, which seem to question a real-time
usage at present [107]. Another drawback of the MILP
method is the restriction to linear models and objective
functions. An important advantage is the utilization of an
established framework with available tool support, which
yjk)T at time k  Dt, and M  0 is a large constant. This
formulations makes use of a well-known modeling
technique to transform a disjunction of inequalities
into a conjunction, which is the standard semantics of
MILP constraints [105]. Only one of the first four
inequalities has to be active, while the others can be
facilitates modeling and implementation. Related work
includes pure linear programming applied to air traffic
control [108] and nonlinear optimization [109].
Elastic Bands Different variants of the elastic-band
method have been applied to path adaptation for
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
coop
f2,6
coop
f2,7
X1,0 dist
f1,4
coop
f1,4
coop
f1,5
Cooperative Group of Vehicles and Dangerous Situations, Recognition of. Figure 20
Visualization of the elastic bands of two oncoming vehicles. Arrows indicate some selected force components
mobile robots [110–112] and to path planning for cars
[79, 113–115]. Recently, the method has been extended
to cooperative motion planning in dangerous situa-
tions [116].
Once again, nodes xi,k = (xi,k, yi,k)T describe the
vehicle positions at discrete points in time. Consecutive
nodes of one vehicle are connected to form an elastic
band. Orientation and velocity of the vehicle are
implicitly represented by the sequence of nodes.
Virtual forces are designed to represent the con-
straints described above in section “Configuration
Space.” The resulting elastic band can be interpreted
as a combination of point masses and springs [114].
The virtual forces show some similarities with potential
field path planners as forces can be interpreted as
derivatives of potential fields. However, potential field
planners only have a local view of the environment
[80], while elastic bands allow collision avoidance
forces to propagate along the band. Internal forces
can be designed to represent the dynamic and kine-
matic constraints of the vehicle.
The total force vector applied to node xi,k is denoted
by fi,k. It is additively composed of several force terms.
The internal forces f dist and f ang implement the
dynamic and kinematic constraints of the vehicle in
longitudinal and lateral direction, respectively. They
correspond to springs between consecutive nodes of
one band, xi,k and xi,k + 1.
External forces repel the band from obstacles, other
vehicles, and the road boundary. For example, the force
f coop acts between nodes of different bands having the
same time index, xi,k and xj,k. The elastic-band frame-
work uses a circular object model. Other object shapes
331
coop
f2,5
coop
f2,4
road X2,0
f1,7
coop
f1,7 ang
f1,9
coop
f1,6
Cooperative Group of Vehicles and Dangerous
Situations, Recognition of. Figure 21
Cooperative obstacle avoidance maneuver planned by the
elastic-band method
are approximated by a covering of circles [113].
A visualization of elastic bands and virtual forces is
shown in Fig. 20.
The nodes of an elastic band have to be initialized in
a meaningful way, e.g., corresponding to a lane-keeping
behavior of the vehicle. The virtual forces are applied to
the band iteratively until an equilibrium is reached in
which the resulting forces vanish for every node:
8i 8k f i;k  0. A continuous state and control trajec-
tory is obtained from the discrete sequence of nodes
using, e.g., cubic spline interpolation [113].
The elastic-band method has been successfully
applied to obstacle avoidance, merging, overtaking,
and intersection scenarios. Figure 21 shows a coopera-
tive maneuver planned by the elastic-band method,
and Fig. 15 shows a space-time visualization of a merging
scenario. Elastic bands exhibit potential for real-time
implementation and a good scalability with the number
of cooperative vehicles, even when a comparatively fine
time discretization is chosen. However, the method can
only find a local optimum, unlike the tree search and
MILP approaches. This means that the result depends on
the initialization of the elastic bands.
332 Cooperative Group of Vehicles and Dangerous Situations, Recognition of

Future Directions vehicles. The evaluation of cooperative maneuvers is

treated in [128].
Many of the methods described in this entry are still at
In the field of automatic maneuvers, a promising
the research stage. They have been verified in simula-
approach to guarantee the safety of a planned motion is
tions or experiments but are not yet ready for market
the concept of inevitable collision states and the partial
deployment. The time to market may be quite long,
motion planner [78]. This is a scheme of iterated plan-
e.g., as shown by platooning technologies that were
ning and execution. The plans selected are guaranteed
demonstrated back in 1997 [117], but a commercial
not to induce a state in which a collision becomes
implementation of such systems is still pending.
inevitable. At least a collision-free contingency maneu-
A major unsolved problem is how to deal with the
ver remains possible [129]. This principle has been
inevitable uncertainty in the environment information
extended to cooperative motion planning [130]. How-
caused by sensor-based perception. On the one hand,
ever, it is not directly applicable to dangerous situations
this can be alleviated by improving sensor technology,
because the assumption of an initial safe state does not
processing algorithms, and multi-sensor fusion. In the
hold there.
context of cooperating vehicles, the integration of per-
The group formation and decision algorithms
ceptions made by multiple vehicles is of particular
described in this entry rely on a designated group
interest. Research on appropriate fusion architectures
coordinator vehicle. Future research should investigate
is in progress [8, 118]; however, to the best of the
to what extent decision making can be distributed. This
authors’ knowledge, cooperative perception has
could result in a better reliability and better distribu-
not yet been implemented in a real-world scenario.
tion of the computational load. Consensus algorithms
On the other hand, situation recognition and planning
or game theoretic methods might possibly be extended
algorithms can be designed which take the uncer-
to handle distributed decision making in dangerous
tainties into account. For situation recognition and
traffic situations [67, 131–133].
prediction, the methods described in this entry can
Regarding wireless vehicular communication, stan-
already cope with uncertain information to a certain
dardization of protocols and bandwidth allocation are
degree. Planning under uncertainty is a very difficult
important prerequisites [134]. Field tests of wireless
problem that is only partially solved. Statistical deci-
vehicular communication and cooperative local danger
sion theory offers a framework to determine the best
warning are currently under way [135]. The US
action under uncertain information [98]. In principle,
Department of Transportation is soliciting new ideas
this approach is equivalent to Bayesian decision net-
on this subject [136]. In Europe, a cooperative driving
works, which establish a connection between situation
challenge has been initiated [137]. Several teams want
assessment and decision making [119, 120]. However,
to show improved driving capabilities based on coop-
its application to cooperative motion planning causes
erative information exchange. This could initiate fur-
a dramatic increase in computational complexity. For
ther research and development in this area, similar to
single vehicles and robots, there are several approaches
the impulse generated by the DARPA Grand Challenges
to considering model based probabilistic uncertainties
and Urban Challenge for autonomous vehicles [3].
[121–124]. A related, but different objective in motion
As real-world testing of cooperative maneuvers is
planning is to reduce the uncertainty of vehicle locali-
a safety-critical task, suitable testing methodologies
zation [125, 126]. Cooperative motion planning
have to be found. Urmson et al. [138] gives a general
with uncertain environment information is still an
overview of testing. For single-vehicle driver assistance,
unsolved problem. The problem complexity may
hardware-in-the-loop simulation and driving tests
require a hierarchical planning, increasing the level of
with mock-up obstacles have been applied [139, 140].
detail step-wise.
Cooperative maneuvers could be tested by starting with
Regarding prediction methods, a comparative eval-
only one real vehicle and several simulated vehicles, but
uation of the numerous approaches would be a topic
including real communication. In order to ensure the
for useful future work. [127] describes a prediction
safety of a system in real use, an enormous number of
method that considers the interaction of several
experiments under a wide range of conditions is
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
required. Since particular dangerous situations may be
rare in real traffic, it is difficult to collect sufficient data.
The rate and handling of false alarms is another key
issue. A major related difficulty is the development of
appropriate certification procedures for vehicles and
their systems [141]. Aspects such as user acceptance
and legal regulations are still to be addressed [142].
These issues will have to be resolved before cooper-
ative technologies may develop their potential to
increase traffic safety and efficiency.
Acknowledgments
The authors’ research has been partially supported by
Deutsche Forschungsgemeinschaft (German Research
Foundation) within the Transregional Collaborative
Research Center 28 “Cognitive Automobiles.”
Bibliography
1. Vahidi A, Eskandarian A (2003) Research advances in intelli-
gent collision avoidance and adaptive cruise control. IEEE
Trans Intell Transp Syst 4(3):143–153
2. Tideman M, van der Voort MC, van Arem B, Tillema F (2007)
A review of lateral driver support systems. In: Proceedings
of the IEEE intelligent transportation systems conference,
Seattle, Washington, pp 992–999
3. Buehler M, Iagnemma K, Singh S (eds) (2009) The DARPA
urban challenge – autonomous vehicles in city traffic. In:
Springer tracts in advanced robotics, vol 56. Springer, Berlin
4. Hartenstein H, Laberteaux K (2008) A tutorial survey
on vehicular ad hoc networks. IEEE Commun Mag 46(6):
164–171
5. Schulze M, Nöcker G, Böhm K (2005) PReVENT: a European
program to improve active safety. In: Proceedings of the 5th
international conference on intelligent transportation systems
telecommunications, Brest
6. Adler C, Straßberger M (2006) Putting together the pieces –
a comprehensive view on cooperative local danger warning.
In: Proceedings of the world congress on intelligent transport
systems and services, London
7. Dao T-S, Ng L, Clark C, Huissoon JP (2008) Realtime experi-
ments in Markov-based lane position estimation using wire-
less ad-hoc network. In: Proceedings of the IEEE intelligent
vehicles symposium, Eindhoven, pp 901–906
8. Tischler K, Vogt H (2007) A sensor data fusion approach for the
integration of negative information. In: Proceedings of confer-
ence on information fusion, Quebec
9. Huang D, Leung H (2005) An expectation-maximization-based
interacting multiple model approach for cooperative driving
systems. IEEE Trans Intell Transp Syst 6(2):206–228
10. Varaiya P (1993) Smart cars on smart roads: problems of
control. IEEE Trans Autom Control 38(2):195–207
333
11. Zambou N, Enning M, Abel D (2004) Nonlinear spacing control
of following vehicles within platoon – a controlled Lagrangian
approach. In: IFAC symposium on nonlinear control systems,
Stuttgart, pp 931–936
12. Dao T-S, Clark C, Huissoon JP (2008) Distributed platoon
assignment and lane selection for traffic flow optimization.
In: Proceedings of the IEEE intelligent vehicles symposium,
Eindhoven, pp 739–744
13. Li L, Wang F-Y (2006) Cooperative driving at blind crossings
using intervehicle communication. IEEE Trans Veh Technol
55(6):1712–1724
14. Schepperle H, Böhm K, Forster S (2007) Traffic management
based on negotiations between vehicles – a feasibility dem-
onstration using agents. In: Workshop on agent mediated
electronic commerce, Honolulu
15. Roughgarden T (2005) Selfish routing and the price of anarchy.
MIT Press, Cambridge, MA
16. Frese C, Beyerer J, Zimmer P (2007) Cooperation of cars and
formation of cooperative groups. In: Proceedings of the IEEE
intelligent vehicles symposium, Istanbul, pp 227–232
17. Frese C, Batz T, Wieser M, Beyerer J (2008) Life cycle manage-
ment for cooperative groups of cognitive automobiles in
a distributed environment. In: Proceedings of the IEEE intelli-
gent vehicles symposium, Eindhoven
18. Schultes D (2008) Route planning in road networks. PhD
thesis, Universität Karlsruhe (TH)
19. Cormen T, Leiserson C, Rivest R (2001) Introduction to algo-
rithms, 2nd edn. MIT Press, Cambridge, MA
20. Vacek S, Nagel R, Batz T, Moosmann F, Dillmann R (2007) An
integrated simulation framework for cognitive automobiles.
In: Proceedings of the IEEE intelligent vehicles symposium,
Istanbul, Türkei, pp 221–226
21. Munaka T, Yamamoto T, Watanabe T (2005) A reliable
advanced-join system for data multicasting in ITS networks.
IEEE Trans Intell Transp Syst 6(4):424–438
22. Chen W, Lee J, Hikita T, Onishi R (2007) Embedded
multicasting with vehicle local peer group for efficient vehicle
communications. In: Proceedings of the IEEE workshop on
vehicle-to-vehicle communications, Istanbul
23. Mocito J, Rodrigues L (2006) Reconfigurable architecture for
group communication support in hybrid networks. In: Koldehofe
B (ed) Proceedings of the 4th MiNEMA workshop, Sintra
24. Miller J (2008) Vehicle-to-vehicle-to-infrastructure (V2V2I) intel-
ligent transportation system architecture. In: Proceedings of the
IEEE intelligent vehicles symposium, Eindhoven, pp 715–720
25. Huang Q, Julien C, Roman G-C (2004) Relying on safe distance
to achieve strong partitionable group membership in ad hoc
networks. IEEE Trans Mob Comput 3(2):192–205
26. Shimura A, Sakaibara T, Hiraiwa M, Aizono T (2003) Proposal of
an autonomous group-management model and its applica-
tion to intelligent transport system. In: Proceedings of the
6th international symposium on autonomous decentralized
systems, Pisa
27. Fujii H, Akiyama M, Tokuda K (1999) Inter-vehicle communica-
tions protocol for group cooperative driving. In: Proceedings
334 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
of the IEEE vehicular technology conference, Amsterdam, The
Netherlands
28. Huang J, Tan H-S (2006) Design and implementation of a
cooperative collision warning system. In: Proceedings of the
IEEE intelligent transportation systems conference, Toronto
29. Clark C (2004) Dynamic robot networks: a coordination plat-
form for multi-robot systems. PhD thesis, Stanford University
30. Premvuti S, Yuta S (1990) Consideration on the cooperation of
multiple autonomous mobile robots. In: Proceedings of the
IEEE/RSJ conference on intelligent robots and systems, Ibaraki
31. Briesemeister L (2001) Group membership and communica-
tion in highly mobile ad hoc networks. Dissertation,
Technische Universität Berlin
32. Pallottino L, Scordio V, Bicchi A (2004) Decentralized cooper-
ative conflict resolution among multiple autonomous mobile
agents. In: Proceedings of the IEEE conference on decision and
control, Atlantis, Paradise Island
33. Ragothaman V, Baloch F, Pendse R (2006) Unassisted aircraft
landing via co-operative data exchange. In: Digital avionics
systems conference, Portland
34. Chisalita I, Shahmehri N (2004) A context-based vehicular
communication protocol. In: Proceedings of the IEEE sympo-
sium on personal, indoor and mobile radio communications,
Barcelona
35. Nett E, Schemmer S (2003) Reliable real-time communication
in cooperative mobile applications. IEEE Trans Comput
52(2):166–180
36. Meier R, Killijian M-O, Cunningham R, Cahill V (2001) Towards
proximity group communication. In: Middleware for mobile
computing, Heidelberg
37. Dao T-S, Clark C, Huissoon JP (2007) Optimized lane
assignment using inter-vehicle communication. In:
Proceedings of the IEEE intelligent vehicles symposium,
Istanbul
38. Li L, Wang F-Y, Zhang Y (2007) Cooperative driving at lane
closures. In: Proceedings of the IEEE intelligent vehicles sym-
posium, Istanbul
39. Haruna T, Okada Y, Shigeno H (2007) A construction method
of road-static network with vehicular ad-hoc networks. In:
Proceedings of the IEEE workshop on vehicle-to-vehicle com-
munications, Istanbul
40. Sheng W, Yang Q, Guo Y (2006) Experimental testbed and
distributed algorithm for cooperative driving in VII simulation.
In: Proceedings of the IEEE intelligent transportation systems
conference, Toronto
41. Judaschke U (1994) Verfahren zur kollisionsfreien Führung
und Koordination mobiler Transportsysteme. PhD thesis,
Universität Dortmund
42. Chiang Y-J, Klosowski J, Lee C, Mitchell J (1997) Geometric
algorithms for conflict detection/resolution in air traffic man-
agement. In: Proceedings of the IEEE conference on decision
and control, San Diego
43. Zhou M (2007) A traffic data exchange protocol in vehicular
ad-hoc network. In: Proceedings of the international work-
shop on intelligent transportation, Hamburg
44. Khan MA, Bölöni L (2005) Convoy driving through ad-hoc
coalition formation. In: Proceedings of the IEEE real time and
embedded technology and applications symposium, San
Francisco, pp 98–105
45. Hallé S, Chaib-draa B (2004) Collaborative driving system using
teamwork for platoon formations. In: Proceedings of work-
shop on agents in traffic and transportation, New York
46. Batz T, Watson K, Beyerer J (2009) Recognition of dangerous
situations within a cooperative group of vehicles. In: Proceed-
ings of IEEE intelligent vehicles symposium, Xi’an, China,
pp 907–912
47. Tan H-S, Huang J (2006) A low-order DGPS-based vehicle
positioning system under urban environment. IEEE ASME
Trans Mechatron 11(5):567–575
48. Tan H-S, Huang J (2006) DGPS-based vehicle-to-vehicle coop-
erative collision warning: engineering feasibility viewpoints.
IEEE Trans Intell Transp Syst 7(4):415–428
49. Wan EA, Van Der Merwe R (2001) Chapter 7: the unscented
kalman filter. In: Kalman filtering and neural networks. Wiley,
New York, pp 221–280
50. Caveney D (2007) Stochastic path prediction using the
unscented transform with numerical integration. In: Proceed-
ings of the 2007 IEEE intelligent transportation systems con-
ference, Seattle, 30 Sept–3 Oct 2007
51. Kvasnica M, Grieder P, Baotić M (2004) Multi-parametric tool-
box (MPT). http://control.ee.ethz.ch/~mpt/
52. Löfberg J (2004) YALMIP: a toolbox for modeling and optimi-
zation in MATLAB, http://control.ee.ethz.ch/~joloef/yalmip.
php. In: Proceedings of the CACSD conference, Taipei,
Taiwan
53. Miller R, Huang Q (2002) An adaptive peer-to-peer collision
warning system. In: Proceedings of the IEEE vehicular technol-
ogy conference, Birmingham, pp 317–321
54. van den Broek T, Ploeg J (2010) Collision warning system
based on probability density functions. In: Proceedings of
the International workshop on intelligent transportation,
Hamburg, pp 141–147
55. Misener JA, Sengupta R, Krishnan H (2005) Cooperative colli-
sion warning: enabling crash avoidance with wireless technol-
ogy. In: Proceedings of the world congress on intelligent
transport systems and services, San Francisco
56. Benmimoun A, Chen J, Suzuki T (2007) Design and practical
evaluation of an intersection assistant in real world tests. In:
Proceedings of the IEEE intelligent vehicles symposium, Istanbul
57. Chen J, Deutschle S, Fuerstenberg K (2007) Evaluation
methods and results of the INTERSAFE intersection assistants.
In: Proceedings of the IEEE intelligent vehicles symposium,
Istanbul, pp 142–147
58. Onken R (1994) DAISY, an adaptive, knowledge-based driver
monitoring and warning system. In: Proceedings of the vehicle
navigation & information systems conference, Yokohama,
pp 3–10
59. Ishida S, Gayko J (2004) Development, evaluation and introduc-
tion of a lane keeping assistance system. In: Proceedings of the
IEEE intelligent vehicles symposium, Parma, pp 943–944
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
60. McCall J, Wipf D, Trivedi M, Rao B (2005) Lane change intent
analysis using robust operators and sparse bayesian learning.
In: IEEE/CVPR workshop on machine vision for intelligent vehi-
cles, San Diego
61. Zhang J, Roessler B (2009) Situation analysis and adaptive risk
assessment for intersection safety systems in advanced
assisted driving. In: Autonome mobile system, Karlsruhe,
pp 249–258
62. ElBatt T, Goel S, Holland G, Krishnan H, Parikh J (2006) Coop-
erative collision warning using dedicated short range wireless
communications. In: ACM VANET, Los Angeles
63. Morsink P, Hallouzi R, Dagli I, Cseh C, Schäfers L, Nelisse M, de
Bruin D (2003) CarTALK 2000: development of a cooperative
ADAS based on vehicle-to-vehicle communication. In: Pro-
ceedings of the world congress on intelligent transport sys-
tems and services, Madrid
64. Girard AR, de Sousa JB, Misener JA, Hedrick JK (2001) A control
architecture for integrated cooperative cruise control and
collision warning systems. In: Proceedings of the IEEE confer-
ence on decision and control, Orlando
65. Uno A, Sakaguchi T, Tsugawa S (1999) A merging control
algorithm based on inter-vehicle communication. In: Proceed-
ings of the IEEE intelligent transportation systems conference,
Tokyo, pp 783–787
66. Bruns T, Trächtler A (2009) Intersection management: trajec-
tory planning by means of dynamic programming. Automati-
sierungstechnik 57(5):253–261
67. Murray R (2007) Recent research in cooperative control of
multivehicle systems. ASME J Dyn Syst Meas Control 129:571–583
68. Dimarogonas D, Kyriakopoulos K (2005) A feedback stabiliza-
tion and collision avoidance scheme for multiple independent
nonholonomic non-point agents. In: Proceedings of the 2005
IEEE international symposium on intelligent control limassol,
Cyprus, pp 820–825
69. Hillenbrand J, Spieker A, Kroschel K (2006) A multilevel colli-
sion mitigation approach – its situation assessment, decision
making, and performance tradeoffs. IEEE Trans Intell Transp
Syst 7(4):528–540
70. Isermann R, Schorn M, Stählin U (2008) Anticollision system
PRORETA with automatic braking and steering. Veh Syst Dyn
46:683–694
71. Schmidt C, Oechsle F, Branz W (2006) Research on trajectory
planning in emergency situations with multiple objects. In:
Proceedings of the IEEE intelligent transportation systems
conference, Toronto, pp 988–992
72. Lachner R, Breitner M, Pesch HJ (2000) Real-time collision
avoidance: differential game, numerical solution, and synthesis
of strategies. In: Filar J, Gaitsgory V, Mizukami K (eds) Advances
in dynamic games and applications. Birkhäuser, Boston
73. Lozano-Pérez T (1983) Spatial planning: a configuration space
approach. IEEE Trans Comput 32(2):108–120
74. Choset H, Lynch K, Hutchinson S, Kantor G, Burgard W, Kavraki L,
Thrun S (2005) Principles of robot motion. MIT Press, Cambridge
75. LaValle S (2006) Planning algorithms. Cambridge University
Press, Cambridge, UK
335
76. Laumond J-P, Sekhavat S, Lamiraux F (1998) Guidelines
in nonholonomic motion planning for mobile robots. In:
Laumond J-P (ed) Robot motion planning and control. Springer,
New York
77. LaValle S, Kuffner J (1999) Randomized kinodynamic planning.
In: Proceedings of the IEEE conference on robotics and auto-
mation, Detroit, pp 473–479
78. Petti S, Fraichard T (2005) Safe motion planning in dynamic
environments. In: Proceedings of the IEEE/RSJ conference
on intelligent robots and systems, Edmonton, pp 3726–3731
79. Hesse T, Sattel T (2007) An approach to integrate vehicle
dynamics in motion planning for advanced driver assistance
systems. In: Proceedings of the IEEE intelligent vehicles sym-
posium, Istanbul
80. Latombe J-C (1991) Robot motion planning. Kluwer,
Boston, MA
81. Sánchez G, Latombe J-C (2002) On delaying collision checking
in PRM planning: application to multi-robot coordination. Int
J Robot Res 21(1):5–26
82. Clark C, Rock S, Latombe J-C (2003) Motion planning for
multiple mobile robots using dynamic networks. In: Proceed-
ings of the IEEE conference on robotics and automation,
Taipei, pp 4222–4227
83. Kant K, Zucker S (1986) Toward efficient trajectory plan-
ning: the path-velocity decomposition. Int J Robot Res
5(3):72–89
84. O´Donnell P, Lozano-Pérez T (1989) Deadlock-free and collision-
free coordination of two robot manipulators. In: Proceedings of
the IEEE conference on robotics and automation, Scottsdale
85. Leroy S, Laumond J-P, Siméon T (1999) Multiple path
coordination for mobile robots: a geometric algorithm.
In: Proceedings of the international joint conference on artifi-
cial intelligence, Stockholm
86. Ghrist R, LaValle S (2006) Nonpositive curvature and pareto
optimal coordination of robots. SIAM J Control Optim
45(5):1697–1713
87. Peng J, Akella S (2005) Coordinating multiple robots with
kinodynamic constraints along specified paths. Int J Robot
Res 24(4):295–310
88. Alami R, Fleury S, Herrb M, Ingrand F, Robert F (1998) Multi-
robot cooperation in the MARTHA project. IEEE Rob Autom
Mag 5(1):36–47
89. Švestka P, Overmars M (1998) Coordinated path planning for
multiple robots. Rob Autom Syst 23:125–152
90. LaValle S, Hutchinson S (1998) Optimal motion planning for
multiple robots having independent goals. IEEE Trans Robot
Automation 14(6):912–925
91. Erdmann M, Lozano-Pérez T (1987) On multiple moving
objects. Algorithmica 2:477–521
92. Freund E, Hoyer H (1988) Real-time pathfinding in multirobot
systems including obstacle avoidance. Int J Robot Res
7(1):42–70
93. Warren CW (1990) Multiple robot path coordination using
artificial potential fields. In: Proceedings of the IEEE conference
on robotics and automation, Cincinnati
336 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
94. Azarm K, Schmidt G (1996) A decentralized approach for the
conflict-free motion of multiple mobile robots. In: Proceed-
ings of the IEEE/RSJ conference on intelligent robots and
systems, Osaka
95. Bekris K, Tsianos K, Kavraki L (2007) A decentralized planner
that guarantees the safety of communicating vehicles with
complex dynamics that replan online. In: Proceedings of
the IEEE/RSJ conference on intelligent robots and systems,
San Diego, pp 3784–3790
96. Bennewitz M, Burgard W, Thrun S (2001) Optimizing sched-
ules for prioritized path planning of multi-robot systems.
In: Proceedings of the IEEE conference on robotics and auto-
mation, Seoul
97. van den Berg J, Overmars M (2005) Prioritized motion plan-
ning for multiple robots. In: Proceedings of the IEEE/RSJ
conference on intelligent robots and systems, Edmonton
98. Berger JO (1993) Statistical decision theory and Bayesian
analysis, Springer series in statistics. Springer, New York
99. Hart P, Nilsson N, Raphael B (1968) A formal basis for the
heuristic determination of minimum cost paths. IEEE Trans
Syst Sci Cybern 4(2):100–107
100. Russell S, Norvig P (2003) Artificial intelligence: a modern
approach. Prentice Hall, New York
101. Frese C, Beyerer J (2010) Planning cooperative motions
of cognitive automobiles using tree search algorithms. In:
Dillmann R, Beyerer J, Hanebeck UD, Schultz T (eds) KI
2010: advances in artificial intelligence. Lecture notes in
artificial intelligence, vol 6359. Springer, Heidelberg,
pp 91–98
102. Schouwenaars T, De Moor B, Feron E, How J (2001) Mixed
integer programming for multi-vehicle path planning. In:
Proceedings of the European control conference, Porto,
pp 2603–2608
103. Borrelli F, Subramanian D, Raghunathan A, Biegler L (2006)
MILP and NLP techniques for centralized trajectory planning
of multiple unmanned air vehicles. In: Proceedings of the
American control conference, Minneapolis
104. Richards A, How J (2002) Aircraft trajectory planning
with collision avoidance using mixed integer linear program-
ming. In: Proceedings American control conference,
Anchorage
105. Sierksma G (1996) Linear and integer programming: theory
and practice. Dekker, New York
106. Earl M, D’Andrea R (2007) Multi-vehicle cooperative control
using mixed integer linear programming. In: Shamma J (ed)
Cooperative control of distributed multi-agent systems.
Wiley, New York, pp 233–259
107. Richards A, How J (2004) Decentralized model predictive
control of cooperating UAVs. In: Proceedings of the IEEE
conference on decision and control, Atlantis
108. Niedringhaus W (1995) Stream option manager (SOM): auto-
mated integration of aircraft separation, merging, stream
management, and other air traffic control functions. IEEE
Trans Syst Man Cybern 25(9):1269–1280
109. İnalhan G, Stipanović D, Tomlin C (2002) Decentralized opti-
mization, with application to multiple aircraft coordination.
In: Proceedings of the IEEE conference on decision and con-
trol, Las Vegas
110. Quinlan S, Khatib O (1993) Elastic bands: connecting path
planning and control. In: Proceedings of the IEEE conference
on robotics and automation, Atlanta
111. Brock O (1999) Generating robot motion: the integration of
planning and execution. PhD thesis, Stanford University
112. Khatib M, Jaouni H, Chatila R, Laumond J-P (1997) Dynamic
path modification for car-like nonholonomic mobile robots.
In: Proceedings of the IEEE conference on robotics and auto-
mation, Albuquerque, pp 2920–2925
113. Hilgert J, Hirsch K, Bertram T, Hiller M (2003) Emergency path
planning for autonomous vehicles using elastic band theory.
In: Proceedings of the IEEE/ASME conference on advanced
intelligent mechatronics, Kobe
114. Brandt T, Sattel T, Wallaschek J (2005) On automatic collision
avoidance systems. In: Proceedings of SAE world congress,
Detroit
115. Gehrig S, Stein F (2007) Collision avoidance for
vehicle-following systems. IEEE Trans Intell Transp Syst
8(2):233–244
116. Frese C, Batz T, Beyerer J (2008) Cooperative behavior of
groups of cognitive automobiles based on a common rele-
vant picture. Automatisierungstechnik 56(12):644–652
117. Thorpe C, Jochem T, Pomerleau D (1997) The 1997 auto-
mated highway free agent demonstration. In: Proceedings
of the IEEE intelligent transportation systems conference,
Pittsburgh, pp 496–501
118. Demmel S, Gruyer D, Rakotonirainy A (2010) V2V/V2I aug-
mented maps: state-of-the-art and contribution to real-time
crash risk assessment. In: Proceedings of the 20th Canadian
multidisciplinary road safety conference, Niagara Falls,
Ontario
119. Jensen FV (2001) Bayesian networks and decision graphs.
Springer, New York
120. Forbes J, Huang T, Kanazawa K, Russell S (1995) The
BATmobile: towards a bayesian automated taxi. In: Proceed-
ings of the international joint conference on artificial intelli-
gence, Montreal, pp 1878–1885
121. Fraichard T, Mermond R (1998) Path planning with uncer-
tainty for car-like robots. In: Proceedings IEEE conference on
robotics and automation, Leuven, pp 27–32
122. Gonzalez JP, Stentz A (2005) Planning with uncertainty in
position: an optimal and efficient planner. In: Proceedings
of the IEEE/RSJ conference on intelligent robots and systems,
Edmonton, pp 2435–2442
123. Melchior N, Simmons R (2007) Particle RRT for path planning
with uncertainty. In: Proceedings of the IEEE conference on
robotics and automation, Roma, pp 1617–1624
124. Toussaint M, Goerick C (2007) Probabilistic inference for
structured planning in robotics. In: Proceedings of the IEEE/
RSJ conference on intelligent robots and systems, San Diego
 Cooperative Group of Vehicles and Dangerous Situations, Recognition of
125. Lambert A, Bouaziz S, Reynaud R (2003) Shortest safe path
planning for vehicles. In: Proceedings of the IEEE intelligent
vehicles symposium, Columbus, OH
126. Gonzalez JP, Stentz A (2007) Planning with uncertainty in posi-
tion using high-resolution maps. In: Proceedings of the IEEE
conference on robotics and automation, Roma, pp 1015–1022
127. Broadhurst A, Baker S, Kanade T (2005) Monte Carlo road
safety reasoning. In: Proceedings of the IEEE intelligent vehi-
cles symposium, Las Vegas, pp 319–324
128. Althoff M, Althoff D, Wollherr D, Buss M (2010) Safety verifi-
cation of autonomous vehicles for coordinated evasive
maneuvers. In: Proceedings of the IEEE intelligent vehicles
symposium, San Diego, pp 1078–1083
129. Bekris K, Kavraki L (2007) Greedy but safe replanning under
kinodynamic constraints. In: Proceedings of the IEEE confer-
ence on robotics and automation, Rome, pp 704–710
130. Bekris K, Tsianos K, Kavraki L (2007) A distributed protocol for
safe real-time planning of communicating vehicles with sec-
ond-order dynamics. In: Proceedings of the conference on
robot communication and coordination, Athens
131. Olfati-Saber R, Fax JA, Murray RM (2007) Consensus and
cooperation in networked multi-agent systems. Proc IEEE
95(1):215–233
132. Georgiev D, Kabamba P, Tilbury D (2008) A new model for
team optimization: the effects of uncertainty on interaction.
IEEE Trans Syst Man Cybern A 38(6):1234–1247
133. Garg D, Narahari Y (2008) Mechanism design for single leader
Stackelberg problems and application to procurement auc-
tion design. IEEE Trans Autom Sci Eng 5(3):377
337
134. Kenney J (2010) Standards and regulations. In: Hartenstein H,
Laberteaux K (eds) VANET: vehicular applications and inter-
networking technologies, Intelligent transportation systems.
Wiley, Chichester, pp 365–429
135. Stübing H, Bechler M, Heussner D, May T, Radusch I, Rechner
H, Vogel P (2010) simTD: a car-to-X system architecture for
field operational tests. IEEE Commun Mag 48(5):148–154
136. US Department of Transportation (2011) Connected vehicle
technology challenge. http://connectedvehicle.challenge.gov/
137. Grand cooperative driving challenge. http://www.gcdc.net/
138. Urmson C, Whittaker WR, Harbaugh S, Clark M, Koon P (2006)
Testing driver skill for high-speed autonomous vehicles.
Computer 39(12):48–51
139. Christen F, Sandkühler D, Benmimoun A, Breuer K
(2004) Verwendung des Verkehrsflusssimulationswerkzeugs
PELOPS mit HiL-Funktionalität bei der Entwicklung von
Fahrerassistenzsystemen. In: VDI Kongress Berechnung und
Simulation im Fahrzeugbau
140. Jansson J, Johansson J, Gustafsson F (2002) Decision making
for collision avoidance systems. In: Society of automotive
engineers technical paper
141. Winner H, Wolf G (2009) Quo vadis, FAS? In: Winner H,
Hakuli S, Wolf G (eds) Handbuch Fahrerassistenzsysteme –
Grundlagen, Komponenten und Systeme für aktive
Sicherheit und Komfort, Vieweg + Teubner, pp 664–673
142. Wahl R, Tørset T, Vaa T (2007) Large scale introduction of
automated transport: which legal and administrative barriers
are present? In: Proceedings of the world congress on intel-
ligent transport systems and services, Beijing
338 Driver Assistance System, Biologically Inspired

Driver Assistance System, Vertical (normal) tire forces They are responsible to
support the weight of the vehicle.
Biologically Inspired
MOUSTAPHA DOUMIATI1, DANIEL LECHNER2 Definition of the Subject
1
B2i Automotive Engineering Company, Massy, France
2 The principal concerns in driving safety with standard
Department of Accident Mechanism Analysis,
vehicles are understanding and preventing risky situa-
IFSTTAR-MA Laboratory, Salon de Provence, France
tions. A close examination of accident data reveals that
losing the vehicle control is the main reason for most
Article Outline
car accidents. To help the driver to prevent such acci-
Glossary dents, vehicle control systems may be used. For their
Definition of the Subject optimal operation, these control systems require cer-
Introduction tain input data concerning vehicle dynamic parameters
Estimation Process Description and vehicle–road interaction. Unfortunately, some fun-
Four-Wheel Vehicle Model damental parameters like the tire-road forces and the
Tire–Road Interface sideslip angle are difficult to measure in a car, for both
Observer’s Design technical and economic reasons. To face this problem,
Experimental Results this entry presents a dynamic modeling and observation
Conclusion and Future Perspectives method to estimate these variables. The ability to accu-
Acknowledgments rately estimate lateral tire forces and sideslip angle is
Bibliography a critical determinant in the performances of many
vehicle control systems. To address nonlinearities and
Glossary unmodeled vehicle dynamics, an observer derived from
Lateral tire forces They are responsible to hold on the unscented Kalman filtering technique is proposed. The
vehicle during a turn. estimation process method is based on the dynamic
Longitudinal tire forces They are responsible to response of a vehicle instrumented with easily available
accelerate/brake the vehicle. and potentially integrable sensors. Performances are
Observer or estimator It models a real system in tested using an experimental car in real driving situa-
order to provide an estimate of its internal state, tions. Experimental results show the potential of the
given measurements of the input and output of the proposed estimation method.
real system.
Sideslip angle It is the angle between the velocity Introduction
heading and the true heading of the vehicle. Vehicle dynamics and stability have been of consider-
Tire forces The developed forces (longitudinal and able interest to automotive engineers, automobile
lateral) are function of tire properties (material, manufacturers, government, public safety groups, and
tread pattern, tread depth, profile, etc.), the normal general public for a number of years. The obvious
load on the tire, and the velocities experienced by dilemma is that people naturally desire to drive faster
the tire. and faster on the roads and highways, yet they expect
Vehicle control systems They provide commands their vehicles to be stable and safe during all normal
and instructions to control the movements of the and emergency maneuvers. For the most part,
vehicle in order to maintain stability and enhance people pay little attention to the limited handling
passengers security and comfort. potential of their vehicles until some unusual behavior
Vehicle dynamics It includes analytical and experi- is observed that often results in fatality. According to
mental technology used to study and understand statistics, worldwide, an estimated 1.2 million people
the dynamical responses of a vehicle in various in- are killed in road crashes each year and as many as
motion situations. 50 million are injured [1]. Preventing car accidents

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3

requires to know what determines vehicle dynamics
during motion [2].
Today, automotive electronic technologies are
developing for safe and comfortable traveling of drivers
and passengers. Nowadays, there are a lot of vehicle
control system such as the Anti-lock Braking System
(ABS) that prevents wheel lock during braking [3], and
the Electronic Stability Control (ESC) that enhances
lateral vehicle stability [4, 5]. These control systems
installation rate is increasing all around the world.
Vehicle control algorithms have made great strides
toward improving the handling and safety of vehicles.
For example, experts estimate that ESC prevents 27%
of loss-of-control accidents by intervening when emer-
gency situations are detected [6]. While nowadays
vehicle control algorithms are undoubtedly a life-
saving technology, they are limited by the available
vehicle state information.
Vehicle control systems currently available on pro-
duction cars rely on available inexpensive measure-
ments such as longitudinal velocity, accelerations, and
the vehicle yaw rate. Sideslip rate can be evaluated
using the yaw rate, lateral acceleration, and vehicle
velocity [7]. Calculating the sideslip angle is possible
from the sideslip rate integration. However, it is prone
to uncertainty and errors from sensor bias. Besides,
these control systems use unsophisticated, inaccurate
tire models to evaluate lateral tire dynamics. In fact,
measuring tire forces and sideslip angles is very difficult
for technical and economic reasons. Therefore, these
important data must be observed or estimated. If con-
trol systems were in possession of the complete set of
lateral tire characteristics, namely, lateral forces, side-
slip angle, and the tire–road friction coefficient, they
could greatly enhance vehicle handling and increase
passenger safety.
As the motion of a vehicle is governed by the forces
generated between the tires and the road, knowledge of
the tire forces is crucial when predicting vehicle
motion. For example, a vehicle can turn because of
the applied lateral tire forces. In fact, what happens is
that when the front wheels of a vehicle are steered, a slip
angle is created, which gives rise to a lateral force. This
lateral force turns or yaws the vehicle. Under normal
driving situations (low slip angle), a vehicle responds
predictably to the driver’s inputs. As the vehicle
approaches the handling limits, for example, during
Driver Assistance System, Biologically Inspired 339
an evasive emergency maneuver, or when a vehicle
undergoes high accelerations, high slip angle occurs
and the vehicle’s dynamic becomes highly nonlinear
and its response becomes less predictable and poten-
tially very dangerous.
Accurate data about tire forces lead to a better eval-
uation of the vehicle possible trajectories, to a better
vehicle control and rollover prevention. Moreover, it
enables the development of a diagnostic tool for eval-
uating the potential risks of accidents related to poor
adherence or dangerous maneuvers.
In the literature, many studies have looked at the
vehicle dynamic states estimation. Several ones have
been conducted regarding the estimation of tire-road
forces [8–20]. For example, in [8], a study of
a 14-degrees-of-freedom (DOF) vehicle model is pro-
posed where the dynamics of the roll center are used to
calculate vertical tire forces. In [9], the authors propose
an estimation method in order to estimate vertical and
lateral forces per axle. The authors in [10–13] estimate
the vehicle vertical forces and other dynamic states for
a four-wheel vehicle model (FWVM) comprising four
DOF. Consequently, lateral tire forces at each tire are
calculated based on the estimated states and using
a quasi-static tire model. In [14], Ray estimates the
vehicle dynamic states and lateral tire forces per axle
for a nine-DOF vehicle model. The author uses
measurements of the applied torques as inputs to his
model. We note that the torque is difficult to get
in practice; it requires expensive sensors. More
recently, authors in [15, 16] propose observers to esti-
mate lateral forces per axle without using torque mea-
sures. In [17], the authors propose an estimation
process based on a three-DOF vehicle model, as
a lateral tire force estimator. In [15–17], lateral
forces are modeled with a derivative equal to random
noise. The authors in [17] remark that such modeling
leads to a noticeable inaccuracy when estimating
individual lateral tire forces, but not in axle lateral
forces. This phenomenon is due to the non-
representation of the lateral load transfer when model-
ing. Studies in [21–24] focus on the tire-road friction
estimation.
The main goal of this entry is to present an estimation
method that uses simple vehicle-road models and a certain
number of valid measurements in order to estimate in real
time and in accurate way the sideslip angle and the lateral
340 Driver Assistance System, Biologically Inspired

force at each individual tire–road contact point. This entry Suspension

Δij
presents two significant particularities: deflection
sensors Block 1

1. First, the estimation process does not use the mea- . Roll plane model +
surements of wheel torques which are very θ coupling of the
ax, aym longitudinal/lateral
expensive. Inertial
dynamics
2. Second, the estimation process uses accurate nor- sensors

mal tire forces, in contrast to many existing .

ψ
Fzij a x, a y
approaches that assume constant vertical forces.
This approach is more realistic since during
cornering, accelerating, and braking, the load dis- Block 2
Magnetic Wij
tribution varies significantly in a car, thus cornering
sensors Four wheel vehicle
stiffness and lateral forces evaluation are directly
model + Dugoff tire
affected. model
δij
The developed estimation process is model based Optical
sensor
and built using Kalman filter technique. The Kalman
filter is known as the most commonly used real-time Fyij β
estimator for linear and nonlinear systems. In order to
Driver Assistance System, Biologically Inspired. Figure 1
show the effectiveness of the estimation method, some
Process estimation diagram
real-time validation tests were carried out on an
instrumented vehicle in realistic driving situations.

Estimation Process Description
The estimation process is shown in its entirety by the
block diagram in Fig. 1, where ax and aym are respec-
tively the longitudinal and lateral accelerations, c_ is the
yaw rate, y_ is the roll rate, Dij (i represents the front 1 or
the rear 2 and j represents the left 1 or the right 2) is the
suspension deflection, wij is the wheel velocity, Fzij and
Fyij are respectively the normal and lateral tire-road
forces, b is the sideslip angle at the center of gravity
(cog). The whole estimation process consists of two
blocks, and its role is to estimate sideslip angle at the
cog, normal and lateral forces at each tire–road contact
point, and consequently evaluate the used lateral fric-
tion coefficient. The following measurements are
needed:
● Yaw and roll rates measured by gyrometers
● Longitudinal and lateral accelerations measured by
accelerometers
● Suspension deflections using suspension deflec-
tions sensors
● Steering angle measured by an optical sensor
● Rotational velocity for each wheel given by mag-
netic sensors
The first block aims to provide the vehicle’s mass,
lateral load transfer, normal tire forces, and the
corrected lateral acceleration ay (by canceling the grav-
itational acceleration component that distorts the
accelerometer signal aym). It contains observers based
on vehicle’s roll dynamics and model that couples lon-
gitudinal and lateral accelerations. Authors have looked
at the first block in some previous studies [10, 11]. This
work focuses only on the second block, whose main
role is to estimate individual lateral tire force and
sideslip angle. The second block makes use of the
estimations provided by the first block. In fact, as will
be shown in the sections, the impact of including
accurate normal forces in the calculation of lateral
forces is fundamental.
One specificity of this estimation process is the use
of blocks in series. By using cascaded observers, the
observability problems entailed by an inappropriate
use of the complete modeling equations are avoided,
enabling the estimation process to be carried out in
a simple and practical way. In the following, the model-
based observer of the second block will be explained in
details. Therefore, the vehicle-road model and the esti-
mation method will be illustrated, respectively.
 Driver Assistance System, Biologically Inspired 341

Four-Wheel Vehicle Model estimation problem in the absence of longitudinal

forces first and include them in later studies. This can
The Four-Wheel Vehicle model (FWVM) is chosen for
be done by focusing on solving the estimation problem
this entry because it is simple and corresponds suffi-
when the vehicle is driven at constant speeds [20].
ciently to our objectives. The FWVM is widely used to
This entry extended the hypothesis of moving in
describe transversal vehicle dynamic behavior [12–14,
a constant speed and addresses the case of a front-wheel
17, 18].
drive, where rear longitudinal forces are neglected rel-
Figure 2 shows a simple diagram of the FWVM
ative to the front longitudinal forces. Longitudinal
model in the longitudinal and lateral planes. In order
front axle forces are considered by assuming that:
to simplify the lateral and longitudinal dynamics,
rolling resistance is neglected. Additionally, the front Fx1 ¼ Fx11 þ Fx12 : ð1Þ
and rear track widths (E) are assumed to be equal. L1
and L2 represent the distance from the vehicle’s center The longitudinal force evolution is modeled with
of gravity to the front and rear axles respectively. The a random walk model, where its derivative is equal to
sideslip at the vehicle center of gravity (b) is the differ- random noise F_ x1 ¼ 0. This is due to the lack of knowl-
ence between the velocity heading (Vg) and the true edge on the longitudinal slip and the effective radius of
heading of the vehicle (c). The yaw rate (c) _ is the the tire [25].
angular velocity of the vehicle about the center of The lateral dynamics of the vehicle can be obtained
gravity. The forward and lateral velocities are by summing the forces and moments about the
respectively V and U. The longitudinal and lateral vehicle’s center of gravity. Consequently, the simplified
forces (Fx,y,i,j) are shown for front and rear tires of the FWVM is formulated as the following dynamic rela-
vehicle. tionships [19]:
Longitudinal forces should be taken into account to 8 " #
>
> 1 Fx1 cosðb  dÞ þ Fy11 sinðb  dÞþ
enable accurate lateral forces estimation during vehicle > _
> Vg ¼ ;
>
>
braking or acceleration. While considering their effect >
> m Fy12 sinðb  dÞ þ ðFy21 þ Fy22 Þ sin b
>
>
>
> 2 3
is certainly important, its inclusion makes solving the >
> L1 ½Fy11 cos d þ Fy12 cos d þ Fx1 sin d
>
>
lateral estimation problem considerably more com- >
> 6 7
>
>
> €¼ 16
c L ½F þ Fy22 þ
6 2 y21
7
7;
plex. Thus, it may be desirable to solve the lateral >
> Iz 4 5
>
< E
½Fy11 sin d  Fy12 sin d
> 2
" #
>
>
>
> 1  Fx1 sinðb  dÞ þ Fy11 cosðb  dÞþ
> _
>b ¼ _
 c;
δ11
>
> mVg Fy12 cosðb  dÞ þ ðFy21 þ Fy22 Þ cos b
>
>
>
>
α11 >
> 1
>
> a ¼ ½F cos d þ Fy12 cos d þ ðFy21 þ Fy22 Þ þ Fx1 sin d;
Fy11 Fx11 >
> y m y11
>
>
y E >
> 1
>
: ax ¼ ½Fy11 sin d  Fy12 sin d þ Fx1 cosd;
x m
z
δ12 ð2Þ
Vg α12
Fy21 Fx21 U Fx12 where m is the vehicle mass, and Iz is the yaw moment
α21 β V
. Fy12 of inertia.
ψ
The tire slip angle (aij) as shown in Fig. 2, is the
difference between the tire’s longitudinal axis and the
Fy22 tire’s velocity vector. The tire velocity vector can be
Fx22 α22
L1 obtained from the vehicle’s velocity (at the cog) and
the yaw rate. Assuming that rear steering angles are
L2 approximately null, the direction or heading of the
rear tires is the same as that of the vehicle. The heading
Driver Assistance System, Biologically Inspired. Figure 2 of the front tires includes the steering angle (d).
Four-wheel vehicle model The front steering angles are assumed to be equal
342 Driver Assistance System, Biologically Inspired

(d11 = d12 = d). The forward velocity V, steering angle d, where Cyij is the cornering stiffness, aij is the slip angle,
yaw rate c,_ and the vehicle body slip angle b are then and f (l) is given by:
used to calculate the tire slip angles aij, where: 
8 " # ð2  lÞl; if l < 1
> V b þ L1 c_ f ðlÞ ¼ ð5Þ
>
> a11 ¼ d  arctan 1; if l < 1
>
> ;
>
> _
V  E c=2
>
> " #
>
> mFzij
>
> V b þ L _
c l¼ : ð6Þ
>
> a ¼ d  arctan
1
; 2Cyij jtanaij j
>
< 12 _
V þ E c=2
" # ð3Þ In the above formulation, m is the friction coefficient
>
> a ¼  arctan V b  L2 c_ ;
>
> and Fzij is the vertical load on the tire. This simplified
>
> 21
_
>
> V  E c=2
>
> " #
tire model assumes pure slip conditions with negligible
>
>
>
>
> V b  L2 c_ longitudinal slip, a uniform pressure distribution,
>
: a22 ¼  arctan :
_
V þ E c=2 a rigid tire carcass, and a constant friction coefficient
for sliding rubber. The original Dugoff tire model has
Tire–Road Interface a constant stiffness in respect to weight transfer. It is
worthy to note that, according to [28], load transfer
As the motion of a vehicle is governed by the forces
affects the cornering stiffness Cyij. It can be represented
generated between the tires and the road, knowledge of
by a second order polynomial with respect to the nor-
the tire forces is crucial in order to predict the vehicle’s
mal force as shown in equation (7):
motion. This section presents the tire–road interaction
phenomenon, especially the lateral tire forces. Since the  
quality of the observer largely depends on the accuracy Cyij ðFz Þ ¼ aFzij  bFzij2 ; ð7Þ
of the tire model, the underlying model must be pre-
cise. Taking real-time calculation requirement, the tire where a and b are respectively the first and second order
model should also be simple. coefficient in the cornering stiffness polynomial. They
are identified once using a set of experimental data
treated offline, where the tires remain in their linear
Dugoff Tire Model operation zone. Therefore, equation (7) is compared to
Many different tire models, based on the physical the calculated cornering stiffness obtained from the
nature of the tire and/or on empirical formulations ratio of the measured lateral force to the measured
deriving from experimental data, can be found in the tire’s slip angle. Hence, a and b are calculated, in such
literature. These models include the Burckhardt, a way that the equation (7) fits the calculated cornering
Dugoff, and Pacejka models [19, 26, 27]. One of the stiffness well.
most commonly used model is the Pacejka’s “Magic This entry proposes a modified Dugoff tire model,
Formula.” It is an effective method for predicting real where the cornering stiffness varies with respect to load.
tire behavior. However, it requires a large number of As shown in equation (4), vertical forces and the
tire-specific parameters that are usually unknown. tire slip angles can be used to find the lateral force on
Another commonly used model is the Dugoff tire each tire. Figure 3, based on the modified Dugoff ’s tire
model. It synthesizes all the tire property parameters model, is a graph of the lateral force versus tire slip
into two constants Cx and Cy, referred to as the longi- angle. It will be noted that as the load increases, the
tudinal and cornering stiffness of the tire. Dugoff ’s peak lateral force occurs at somehow higher slip angle.
model is the one used in this entry. Neglecting longi- It is clear that for small slip angles the force profile can
tudinal forces, the simplified nonlinear lateral tire be defined by a linear region. When operating in this
forces are given by: region, avehicle responds predictably to the driver’s inputs.
As the slip angle continues to grow, the tire begins
Fyij ¼ Cyij tanaij f ðlÞ; ð4Þ to saturate and reaches a peak value; this area is
 Driver Assistance System, Biologically Inspired 343

9000 Load: Fz=10000 N.

8000
Load: Fz=8000 N.
7000
Lateral force: - Fy (N)
6000 Load: Fz=6000 N.

5000
Load: Fz=4000 N.
4000

3000

2000

1000

0
0 5 10 15 20
Sideslip angle (°)

Driver Assistance System, Biologically Inspired. Figure 3

Generic tire curve: lateral force versus slip angle

commonly called the nonlinear region of the tire curve.
It represents the tire limits and it is rarely reached
under normal driving conditions. If the front tires
saturate first, the vehicle is said to display understeer,
and may plow out of a bend. If the rear tires saturate
first, the vehicle limit oversteers and may spin out.
Because most drivers are not accustomed to operate
in the nonlinear handling regime, both of these
responses are potentially very dangerous. Noting that
oversteer situation is much more difficult to be con-
trolled than the understeer [29].
The input vector, U, comprises the steering angle
and the normal forces considered estimated by the first
block (see Fig. 1):
U ¼ ½d; Fz11 ; Fz12 ; Fz21 ; Fz22 T
ð9Þ
¼ ½u1 ; u2 ; u3 ; u4 ; u5 T :
The measure vector, Y, comprises yaw rate, vehicle
velocity (approximated by the mean of the rear wheel
velocities calculated from wheel-encoder data), and
longitudinal and lateral accelerations:

Observer’s Design
This section presents a description of the observer
devoted to lateral tire forces and sideslip angle. The
state-space formulation, the observability analysis,
and the estimation method will be presented.
_ Vg ; ax ; ay T ¼ ½y1 ; y2 ; y3 ; y4 T :
Y ¼ ½c;
ð10Þ
The state vector, X, comprises yaw rate, vehicle
velocity, sideslip angle at the cog, lateral forces, and
the sum of the front longitudinal tire forces:
_ Vg ; b; Fy11 ; Fy12 ; Fy21 ; Fy22 ; Fx1 T
X ¼ ½c;

Stochastic State-Space Representation ¼ ½x1 ; x2 ; x3 ; x4 ; x5 ; x6 ; x7 ; x8 T :

ð11Þ
The nonlinear stochastic state-space representation of
the system described in previous section is given as: Authors would like to emphasize that the consider-
( ation of the lateral forces as states allows:
_
XðtÞ ¼ f ðXðtÞ; U ðtÞÞ þ wðtÞ
ð8Þ 1. A better evaluation of the tire forces. In fact, what-
Y ðtÞ ¼ hðXðtÞ; U ðtÞÞ þ vðtÞ
ever the complexity of the tire models, there are
344 Driver Assistance System, Biologically Inspired

several reasons why such models do not match The observation function h(.) is:
the actual tire forces perfectly [30]. From these 8
reasons, we can cite especially the changes in the >
> h1 ¼ x1 ;
>
>
tire’s pressure and temperature and the changes in >
> h2 ¼ x2 ;
>
>
the road characteristics. Therefore, authors believe >
< 1
that according to the closed loop observer theory, h3 ¼ ½x4 sin u1  x5 sin u1 þ x8 cos u1 ;
> m
>
>
the integration of the tire forces in the state vector >
> ¼
1
½x4 cos u1 þ x5 cos u1 þ ðx6 þ x7 Þ
>
> h4
may lead to better results than just using an open >
> m
:
loop tire model. þ x6 sin u1 :
2. A better understanding of the tire behaviors using ð13Þ
the relaxation-length formulation, especially in
transient maneuvers [31]. The state vector X (t) will be estimated by applying
3. The forces reconstruction to be done robustly with the unscented Kalman filter technique.
respect to some parameter variations. In fact, it is
well known that the Kalman filters have proven to
Observability
be robust to parameter changes.
Observability is a measure of how well the internal
Taking these observations in mind, one can infer
states of a system can be inferred from knowledge of
the contribution of this entry with respect to other
its inputs and external outputs. This property is often
existing studies in the literature like [12, 13], which
presented as a rank condition on the observability
estimate dynamic variables of the vehicle, and then
matrix. Using the nonlinear state-space formulation
assess the tire forces using a properly adjusted tire
of the system presented in section “Stochastic State-
model.
Space Representation,” the observability definition is
The process and measurements noise vectors,
local and uses the Lie derivative [32]. An observability
respectively w and v, are assumed to be white, zero
analysis of this system was undertaken in [33]. It was
mean and uncorrelated.
shown that the system is observable except when:
Consequently, the particular nonlinear function
f(.) of the state equations is given by: ● Steering angles are null
8 2 3 ● The vehicle is at rest (Vg = 0)
>
> L1 ½x4 cos u1 þ x5 cos u1 þ x8 sin u1 
>
>
16 7
> For these situations, we assume that lateral forces
>
> 6 L2 ½x6 þ x7 þ 7
>
> f1 ¼ 6 7; and sideslip angle are null, which approximately
>
> Iz 4 5
>
> E
>
> ½x4 sin u1  x5 sin u1  corresponds to the real cases.
>
>
>
> "2 #
>
>
>
>
> f2 1 x8 cosðx3  u1 Þ þ x4 sinðx3  u1 Þþ
>
> ¼ ; Estimation Method: Kalman Filter Algorithms
>
> m x5 sinðx3  u1 Þ þ ðx6 þ x7 Þ sinðx3 Þ
>
> " #
>
> The aim of an observer or a virtual sensor is to estimate
>
> 1  x8 sinðx3  u1 Þ þ x4 cosðx3  u1 Þ
<f ¼  x1 ;
3 a particular unmeasurable variable from available mea-
mx2 þ x5 cosðx3  u1 Þ þ ðx6 þ x7 Þ cos x3
>
> surements and a system model in a closed loop
>
>
> x2  
>
> f4 ¼ x4 þ Fy11 ða11 ; u2 Þ ; observation scheme, as illustrated in Fig. 4. Because of
>
> s2
>
> x2   the vehicle system-model mismatches (unmodeled
>
>
>
> f5 ¼ x5 þ Fy12 ða12 ; u3 Þ ;
>
> s2 dynamics, parameter variations, etc.) and the presence
>
> x2  
>
> of unknown and unmeasurable disturbances, the cal-
>
> f6 ¼ x6 þ Fy21 ða21 ; u4 Þ ;
>
> s2 culation obtained from vehicle’s model would deviate
>
> x2  
>
>
>
> f7 ¼ x7 þ Fy22 ða22 ; u5 Þ ; from the actual values over time. In order to reduce the
>
> s2
>
: estimation error, at least some of the measured outputs
f8 ¼ 0:
are compared to the same variables estimated by the
ð12Þ observer. The difference is fed back into the observer
 Driver Assistance System, Biologically Inspired 345

Measurements
Inputs

+
X=f(X,U) h(X,U) −
sensors
evolution
+
+ K
X
correction Kalman Gain

observer

Driver Assistance System, Biologically Inspired. Figure 4

Process estimation diagram

after being multiplied by a gain matrix K, and so we The UKF can be formulated as follows [37–40]:
have a closed loop observer (see Fig. 4). All developed Initialization
observers are implemented in a first-order Euler (
X 0 ¼ E½X0 
approximation discrete form. At each iteration, the ð15Þ
state vector is first calculated according to the evolution P0 ¼ E½ðX0  X 0 ÞðX0  X
 0 ÞT 
equation and then corrected online with the measure-
where X 0 and P0 are respectively the initial state and the
ment errors (innovation) and filter gain K in a recursive
prediction-correction mechanism. In this study, the initial covariance.
gain is calculated using the Kalman filter method Sigma points calculation and time update
8 h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
which is a set of mathematical equations and is widely >  k1 ; X k1  ðn þ lÞPk1
>
> w k1 ¼ X
represented in [34–36]. >
>
>
>
The EKF (Extended Kalman Filter) is probably the >
> wkjk1 ¼ f ðwk1 ; Uk1 Þ
>
> X2n
most commonly used estimator for nonlinear systems. >
>
>
> xkjk1 ¼ w m wi;kjk1
>
> i¼0 i
However, in this study the UKF (Unscented Kalman >
> X2n  
>
> Pkjk1 ¼  
Filter) [38–40] is chosen for the following fundamental < w
i¼0 i
c
w i;kjk1  X kjk1
reasons:  T
>
>  kjk1 þ Q
>
>  gi;kjk1  X
● The high nonlinearities of the vehicle-road model >
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii
>
> h
● The calculation complexity of the Jacobian matrices >
> ¼  k1 ; X
 k1  ðn þ lÞPkjk1
>
> w kjk1 X
>
>
which causes implementation difficulties >
>
>
> gkjk1 ¼ hðwkjk1 Þ
>
> X2n
>
: Y
kjk1 ¼ w mg
i¼0 i i;kjk1
UKF Algorithm In this subsection, the principle of
the UKF is summarized. Consider the general discrete ð16Þ
nonlinear system: where
(
Xkþ1 ¼ f ðXk ; Uk Þ þ wk 8
ð14Þ >
>
l
> w0 ¼
m
Yk ¼ hðXk Þ þ vk >
>
> n þ l
>
>
where Xk 2 Rn is the state vector, Uk 2 Rr is the known >
< wc ¼ l
þ ðn  a2 þ bÞ
input vector, Yk 2 Rm is the output vector at time k.
0
nþl ð17Þ
>
>
>
>
1
> wi ¼ wi ¼ 2ðn þ lÞ i ¼ 1; . . . ; 2n
m c
wk and vk are, respectively, the disturbance and sensor- >
>
>
noise vector, which are assumed to be Gaussian white >
:
noise with zero mean and uncorrelated. l ¼ nða2  1Þ
346 Driver Assistance System, Biologically Inspired

Measurement update Experimental Results

8 X2n In this section, the experimental car used to evaluate
>
> P Y k Y k ¼ w c ðg  Y kjk1 Þ
>
> i¼0 i i;kjk1 the observer performances is presented. Moreover, the
>
>
>
>
>  ðgi;kjk1  Y kjk1 ÞT þ R test conditions and the results of the previously devel-
>
> X2n
>
>  kjk1 Þ oped observers are discussed and analyzed.
>
< PX k Y k ¼ i¼0 i
w c ðwi;kjk1  X
 ðgi;kjk1  Y kjk1 ÞT ð18Þ
>
>
>
> Experimental Car
>
> Kk ¼ PX k Y k PY1
>
>  Y
>
>
k k
The experimental vehicle shown in Fig. 4 is the
>
> ¼  Kk PY k Y k KkT
>
> Pk Pkjk1
: INRETS-MA (Institut National de Recherche sur les
k ¼ X
X  kjk1 þ Kk ðYk  Y kjk1 Þ
Transports et leur Sécurité – Département Mécanismes
d’Accidents) Laboratory’s test vehicle [7]. It is
k and Ykjk1
where the variables are defined as follows: X
a Peugeot 307 equipped with:
are the estimations respectively of the state and of the
real measurement at each instant k. wi is a set of scalar ● Gyrometers and accelerometers that measure
weights, and n is the state dimension; the parameter a respectively the rotations (roll, pitch, and yaw
determines the spread of the sigma points around X  rates) and accelerations (longitudinal, lateral, and
and is usually set to 1e  4 < a < 1. The constant b is vertical) of the car body
used to incorporate part of the prior knowledge of the ● Suspension sensors that measure the distances
distribution of X, and for Gaussian distributions, b = 2 between the wheels and the car body
is optimal. Q and R are respectively the disturbance and ● Three Correvit non-contact optical sensors:
sensor-noise covariance: R takes into account the 1. One is located in chassis rear overhanging posi-
uncertainty in the measured data and Q is tuned tion, and it measures longitudinal and lateral
depending on the model quality. Remember that the vehicle speeds.
computation of the Kalman gain is a subtle mix 2. The other two are installed on the front right
between process and observation noises. The less and rear right tires and they measure front and
noise in the operation compared to the uncertainty in rear tires’ longitudinal/lateral velocities and
the model, the more the variables will be adapted to sideslip angles.
follow measurements. Since the lateral forces are ● Dynamometric wheels fitted on all four tires, which
modeled using a relaxation model based on reliable are able to measure tire forces and wheel torques in
tire models, the uncertainty affected to them is not and around all three dimensions
too high. However, the longitudinal force per ● Steering-rack displacement sensor that is used to
front axle is not modeled at all; hence, it is represented determine the steering angle
by a high noise level. The other states (yaw rate, ● Magnetic sensors that measure rotational velocity
longitudinal and lateral vehicle velocity) are modeled for each wheel
using the vehicle’s equations. Therefore, they are
It is important to note that the Correvit and the
said to have an average noise. On the other hand,
wheel-force transducers (see Fig. 5) are very expensive
since the embedded sensors have good accuracy,
sensors. They are used in this study as a reference
the noises on the measurements are quite small.
for validating the estimation process. The sampling
In order to reduce the complexity of the problem,
frequency of the different sensors is 100 Hz.
both measurement covariance matrix, R, and the
process covariance matrix, Q, are assumed to be con-
Test Conditions
stant and diagonal. The off-diagonal elements are
set to 0. That means that both the measurement Test data from nominal as well as adverse driving con-
noises and the process noises are supposed ditions were used to assess the performance of the
uncorrelated. observer presented in section “Observers Design,” in

Wheel force transducer
Driver Assistance System, Biologically Inspired. Figure 5
Wheel-force transducer and sideslip sensor installed at the tire level
realistic driving situations. We report a lane-change
maneuver where the dynamic contributions play an
important role. Figure 6 presents the Peugeot’s trajec-
tory (on a dry road), its speed, steering angle, and “g-g”
acceleration diagram during the course of the test. The
acceleration diagram, which determines the maneuver-
ing area utilized by the driver/vehicle, shows that large
lateral accelerations were obtained (absolute value up
to 0.6 g). This means that the experimental vehicle was
put in a critical driving situation.
The estimation process algorithm was written in
C++ and has been integrated into the laboratory car
as a DLL (Dynamic Link Library) that functions
according to the software acquisition system.
Validation of Observers
The observer results are presented in two forms: as
tables of normalized errors and as figures comparing
Driver Assistance System, Biologically Inspired 347
Correvit
the measurements and the estimations. The normal-
ized error for an estimation z is defined as:
k zobs  zmeasured k
Ez ¼ 100  ð19Þ
max ðk zmeasured kÞ
where zobs is the variable calculated by the observer,
zmeasured is the measured variable, and max(||zmeasured||)
is the absolute maximum value of the measured vari-
able during the test maneuver.
Figures 7 and 8 show lateral forces on the front and
rear wheels. According to these plots, the observers are
relatively good with respect to measurements. Some
small differences during the trajectory are to be noted.
These might be explained by neglected geometrical
parameters, especially the camber angles, which also
produce a lateral forces component [41]. It is also
shown that the lateral forces on the right-hand tires
exceed those on the left-hand tires. This result is clearly
348 Driver Assistance System, Biologically Inspired

200

110

Speed (km.h1)
Y position (m)

100
100

0 90

80
−100

0 500 1000 0 20 40
X position (m) Time (s)

0.04
0.1

Longitudinal acceleration (g)

Steering angle (rad)

0.02 0.05

0 0

−0.05
−0.02

−0.1
−0.04
0 20 40 −0.4 −0.2 0 0.2 0.4 0.6
Time(s) Lateral acceleration (g)

Driver Assistance System, Biologically Inspired. Figure 6

Experimental test: vehicle trajectory, speed, steering angle, and acceleration diagrams for the lane-change test

a consequence of the load transfer produced
during cornering from the left to the right-hand side
of the vehicle. In fact, lateral force increases as normal
force increases. Figure 9 shows how sideslip angle
changes during the test. Reported results are relatively
good.
Table 1 presents maximum absolute values, nor-
malized mean errors, and normalized std (standard
deviations) for lateral tire forces and sideslip angles.
Despite the simplicity of the model, we can deduce that
for this test, the performance of the observer is satis-
factory, with normalized error globally less than 8%.
Given the vertical and lateral tire forces at each tire–
road contact level, the estimation process is able to
evaluate the used lateral friction coefficient m. This is
defined as the ratio of friction force to normal force,
and is given by [29, 41]:
Fyij
mij ¼ ð20Þ
Fzij
The lateral friction coefficients in Fig. 10 show that
the estimated mij are close to the measured values.
A closer investigation reveals that the used lateral fric-
tion coefficients m12 and m22 corresponding to the
overloaded tires during cornering are lower than m11
and m21. This phenomenon is due to the tire load
sensitivity effect: the lateral friction coefficient is nor-
mally higher for the lighter loads, or conversely, falls off
as the load increases [29, 41].
 Driver Assistance System, Biologically Inspired 349

Front left lateral force Fy11 (N)

2000
measurement
1000 Estimation

−1000

−2000
0 10 20 30 40 50

Front right lateral force Fy12 (N)

4000

2000

−2000
0 10 20 30 40 50
Time (s)

Driver Assistance System, Biologically Inspired. Figure 7

Estimation of front lateral tire forces

Rear left lateral force Fy21 (N)

measurement
1000
Estimation

−1000

0 10 20 30 40 50

Rear right lateral force Fy22 (N)

3000

2000

1000

−1000
0 10 20 30 40 50
Time (s)

Driver Assistance System, Biologically Inspired. Figure 8

Estimation of rear lateral tire forces
350 Driver Assistance System, Biologically Inspired

Sideslip angle at the cog (rad)

0.01
measurement
Estimation
0.005

−0.005

−0.01

−0.015

−0.02

−0.025

10 20 30 40 50 60
Time (s)

Driver Assistance System, Biologically Inspired. Figure 9

Estimation of the sideslip angle at the cog

Driver Assistance System, Biologically Inspired. Table 1 stability and the risk of leaving the road. Consequently,
Maximum absolute values, normalized mean errors, and the developed observer could feed control systems
normalized std with fundamental vehicle-dynamics data in order to
enhance vehicle safety.
Max || || Mean % Std %
The proposed observer is derived from a simplified
Fy11 2180 (N) 8.23 4.80
four-wheel vehicle model and is based on unscented
Fy12 4070 (N) 3.70 3.74 Kalman filtering technique. Tire–road interaction is
Fy21 1441 (N) 7.51 3.52 represented by the Dugoff model. A comparison with
Fy22 2889 (N) 1.91 1.77 real experimental data demonstrates the potential of the
estimation process. It is shown that it may be possible
b 0.027 (rad) 8.32 6.41
to replace expensive correvit and dynamometric hub
sensors by real-time software observers. This is one of
This test also demonstrates that m11 and m21 are the important results of this report. Another important
high, especially for lateral accelerations up to 0.6, result concerns the estimation of individual lateral forces
and that they attain the limit for the dry road acting on each tire. This can be seen as an advance with
friction coefficient. In fact, dry road surfaces show a respect to the current vehicle-dynamics literature.
high friction coefficient in the range 0.9–1.2 (implying Future studies will improve the vehicle-road
that driving on these surfaces is safe), which means that model in order to widen validity domains for the
for this test the limits of handling were reached. observer (take into account road irregularities and
road bank angle), and make it adaptive with the road
Conclusion and Future Perspectives
conditions (especially the road friction). Moreover, it
This entry presents an interesting method for estimat- will be of major importance to study the effect of
ing lateral tire forces and sideslip angle, that is to say coupling longitudinal/lateral dynamics on lateral tire
two of the most important parameters affecting vehicle behavior.
 Driver Assistance System, Biologically Inspired 351

μ11 μ12
1
0.5

0.5
0

0
−0.5

0 20 40 0 20 40
Time (s) Time (s)
μ21 μ22
1 0.5

0.5
0

0
−0.5

0 20 40 0 20 40
Time (s) Time (s)

Driver Assistance System, Biologically Inspired. Figure 10

Used lateral friction coefficients developed by the tires

Acknowledgments of the 6th international symposium on advanced vehicle con-

This work was done at Heudiasyc Laboratory UMR
CNRS 6599, UTC University (Compiègne, France), in
collaboration with Alessandro Victorino and Ali
Charara.
Bibliography
Primary Literature
1. FARS (2007) Fatality analysis reporting system (FARS). Techni-
cal report, National Highway Traffic Safety Administration,
National Center for Statistics and Analysis
2. Aparicio F, Paez P, Moreno F, Jimenez F, Lopez A (2005)
Discussion of a new adaptive speed control system incorpo-
rating the geometric characteristics of the road International.
J Veh Auton Syst 3(1):47–64
3. SAE Society of Automotive Engineers (2004) Bosch automo-
tive handbook, 6th edn. SAE Society of Automotive Engineers,
Warrendale
4. ESC (2003) Electronic stability control fact sheet. Technical
report, ESC Coalition and DEKRA Automotive research
5. Van Zanten A (2002) Evolution of electronic control systems
for improving the vehicle dynamic behavior. In: Proceedings
trol (AVEC), Hiroshima, pp 7–15
6. Hsu Y, Gerdes JC (2005) Stabilization of a steerby-wire
vehicle at the limits of handling using feedback
linearization. In: Proceedings of IMECE 2005, AMSE interna-
tional mechanical engineering congress and exposition,
Florida
7. Lechner D (2008) Embedded laboratory for vehicle dynamic
measurements. In: International symposium on advanced
vehicle control, Kobe
8. Shim T, Ghike G (2007) Understanding the limitations of dif-
ferent vehicle models for roll dynamics studies. Veh Syst Dyn
45:191–216
9. Doumiati M, Baffet G, Lechner D, Victorino A, Charara A (2008)
Embedded estimation of the tire/road forces and validation in
a laboratory vehicle. International symposium on advanced
vehicle control, Kobe
10. Doumiati M, Victorino A, Charara A, Lechner D, Baffet G (2008)
An estimation process for vehicle wheelground contact nor-
mal forces. In: IFAC world congress, Seoul
11. Doumiati M, Victorino A, Charara A, Lechner D (2009) Lateral
load transfer and normal forces estimation for vehicle safety
experimental evaluation. Veh Syst Dyn 47(12):1511–1533
12. Wenzel TA, Burnham KJ, Blundell MV, Williams RA (2006)
Estimation of the nonlinear suspension tyre cornering forces
from experimental road test data. Veh Syst Dyn 44:153–171
352 Driver Assistance System, Biologically Inspired
13. Dakhlallah J, Glaser S, Mammar S, Sebsadji Y (2008) Tire-road
forces estimation using extended Kalman filter and sideslip
angle evaluation. American Control Conference, Seatle
14. Ray LR (1997) Nonlinear tire force estimation and road friction
identification: simulation and experiments. Automatica
33(10):1819–1833
15. Baffet G, Charara A, Dherbomez G (2007) An observer of tire
road forces and friction for active-security vehicle systems.
Mechatronics, IEEE/ASME 12:651–661
16. Baffet G, Charara A, Lechner D, Thomas D (2008) Experimental
evaluation of observers for tire-road forces, sideslip angle and
wheel cornering stiffness. Veh Syst Dyn 45:191–216
17. Wilkin MA, Manning WJ, Crolla DA, Levesley MC (2006) Use of
an extended Kalman filter as a robust tyre force estimator. Veh
Syst Dyn 44:50–59
18. Venhovens PT, Naab K (1999) Vehicle dynamics estimation
using Kalman filters. Veh Syst Dyn 32:171–184
19. Kiencke U, Daiss A (1997) Observation of lateral vehicle
dynamics. Control Eng Pract 5(8):1145–1150
20. Hsu YHJ (2009) Estimation and control of lateral tire forces
using steering torque. Ph.D., Department of mechanical engi-
neering, Stanford University
21. Haffner L, Kozek M, Shi J, Jorgel HP (2008) Estimation of the
maximum friction coefficient for a passenger vehicle using the
instantaneous cornering stiffness. In: American control confer-
ence, Seatle
22. Gustafsson F (1997) Slip-based tire-road friction estimation.
Automatica 33(6):1087–1099
23. Hahn JO, Rajamani R, Alexander L (2002) GPS-Based real time
identification of tire-road friction coefficient. IEEE Trans Con-
trol Syst Technol 10(3):331–343
24. Uchanski MR (2001) Road friction estimation for automobiles
using digital signal processing methods. Ph.D. thesis,
University of California, Berkley
25. Koo SL, Tan HS (2007) Tire dynamic deflection and its impact
on vehicle longitudinal dynamics and control. Mechatronics
IEEE/ASME 12(6):623–631
26. Dugoff J, Fanches P, Segel L (1970) An analysis of tire proper-
ties and their influence on vehicle dynamic performance.
Society of Automotive Engineers paper: 700377
27. Pacejka HB (2002) Tyre and vehicle dynamics. Elsevier, Amsterdam
28. Gillespie TD (1992) Fundamental of vehicle dynamics. SAE,
Warrendale
29. Rajamani R (2005) Vehicle dynamics and control. Springer,
New York
30. Lidner L (1993) Experience with the magic formula tyre model.
In: Proceedings of the 1st international colloquium on tyre
models for vehicle dynamic analysis, Amsterdam
31. Svendenius J (2007) Tire modeling and friction estimation.
Ph.D. thesis, Lund University, Sweden
32. Nijmeijer H, Van der Schaft AJ (1991) Nonlinear dynamical
control systems. Springer, New York
33. Doumiati M, Victorino A, Charara A, Lechner D (2009) Unscented
Kalman filter for real-time vehicle lateral tire forces and sideslip
angle estimation. In: IEEE intelligent vehicles symposium, Xi’an
34. Kalman RE (1960) A new approach to linear filtering
and prediction problems. TransAS ME-J Basic Eng 82(D):35–45
35. Welch G, Bishop G (2001) An introduction to the Kalman filter.
Course 8 University of North Carolina, Department of Com-
puter Science
36. Durrant-Whyte H (2001) Multi sensor data fusion. Technical
report, Australian center for field robotics, University of
Sydney
37. Mohinder GS, Angus PA (1993) Kalman filtering theory and
practice. Prentice Hall, Englewood Cliffs
38. Julier SJ, Uhlman JK (1997) A new extension of the Kalman
filter to nonlinear systems. In: International symposium aero-
space/defense sensing, simulation and controls, Orlando
39. Julier SJ, Uhlman JK, Durrant-Whyte HF (1995) A new exten-
sion approach for filtering nonlinear systems. In: American
control conference, Seattle
40. Wan EA, Merwe RVD (2000) The unscented Kalman filter
for nonlinear estimation. In: Proceedings of the adaptive
systems for signal processing, communications, and control
symposium, Atlanta, pp 153–158
41. Milliken W, Milliken DL (1995) Race car vehicle dynamics. SAE,
Warrendale
Books and Reviews
Aga M, Okada A (2003) Analysis of vehicle stability control (VSC)’s
effectiveness from accident data. In: Proceedings of the 18th
international technical conference on the enhanced safety of
vehicles, Nagoya
Aguirre LA, Letellier C (2005) Observability of multivariate differ-
ential embeddings. J Phys A Math Gen 38:6311–6326
Bakker E, PAcejka HB, Lidner L (1987) A new tire model with
application in vehicle dynamic systems. SAE paper 890087
Besançon G (2007) Nonlinear observers and applications. Springer,
Berlin
Bevly DM, Gerdes JC, Wilson C, Zheng G (2000) The use of GPS
based velocity measurements for improved vehicle state esti-
mation. In: Proceedings of the American control conference,
Chicago
Brown T, Hac A, Martens J (2004) Detection of vehicle rollover.
In: SAE world congress, Michigan
Doumiati M, Sename O, Martinez J, Poussot C, Dugard L (2010)
Gain scheduled steering and braking control for improvement
of vehicle handling and stability. In: IEEE conference on deci-
sion and control, Georgia
Doumiati M, Sename O, Martinez J, Dugard L, Gaspar P, Szabo Z,
Bokor J (2011) Vehicle yaw control via coordinated use of
steering/braking systems. In: IFAC world congress, Milano
Heydinger G, Howe J (2000) Analysis of vehicle response data
measured during severe maneuvers. SAE paper 2000-01-1644
Jazar RN (2008) Vehicle dynamics, theory and application.
Springer, New York
Smith ND (2004) Understanding parameters influencing tire
modeling. Colorado State University, Formula SAE Platform
 Driver Assistance Systems, Automatic Detection and Site Mapping 353

Driver Assistance Systems, Safety barrier Infrastructure element for physically

separating traffic from an opposing lane or work
Automatic Detection and Site zone, e.g., crash barrier/guard rail, Jersey barrier.
Mapping
ANDREAS WIMMER, KLAUS C. J. DIETMAYER Definition of the Subject and Its Importance
Institute of Measurement, Control, and Road construction sites are often the reason for traffic
Microtechnology, University of Ulm, Ulm, Germany congestion and accidents on highways and freeways.
This causes great economic and ecological costs to the
society and environment through increasing travel
Article Outline time and additional fuel consumption. Driver assis-
Glossary tance systems specifically designed for work zones will
Definition of the Subject and Its Importance help to reduce the negative impact of construction sites
Introduction for traffic flow. At road works, the lane width usually is
Sensors and Measurement Vehicle reduced, which makes lane keeping a challenging task,
Mapping of Safety Barriers especially for heavy duty vehicles. It often happens that
Detection of Beacons and Traffic Pylons truck drivers slightly ride over the lane markings, thus
Detection and Mapping of Road Markings preventing other vehicles to use the neighboring lane at
Highly Accurate Road Work Map dual carriageways. The aim is to provide lane keeping
Future Directions support for vehicles even in complex scenarios like
Bibliography road construction sites. An assistance system which
laterally controls a heavy-duty vehicle highly depends
Glossary on a robust and accurate estimation of the position of
the vehicle within its lane. Depending on weather and
Advanced driver assistance system Driver assistance lighting condition, this may not be achievable with
system with environmental perception which warns a camera sensor only. Therefore, a system is proposed
or informs the driver or even activates actuators for, which additionally uses a laser scanner and a precise
for example, braking or steering. and detailed digital road work map, which might be
Beacon, traffic pylon, guiding reflector post Infra- regarded as an extension to standard navigation maps.
structure elements, which are often used at road This highly accurate map may be sent to vehicles via
construction sites for guiding motorists at areas wireless communication technique as soon as they
where the traffic routing has changed or at poten- approach the road construction site. The following
tially dangerous spots. chapters describe how the layout and all typical ele-
Environmental perception Concept of measuring and ments of work zones on highways and freeways can be
evaluating the environment, here: the surrounding detected with modern sensors. Then, this data is used
of a vehicle. to automatically generate the above-mentioned highly
Global Navigation Satellite System (GNSS) Satellite- accurate road work map.
based system for positioning in a global reference
system, e.g., GPS, GLONASS. Introduction
Laser scanner Range sensor for environmental per-
ception with a rotating laser beam and high angular Mobility plays an important role in the society: people
resolution. The measurement principle is based on travel to work and goods are transported. The freight
the time of flight of light. transport volume on roads constantly increases in
Road construction site Area where a road is built or many countries worldwide [1, 2]. On the one hand,
rehabilitated, also: road works, work zone. the additional number of cars contributes to a higher
Road marking Marking on the road to inform drivers, possibility of congestion. On the other hand, heavy-
e.g., to delineate the traffic lanes. duty vehicles cause high stress on the road surface,

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
354 Driver Assistance Systems, Automatic Detection and Site Mapping
more than much lighter passenger cars. This makes it
necessary to constantly maintain and repair the car-
riageways. On account of this, a road construction sites
is set up.
Additionally, constructing new highways and free-
ways as well as the rehabilitation of existing ones is an
accepted part of economic stimuli programs, as invest-
ments in the infrastructure are believed to have a high
return on investment [3]. As a result, the number of
road work zones is likely to increase in the future.
Road construction sites, on the contrary, account
for additional stress and annoyance of car drivers due
to the higher risk of traffic jams and delays caused by
reduced speed limits. There is a survey, which reports
an increased nervousness of drivers while traveling
through a work zone [4]. Another survey on 1,700 car
drivers [5] revealed that 41% regard a construction site
as source of danger. Especially the reduced lane width is
a reason why drivers feel uncomfortable.
If the number of lanes is reduced, the total through-
put capacity may decrease as well, which is another
reason for congestion. But not only discomfort of
road users is an issue, there are several studies which
identify a higher accident rate at a road work zones.
Khattak et al. [6] extensively evaluated the effect of
work zone presence on crashes in the USA. Approxi-
mately 24,000 injury crashes and 700 fatalities occur at
US road construction sites every year. In this work, 36
observations have been examined, both on prework
zone and during-work zone. They ascertained that the
total work zone crash rate on highways is 21.5% higher
than the prework zone crash rate. Similar analyses have
been performed in other countries. The ARROWS pro-
ject (Advanced Research on Road Work Zone Safety
Standards in Europe [7]) states that “work zones gen-
erally have higher accident rates than the same road
sections without a work zone”. In detail, however, the
data differs between countries [7, 8]. The main reasons
for these accidents are driving errors and have been
examined in [9]. Motorists have to recognize the work
zone, assess geometry of their lane, and predict the
behavior of other road users in dense traffic. An inad-
equate or late response may easily result in crashes [9].
The question arises, what can be done to make
work zones less dangerous and stressful for drivers.
Accidents statistics [10] show that rear-end collisions
represent a high relative portion. This issue is
addressed by driver assistance systems such as ACC
(Adaptive Cruise Control) or AEB (Automatic
Emergency Break) [11, 12], which are either already
in series production or about to be released. But at
work zones, sideswipe collisions occur more often
than on regular roads. An assistance system which
helps centering the vehicle within its lane will reduce
the possibility for accidents of this type. There are
several systems available, such as Lane Keeping Assist
(LKA) which are designed for standard highways
and freeways. The layout of a road construction site
usually is much more complex and thus poses
a great challenge for driver assistance systems. The
following contribution deals with the special needs to
automatically interpret work zones for future driver
assistance systems. This concept should serve several
purposes: aligning the vehicle within its lane reduces
the work load for the driver and thus enabling him to
concentrate on the surrounding traffic, which reduces
the risk of accidents. If a heavy duty vehicle is optimally
laterally controlled, there will be enough space on
the neighboring lane so that other vehicles can easily
pass. This again reduces the stress on drivers and
additionally helps to maintain the full throughput
capacity as all lanes can be used.
The above-mentioned goals rely on a robust and
accurate estimation of the position of the vehicle within
the work zone and especially within its lane. The
required robustness can be achieved by fusing informa-
tion of different sensors. A single video camera
designed for lane detection may not be able to detect
and interpret the complex scenery of a road construc-
tion site if the lighting condition and weather is disad-
vantageous. Rain and thus a wet surface may produce
reflections on the road, which might be misinterpreted
as road marking. Adverse lighting, such as direct sun-
light or fog, often decreases the contrast of the image.
Longitudinal groove zones might erroneously be
interpreted as lane markings and therefore lead to
false detections. In dense traffic, some road markings
may also be not detectable due to occlusions from
other vehicles. If the road surface is covered by snow,
a video-based detection of the road markings is not
possible. For these reasons, a different approach has to
be chosen. The idea is to use a detailed and highly
accurate digital map (Road Work Map), which con-
tains the positions of all important elements of a road

construction site, for example, road markings, crash
barriers, and pylons/beacons. This additional informa-
tion is then used to fully reconstruct the environmental
model of the vehicle, if parts of the surrounding cannot
be measured with on-board sensors. For a robust detec-
tion, active sensors, such as lasers scanners or radar, are
suitable as they are less susceptible to weather condi-
tions. The laser scanner is not only able to detect other
vehicles but also beacons, traffic pylons, and guard
rails, which are commonly used at work zones. This
data then is fused with the data of a video camera,
a position estimate derived from GPS, and the detailed
Road Work Map described above. Therefore, robust
and reliable data can be supplied for assistance systems
such as lane keeping.
Controlling a heavy-duty vehicle laterally is
a demanding task [13–15]. Many systems require
a look-ahead capability to enable a robust control.
This can be easily provided, as the future road course
can be extracted from the proposed Road Work Map.
The concept described above is divided into two
parts (Fig. 1). First, the highly accurate map has to be
created. For this, a measurement vehicle is equipped
with additional sensors. Each time when a new road
construction site is set up, the vehicle will drive through
the scenery and automatically detect and map all
important elements. Here, post-processing and the
use of additional and more accurate hardware are
Detection of Vertical
safety barriers laser scanner
Detection of Horizontal
beacons/pylons laser scanner
Detection of Video
road markings camera
Mapping, post-processing
Highly accurate
positioning Road work map
(RTK-GPS)
Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 1
Overview over the modules for creating a highly accurate Road Work Map (left). Proposed system setup, which uses the
Road Work Map to supply reliable data for advanced driver assistance systems (right)
Driver Assistance Systems, Automatic Detection and Site Mapping 355
possible. This generated Road Work Map is transferred
to facilities at the entrance of the road work zone.
The second part deals with the application for all
vehicles driving through the construction site. It is
assumed that cars and trucks receive a current map
from the work zone via wireless communication. If
the vehicles are equipped with a future sensor, for
example, a laser scanner, they can match their current
environmental model to the map and thus profit from
additional information for assistance systems such as
lane keeping assist.
The automatic generation of the Road Work Map
will be presented in the following sections.
Sensors and Measurement Vehicle
The measurement vehicle is a standard car of series
production, but additionally equipped with several
sensors for environmental perception (Fig. 2). A
front-facing gray-scale camera with a resolution of
640
480 pixels is used for the detection of road
markings. A wide-angle lens allows the measurement
of markings not only from the lanes of the host vehicle,
but also from neighboring lanes.
Two Ibeo ALASCA XT laser scanners [16] are used
for acquiring a horizontal and a vertical distance profile
of the surrounding area. This type of sensor sends out
a rotating laser beam and detects the echo reflected on
Standard Series Laser
DGPS ESP scanner
Road work map
Map matching
Assistance systems, e.g., lateral control
356 Driver Assistance Systems, Automatic Detection and Site Mapping
RTK-GPS
Vertical laser scanner
Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 2
The measurement vehicle with two laser scanners, a video camera, and a highly accurate RTK-GPS
obstacles. With the speed of light and the time of flight,
the distance to an object can be derived. The laser
scanner has an angular resolution between 0.125 and
0.5 , depending on the angle and scanning frequency.
The total horizontal field of view is up to 240 ; at this
vehicle this is limited to 180 due to constraints of the
housing and mounting position. The laser scanner
provides four layers with a vertical angle shifted by
0.8 , resulting in a vertical opening angle of 3.2 . That
feature enables the laser scanner to detect objects even
in cases where the vehicle pitches, thus making tracking
more robust. The detection range is up to 200 m,
depending on the reflectivity. The good distance reso-
lution of 4 cm enables a highly accurate detection and
mapping. The front-facing laser scanner is used to
detect beacons and traffic pylons as well as detecting
the free space in front of the vehicle. At the rear of the
vehicle, another laser scanner is mounted to acquire
a vertical distance profile. This is used to detect and
map safety barriers such as guard rails and Jersey
barriers, which limit the road to the left and right.
Besides the novel environmental sensors, series sen-
sors are used for estimating the position and movement
of the host vehicle. Wheel encoders and a MEMS
(micro-electro-mechanical systems) yaw rate sensor
of a modern ESP (Electronic Stability Program) system
are used for calculating the relative host vehicle move-
ment between successive measurement frames. A stan-
dard DGPS (Differential Global Positioning System)
provides global position information of about 3–10 m
accuracy, depending on the measurement conditions of
Camera
Horizontal laser scanner
the satellites. This type of sensor is already part of series
cars equipped with an on-board navigation system.
For generating the Road Work Map, a highly accu-
rate position of the measurement vehicle must be
known in a global coordinate system, for example,
WGS84 [17] or UTM [18]. This position is acquired
using a RTK-GPS (Real Time Kinematic GPS), which
additionally uses online correction data of the land
survey administration, thus enabling a horizontal posi-
tion accuracy of 1 cm (RMS). This system not only uses
GPS, but also GLONASS, the Russian Global Naviga-
tion Satellite System (GNSS). High accuracies can only
be achieved under good measurement conditions. In
cases where no or only a little number of satellites is
visible (e.g., in tunnels or in street canyons), an IMU
(inertial measurement unit), which contains highly
accurate yaw velocity and acceleration sensors, is used.
Both data is fused to obtain best position estimation
(ADMA: Automotive Dynamic Motion Analyzer, [19]).
All data of position sensors and environmental
sensors must be transformed to a global coordinate
system. For that reason, the mounting position and
alignment of each sensor relative to the vehicle must
be known. Laser scanner and video camera have to be
calibrated [20] and the position of the ADMA has to be
known.
Besides this, timing of the sensors is similarly
important on a moving platform like the measurement
vehicle. Typical speeds at road work zones on highways
are 80 km/h. This means, that a timing inconsistency of
5 ms already accounts for an error of over 11 cm.

Trigger
ECU
Horizontal
laser scanner
Fusion
ECU
Vertical
laser scanner
Video camera
Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 3
Schematic of sensors and modules of the measurement vehicle. All important sensors have trigger capability to provide
accurate timing information
For that reason, laser scanner and video camera are
hardware-triggered; the measurement frequency is set
to 12.5 Hz. The highly accurate GPS system provides
a trigger pulse every 20 ms (50 Hz). Both pulses are
input to a trigger message controller which generates
a time stamp message on the CAN bus (see Fig. 3). The
data of the ESP is already present at the CAN bus. All of
these messages are used to align sensor data from
different sources to a common time reference.
Mapping of Safety Barriers
High speeds at highways and freeways even at road
construction sites demand a mechanically robust pro-
tection system both for workers as well as motorists.
Often, the traffic routing is changed significantly.
A lane may be omitted, the width will be reduced and
even one lane may be routed to the side of the opposing
traffic. This poses a potential danger for all road users.
For that reason, different safety barriers have been
developed [21]. Crash barriers (guard rails) can be
found on almost every dual carriageway. Especially
for road construction sites, removable barriers are
used. A commonly used type is the Jersey barrier,
which is made of concrete and exists in different sizes.
But steel barriers are also frequently used. Safety bar-
riers in general serve different needs, which may con-
tradict each other. On the one hand, these objects
should have a preferably small profile to offer enough
space for the actual traffic and the area of work. On the
Driver Assistance Systems, Automatic Detection and Site Mapping 357
Trigger message
controller
ESP
system
Mapping RTK-GPS/IMU
system (ADMA-System)
Standard
DGPS
other hand, these barriers have to withstand impacts of
heavy duty vehicles, so that these trucks can be kept
within their lane in the event of a crash. Additionally,
the barriers are designed to absorb some amount of
crash energy, so that vehicles do not directly bounce off
the wall. There are regulations which specify the phys-
ical characteristics in the event of a crash, subject to the
total speed of the vehicle, its tonnage, and the angle
under which it hits the barrier. Depending on the
maximum speed limit and other operation purposes,
different rules may apply [22]. The actual shape or
material, however, is not regulated. For that reason,
a detection system has to deal with numerous different
outlines [21].
A vertically scanning laser scanner mounted at the
rear of the vehicle has been chosen as sensor setup for
detecting and mapping safety barriers. The measure-
ment principle is shown in Fig. 4. Because of the high
angular resolution and the four layers, a detailed
outline can be obtained. This data undergoes
further analysis to distinguish between background
(e.g., vegetation) and actual work zone infrastructure.
As a first step, the data is filtered to reduce mea-
surement noise and to sharpen the contour [23].
A third-order Gaussian noise filter is applied on all
raw laser scan points. On average roads, the mapping
vehicle is subject to slight rolling, which distorts
the measurement. Since the laser scanner acquires a
distance profile of the road directly behind the rear of
the vehicle, a regression line can be fit into the data.
358 Driver Assistance Systems, Automatic Detection and Site Mapping

It can be assumed that the area which the vehicle is
driving on is even. Therefore, a compensation of the
roll angle is possible.
The structures of interest extend in vertical direc-
tion, as is illustrated in Fig. 5. That fact is exploited to
extract suitable candidates for safety barriers. A
histogram is applied on the data, where several vertical
slots accumulate the number of scan points in this
region. The distance between successive scan points
increases in further distances from the vehicle due to
the radial measurement principle of the scanner. This
Laser scanner
Jersey barrier
Distance profile
Driver Assistance Systems, Automatic Detection and Site
Mapping. Figure 4
Measurement principle of a vertically mounted laser
scanner. A vertical distance profile enables the detection of
barriers, which are located left and right of the vehicle
effect can be compensated before the accumulation is
performed by applying different weights on the points
based on the geometry [24]. The bins with a higher
rating are regarded as valid candidates and are further
examined.
Automatically detecting safety barriers requires
distinguishing between background and actual infra-
structure. Grass and bushes often grow underneath and
around guard rails, which is also measured by the laser
scanner. The Road Work Map, however, should not
include background data, as this is not the border of
the road. Preliminary tests have shown, that a rule-
based separation of infrastructure and background
only yields low detection rates. For that reason,
a more sophisticated approach had to be chosen. As
mentioned above, there is a variety of different safety
barrier types. Extensive tests in Germany have shown
that all different barriers can be grouped into a few
different classes, which can be seen in Fig. 9. A similar
analysis can be performed for other regions and barrier
types. These types may not be totally the same struc-
ture, but they have a similar sensor reading, hence an
analog outline. Separating between different classes
leads to the use of a Bayesian classifier.
For making use of a Bayesian classifier, features have
to be found, which best describe the different classes.
The vertical distance profile of the laser scanner is
processed to find a representation of the typical outline
of safety barriers. As the scan points are subject to

Crash barrier Jersey barrier

Characteristic
points
0.2 m
0.2 m

Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 5

Real scan data example of a vertical scan. A guard rail is located on the left side, whereas a Jersey barrier is on the right side.
As a result of the detection algorithm, three characteristic points can be extracted to incorporate them into the Road Work
Map
 Driver Assistance Systems, Automatic Detection and Site Mapping 359

Length of Variance of
Depth of Height of Length of Height of vertical parts points to
grooving grooving hull curve structure of hull curve regression line

Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 6

Several feature values can be calculated from the vertical scan based on the geometry of the structure. These features are
used as an input for the Bayesian classifier

noise, a hull curve is fit into the measurement points.
This hull curve is designed to represent the outer hull
(closest to the vehicle), as a conservative approxima-
tion of the barrier with respect to the application. The
hull curve is the basis for further feature extraction
algorithms, which are based on the geometry of the
structure. One obvious useful feature is the height of
the barrier. Tall Jersey barriers reach approximately
1.2 m whereas guard rails have a typical height over
ground of 0.8 m. Two other features are directly derived
from the above-mentioned hull curve. One feature is
the total length of the contour, the other one is the
length of all vertical parts of the hull curve. Based on an
angle interval criterion only those segments are accu-
mulated, which are regarded as approximately vertical.
Another two features are derived from the shape of
guard rails. These structures show a typical grooving
in the middle of the metal crash barrier. The height of
this grooving is taken as one feature; the calculated
width (horizontal distance of closest and furthest
point to the vehicle) is taken as the other feature. For
separating artificial structures like Jersey barriers from
background vegetation a sixth feature is extracted:
a regression line is fit into the vertical scan points.
The variance of the horizontal distance of the points
is regarded as a means for the smoothness of the struc-
ture. An illustration of all features can be seen in Fig. 6.
The features described above are evaluated for all
different kinds of safety barriers. Several hundreds of
data frames have been labeled manually to gain the
typical distribution of the feature values. The result
Number of occurences
20
15
10
5
0
–0.1 0 0.1 0.2 0.3
Feature value: variance to regression line [m]
Driver Assistance Systems, Automatic Detection and Site
Mapping. Figure 7
Number of occurrences of the feature values (“variance
of points to regression line”) for one barrier type
(metal crash barrier)
for one class and one feature value is exemplarily
shown in Fig. 7. It can be seen that the class-conditional
probability density function pðxjCi Þ for a feature x and
class Ci corresponds to a Gaussian distribution.
Hence, the density functions can be approximated
by the mean m and variance s2 of the features for every
class. This fact motivates the use of a Bayesian classifier
[25]. It is assumed, that all types of barriers (classes Ci )
have equal prior probabilities PðCi Þ. PðCi jxÞ is the
probability that class i is true for the given feature
value x (posterior probability) and should be calcu-
lated based on the measurement of all feature values
and the predetermined class-conditional probability
density functions.
360 Driver Assistance Systems, Automatic Detection and Site Mapping

This can be calculated for a one-dimensional valid, which lie in between the 2s interval of its mean.
feature and N classes: This value has been found as best compromise between
detection rate and the rate of non-detected safety bar-
pðxjCi ÞP ðCi Þ
P ðCi jx Þ ¼ PN     riers. Approximately 3,300 measurements have been
j¼1 p xjCj P Cj labeled manually for evaluating the performance of
The class-conditional probability density functions the classifier. The overall detection rate is 97%, which
of all types of barriers for the feature “length of hull indicates the differentiation between background and
curve” can be seen in Fig. 8. For the six-dimensional safety barriers. This number includes some misclassifi-
(d ¼ 6) feature vector x (mean vector m and covariance cations between different classes, that is, a Jersey barrier
matrix K ), the class-conditional probability density is: might have been mistaken as steel barrier (Fig. 9).
 Regarding correct class assignment, the correct classifi-
 T  
exp 0:5 x  m K 1
xm cation rate is about 95%. Non-detected elements occur
P ðxjCi Þ ¼ p ffiffiffiffiffiffi
ffi in 3% of the measurements, where the false positive
ð2pÞ0:5d jK j rate (background erroneously detected as safety bar-
rier) is only 2%.
The extension to the multidimensional case results
After the classification of all candidates, a suitable
in a most probable class (type of barrier) for the given
representation of the structure is calculated. The top
feature vector. This does not yet include the case, where
most and bottom point is extracted, as well as the
the candidate structure is measured in the background.
position of the typical grooving for barriers which are
Therefore, only those measurements are regarded as
classified as guard rails. For all other barriers, top and
bottom points are derived as well, but additionally, the
position of the bend of Jersey barriers is extracted. All
safety barriers are now represented by three character-
Class conditional probability density istic points (see Fig. 5). This enables a lower amount of
data to be stored than using the whole hull curve. But
5
on the other hand, the structure is still represented with
high accuracy.
4 The above-mentioned process has been performed
on a single frame basis. As the detection of safety
3 barriers is meant for mapping, the connection of suc-
cessive measurement points can be done with post-
2
processing calculations. First of all, all measurement
1
points are transformed from the local sensor coordi-
nate system to a global frame using the position of the
0 vehicle in global coordinates derived from the ADMA
–0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
system. Then, all nodes which originate from the same
Length of hull curve [m]
class are grouped to segments. An outlier detection
Jersey barrier Guard rail based on distance criteria removes false points of the
(tall size)
classification process. All data of one segment is inter-
Jersey barrier Steel safety
(medium size) barrier polated by a smoothing spline [26, 27]. On the one
Group of small safety barriers hand, this helps to bridge gaps between non measured
areas (due to e.g., occlusion) and on the other hand,
Driver Assistance Systems, Automatic Detection and Site additional smoothing of noisy data is possible.
Mapping. Figure 8 A smoothing spline can be regarded as a combination
Class conditional probability density functions for five of an interpolating spline, which connects all nodes by
different types of barriers for the feature “length of hull a piecewise continuous polynomial, and a regression
curve” line. A regression line minimizes the squared distance
 Driver Assistance Systems, Automatic Detection and Site Mapping 361

a b c

d e

Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 9

Examples of all five different types of barriers used for the classification algorithm: (a) tall Jersey barrier, (b) medium-size
Jersey barrier, (c) standard guard rail, (d) steel safety barrier, (e) example of a group of different small safety barriers

100
Number of nodes [percent]

of all nodes to the function, where the spline, in con-

trast, minimizes the curvature of the function. The
trade-off between those characteristics can be chosen
by the smoothing factor l. Si ðxÞ is the polynomial in 50
segment i and yi the node with an assumed standard
deviation of si . Minimizing the function L gives the
desired smoothing spline:
0
0 2 4 6 8
n 
X  Zxn
yi  Si ðxÞ 2 Maximum loss of accuracy [cm]
fSi 00 ðxÞg dx:
2
L¼l þ ð1  lÞ
i¼0
si Driver Assistance Systems, Automatic Detection and Site
x0
Mapping. Figure 10
As the Road Work Map is designed to be transferred Exemplary relationship between the parameter “maximum
via wireless communication, the amount of data loss of accuracy” and the number of nodes necessary for
should be as low as possible. As the spline offers representing the barrier
a very flexible representation of arbitrary curves, the
user is able to omit nodes without losing too much
accuracy. A maximum distance between the original
node and the spline reconstruction of the barrier with Besides the detection rate, the overall accuracy
less supporting points can be set to a value which is which can be achieved with that concept is of interest.
determined by the application. The exemplary rela- For evaluating the preciseness of the algorithms men-
tionship between the maximum loss of accuracy and tioned above, the exact position of a guard rail has been
the number of nodes of a sample barrier is shown manually derived by a static RTK-GPS measurement,
in Fig. 10. which yields to a position accuracy of about 3 mm
362 Driver Assistance Systems, Automatic Detection and Site Mapping
(1 s interval) [28]. This is compared to the results of
the automatic detection and mapping process. The
evaluation shows a mean squared error of 7.1 cm and
a standard deviation of 5.7 cm. This means that the
proposed set-up produces results accurate enough for
the generation of a detailed Road Work Map.
Detection of Beacons and Traffic Pylons
At the beginning and ending of road construction sites
traffic pylons and guiding reflector posts (beacons) are
often used (Fig. 11). They commonly mark the lead-in
and exit tapers and guide drivers when traffic routing is
changed. Pylons and Beacons are used especially for
works of short duration [22], for example, when new
road markings are applied on the road, as this is less
time consuming than erecting safety barriers.
Both elements must have a reflecting surface. This
fact is exploited when pylons and beacons should be
detected with a laser scanner. The laser sensor not only
measures the distance to objects but also the so called
echo pulse width, which is proportional to the amount
of energy returned from targets. The total amount of
energy decreases with the distance of objects from the
sensor. By comparing the data of reflecting and non-
reflecting materials for different distances, a character-
istic can be deduced, which is shown in Fig. 12. This
characteristic enables the classification of each scan
Beacon, Guiding Traffic pylon,
reflector post Traffic cone
Driver Assistance Systems, Automatic Detection and Site
Mapping. Figure 11
Two infrastructure elements frequently used at road
construction sites: beacon or guiding reflector post (left),
traffic pylon or cone (right)
point to be reflector or non-reflector and is only
slightly influenced by weather conditions.
Besides the reflector attribute, the size of the
measured objects is taken into account as well. For
extracting suitable beacon candidates, a measurement
grid of the laser scan is used. This concept has been
introduced in the field of robotics [29] and is now also
used for automotive applications [30]. The area in
front of the vehicle is segmented into a grid of squares
with, for example, 20 cm edge length and initialized
with probability value 0.5. For each scan point within
one cell, the value is increased, depending on its dis-
tance and radial position to account for measurement
noise. The area between the sensor and a measurement
is assumed to be unoccupied, so the value of the cells on
the way of the laser beam is decreased. Due to noise,
this is also done distance dependent. The grid cell
values can be visualized in grey scale color, where
black means occupancy of 1 and white means 0 proba-
bility of occupancy (see also Fig. 13). As a result, the
grid contains dark areas which are presumably occu-
pied and hence denote the position of obstacles. Other
grid cells have lighter color which means they are
25
20
Echo pulse width [m]
15
10
0
0 20 40 60 80 100
Distance [m]
Driver Assistance Systems, Automatic Detection and Site
Mapping. Figure 12
Distinguishing between reflecting and nonreflecting scan
points based on the echo pulse width and the distance
from the sensor. Scan points which originate from
reflecting surfaces are drawn in red; points from
nonreflecting surfaces are colored in blue. The black line
shows the limiting curve which is used in the algorithm

Beacon (candidate)
Scan
points
5m Beacon (verified)
Unknown area
Free space
Driver Assistance Systems, Automatic Detection and Site
Mapping. Figure 13
Detection of beacons using a front-facing laser scanner.
Exploiting both size and reflectivity of objects enables the
extraction of suitable candidates. A tracking algorithm
verifies true beacons
assumed to be on free space or the region is unknown,
that is, the value remains 0.5. Based on this measure-
ment grid, a contour extraction algorithm is applied to
find the position of free standing objects with a certain
size. If these objects additionally contain scan points
marked as reflectors, they are regarded as candidates for
pylons and beacons. After the object generation based
on single frame, a tracking over time is applied. The
position of all candidates of one measurement time is
transformed into the next frame by compensating the
host vehicle movement, as the pylons are assumed to be
stationary. An association algorithm finds correspon-
dences between last and current frame. In this way,
random detections in the background can be filtered
out, if no new measurements are found in this area.
A track confirmation finally verifies pylon and beacon
candidates. These positions can either be provided for
online use while driving, or for incorporating the
objects into the Road Work Map.
For assessing the preciseness of the detection and
tracking algorithm, the position of some beacons and
pylons has been determined manually with a static
RTK-GPS measurement. The accuracy of this reference
positions is approximately 3 mm (1 s interval). This is
compared to the automatic mapping process. The
mean squared error evaluates to 7.3 – 7.7 cm for bea-
cons and pylons respectively (see Fig. 14). With respect
to the sensor resolution, this is a good value.
Driver Assistance Systems, Automatic Detection and Site Mapping 363
Tangent plane north coordinate [m]
0.15
0.1
0.05
0
–0.05
–0.1
–0.2 –0.1 0 0.1 0.2
Tangent plane east coordinate [m]
Driver Assistance Systems, Automatic Detection and Site
Mapping. Figure 14
Accuracy of the mapping algorithm for traffic pylons. The
true position (reference) is shown by a black circle with
a diameter of 25 cm, which is the size of the bottom of the
pylon. Each single measurement is drawn as a blue circle.
The tracked position, which would be incorporated into
the map, is shown as red star
Detection and Mapping of Road Markings
Road markings are used for guiding drivers on all
major roads. They instruct motorists, if a lane change
is possible or not. At lead-in and exit tapers, they are
especially important for a safe and smooth guidance of
all vehicles. At work zones the original marking (e.g.,
white color) may be invalidated by markings of another
color (e.g., yellow), while the primary lines still remain
on the road. That is the reason why lane recognition in
road construction sites is such a demanding task. The
Road Work Map described above should provide an
interpreted description of the whole environment,
including road markings.
Lane recognition with video cameras already has
a long history in research. For well-structured scenarios
such as highways and freeways, assistance systems for
lane departure warning already exist on the market.
The specific needs of such systems for road work
zones, however, are not yet addressed. Current lane
detection approaches offer a wide variety of concepts
[31]. The first step usually is a feature extraction,
followed by road modeling and finally a temporal
364 Driver Assistance Systems, Automatic Detection and Site Mapping
tracking. These steps can be performed either based on
monocular [32] or stereo [33] images. One of the most
common concepts for feature extraction is to find edges
in the image caused by a light-to-dark or dark-to-light
transition of white or yellow markings on a darker road
surface. Other approaches use color information or try
to find areas of similar texture [34]. A template
matching such as the Hough transform [32] can be
applied to fit the data to a predefined model.
Depending on the type of road which is mostly
addressed by the application, an adapted mathematical
curve is fit to the data. Commonly used are parabolic
[31] or cubic curves, the assumption of constant cur-
vature [32], or approximations for clothoids [35, 36].
The assumption of a flat world [31] may hold for many
applications; otherwise, the road has to be modeled
three dimensional [37]. A temporal filter such as
a Kalman filter or particle filter can be applied to gain
additional robustness against single frame detections
[32, 34, 38].
The road markings in work zones have to be treated
in a special case. The following paragraphs will describe
all algorithms and adaptations for creating a highly
accurate environmental model of road markings for
the purpose of mapping. All lane markings should
be detected at one single drive-through; therefore
a video camera with wide-angle lens is used. This
results in a detection range of about 20 – 35 m which
is enough for mapping. Additionally, modeling a flat
world in this range is a valid assumption on highways
and freeways.
For edge detection, a Canny filter [39] has been
chosen. This commonly used algorithm consists of
multiple stages, among them are a Gaussian filter for
noise reduction, an operator for determining magni-
tude and direction of the edge, and a non-maximum
suppression. Road markings are assumed to have
a left (dark to light transition) and right (light to dark
transition) edge. This is exploited for finding appro-
priate pairs of edge points. The magnitude G
and direction y of the edge is calculated with the
horizontal (x) and vertical (y) Sobel operator [39] on
the image I:
2 3
1 0 1
Gx ¼ 4 2 0 2 5  I
1 0 1
2 3
1 2 1
Gy ¼ 4 0 0 0 5I
1 2 1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
G¼ Gx2 þ Gy2
 
Gy
y ¼ arctan
Gx
Due to image noise and blur there may be several
edges in close distance. A local maximum gradient
search finds the most probable position of the edge.
These points are projected from image domain to vehi-
cle coordinate system using a flat world assumption.
The distance between left and right edge is the mea-
sured width of the road marking. Edge point pairs are
only considered valid, if the distance matches the reg-
ulations for road marking widths [40] including some
noise margin. Finally, only a little number of false
detections remains.
After that, those point pairs which originate from
the same road marking have to be associated. A linear
extrapolation in the driving direction is calculated
based on a least-square fit to associate adjacent mea-
surement nodes. Point pairs are associated, if they have
a similar width and an appropriate distance to the
preceding point pair. The measurement of lengths
and widths is subject to noise. These errors can have
several reasons and must be considered. The quantiza-
tion through pixels results in different error values in
the road plane, that is, they will be little in near distance
from the vehicle and bigger in long distances. Addi-
tionally, pitching and rolling of the vehicle will result in
errors because of the projection from image to flat road
surface. These influences are taken into account before
association. Point pairs, which cannot be associated are
regarded as outliers and are neglected.
For separating one lane from the neighboring lane,
dashed road markings are commonly used. The line
length usually is 6 m, the gap between two line seg-
ments can be either 6 or 12 m [40, 41]. The detection
algorithm only finds edges on the actual markings, so it
is desirable to associate adjacent lines. Therefore,
a higher order regression polynomial fM is fit into the
whole data of each line segment:
fM ðt; aÞ ¼ a0 þ a1 t þ ::: þ an t n
 Driver Assistance Systems, Automatic Detection and Site Mapping 365

where a0 ; :::; an are the polynomial coefficients and n
the order of the polynomial, which can be adapted
depending on the number of nodes. Finding the opti-
mal parameters ^ a on a set of m data points yi means
minimizing the squared error sum S[42]:
Xm
min S ¼ min ðyi  fM ðti ; aÞÞ2
a a i¼1
This problem can be rewritten:
0 10 1 0 1
1 t1 t12    t1n a0 y1
B 1 t2 t22    t2n CB a1 C B y2 C
B CB C B C
B .. .. .. . . .. CB .. C ¼ B .. C
@. . . . . A@ . A @ . A
1 tm tm    tmn
2 an ym
a¼y
B^
a ¼ ðB T BÞ1 B T y
^ steps of the safety barrier mapping, a suppression of
The results are two polynomials which can be
extrapolated and compared via distance criteria. If
they overlap, these two lines are assumed to be part of
the same lane marking. This method also helps to
bridge gaps between non-detected segments (e.g., due
to occlusion) of road markings. As road markings can
have a curvature, especially when the traffic routing
changes, this method assures a high flexibility for arbi-
trary curvatures within the field of view of the camera.
Modeling curves for longer distances is done by splines
in the post-processing step.
The detections based on single frames (illustrated in
Fig. 15) are handed over to several post-processing
stages. First, all data of the same road marking, which
has been measured several times in different time steps
has to be associated. This is based on distance criteria
under consideration of the regulations for lane mark-
ings. As a result, there will be a lot of nodes of the same
line, depending on the performance of the detection
algorithm. For further processing, these nodes need to
be clustered and, if necessary, the number of nodes can
be reduced. That is done by an algorithm inspired by
the k-means clustering [43] with fixed cluster size,
which preserves enough data points in areas with
sparse number of points. After that, an outlier detec-
tion based on geometric parameters is applied. Back-
ground and beacons, which sometimes produce
erroneous detections, can thus be removed, as they
appear only in a limited number of frames and in
inconsistent directions. Similar to the post-processing
noise is achieved through the calculation of
a smoothing spline and the number of nodes is reduced
by an iterative process with predefined maximum error.
Highly Accurate Road Work Map
All three different elements of road construction sites
can now be incorporated into the map. Beacons and
traffic pylons have been measured by a front-facing
laser scanner. Safety barriers are detected by
a vertically scanning laser sensor and road markings
are identified with a video camera. This data can now
be interpreted with respect to valid and invalid lane
markings. The idea is that vehicles which approach the

Driver Assistance Systems, Automatic Detection and Site Mapping. Figure 15

Examples of the road marking detection algorithm. The images show the performance for different lighting and surface
conditions
366 Driver Assistance Systems, Automatic Detection and Site Mapping
work zones will receive a current and detailed Road
Work Map. Usually, cars will not be equipped with
highly accurate RTK-GPS sensors, but with standard
DGPS already used for navigation applications.
A matching algorithm aligns the map data with the
current environmental perception, thus enabling
a more accurate and extended model of the construc-
tion site. This is used as input for driver assistance
systems such as lane keeping support.
Future Directions
Increasing traffic density makes driving more and more
challenging. Assistance systems help to support motor-
ists in their basic tasks. In the future, semiautomated
vehicles are not completely unrealistic. Driving long
distances on well-structured environments like free-
ways could partly be handed over from human to
intelligent vehicles equipped with longitudinal and lat-
eral control. More complex scenarios like urban traffic
with pedestrians and cyclists sharing the same traffic
area will still require human supervision, but there are
already concepts for supporting the driver in more and
more tasks.
The proposed concept of transferring maps to
approaching cars and trucks requires infrastructure-
to-vehicle communication. All major vehicle manufac-
tures are involved in the research of applications for
vehicle-to-vehicle or infrastructure-to-vehicle commu-
nication applications. The technical details are not yet
fully specified, but the development shows a promising
progress. This would enable the use of highly accurate
infrastructure maps for many different future driver
assistance systems [44]. These maps could also be
extended by dynamic objects, such as pedestrians at
intersections in urban areas.
Bibliography
1. Bureau of Transportation Statistics (2010) U.S. Department of
Transportation, National Transportation Statistics, Washington
2. European Commission (ed) (2009) Panorama of transport,
office for official publications of the European communities,
Luxembourg
3. Helling A (1997) Transportation and economic development:
a review. In: Little RG(ed) Public works management and
policy, vol 2 (1). Sage, Thousand Oaks, CA, USA, pp 79–93
4. Benekohal R, Hashmi A, Orloski R (1993) Drivers’ opinion
on work zone traffic control. In: Transportation quarterly,
47 (1). ENO Foundation, Eno Transportation Foundation,
Washington, DC, USA, pp 19–38
5. DEKRA (ed) (2009) Road works cause anxiety – Baustellen
sorgen fuer Nervenflattern, DEKRA press and information,
Stuttgart
6. Khattak Asad J, Khattak Aemal J, Council FM (2002) Effects of
work zone presence on injury and non-injury crashes. In:
Accident analysis and prevention, vol 34(1). England, Elsevier
Science Ltd, Amsterdam, Netherlands, pp 19–29
7. ARROWS Consortium (1999) Advanced research on road work
zone safety standards in Europe (arrows), road work zone
safety, practical handbook, background report. National Tech-
nical University Athens
8. Board of Trustees for Traffic Safety (ed) (2007) Hazard area
freeway road works? – Gefahrenstelle Autobahnbaustelle?
Kuratorium für Verkehrssicherheit – Board of trustees for
traffic safety, Austria
9. Twisk D, Mesken J (2007) PREVENT – Education to improve
road safety around work zones. In: Larsson TJ Safety science
monitor, vol 11(2). Perth, Australia
10. Wang J, Hughes WE, Council FM, Paniati JF (1996) Investigation
of highway work zone crashes: what we know and what we
don’t know. In: Transportation research record 1529. TRB,
National Research Council, Washington
11. Marsden G, McDonald M, Brackstone M (2000) Towards an
understanding of adaptive cruise control. In: Transportation
research part C. Elsevier Science, Pergamon
12. Bishop R (2000) A survey of intelligent vehicle applications
worldwide. In: Proceedings of the IEEE intelligent vehicles
symposium, Dearborn, 2000
13. Martini S, Murdocco V (2003) Lateral control of tractor-trailer
vehicles. In: Proceedings of 2003 IEEE Conference on Control
Applications, CCA
14. Wang J-Y (1999) Robust H1 Lateral control of heavy-duty
vehicles in automated highway system. In: Proceedings of
the American control conference, San Diego
15. Fritz H (1999) Longitudinal and lateral control of heavy
duty trucks for automated vehicle following in mixed
traffic: experimental result from the CHAUFFEUR project.
In: Proceedings of the 1999 IEEE international conference
on control applications. Kohala Coast-Island of Hawai’i,
Hawai’i
16. Ibeo Automobile Sensor GmbH (2006) ALASCA user manual,
Hamburg
17. European Organisation for the Safety of Air Navigation
(1998) WGS 84 implementation manual, European
organisation for the safety of air navigation, Brussels,
Belgium and Institute of Geodesy and Navigation
University FAF Munich, Germany
18. Osborne P (2008) The mercator projections, edinburgh, http://
mercator.myzen.co.uk/
19. GeneSys GmbH (2010) ADMA user manual, Offenburg

20. Kaempchen N (2007) Feature-level fusion of laser scanner and
video data for advanced driver assistance systems. Ph.D.
dissertation, University of Ulm, Ulm, Germany
21. Knoll E (ed) (2004) Der elsner – handbook for traffic and
transportation. Otto Elsner Verlagsgesellschaft mbH & Co.
KG, Dieburg, Germany
22. German Ministry of Transport (Bundesministerium für Verkehr,
Bau- und Wohnungswesen, Abteilung Straßenbau (BMVBS))
(1995) Safety instructions for road construction sites –
Richtlinien für die Sicherung von Arbeitsstellen an Straßen,
Bonn, Germany
23. Wimmer A, Weiss T, Flögel F, Dietmayer K (2009) Automatic
detection and classification of safety barriers in road construc-
tion sites using a laser scanner, In: IEEE proceedings of intelli-
gent vehicles symposium, Xi’an, China
24. Weiss T, Dietmayer K (2007) Automatic detection of traffic
infrastructure objects for the rapid generation of
detailed digital maps using laser scanners. In: Proceedings
of the 2007 IEEE intelligent vehicles symposium, Istanbul,
Turkey
25. Duda RO, Hart PE, Stork DG (2000) Pattern classification. Wiley,
New York
26. Reinsch CH (1967) Smoothing by spline functions. In: Reinsch
CH (ed) Numerische mathematik, vol 10. Oxford University
Press, New York, pp 177–183
27. Pollock D S G (1993) Smoothing with cubic splines, Depart-
ment of Economics, Queen Mary College, University of London
28. NovAtel Inc (2007) Datasheet for receiver OEMV-3. NovAtel
Inc, Canada
29. Thrun S, Fox D, Burgard W (2005) Probabilistic robotics. MIT
Press, Cambridge
30. Weiss T, Schiele B, Dietmayer K (2007) Robust driving
path detection in Urban and highway scenarios using a laser
scanner and online occupancy grids. In: IEEE proceedings of
intelligent vehicles symposium, Istanbul, Turkey
31. McCall JC, Trivedi M (2006) Video-based lane estimation and
tracking for driver assistance: survey, system, and evaluation.
In: IEEE transactions on intelligent transportation systems,
Piscataway, NJ, USA
32. Taylor C, Koŝecká J, Blasi R, Malik J (1999) A comparative study
of vision-based lateral control strategies for autonomous
highway driving. I J Robotics Res, Sage Publications, Thousand
Oaks, CA, USA
Driver Assistance Systems, Automatic Detection and Site Mapping 367
33. Bertozzi M, Broggi A (1998) GOLD: a parallel real-time
stereo vision system for generic obstacle and lane
detection. In: IEEE transactions on image processing,
Piscataway, NJ, USA
34. Apostoloff N, Zelinsky A (2003) Robust vision based lane
tracking using multiple cues and particle filtering. In:
Proceedings of IEEE intelligent vehicles symposium,
Piscataway, NJ, USA
35. Dickmanns ED, Mysliwetz BD (1992) Recursive 3-D road and
relative ego-state recognition. In: IEEE transactions on pattern
analysis and machine intelligence, vol 14(2), IEEE Computer
Society, Washington, DC, USA, pp 199–213
36. Gumpp T, Nienhüser D, Liebig R, Zöllner JM (2009)
Recognition and tracking of temporary lanes in motorway
construction sites. In: IEEE proceedings of intelligent vehicles
symposium, Xi’an, China, IEEE, Piscataway, NJ, USA
37. Nedevschi S, Schmidt R, Graf T, Danescu R (2004) 3D
Lane detection system based on stereovision. In: IEEE
intelligent transportation systems conference, Piscataway,
NJ, USA
38. Loose H, Franke U, Stiller C (2009) Kalman particle filter for lane
recognition on rural roads. In: IEEE proceedings of intelligent
vehicles symposium, Xi’an, China, Piscataway, NJ, USA
39. Jähne B (2005) Digitale bildverarbeitung. Springer, Berlin
40. Bundesministerium für Verkehr (ed) (1993), Instructions for the
construction of road markings – Richtlinien für die Markierung
von Straßen, Bundesministerium für Verkehr – German Minis-
try of Transport, Köln
41. Research Association for Road and Transportation (ed) (1996)
Instructions for the construction of roads – richtlinien für die
anlage von Straßen, Forschungsgesellschaft für Straßen- und
Verkehrswesen – Research association for road and transpor-
tation, Köln
42. Ljung L (1999) System identification: theory for the user. Pren-
tice Hall, New Jersey
43. MacQueen J (1967) Some methods for classification and anal-
ysis of multivariate observations. In: Proceedings of fifth Berke-
ley symposium on mathematical statistics and probability,
vol 1. University of California Press
44. Blervaque V, Mezger K, Beuk L, Loewenau J (2006) ADAS
horizon – how digital maps can contribute to road safety.
In: Proceedings of advanced microsystems for automotive
applications Berlin
368 Driver Behavior at Intersections
Driver Behavior at Intersections
TOSHIHISA SATO, MOTOYUKI AKAMATSU
Human Technology Research Institute, National
Institute of Advanced Industrial Science and
Technology (AIST), Tsukuba-shi, Ibaraki, Japan
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Behavioral Data Collection in an Actual Road
Environment
Analysis of Data Distribution Based on Traffic and
Road Environments
Modeling of Driver’s Preparatory Behavior Before
Making Driver’s Side Turn
Future Directions
Bibliography
Glossary
Driving simulator An indoor system providing
a multisensory environment for a driver to perceive
and control virtual vehicle movements. A standard
simulator has a vehicle cockpit; a visual system,
including screens and image generators; an audio
system; and a motion system that gives the driver
vehicle vibration and motion linked with the
driver’s operations.
HMI (human–machine interface/interaction) Inter-
face or interaction between users and computer-
based systems. The systems provide the users with
visual, auditory, and/or haptic information. The
users operate the systems using input devices,
including a remote controller, a touch panel, and
their voice.
Human-centered design Interface design of informa-
tion contents or information devices that is adap-
tive to human cognitive and/or physical functions.
In the automotive technology, research on the
driver characteristics in perceptual, cognitive, and
operational functions while driving is conducted
for the purpose of applying the research findings
to the interface design development of driver assis-
tance systems, such as car navigation systems.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Instrumented vehicle A passenger vehicle equipped
with various sensing technologies to detect and
track internal and external conditions and
a driving recorder system to save the measured data.
ITS (intelligent transport systems) Application of
information, communication, and sensor technol-
ogies to multiple modes of transportation, includ-
ing road, rail, air, and waterborne transports.
Expected benefits by the introduction of ITS are
reduction of traffic accidents, mitigation of traffic
congestion, environmental improvement, positive
economic impact, etc.
Naturalistic driving behavior Observation of driving
behavior that takes place in its natural setting. The
drivers are given no special instructions, no exper-
imenter is present, and the data collection instru-
mentation is unobtrusive.
Preparatory behavior Driving behavior that occurs
before making a turn at an intersection. This behav-
ior includes activation of turn signal, release of the
accelerator pedal, movement of driver’s right foot
to cover the brake pedal, and onset of pressure on
the brake pedal.
Definition of the Subject and Its Importance
Vehicle navigation and communication systems play
a role of an information center in road-traffic environ-
ments and a key component for realizing ITS (intelli-
gent transport systems). The most popular function of
the in-vehicle navigation systems is route guidance.
The route guidance system presents drivers real-time,
step-by-step driving instructions, such as preparation
for turns, exits, or road changes. This function helps
drivers choose and maintain efficient routes with less
mental effort.
However, an inadequate interface design may lead
to driver distraction or inattention to the primary
driving tasks. This results in a reduction of driver
acceptance of the route guidance as well as
a reduction of driving safety. It is important to
develop interface design of route guidance system in
order for drivers to recognize the displayed contents
correctly.
The in-vehicle navigation systems provide drivers
with visual and auditory route guidance instruction as
# 2012, DOI 10.1007/978-1-4419-0851-3

well as a digital map with road information. The audi-
tory message of the route guidance is usually presented
three times while approaching the target intersection
[1]. First, about 700 m before the target intersection,
auditory information is provided for drivers to select
the lane appropriate for the next turning direction.
Second, guidance about 300 m before the intersection
is presented when drivers can recognize a landmark.
Third, just prior to reaching the intersection, a voice
message (e.g., “turn to the right soon!”) attracts atten-
tion to make the turn and can trigger a change in the
driver’s maneuvers from moving straightforward to
preparing to make the turn. This essay focuses on the
presentation timing of the last voice guidance. This last
instruction is very important for the maneuver of trac-
ing the provided route, especially where drivers have
difficulty in recognizing the landmarks around the
turning point due to complex road-traffic environ-
ments. Too early or too late presentation timing of
the last message may lead to drivers’ annoyance with
the provided guidance and to drivers’ navigational
error, such as passing over the turning point. It is
important to develop the presentation timing of the
last voice guidance that enables drivers to begin pre-
paring to make a turn at the same location as in their
typical driving pattern. Understanding of the typical
onset location of driver preparatory behavior for mak-
ing a driver’s side or curb side turn is necessary in the
development of the optimal presentation timing.
When drivers missed correct turn, the navigation
systems search the route again and update the guidance
information. Rerouting requires a little time, and this
delay may lead to a reduction in driver utilization of the
route guidance. Prediction of the onset location of
the driver’s preparatory maneuvers prior to reaching
the target intersection can contribute to enhancing
navigation rerouting functions. For example, if
a driver did not begin to prepare to make a turn after
reaching the predicted onset point, the in-vehicle sys-
tem can assess the driver’s navigational error, deter-
mining whether he/she made an error by identifying
an incorrect turning point or did not notice the infor-
mation provided. The system can then reissue visual
and/or auditory assistance information about the loca-
tion of the target intersection or search the alternative
route soon and present the updated route guidance.
Modeling of the typical preparation behavior based on
Driver Behavior at Intersections 369
the driving behavior data is necessary to predict the
onset location where the driver begins to prepare to
make a turn under a specific road-traffic situation.
This essay describes a methodology of understand-
ing drivers’ typical preparatory behavior in a real road-
traffic environment and of constructing preparatory
behavior model using the behavioral data measured
on the real road.
Introduction
Driving tasks have three levels of hierarchical structure:
strategic, tactical, and operational [2]. The strategic
level involves trip planning, determination of trip
destination, and route choice, as well as general
considerations about driving, including an evaluation
of the costs, risks, satisfaction, and comfort.
Tactical behaviors involve decision-making, such as
the driving speeds, headway distances, and gaps in
traffic. Operational behaviors involve the stable driving
of the vehicle, including acceptable steering, and mod-
erate acceleration and braking. The behavior at higher
level has an effect on that at lower level, e.g., maneuvers
at the tactical level (turning, overtaking, obstacle avoid-
ance, and gap acceptance) must meet the criteria
derived from the general goals set at the strategic
level, and vice versa: e.g., a driver selects a route that
includes many roads with less traffic volume when the
driver likes a large margin to vehicles in the vicinity of
his/her car.
Route guidance of in-vehicle navigation systems
supports drivers’ trip planning at the highest level of
the driving tasks. In Japan, the route guidance systems
presented a digital map, vehicle’s current location on
the map, and visual and auditory route guidance
instructions to a destination as HMI in 1990s when
the navigation systems began in widespread use. On the
other hand, turn-by-turn displaying methods just indi-
cating the next turning direction on the display were
introduced in U.S. and Europe. Historically, the first
type of the route guidance systems was developed by
Honda in 1981. The contemporary navigation system
indicating the vehicle’s location on a map shown on
a display device was mounted on the Crown of Toyota
in 1987. Then, the first practical route guidance system
which had real-time presentation of the vehicle loca-
tion on a road map was the navigation system mounted
370 Driver Behavior at Intersections
on the Nissan Cima, developed by Sumitomo Electric
Industries, Ltd. in 1989. Please see literature [3] for
development and commercialization process of navi-
gation systems in Japan.
Route guidance systems using a digital map are now
on the market all over the world. Recently, the naviga-
tion systems have more sophisticated functions,
including real-time update of route guidance based
on dynamic road-traffic information. These systems
are integrated with several kinds of road sensors and
road-traffic management center via multiple media.
The systems provide the driver with the route guidance
which takes into account current traffic jam conditions.
In addition, the navigation systems present drivers with
information on various facilities around the driving
area, such as a car park.
The advanced functions of the in-vehicle navigation
systems are expected to enhance road safety as well as to
assist drivers in choosing and maintaining efficient
routes with lower mental workload. However, the use
of the new systems has strong concerns about driver
confusion and distraction that can lead to an interfer-
ence with the primary driving task and a reduction of
driver acceptance of the provided information. There-
fore, research on in-vehicle navigation systems with
a human-centered design approach has been
conducted and focused on the development of the
interface design that is adapted to driver behavior in
driving contexts [4].
The route guidance of in-vehicle navigation sys-
tems consists of visual and auditory information. The
systems display a digital map with the route to a
destination in real time while traveling. About 300 m
before the target intersection, the systems present
a magnified map of the vicinity of the intersection
and the direction of the upcoming turn. The provision
of the magnified map helps drivers identify the loca-
tion of their next turn. Thus, visual information
accompanying the magnified map is helpful in reduc-
ing drivers’ misunderstanding of the target intersec-
tion. Human factor studies have investigated the
information contents based on drivers’ cognitive
map and landmark designs that are suitable and
acceptable to drivers [5–9].
In addition to the visual information, the naviga-
tion systems present auditory route guidance three
times while approaching an intersection where to
turn. The presentation timing is about 700 m before
the intersection, about 300 m before it, and just prior to
reaching it. The distance required for a lane change
while approaching the target intersection, investigated
by an empirical survey on urban roads [10], is “700 m.”
The distance where a landmark around the intersection
(e.g., intersection sign, traffic signal, configuration of
intersection) becomes recognizable and drivers identify
the turning point is “300 m.” The last guidance is
important for drivers to leave the current road correctly
because drivers frequently cannot recognize the land-
marks around the turning point because of complex
road-traffic environments, especially in urban areas. It
is important to present the last guidance with appro-
priate timing that helps drivers begin preparing to
make a turn at the same location as in their usual
driving pattern. However, there may be some variations
in the driving operations because the driving behavior
is influenced by the road-traffic conditions [11]. Mea-
surement and analysis of driver preparatory behavior
prior to turning in an actual road environment are
required and should clarify the influence of various
road-traffic conditions – such as the presence of vehi-
cles to the front and rear, the structure of the intersec-
tion, and the condition of the road surface – on the
driver preparatory maneuvers. This essay focuses on
the presentation timing of the last route guidance
before making a driver’s side turn at an intersection
and deals with the existence of a lead and following
vehicle and the intersection type (standard intersec-
tion, T-junction, and intersection after a curve). The
behavior data described in this essay were collected on
roads in Japan, where driving is on the left side of the
road. Therefore, the right turn as driver’s side in this
essay corresponds to a left turn where driving is on the
right side of the road, e.g., the United States.
Current navigation systems reroute when a driver
failed to identify the turning point or failed to notice
the provided guidance and he/she passed over the tar-
get intersection. Prediction of the onset location of the
driver preparatory behavior can contribute to detecting
the driver’s navigational error before passing through
the intersection at which to turn. The in-vehicle system
can assess driver’s failure of identifying the required
turning point or his/her unawareness of the informa-
tion provided if the driver does not begin to prepare for
making a turn after reaching the predicted location.
 Driver Behavior at Intersections 371

Thus, the preparatory behavior prediction can lead to
further advanced functions of the route guidance sys-
tems, such as restating the route guidance and quick
rerouting and providing the updated route. The former
function is expected to reduce the driver’s error of
passing over the target intersection, and the latter is
expected to enhance usability of the route guidance
systems. The influence of the road-traffic environments
on the driver preparatory behavior should be evaluated
quantitatively in order to predict the onset of the
behavioral event. To accomplish this, a driving behav-
ior model should be constructed based on naturalistic
behavioral data measured repeatedly in a real
environment.
Figure 1 presents a design concept of the presenta-
tion timing adapted to driver’s typical preparatory
behavior and the detection of driver navigational
error based on prediction of the preparatory behavior
onset. This essay describes a method of estimating
typical onset location of driver behavior for preparing
to make a driver’s side turn and of predicting the onset
location using statistical driver model.
Behavioral Data Collection in an Actual Road
Environment
Overview of Data Collection Methods
Driving a vehicle in traffic is a task that is influenced by
road-traffic environments. The preparatory behavior
before making a turn is also influenced by traffic con-
ditions while approaching an intersection where to
turn and by road environments of the target intersec-
tion. It is essential to collect data on the external factors
that affect the driver preparatory behavior in addition

Intersection
(Turning point)

Restatement of route guidance

Presentation of the updated guidance quickly

Detection of driver navigational

errors when the driver does not
Estimation of typical onset begin to prepare to make a right
location of driver preparatory turn at the predicted onset
behavior for making a right turn location

Driver reaction time to

the provided information

Determination of presentation timing

of route guidance instruction
Prediction of onset location of
Turn to the right Driver’s
the driver preparatory behavior,
soon! vehicle
taking into account road traffic
conditions

Driver Behavior at Intersections. Figure 1

Design concept of the presentation timing of last voice guidance compatible with driver behavior and the detection
of driver navigational error. The typical onset location of driver preparation before making a driver’s side turn is estimated
based on measurement of naturalistic behavior data while approaching a target intersection. The driver navigational
error is detected based on comparison between the actual behavior and prediction of the onset location using statistical
driver model
372 Driver Behavior at Intersections
to the data on the driver’s operations for the prepara-
tions for making a turn.
A driver experiences only a few kinds of traffic
conditions for a short term. For example, 1 day, there
is a lead vehicle driving slowly in front of the driver’s
vehicle on one leg of a route; another day, there is a lead
vehicle driving faster on the same leg; and the other
days, he/she has no preceding vehicles on the same leg.
Thus, a few days are too short to understand the
driver’s typical preparatory behavior under real traffic
conditions. It is necessary to carry out repeated exper-
iments on the same roads over a long term. The long
term for the data collection is effective in recording the
driver preparatory behavior on various kinds of road
conditions, including dry, wet, and rainy roads. Anal-
ysis of the driving behavior that is measured on differ-
ent intersections with various structures is necessary to
investigate the influence of the road structures on the
driver preparatory behavior. A driving route, which
includes several kinds of intersection types – e.g., inter-
section on a straight road, intersection at the end of the
road (T-junction), intersection after a curved road, and
intersection after a slope – should be taken into account
when designing and conducting an experiment.
Methodologies for the behavior data collection are
point measurements, driving simulator experiments,
and field experiments. Traffic-flow point measure-
ments are conducted in which the number of vehicles,
the time intervals between passing vehicles, and the
traveling speeds are recorded via video cameras fitted
at the site [12, 13]. However, the point measurements
cannot clarify onset location and timing of driver decel-
erating operations.
Driving simulators have been used as experimental
tools to measure driver operations and to examine the
relationships between the external factors and the driv-
ing behavior [14–16]. Recently, image generation tech-
nologies, such as construction of geometric road
structures based on CAD data and construction of
scenery along the roads using texture mapping, have
been applied to the development of realistic road envi-
ronments [17]. However, the simulator cannot fully
reproduce natural driving behavior in a real road-
traffic environment due to mechanical restrictions,
e.g., the lack of longitudinal movement during decel-
eration and the lack of speed and distance perception
[18, 19]. These limitations negatively influence the
driver operations before making a driver side turn.
For example, a driver tends to apply the brake pedal
several times before coming to a full stop at the target
intersection in the driving simulator environment,
whereas the stroke of the brake pedal is executed once
and the driver presses down on the brake pedal gradu-
ally in the real road environment. Therefore, the onset
location and timing of the preparatory behavior at
a real intersection differ from those at the intersection
of a driving simulator where the geometric structure
and roadside landmarks are represented exactly similar
to the real intersection.
Although repeated field experiments for a long
period have a risk for traffic accidents, the real-world
experiments are the most effective in collecting natu-
ralistic data on the driver preparatory behavior. The
recent development of memory devices to record large-
scale data and sensing technologies to detect and track
external conditions contributes to the realization of the
field data collection. The following two approaches are
used for the repeated experiments on an actual road:
(1) installation of memory devices and sensors into
drivers’ owned passenger vehicles and (2) development
of instrumented vehicles equipped with various sensors
and a driving recorder system.
The representative project of the former method is
the 100-Car Naturalistic Driving Study by Virginia
Tech Transportation Institute [20]. In this project, sev-
eral data collection devices, including speed sensor, GPS,
accelerometers, front and rear radar sensors, and five
channels of digital video, are installed into participants’
owned vehicle. The large-scale behavior data have been
recorded over 13 months with the primary purpose of
collecting precrash and near-crash naturalistic driving
data. The advantages of using privately owned vehicles
are to eliminate the participants’ unfamiliarity with an
experimental vehicle and to eliminate their awareness of
being monitored. This method has limitations from the
viewpoint of driver preparatory behavior analysis in
extracting a large amount of behavior data around the
target intersections from the large-scale data.
Use of Instrumented Vehicles to Collect
Behavioral Data
Instrumented vehicle is developed to collect driver behav-
ior data as well as to measure the vehicle status and the
 Driver Behavior at Intersections 373

CCD camera for

recording front scene CCD cameras for driver monitor
Laptop PC & Driving
recorder system

CCD camera for

recording left line
D-GPS sensor Laser radar for measuring
leading vehicle

Laser radar for measuring

following vehicle
Pedal sensors for foot
Speed sensor movement detection
Steering sensor
G-sensor & CCD camera for
& lever sensor
gyro sensor recording right line

CCD cameras for

recording rear scene

Driver Behavior at Intersections. Figure 2

Overview of the AIST instrumented vehicle. Several sensors and a driving recorder system are fixed inside the trunk where
the drivers cannot see the instruments, thus encouraging naturalistic driving behavior. A participant drives alone this
vehicle during measurement trials

traffic conditions around the vehicle. Figure 2 presents
the AIST instrumented vehicle [21]. The following data
sets are collected using the instrumented vehicle:
● Driving speed
● Vehicle acceleration (longitudinal, lateral, and
vertical)
● Angular velocity (roll, pitch, and yaw)
● Geographical position of the vehicle
● Relative distance and speed to leading vehicle
● Relative distance and speed to following vehicle
● Application of accelerator and brake pedals
● Position of driver’s right foot (covering the pedal
without pressing)
● Steering wheel angle
● Turn signal activation
● Visual images (forward and backward traffic scenes,
left and right lane positions, and the driver’s face)
A speed sensor detects the driving speed using
a speed pulse signal. A G-sensor detects the vehicle
acceleration in longitudinal, lateral, and vertical direc-
tions, and a gyro sensor measures the angular velocity
in roll, pitch, and yaw directions. A D-GPS sensor
obtains the geographical position of the vehicle. Laser
radar units fixed within the front and rear bumpers
record the relative distance and relative speed to the
leading and following vehicles. Potentiometers mea-
sure the applications of the accelerator and brake
pedals. Laser sensors fitted above the pedal surfaces
detect the position of the driver’s right foot (covering
the accelerator or brake pedal without pressing).
Encoders added to the steering wheel and turn signal
lever record the steering wheel angle and turn signal
activation. A total of five CCD cameras are used to
record visual images, including the forward and back-
ward traffic scenes, left and right lane positions, and the
374 Driver Behavior at Intersections

driver’s face. These data are recorded on a laptop com- preparatory behavior differs between encountering
puter and mobile hard disks via a driving recorder green and red traffic lights. It is recommended to
system. The sampling frequency of all data except the exclude the behavioral data recorded at the red traffic
visual images is 30 Hz; the visual images are sampled at light when developing the presentation criteria of route
eight frames/s. The abovementioned instruments are guidance based on the measurement of the typical
arranged so as to be as unobtrusive as possible (the preparatory behavior [24].
recorder system is fixed inside the trunk of the vehicle)
in order to encourage naturalistic driving behavior. Influence of Vehicle Velocity on Driver Preparatory
The disadvantage of using the instrumented vehicle Behavior
is participants’ unfamiliarity with the experiment vehi-
Vehicle velocity while approaching an intersection influ-
cle. Practice drives before measurement trips are
ences the driver decelerating operation before making
important in order for the participating drivers to be
a driver’s side turn. In contrast, the onset location of the
familiar with the instrumented vehicle and the experi-
turn signal activation is almost stable regardless of the
mental atmosphere as well as to drive from the start to
vehicle velocity. Figure 3 presents the onset locations of
the end without using any assistance, such as a map,
driver preparatory behavior in the five categorized
a vehicle navigation system, and passengers. The driv-
vehicle velocities. Field experiments using the AIST
ing route is predetermined, which included several left
instrumented vehicles were conducted, and the prepa-
and right turns. The participant rides alone in the
ratory behavior data were collected before turning at an
vehicle during the experiment trials, whereas the exper-
intersection with two traffic lanes and a designated lane
imenter rides with the participants on the first day of
for making a driver’s side turn. The number of partic-
the trials in order to confirm that the participants are
ipants was four (three males and one female). The
following the correct route. Thus, the data from the
average age of the participants was 34.8 years, and the
first day are not used for subsequent data analyses and
average driving experience was 16.3 years. The total
data modeling. The participating drivers are instructed
trips for each participant were 40 over a period of
only to drive in their typical manner.
about 2 months (weekdays). The four participants
started driving on the identical route (total mileage,
Analysis of Data Distribution Based on Traffic and 15 km) at 10-min intervals. Please see literature [24] for
Road Environments detailed experiment methods and analyses. The vehicle
velocity is measured when the driver moves the right
Driver preparatory behavior before making a turn is
foot from the accelerator pedal to the brake pedal
composed of turn signal activation and deceleration.
because the velocity, when covering the brake pedal
The decelerating operations involve releasing the accel-
and when activating the turn signal, is almost the same.
erator pedal, moving the right foot from the accelerator
The remaining distances to the center of the inter-
pedal to the brake pedal, and pressing the brake pedal.
section when covering the brake pedal are directly
In this section, data analysis of the turn signal use and
proportional to the driving speeds, and the onset loca-
the foot movement to cover the brake pedal is intro-
tions of the foot movement become closer to the inter-
duced as representative maneuvers of the driver prepa-
section when approaching it more slowly. In contrast,
ratory behavior. The accelerator pedal is released not
the proportional relation between the speed and the
only to decelerate for making turns but also to main-
turn signal use is not remarkable. The onset locations
tain an appropriate driving speed [22]. The brake pedal
of the turn signal operation do not change in any
application is an operation after driver’s decision-
categorized driving speeds.
making on leaving the current road, and it is influenced
by driver factors (such as driving skill) other than the
Influence of Traffic Conditions on Driver
decision to leave the current road [23]. Only the data
Preparatory Behavior
acquired when there was a green light at the target
intersection are analyzed in the following sections. Figure 4 presents the onset location of driver prepara-
The influence of traffic conditions on driver tory behavior in the four categories of the relative
 Driver Behavior at Intersections 375

Cover a brake pedal Activate a turn signal

180 180

160 160
center of intersection (m)

center of intersection (m)

140 140
Remaining distance to

Remaining distance to
120 120

100 100

80 80

60 60

40 40

20 20

0 0
13-14 15-16 17-18 19-20 21- 13-14 15-16 17-18 19-20 21-
Vehicle velocity (m/sec) Vehicle velocity (m/sec)
a b

Driver Behavior at Intersections. Figure 3

Relationship between the onset location of driver preparatory behavior and vehicle velocity. The graphs present
averages and standard deviations of the remaining distances to the center of the intersection when each behavioral event
occurs in each category of the vehicle velocity. The data were collected at an intersection with two traffic lanes and
a designated turn lane

Cover a brake pedal Activate a turn signal

160 160

140 140
center of intersection (m)

center of intersection (m)

Remaining distance to

Remaining distance to

120 120

100 100

80 80

60 60

40 40

20 20

0 0
0-19m 20-39m 40m- No leading 0-19m 20-39m 40m- No leading
vehicles vehicles
a Distance to a leading vehicle b Distance to a leading vehicle

Driver Behavior at Intersections. Figure 4

Relationship between the onset location of driver preparatory behavior and relative distance to a leading vehicle. The
graphs present averages and standard deviations of the remaining distances to the center of the intersection when each
behavioral event occurs in each category of the relative position to the leading vehicle. The data were collected at the
intersection with two traffic lanes and a designated turn lane
376 Driver Behavior at Intersections

distance to the lead vehicle, including the absence of
a leading vehicle. The data used are the same as those in
Fig. 3. The relative distances from the leading vehicle
are measured before the driver enters the designated
turn lane. Here, the leading vehicle is a forward vehicle
that travels straight toward the target intersection. The
remaining distances to the center of the intersection
when covering the brake pedal while driving with
a leading vehicle are shorter than those during drives
without a leading vehicle. In addition, shorter relative
distances to the leading vehicle lead to shorter
remaining distances at the onset of covering the brake
pedal. On the other hand, the onset location of turn
signal activation is at a remaining distance of between
80 and 90 m to the center of the target intersection. The
turn signal use is independent of the existence of and
relative distance to a leading vehicle.
Figure 5 presents the onset location of driver pre-
paratory behavior in the four categories of the relative
distance from a following vehicle, including the
absence of a following vehicle. These data are also
the same as those in Fig. 3. The relative distances
from the following vehicle are measured before the
driver enters the designated turn lane. Here, the
following vehicle is a vehicle that follows the driver’s
vehicle in the same driving lane. The remaining dis-
tances when covering the brake pedal while driving
with a following vehicle are shorter compared to driv-
ing without a following vehicle, particularly for
a shorter relative distance between vehicles. On the
other hand, the remaining distances when the turn
signal is activated are almost stable in the four catego-
ries of the relative distance from the following vehicle,
ranging from 80 to 90 m.
The vehicle velocity while approaching the inter-
section is correlated with the relative distance between
the driver’s vehicle and a leading or following vehicle. It
is hypothesized that only the slower driving speed leads
to a shorter remaining distance when covering the
brake pedal. At a specific range of vehicle velocity
(from 16.0 to 17.5 m/s), the onset locations when
covering the brake pedal were compared among the
four situations: driving with both leading and follow-
ing vehicles, driving with only a leading vehicle, driving
with only a following vehicle, and driving without
either a leading or following vehicle. The remaining
distance at the onset of covering the brake pedal was
the shortest when driving with both leading and

Cover a brake pedal Activate a turn signal

160 160

140 140
center of intersection (m)

center of intersection (m)

120 120
Remaining distance to

Remaining distance to

100 100

80 80

60 60

40 40

20 20

0 0
0-19m 20-39m 40m- No following 0-19m 20-39m 40m- No following
vehicles vehicles
a Distance to a following vehicle b Distance to a following vehicle

Driver Behavior at Intersections. Figure 5

Relationship between the onset location of driver preparatory behavior and relative distance to a following vehicle.
The graphs present averages and standard deviations of the remaining distances to the center of the intersection
when each behavioral event occurs in each category of the relative position from the following vehicle. The data were
collected at the intersection with two traffic lanes and a designated turn lane

following vehicles. The onset location of covering the
brake pedal was closer when driving with only a leading
vehicle than that when driving with only a following
vehicle.
Drivers begin to decelerate at a point closer to the
center of the intersection under car-following condi-
tions compared to free-running conditions at the same
driving speed. The preparation point is the closest to
the target intersection when drivers prepare to make
a turn with a following vehicle in addition to a leading
vehicle. Drivers following a forward vehicle pay atten-
tion to that vehicle in order to respond quickly to its
actions [25, 26]. Additionally, they pay attention both
forward and backward in so-called platoon conditions
under which they travel a short distance to leading and
following vehicles. Intense concentration on vehicles
surrounding the driver’s vehicle may lead to a delay in
the onset of driver decelerating maneuvers before mak-
ing a driver’s side turn at an intersection.
The analyses of turn signal use within the specific
range of the vehicle velocity reveal the fixed location of
turn signal activation regardless of the traffic condi-
tions surrounding the driver’s vehicle. As mentioned
from Figs. 3–5, the remaining distances to the center of
the target intersection when activating the turn signal
are almost constant in all ranges of the vehicle velocity
and of the relative distance from the leading or follow-
ing vehicle. The analyzed intersection has a designated
turn lane (the distance between the entrance of the
designated lane and the center of the intersection is
70 m). This road alignment feature may contribute to
the constant location at which the drivers activate the
turn signal.
Influence of Road Structure on Driver Preparatory
Behavior
Traffic conditions (the existence of leading and/or fol-
lowing vehicles) influence the driver decelerating
maneuver at a specific intersection with two traffic
lanes and a designated turn lane. Moreover, the road
structures of intersection where to turn influence the
relations between the traffic conditions and the driver
preparatory behavior for decelerating before a driver’s
side turn. Figure 6 presents the onset locations of cov-
ering the brake pedal in the four categorized traffic
conditions at three kinds of intersections. The data
Driver Behavior at Intersections 377
collection methods were as follows: number of partic-
ipants, eight (five males and three females); average age
of the participants, 38.1 years; average driving experi-
ence, 18.1 years; total trips for each participant, 40 over
a period of about 2 months (weekdays). Four partici-
pants started driving on the route, including the target
intersections, at 10-min intervals. The other drivers
made recorded trips at another period following the
same procedures. Please see literature [27] for detailed
experiment methods and analyses. Intersection A is
a standard right-angle intersection on a straight road
with one traffic lane. Intersection B is a T-junction at
the end of the road. Intersection C is also a T-junction,
and this intersection exists after drivers drive along
a curved road (curve radius, 126.9 m).
At intersection A, the remaining distances while
driving with leading and following vehicles and while
driving with only a leading vehicle are shorter than
those while driving without a leading or following
vehicle. At intersection B, the onset locations of cover-
ing the brake pedal tend to be longer to the center of the
intersection when there is a following vehicle compared
to only a leading vehicle and without a leading or
following vehicle. At intersection C, the remaining
distances to the center of the intersection when cover-
ing the brake pedal are longer when there is a vehicle in
front. The onset location is the longest among the four
traffic conditions when driving with both leading and
following vehicles.
The comparison between the standard intersection
and the intersection with a designated turn lane indi-
cates that the number of traffic lanes has less influence
on the relationship between the traffic conditions and
the driver decelerating behavior before making
a driver’s side turn. The behavioral analysis at the
intersection after a curved road suggests that drivers
begin to prepare for deceleration at an earlier point
while approaching the intersection with leading and
following vehicles and with only a leading vehicle com-
pared to without such vehicles. Drivers cannot see the
landmarks, including traffic lights, and cannot recog-
nize the remaining distance to the turning point until
just before the intersection after a curve. At the stan-
dard intersection, drivers can see and recognize the
location of the intersection, and they may adapt their
behavior to the leading vehicle until they are closer to
the intersection. In contrast, drivers change their
378 Driver Behavior at Intersections

100

center of intersection (m)

Remaining distance to
80

60

40

20

0
With leading With a following
and following vehicle
vehicles Without leading
With a leading and following
vehicle vehicles
a Intersection A

center of intersection (m) 100

Remaining distance to

80

60

40

20

0
With leading With a following
and following vehicle
vehicles Without leading
With a leading and following
vehicle vehicles
b Intersection B
center of intersection (m)

60
Remaining distance to

40

20

0
With leading With a following
and following vehicle
vehicles Without leading
With a leading and following
vehicle vehicles
c Intersection C

Driver Behavior at Intersections. Figure 6

Relationship between the onset locations of driver preparatory behavior and traffic conditions at three kinds of
intersections. The graphs present averages and standard deviations of the remaining distances to the center of the
intersection when the drivers move their right foot from the accelerator pedal to the brake pedal in each category of the
traffic conditions

driving mode from following the leading vehicle to
preparing to make a driver’s side turn at an earlier
point when the intersection cannot be recognized in
advance.
Drivers begin to decelerate farther from the
T-junction on a straight road when they have
a vehicle following. In this case, drivers may intend to
use their brake lights as a signal to the following driver
to avoid a rear-end collision by the following vehicle. At
T-junction, the leading vehicle always decelerates and
turns right or left. Drivers can anticipate the decelera-
tion of the lead vehicle, and they may pay more atten-
tion to the following vehicle compared to the
movement of the lead vehicle while approaching the
T-junction with leading and following vehicles.
A driver’s recognition of a following vehicle may lead
to earlier preparation for deceleration. Intersection C is
also a T-junction, but the onset location of covering the
brake pedal when driving with only a following vehicle
is similar to driving without leading or following vehi-
cles. Drivers should pay attention to the inside of the
curve during curve negotiation [28]. Thus, drivers
might not recognize the presence of a following vehicle
while approaching an intersection after a curve. When
drivers notice the following vehicle before entering the
curve, they may allocate more resources to the curve
negotiation and may not pay attention to the rear
vehicle.
Recommendations for Presentation Timing of Route
Guidance
The presentation timing of route guidance instruction
compatible with driver’s typical preparatory behavior
is recommended to enhance driver acceptance of the
provided information. Data analyses described in sec-
tion “Influence of Vehicle Velocity on Driver Prepara-
tory Behavior, Influence of Traffic Conditions on
Driver Preparatory Behavior, Influence of Road Struc-
ture on Driver Preparatory Behavior” suggest that it is
effective to change the timing of the last route guidance
presentation based on the vehicle velocity, the existence
of and relative distance to leading and following vehi-
cles, and the road structures of the turning point.
First of all, the presentation timing should be
changed corresponding to the driving speeds while
approaching a target intersection. Driving faster
Driver Behavior at Intersections 379
requires earlier presentation timing compared to driv-
ing slower. In addition to the vehicle velocity, the traffic
conditions surrounding the driver’s vehicle should be
taken into account. Presenting the instruction closer to
the intersection would be effective for guidance and
compatible with driver preparatory behavior while
approaching right-angle intersections on straight
roads under close car-following conditions, whereas
earlier presentation would be acceptable while driving
with a leading and/or following vehicle at long range
and while driving without leading or following vehi-
cles. Moreover, the route guidance presentation with
different timings at intersections with T-junctions and
at intersections after curves is recommended. Earlier
guidance would effectively enhance drivers’ utilization
of the instruction while approaching a T-junction on
a straight road when driving with a following vehicle,
independent of the existence of the lead vehicle. At
intersections after curves, drivers cannot see or recog-
nize the relation between their vehicle and the turning
point until just before the target intersection. In these
cases, the route guidance instructions should be pro-
vided earlier in car-following situations to improve the
drivers’ acceptance of the presented information.
Modeling of Driver’s Preparatory Behavior Before
Making Driver’s Side Turn
Preparatory Behavior Model Using Structural
Equation Modeling
Data analyses of driver preparatory behavior before
making a driver’s side turn at a specific intersection
with a designated turn lane reveal linear relationships
among the vehicle velocity, the traffic conditions in the
vicinity of the driver’s vehicle, and the preparatory
behavior (sections “Influence of Vehicle Velocity on
Driver Preparatory Behavior and Influence of Traffic
Conditions on Driver Preparatory Behavior”). The
onset locations of the foot movement to cover the
brake pedal are positively correlated with the driving
speeds or the remaining distances to leading and fol-
lowing vehicles. In contrast, the onset locations of the
turn signal operation are almost constant, independent
of the speeds and the traffic conditions. Structural
equation modeling is an efficient modeling tech-
nique for quantifying the impact of independent vari-
ables on the dependent variables that suggests linear
380 Driver Behavior at Intersections
relations [29]. Thus, the structural equation model can
be applied to describe quantitatively the relationships
among the vehicle velocity, the relative distances to
the leading and following vehicles, and the onset loca-
tions of covering the brake pedal and activating the
turn signal.
Structural equation modeling represents the inter-
relations among a set of variables comprising a model
hypothesized by theory and empirical research.
Observed and latent variables are used in the structural
equation model: observed variables are directly col-
lected and measured, and latent variables are inferred
from the set of observed variables. The parameters
specified in the structural equation model are
estimated so as to minimize differences in the
variance–covariance matrices between the hypothesized
theoretical model and the measurement data. Various
kinds of model fit indices evaluate whether or not the
covariance matrix of the hypothesized model is as close
as possible to that of the observed variables [30].
In the structural equation model of driver prepara-
tory behavior, the following five observed variables are
used: the driving speed, the relative distance to the
leading vehicles, the relative distance to the following
vehicles, the remaining distance to the center of the
target intersection when covering the brake pedal, and
the remaining distance to the center of the target inter-
section when activating the turn signal. Subsequently,
the constructed model is applied to a prediction of the
onset location of the driver preparatory maneuvers.
Therefore, the independent variables, the driving
speed and the headway and rear distances, at the
onset of releasing the accelerator pedal are used in the
model construction. Here, the release of the accelerator
pedal is defined as the final release maneuver of the
accelerator pedal, leading to right foot movement to
cover the brake pedal. The cases where there are no
leading and/or following vehicles when releasing the
accelerator pedal have missing data for the relative
distances to the leading and/or following vehicles.
To overcome this limitation, having no leading and/or
following vehicles around the driver’s vehicle is defined
as the condition in which leading and/or following
vehicles are such a long way from the driver’s vehicle
that the laser radar cannot detect and measure them.
The relative distances to the leading and following
vehicles in the case of no leading and/or following
vehicles are set to 100 m, a value which is beyond the
detection limits of the laser radar unit.
The latent variables are important to obtain an
acceptable model to data fit in the structural equation
model specification of driver preparatory behavior.
This is because the velocity and the headway and rear
distances influence, not independently but interac-
tively, the onset locations of driver preparations for
making a driver’s side turn. A latent variable, called
the free-driving condition level, is input to the model
specification as a latent factor describing the interac-
tion between the leading or following vehicles and the
driver’s vehicle. This latent variable uses two indicator
variables: the relative distance to the leading vehicle
and the relative distance to the following vehicle.
A higher free-driving condition level denotes that
a driver drives without leading and following vehicles
or remains a long distance from the front and rear
vehicles. Conversely, a lower free-driving condition
level denotes that a driver drives under close car-
following conditions. In addition, a latent variable,
called the location of transition to preparatory behav-
ior, is introduced as a latent factor related to the onset
location of the change in driving behavior from straight
(going forward along the roadway) mode to the prep-
aration mode. This latent variable comprises two indi-
cator variables: the remaining distance to the center of
the intersection when covering the brake pedal and the
remaining distance to the center of the intersection
when activating the turn signal. The location of transi-
tion to preparatory behavior denotes the degree to
which a driver begins to prepare for making a driver’s
side turn at a long distance from the center of the
intersection. A closer location of transition to prepara-
tory behavior denotes that the onset location of the
behavioral event is closer to the target intersection.
Please see literature [31] for the details of the model
construction.
Figure 7 presents a path diagram of the proposed
structural equation model, the estimated path coeffi-
cients and factor loadings (standardized weights), and
the results of model fit indices. The results of the model
fit indices suggest that the estimated structural equa-
tion model fits the observed data well. The vehicle
velocity and the free-driving condition level are posi-
tively related to the location of transition to prepara-
tory behavior. The relation between the free-driving

Independent variables
Vehicle
velocity
0.60
0.89
Relative distance 0.69
E3 0.29
to leading vehicle Free-driving
condition level
E4 Relative distance
0.47
to following vehicle
D1
Result of model fit indices
χ2 (df): 0.349(3), p value: 0.951
Goodness-of-Fit Index: 0.999
Adjusted Goodness-of-Fit Index: 0.994 Root Mean Squre Error of Approximation: 0.000
Driver Behavior at Intersections. Figure 7
Structural equation model of driver preparatory behavior. The values in the path diagram present the standardized path
coefficient (among vehicle velocity, free-driving condition level, and location of transition to preparatory behavior) and
factor loadings (between the latent variables and each indicator variable). The model was estimated using the maximum
likelihood method. Total data sets are 119 for 4 drivers
condition level and each indicator variable (relative
distances to leading and following vehicles) is also
positive, suggesting that the location of transition to
preparatory behavior is directly proportional to the
headway and rear distances. The factor loading from
the location of transition to preparatory behavior to
the remaining distance at the onset of covering the
brake pedal is higher than the factor loading for the
remaining distance when activating turn signal. This
means that higher driving speeds and free-driving con-
ditions contribute to an earlier onset location when
covering the brake pedal and that the independent
variables have less effect on turn signal activation.
At T-junctions and intersections after curves, the
relationships between the traffic conditions and the
onset location of driver decelerating maneuver are dif-
ferent from those at the right-angle intersections on
straight roads. In these cases, the structural equation
model should be estimated at each intersection. The
path diagram is the same among the intersections, and
the path coefficients and/or the factor loadings will be
changed according to the relationships observed at
each intersection. At T-junctions, the factor loading
Driver Behavior at Intersections 381
Dependent variables
Remaining distance to
center of intersection
0.99 E1
when covering brake
pedal
Location of transition of
preparatory behavior
Remaining distance to
0.56
center of intersection
D2 E2
when activating turn
signal
E1-E4: Measurement error D1, D2: Prediction error
Comparative Fit Index: 1.000
between the free-driving condition level and the rela-
tive distance to the following vehicle will be negative. At
intersections after curves, the path coefficients from the
vehicle velocity and the free-driving condition level to
the location of transition to preparatory behavior will
be negative.
Prediction of Onset Location of Driver Preparatory
Behavior
The path coefficients and factor loadings estimated in
the structural equation model are used to predict the
onset locations of driver preparatory operations.
Figure 8 presents the method for predicting the onset
locations of the driver preparatory behavior. The influ-
ences of the vehicle velocity on the location of transi-
tion to preparatory behavior are divided into direct
influence and indirect influence via the free-driving
condition level (Fig. 8a). The factor score for the
free-driving condition level is calculated by using the
factor loadings for the two indicator variables because
the relationship between the free-driving condition
level and the relative distances to the leading and
382 Driver Behavior at Intersections

Vehicle velocity X1
a21

a11 Location of transition to

F2-1
preparatory behavior

Free-driving b11
condition level F1

F2−1 = (a21 + a11 ⫻ b11) ⫻ X1

a Direct and indirect influence of vehicle velocity

Relative distance
X2
to leading vehicle
c21

Free-driving Location of transition to

F1 F2-2
condition level b11 preparatory behavior

c31
Relative distance
X
to following vehicle 3

F1 = XR−1C X is a matrix of standard scores of X2 and X3

R is a correlation matrix of X
C is a matrix of the factor loadings of c21 and c31

F2−2 = b11 ⫻ F1

Influence of relative distances to leading and

b following vehicles using factor score estimation

Remaining distance to
c12 intersection when Y1
covering brake pedal
Location of transition to
preparatory behavior F2−1 + F2−2

Remaining distance to
c22
intersection when Y2
activating turn signal

Y1 = c12 ⫻ (F2−1 + F2−2)

Y2 = c22 ⫻ (F2−1 + F2−2)

Prediction of onset locations of foot movement

c to cover brake pedal and turn signal activation

Driver Behavior at Intersections. Figure 8

Procedures for predicting the onset locations of driver preparatory maneuvers. The procedures are divided into three
steps: the effect of vehicle velocity on location of transition to preparatory behavior, the effect of headway and rear
distances on location of transition to preparatory behavior, and the calculation of remaining distances to intersection
when covering brake pedal and activating turn signal from the latent variable. (a) Direct and indirect influence of vehicle
velocity. (b) Influence of relative distances to leading and following vehicles using factor score estimation. (c) Prediction of
onset locations of foot movement to cover brake pedal and turn signal activation

following vehicles corresponds to a confirmatory factor
model [32]. Then, this score is used to calculate the
direct effect of the free-driving condition level on the
location of transition to preparatory behavior (Fig. 8b).
Finally, the remaining distances to the center of the
intersection when covering the brake pedal and when
activating the turn signal are predicted by using the
factor loadings between the location of transition to
preparatory behavior and the onset location of each
behavioral event (Fig. 8c).
The onset locations of covering the brake pedal and
activating the turn signal can be predicted by applying
estimated path coefficients and factor loadings to the
recorded vehicle velocity and relative distances to lead-
ing and following vehicles. The estimated structural
equation model uses the driving speed and the headway
and rear distances when drivers release the accelerator
pedal. The onset location of releasing the accelerator
pedal may be almost the same as the last route guidance
presentation that is determined based on drivers’ typ-
ical preparatory behavior. The route guidance system
linked with laser radar units and a speed pulse signal
source can predict the driver decelerating maneuver
and turn signal operation when the last voice message
is presented. Sometimes a driver does not begin prep-
aration after reaching the predicted onset location.
In such cases, the system can infer a driver error that
he/she did not accurately identify the turning point due
to a misunderstanding of the provided information or
did not notice the information provision. The systems
could then restate the voice route guidance or quickly
reroute and present the updated route.
Future Directions
Further data analyses reveal that the relationships
among the onset location of driver preparatory behav-
ior, the traffic conditions, and the road environments
before making a curb side turn are similar to those
before making a driver’s side turn described in this
essay [27]. Thus, the findings of understanding and
modeling of typical preparatory maneuvers for
a driver’s side turn can be applied to the route guidance
presentation timing and restatement before a curb side
turn.
Understanding the driver usual preparatory behav-
ior and developing the structural equation model need
Driver Behavior at Intersections 383
a large amount of measured data. Recording of the
large-scale data by one vehicle requires a few weeks.
After a driver bought an in-vehicle navigation system
and used it in his/her daily lives for a few weeks, the
route guidance system becomes sophisticated: the pre-
sentation timing becomes compatible with his/her typ-
ical preparatory behavior, and the advanced functions
of the restatement of route guidance and the quick
update of route become available when he/she misses
the turning point.
Data analysis and modeling approach can be
applied to driving tasks other than preparations for
turning at intersections. The linear relations between
the driving behavior and the traffic conditions are
found in the driver’s side turn behavior, and the struc-
tural equation modeling is adopted. If the measured
data sets suggest nonlinear relationships, probabilistic
models, including a Markov dynamic model and
a Bayesian network model, will be used for modeling
and predicting the driving behavior in a real environ-
ment (e.g., [33, 34]).
Bibliography
Primary Literature
1. Kimura K, Marunaka K, Sugiura S (1997) Human factors con-
siderations for automotive navigation systems – legibility,
comprehension, and voice guidance. In: Noy YI (ed) Ergonom-
ics and safety of intelligent driver interfaces. Lawrence
Erlbaum Associate, Mahwah, pp 153–167
2. Michon JA (1985) A critical view of driver behavior models:
What do we know, what should we do? In: Evans L, Schwin RC
(eds) Human behavior and traffic safety. Plenum, New York,
pp 485–520
3. Ikeda H, Kobayashi Y, Hirano K (2011) How car navigation
systems have been put into practical use – development
management and commercialization process-. Synthesiology –
English Edition 3:280–289
4. Pauzie A (2005) Methodologies for driver-centred evaluation
of perceptive and cognitive requirements of in-vehicle display.
In: Proceedings of 11th international conference on human-
computer interaction, Las Vegas, CD-ROM
5. Pauzie A, Daimon T, Bruyas MP (1997) How to design
landmarks for guidance systems? In: Proceedings of 4th
world congress on intelligent transport systems, Berlin,
CD-ROM
6. Daimon T, Nishimura M, Kawashima H (2000) Study of driver’s
behavioral characteristics for designing interfaces of in-vehicle
navigation systems based on national and regional factors.
JSAE Rev 21:379–384
384 Driver Behavior at Intersections
7. Kawai M, Kawasumi M, Nakano T, Yamamoto S,
(2003) Displaying method in on-board display – route guid-
ance method based on sex differences. In: Proceedings of IEEE
intelligent transportation systems council, Columbus, OH,
pp 357–360
8. Ross T, May AJ, Grimsley PJ (2004) Using traffic light informa-
tion as navigational cues: implications for navigation system
design. Transp Res Part F 7:119–134
9. May AJ, Ross T (2006) Presence and quality of navigational
landmarks: effect on driver performance and implications for
design. Hum Factors 48:346–361
10. Kishi K, sugiura S (1993) Human factors considerations for
voice route guidance. In: SAE SP-964 automotive display
systems and IVHS. Society of Automotive Engineers,
Warrendale, pp 99–109
11. Yoshioka M, Akamatsu M, Matsuoka K, Sato T (2003) Aim of
“behavior-based human environment creation technology”.
In: Proceedings of the XVth triennial congress of the interna-
tional ergonomics association, Seoul, CD-ROM
12. Vogel K (2002) What characterizes a “free vehicle” in an urban
area? Transp Res Part F 5:15–29
13. Liu BS (2007) Association of intersection approach speed
with driver characteristics, vehicle type and traffic conditions
comparing urban and suburban areas. Accid Anal Prev
39:216–223
14. Akamatsu M, Okuwa M, Onuki M (2001) Development of
hi-fidelity driving simulator for measuring driving behavior.
J Robot Mechatron 13:409–418
15. Kappe B, Winsum W, Wolffelaar P (2002) A cost-effective
driving simulator. In: Proceedings of DSC (driving simulation
conference) 2002. INRETS-RENAULT, Paris, pp 123–134
16. The National Advanced Driving Simulator (2010) The
University of Iowa. http://www.nads-sc.uiowa.edu/media/
pdf/NADS_2010Overview.pdf. Accessed 19 May 2011
17. Akamatsu M, Onuki M (2008) Trends in technologies for
representing the real world in driving simulator environment.
Rev Automot Eng 29:611–618
18. Pinto M, Cavallo V, Ohlmann T, Espie S, Roge J (2004) The
perception of longitudinal accelerations: what factors influ-
ence braking maneuvers in driving simulators?. In: Proceed-
ings of DSC (driving simulation conference) 2004 Europe.
INRETS-RENAULT, Paris, pp 139–151
19. Kemeny A, Panerai F (2003) Evaluating perception in driving
simulation experiments. Trends Cogn Sci 7:31–37
20. Dingus TA, Klauer SG, Neale VL, Petersen A, Lee SE, Sudweeks J,
Perez MA, Hankey J, Ramsey D, Gupta S, Bucher C, Doerzaph ZR,
Jermeland J, Knipling RR (2006) The 100-car naturalistic driving
study, phase 2 – results of the 100-car field experiment, Report
no. DOT HS 810 593
21. Akamatsu M, Sakaguchi Y, Okuwa M, Kurahashi T, Takiguchi K
(2003) Database of driving behavior measured with equipped
vehicles in the real road environment. In: Proceedings of the
XVth triennial congress of the international ergonomics
association, Seoul, CD-ROM
22. Sato T, Akamatsu M (2006) Analysis of drivers’ behaviour
before making a right turn at an intersection based on
driving behavior data on an actual road. RevAutomot Eng
27:287–294
23. Sato T, Akamatsu M, Daimon T (2006) Investigation on drivers’
behaviour while approaching an intersection: comparison
between real and simulated driving. In: Proceedings of driving
simulation conference Asia/Pacific (DSC-AP 2006), Tsukuba,
CD-ROM
24. Sato T, Akamatsu M (2007) Influence of traffic conditions on
driver behaviour before making a right turn at an intersection:
analysis of driver behaviour based on measured data on an
actual road. Transp Res Part F 10:397–413
25. Brackstone M, McDonald M (2007) Driver headway: how close
is too close on a motorway? Ergonomics 50:1183–1195
26. Ranney TA (1999) Psychological factors that influence car-
following and car-following model development. Transp Res
Part F 2:213–219
27. Sato T, Akamatsu M (2009) Analysis of drivers’ preparatory
behavior before turning at intersections. IET Intell Transp
Syst 3:379–389
28. Land MF, Lee DN (1994) Where we look when we steer. Nature
369:742–744
29. Schumacker RE, Lomax RG (2004) SEM Basics. In: Riegert D (ed)
A beginner’s guide to structural equation modeling. Lawrence
Erlbaum Associates, Mahwah, pp 61–78
30. Hu L, Bentler PM (1995) Evaluating model fit. In: Hoyle R (ed)
Structural equation modeling: issues, concepts, and applica-
tions. Sage, Newbury, pp 76–99
31. Sato T, Akamatsu M (2008) Modeling and prediction of driver
preparations for making a right turn based on vehicle velocity
and traffic conditions while approaching an intersection.
Transp Res Part F 11:242–258
32. Bollen KA (1989) Confirmatory factor analysis. In: Structural
equations with latent variables. Wiley, New York, pp 226–318
33. Akamatsu M, Sakaguchi Y, Okuwa M (2003) Modeling of driv-
ing behavior when approaching an intersection based on
measured behavioral data on an actual road. In: Proceedings
of the human factors and ergonomics society 47th annual
meeting, Denver, CL, pp 1895–1899
34. Pentland A, Liu A (1999) Modeling and prediction of human
behavior. Neural Comput 11:229–242
Books and Reviews
Cacciabue C (2007) Modelling driver behavior in automotive envi-
ronments: critical issues in driver interactions with intelligent
transport systems. Springer, New York
Fuller R, Santos JA (2002) Human factors for highway engineers.
Pergamon, Oxford, UK
Shinar D (2007) Traffic safety and human behavior. Emerald, Bingley
Takeda K, Hansen JHL, Erdogan H, Abut H (2009) In-vehicle corpus
and signal processing for driver behavior. Springer, New York

Driver Characteristics Based on
Driver Behavior
JIANQIANG WANG1, LEI ZHANG1, XIAOJIA LU2, KEQIANG LI1
1
State Key Laboratory of Automotive Safety and
Energy, Tsinghua University, Beijing, China
2
China Agricultural University, Beijing, China
Article Outline
Glossary
Definition of the Subject
Introduction
Driver Behavior Questionnaire Design
Data Statistical Analysis and Quantification Analysis
Comparison and Discussion
Future Directions
Acknowledgments
Bibliography
Glossary
Driver Assistance System (DAS) Also called ADAS
(Advanced Driver Assistance Systems), can prevent
accidents caused by human errors or alleviate the
damage by taking control of the vehicle just before
an accident happens. It can contribute to improving
traffic safety, environmental impact, energy effi-
ciency, traffic homogeneity, and driver comfort
and convenience. An adaptive Driver Assistance
System ensures safe driving by assisting the driver
in the following ways: reducing driver fatigue and
maintaining driver performance by supporting
a driver’s “recognitions,” “judgments,” and
“actions” displaying warning signs in the case that
the actions are judged to be dangerous and taking
control of the vehicle in the case that the driver is
unable to avoid a collision.
Driver Characteristics The characteristics performed
when driving, including the driver’s operations to
the vehicle (e.g., car-following and lane-changing),
the physiological characteristics (e.g., response,
cognition, and fatigue), and the psychological char-
acteristics and so on. It has been studied mainly in
two main aspects, i.e., driving skill and driving style.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Driver Characteristics Based on Driver Behavior 385
Driver Behavior Questionnaire (DBQ) One of the
most widely implemented measurement scales to
examine self-reported driving behaviors, developed
by Reason et al.
Factor Analysis Factor analysis was invented by psy-
chologist Charles Spearman. It is a collection of
methods used to examine how underlying con-
structs influence the responses on a number of
measured variables. It is related to the principal
component analysis but not identical; it estimates
how much of the variability is due to common
factors. It is performed by examining the pattern
of correlations (or covariances) between the
observed measures. Measures that are highly corre-
lated (either positively or negatively) are likely
influenced by the same factors, while those that
are relatively uncorrelated are likely influenced by
different factors.
Human Factors It is a theory that studies the physical
and mental state of human beings when given dif-
ferent environments, products, and services.
Reliability It is the quality of being dependable or
reliable. And it is defined as the proportion of the
real variance in the total variance in the statistics. At
present, the Cronbach’s alpha reliability coefficient
method is a commonly used measure. It is specially
used for the reliability estimation of the accumula-
tive Likert scale. It is defined as: the ratio of the sum
variance of each item and the total variance of the
whole scale.
T-Test The T-test method is a common method for
hypothesis testing. It is used to determine whether
there are differences between the two samples (such
as A and B) with an unknown but equal population
variance. It can calculate the values simply, and is
suitable for small sample testing.
Validity It is defined as the ratio of the variance
of the scores that related to the measure pur-
poses and the total variance. The content validity
and discriminate validity are the two main
indexes.
Definition of the Subject
According to the road traffic statistics and analysis
information, most accidents can be directly attributed
# 2012, DOI 10.1007/978-1-4419-0851-3
386 Driver Characteristics Based on Driver Behavior
to the driver mistake and violation behavior. In driving
behavior research, the driving skill and style can be seen
as the two main components of human factors [1]. The
deficient driving skill and aggressive driving style may
both lead to serious crash risk. Therefore, different
kinds of Advanced Driver Assistance Systems based
on intelligent technologies are developed by automo-
tive manufacturers to increase driver skill and decrease
driver workload, such as Forward Collision Warning
System and Adaptive Cruise Control System [2]. The
governments also conduct much work on the driver
education and regulation to avoid aggressive behaviors.
In order to optimize the effects of technical driving
assistance and driver normalization, understanding
and measuring of driver skill and style have become
important research topics. However, because of the
complication and variability of human drivers, it is
challenging to convert the descriptive concepts of skill
and style to quantitative mathematic variables.
Introduction
A self-reported survey is an effective method for study-
ing the driver behavior, and it can be used to analyze
the driver behavior, especially the relationship between
abnormal behavior and driver characteristics.
A number of driver measurement scales have been
designed to investigate the driver characteristics and
behavior, such as the Driver Anger Scale [3] and the
Driving Skill Inventory [4]. In these methods, the
Driver Behavior Questionnaire (DBQ) developed by
Reason et al. [5] has recently become one of the most
widely implemented measurement scales to examine
self-reported driving behaviors. Based on this DBQ and
its improved editions, many researches on driver
behavior were carried out from different perspectives,
e.g., the culture comparison [4], region diversity [6],
and the gender and age influences [7]. Although the
self-reported survey may affect the data of behavior
scale by some subjective factors and objective factors
(i.e., the uncertain factors from the outside), it cannot
reflect the driver characteristics absolutely [8]. The
existing research results show that this method has
obvious advantages than previous methods, because
a great quantity of behavior data about abnormal driv-
ing that cannot be obtained by experiments can
be collected quickly and effectively. Meanwhile, the
statistical methods that are involved in this paper
(e.g., factor analysis, correlation analysis, and cluster
analysis) provide an effective tool and basis for driver
characteristics analysis. Thus, the indicated factors
extracted from the behavior scale can accurately
describe the abnormal driver characteristics of driver
individual and population to a certain extent.
In this research, the DBQ is applied to explore
a quantification method of driver characteristics.
A self-reported survey based on DBQ is implemented
with 33 Chinese driver participants. Factor analysis is
used as the data processing method and a two-
dimensional factor structure is established to describe
the driver’s characteristics of driving skill and driving
style respectively. Some comparisons are also executed
to validate the reasonability of this method.
Driver Behavior Questionnaire Design
Participants
The self-reported data in this survey are collected from
33 Chinese drivers with different occupations. There
are 29 males (87.9%) and 4 females (12.1%). The
average age of the participants is 45.54 years (S.D. =
9.86), and the age range is from 30 to 62 years. On
average the participants hold their driving licenses for
13.21 years (S.D. = 8.29, range 3–38 years). The average
driving mileage is 180.8000 km (S.D. = 18.33), and
the driving range is from 20,000 to 600,000 km. Seven-
teen participants (51.5%) have traffic accidental
history, and to them the average accident amount is
3.24 (S.D. = 2.77).
Driver Behavior Questionnaire Method
A self-reported survey is designed based on the Driver
Behavior Questionnaire [9] and some modifications
are made to adjust the items for Chinese traffic and
driver conditions. The survey consists of three parts:
1. Individual information: including gender, age,
driving experience (in terms of driving years), and
traffic accidental history.
2. A self-evaluation on the driving skills and driving
style, which uses an estimation method of Social
Comparison [8]. The purpose of self-evaluation
is to validate the quantification method. The par-
ticipants were required to indicate on a six-point
 Driver Characteristics Based on Driver Behavior 387

scale (0 = never to 5 = nearly all the time) how often Driver Characteristics Based on Driver Behavior. Table 1
they commit each of the abnormal driving behav- (Continued)
ior. The skill is rated as five-point scales (0 = very r with
strong and 4 = very weak), and so is the driving style total
(0 = very prudent and 4 = very aggressive). No Item Mean S.D. score
3. The main body of the survey, including 20 items 9 Sound your horn to 1.3636 0.7424 0.414*
describing kinds of driver behaviors, which are indicate your annoyance
extracted and transferred from the original to another road user
questionnaire, as shown in Table 1. According to 10 Stay in a closing lane and 0.5758 0.6629 0.439*
the definition and analysis of the original DBQ, force your way into
another
the first to the tenth items are aggressive viola-
tion behaviors. The 11th to the 14th items are 11 Fail to notice pedestrians 0.5455 0.6170 0.367*
fatal errors which may cause serious accidents, are crossing in your path
of traffic
and the 15th to the 20th items are just lapses,
which are not very dangerous but may bring 12 Fail to check your 0.5758 0.7513 0.438*
rearview mirror before
some troubles.
pulling out, changing
lanes, etc.
13 Underestimate the 1.0303 0.8095 0.321
Driver Characteristics Based on Driver Behavior. speed of an oncoming
vehicle when overtaking
Table 1 Driver behavior questionnaire and statistical
analysis results 14 Brake too quickly on 0.5152 0.5658 0.473**
a slippery road or steer
r with the wrong way in a skid
total
No Item Mean S.D. score 15 Get into the wrong lane 1.2121 0.7398 0.493**
approaching
1 Drive so close to the car 0.8485 0.7953 0.471** a roundabout or
in front that it would be a junction
difficult to stop in an
emergency 16 Misread the signs and 1.2424 0.9024 0.692**
exit from a roundabout
2 Cross-junction knowing 0.5152 0.6185 0.289 on the wrong road
traffic lights have already
turned 17 Forget where you left 1.2121 1.0828 0.587**
your car in a car park
3 Disregard the speed limit 0.9697 0.8833 0.473**
on a residential road 18 Hit something when 1.0303 0.6840 0.460**
reversing that you had
4 Disregard the speed limit 0.9394 0.8993 0.411* not previously seen
on a intercity highway
19 Intending to drive to 0.5758 0.6629 0.567**
5 Become angered by 1.6364 1.0252 0.485** destination A, find
another driver and show yourself on the road to
anger to him destination B
6 Overtake a slow driver on 1.5152 0.7953 0.576** 20 Have no clear 1.0303 0.9180 0.444**
the inside recollection of the
7 Race away from traffic 0.5152 0.6185 0.460** road along which
lights to beat the car you have just been
beside you traveling

8 Become angered by 0.8485 0.7550 0.524** Total score 18.6970 7.3802 1.000
another driver and give *
p < 0.05
chase **
p < 0.01
388 Driver Characteristics Based on Driver Behavior

Questionnaire Validity and Reliability Analysis In general, the reliability of scale is estimated based on
the sample values that measured, because the variance is
In the statistical field, reliability, and validity are the
hard to accurately calculate. At present, the Cronbach’s
two main indexes which can determine whether the
alpha reliability coefficient method is the most com-
results of such self-reported questionnaire are reliable
monly used method [11]. This method is specially
and accurate. A questionnaire with high reliability and
used for the reliability estimation of the accumulative
validity can be considered as the standard scale to test
Likert scale. It is defined as: the ratio of the sum variance
and quantify some characteristics of the participants
of each item and the total variance of the whole scale.
[10]. In this section, reliability and validity have been
According to the items shown in Table 1 and the total
analyzed and verified based on the simplified DBQ.
variance, the Cronbach’s alpha reliability coefficient of
In statistics, reliability is defined as: the proportion
the simplified questionnaire can be obtained as 0.8545,
of the real variance in the total variance [10]. When the
which indicates that the reliability of this scale is very
scale has reliability larger than 0.6, it is persuasive.
high. Therefore, this scale is relatively persuasive.

Driver Characteristics Based on Driver Behavior. Table 2 DBQ score analysis of the driver groups
N Mean S.D. T value T Sig. Mean difference
Male 29 18.310 7.517
0.806 0.426
3.190
Female 4 21.500 6.455
Age < 45 years 18 19.333 7.348 0.537 0.595 1.400
Age  45 years 15 17.933 7.601
Driving years < 10 years 15 19.733 5.587 0.731 0.47 1.900
Driving years  10 years 18 17.833 8.665
No accidental history 16 15.750 7.344 2.382 0.024*
5.721
With accidental history 17 21.471 6.443
*
p < 0.05

20 100
18 90
Accumulation Frequency [%]

16 80
14 70
Frequency [%]

12 60
10 50
8 40
6 30
4 20
2 10
0 0
0 1 2 3 4 5 0 1 2 3 4 5
a Item Scores b Item Scores

Driver Characteristics Based on Driver Behavior. Figure 1

Statistical results of item score frequency. (a) Frequency distribution and (b) accumulation frequency distribution
 Driver Characteristics Based on Driver Behavior 389

6 Validity is mainly used to express the ability that the

5 scale is able to truly measure the desired information,
i.e., the effective degree of the scale. It is defined as the
4 ratio of the variance of the scores that related to the
Eigenvalue

3 measure purposes and the total variance. In this sec-

tion, the content validity and discriminate validity of
2 the DBQ (the two items of the validity) are assessed by
1
mainly utilizing the correlation analysis and average
analysis method.
0 The content validity analysis can be used to dem-
1 3 5 7 9 11 13 15 17 19
2 4 6 8 10 12 14 16 18 20 onstrate the validity of the scale by the correlation
Component number analysis of each item score and the total score. The
more the correlation between each item score and the
Driver Characteristics Based on Driver Behavior.
total score, the more the items involved in the scale
Figure 2
reflect the same theme. The measure index of content
Scree plot of factor analysis
validity is defined as the Pearson correlation coefficient
(in terms of rxy) between each item score and the total
score, whose absolute value is less than 1. The judging
rule is that: the closer |rxy| approaches to 1, the more
Driver Characteristics Based on Driver Behavior.
significant the linear relation is. By analyzing the con-
Table 3 Factor loading matrix
tent validity of the DBQ in this paper, it can be seen that
Item No. Behavior Factor 1 Factor 2 the correlation coefficients are all positive. In addition,
16 Lapse 0.795 90% of the items have high positive interrelations with
20 Lapse 0.749 the total score under 0.05 and 0.01 significance levels.
Therefore, this DBQ has a good content validity.
19 Lapse 0.746
Discriminate validity is expressed as the ability of
11 Error 0.641 discriminating the characteristics of the measured sam-
14 Error 0.638 ple from the scale. In this paper, it is desired that the
18 Lapse 0.574 scale can be used to analyze the driver characteristics,
especially the driver behavior characteristics in traffic
17 Lapse 0.554
accidents. Therefore, the participants are divided into
15 Lapse 0.507 groups based on gender, age, driving years, mileage,
13 Error 0.469 and accidental history respectively, and in order to
12 Error 0.466 analyze the discriminate validity of the DBQ, T-test is
carried out to investigate the difference between the
7 Violation 0.804
groups. As Table 2 shown, the significant probability
8 Violation 0.693 between groups divided by the accidental history is less
4 Violation 0.682 than 0.05, which indicates that there is significant dif-
3 Violation 0.640 ference. Meanwhile, the total score of the group with
9 Violation 0.612
accidental history is obviously higher than the group
with no accidental history. It indicates that the partic-
6 Violation 0.609
ipants having higher frequency behaviors are more
1 Violation 0.595 likely to be involved in accidents. In addition, in groups
2 Violation 0.516 divided by gender, age, and driving years, there are
5 Violation 0.366 some differences between the average scores: the aver-
age score of the low age group is higher than that of the
10 Violation 0.350
high age group, and average score of the low driving
390 Driver Characteristics Based on Driver Behavior

Driver Characteristics Based on Driver Behavior. abnormal driving behaviors in this DBQ are rare. The
Table 4 The factor score coefficient main reason for this phenomenon may be that most of
the participants have high education background and
Item No. a b
good driving experience.
1
0.001 0.154
2
0.040 0.140
Quantification Analysis
3
0.014 0.168
4
0.054 0.185 Factor Analysis Factor analysis theory is applied to
the data processing and extraction of this 20-item DBQ.
5 0.050 0.087
Principle components analysis with oblique rotation is
6 0.034 0.153 implemented as the factor analysis method to investigate
7
0.061 0.218 the factor structure of the DBQ. A two-dimensional
8 0.002 0.180 factor structure is established, too. This structure is
also suggested in the scree plot shown in Fig. 2, and
9
0.018 0.161
the curve becomes smooth from the third component.
10 0.060 0.082 The two factors could account for 40.1% of the total
11 0.163
0.063 variance. Table 3 shows the factor load matrix and the
12 0.112 0.001 items with load less than 0.35 are omitted.
This factor analysis method distinguishes the differ-
13 0.119
0.043
ent types of behaviors clearly. The first factor accounts
14 0.155
0.010 for 20.8% of the total variance and contains ten items
15 0.117 0.033 which are all lapse and error behaviors. The second
16 0.188 0.022 factor contains the other ten items which describe vio-
17 0.125 0.059
lation behaviors. It could be assumed that both of the
error and lapse are attributed to the driver’s
18 0.139
0.010
unintentional passive characteristics caused by the lack
19 0.181
0.013 of driving skill. Taking the 12th item “Fail to check your
20 0.189
0.066 rearview mirror before pulling out, changing lanes, etc.,”
as an example: the behavior occurs because the driver is
short of training on this “mirror-checking” experience.
As viewed from the behavior’s property, this behavior is
years is higher than that of the high driving years’ similar to the lapse behavior such as the 18th item “Hit
group, but the difference is not significant. The average something when reversing that you had not previously
analysis indicates that the DBQ have a significant dis- seen,” and the diversities between these items are only
criminate validity in the aspect of accidental history. the different driving skills presented by the behaviors.
Therefore, the first factor could be defined as Fskill to
Data Statistical Analysis and Quantification describe the driver’s characteristics on driving skills.
Analysis The violation behaviors indicate the driver’s inten-
tional active characteristics attributed to driving style.
Data Statistical Analysis
The more aggressive style the driver is inclined to, the
The item scores of the questionnaire were analyzed, and more violation behaviors may occur during his/her
the mean score, the standard deviation of the tested driving. Similarly, the second factor is defined as Fstyle
samples and that of the total scores are shown in to describe the driver’s style characteristics. Based on
Table 1. Figure 1 is the statistical results of item score the principle of factor analysis, these two factors are
frequency. It can be seen that the scores of the items are independent and represent the human driver’s
generally low, and almost all less than 2, more than 50% unintentional (not purposive) and intentional (purpo-
of these less than 1, which indicates that most of the sive) behaviors respectively.

Driver Characteristics Quantification The factor
analysis converts the 20 items to two factors Fskill and
Fstyle, and provides a method to quantify individual
driver’s characteristics of driving skill and style
based on the scale point data of self-reported question-
naire. The quantification of Fskill and Fstyle is calculated
as the equation below. X1, X2,···,XM are the standard-
ized scale points of the DBQ items provided by
the participants during self-reported survey, and a1,
a2 ··· aM, b1, b2 ··· bM are the factor score coefficients
(M =20).
Fskill ¼ a1 X1 þ a2 X2 þ    þ aM XM
Fstyle ¼ b1 X1 þ b2 X2 þ    þ bM XM
Based on the original data of the 33 participants, the
coefficients a1, a2 ··· aM, b1, b2 ··· bM could be estimated
with regression analysis and the results are shown in
Table 4.
The quantified factors are normalized that the
mean is 0 and the standard deviation is 1. The drivers
with larger driving skill factor Fskill have higher fre-
quency of error and lapse behaviors, and the larger
value of driving style factor Fstyle means more violation
behaviors. The data points of the 33 participants’ char-
acteristics factors distribute in all of the four quadrants,
2.5
2
II
1.5
0.5
Fstyle
–0.5
–1
III
–1.5
–2
–2 –1.5 –1 –0.5
Driver Characteristics Based on Driver Behavior. Figure 3
Factor score distribution
Driver Characteristics Based on Driver Behavior 391
which are shown in Fig. 3. There are six participants
(18.2%) in the first quadrant, 11 participants (33.3%)
in the second quadrant, seven participants (21.2%) in
the third quadrant, and nine participants (27.3%) in
the fourth quadrant. This proportion of distribution
indicates that the drivers are inclined to two main
tendencies: one is aggressive style with better skill
(quadrant II), and the other is prudent style with
worse skill (quadrant IV). This result is coincident
with the human psychological features.
Comparison and Discussion
The participants’ self-evaluations on driving skill
and style are collected during the survey, and the sta-
tistical results are shown in Table 5. According to the
five-point scales, most participants are confident with
Driver Characteristics Based on Driver Behavior.
Table 5 Driver self-evaluation statistical results
Skill Style
Mean 1.212 1.879
S.D. 0.600 1.023
I
IV
0 0.5 1 1.5 2 2.5
Fskill
392 Driver Characteristics Based on Driver Behavior

Driver Characteristics Based on Driver Behavior. Table 6 Correlative coefficients analysis results of factor score and
self-evaluation

Self-evaluation of Self-evaluation of Driving skill Driving style factor

driving skill driving style factor Fskill Fstyle
Self-evaluation of 1 r =
0.262 r =
0.542** r =
0.294
driving skill
Sig = 0.140 Sig = 0.001 Sig = 0.096
Self-evaluation of – 1 r = 0.225 r = 0.481**
driving style
Sig = 0.208 Sig = 0.005
Driving skill factor Fskill – – 1 r = 0.000
Sig = 1
Driving style factor Fstyle – – – 1
**
p < 0.01

Driver Characteristics Based on Driver Behavior. Table 7 Fskill analysis of the driver groups

N Mean S.D. T value T Sig. Mean difference

Male 29
0.129 0.987
2.101 0.044 **

1.065
Female 4 0.936 0.487
Age < 45 years 18 0.047 0.924 0.290 0.774 0.103
Age  45 years 15
0.056 1.115
Driving years < years 15 0.231 0.894 1.222 0.231 0.424
Driving years  10 years 18
0.192 1.067
No accidental history 16
0.301 0.971
1.726 0.094*
0.584
With accidental history 17 0.283 0.970
*
p < 0.1
**
p < 0.05

Driver Characteristics Based on Driver Behavior. Table 8 Fstyle analysis of the driver groups
N Mean S.D. T value T Sig. Mean difference
Male 29 0.049 0.975 0.754 0.456 0.405
Female 4
0.356 1.263
Age < 45 years 18 0.102 0.950 0.633 0.532 0.223
Age  45 years 15
0.122 1.077
Driving years < 10 years 15
0.034 0.832
0.177 0.860
0.063
Driving years  10 years 18 0.029 1.144
No accidental history 16
0.229 0.893
1.287 0.208
0.444
With accidental history 17 0.215 1.073

their driving skill and consider themselves as prudent
drivers.
For the purpose of investigating the relationships
between the driver background information, self-
evaluation and the quantified factor scores, correlation
analysis of these data is carried out. Table 6 shows the
correlative coefficients analysis results of factor score
and self-evaluation.
It can be seen that there are all positive correlations
that with the significance level less than 0.01 between
the driving skill factor Fskill and driving skill self-
evaluation, the driving style factor Fstyle and driving
style self-evaluation. These results show that the quan-
tified factor scores are in accordance with the driver’s
self-evaluation.
The comparisons between the groups divided by
gender, age, driving years, and accidental history are
also conducted with T-test of Fskill and Fstyle, which are
shown in Tables 7 and 8 respectively.
It can be seen that there are certain differences
between the average values of common factors with
the different groups. As for the group divided by gen-
der, the average values of the driving skill factor Fskill of
the female group are all higher than those of the male
group. And the difference significant level is lower than
0.05, which indicates that there are certain differences
for the driving ability of female participants compared
with that of male participants. At the same time, the
average values of driving style factor Fstyle with the
female group are all lower than those of the male
group, which indicates that the driving style of the
female participant tends to be more conservative com-
pared with the male participant. It accords with the
gender characteristics to a certain extent.
As for the group divided by age and driving years,
the average values of the driving skill factor Fskill of the
low age group and the short driving years group is
higher than those of the high age group and the long
driving years’ group. In addition, the average values of
the driving style factor Fstyle with the low age group is
higher than those of the high age group, and there is
almost no difference between the short driving years
group and the long driving years group. Then the
difference also accords with the conventional under-
standing, but there is no significant difference among
the most average values of the driving skill factor Fskill,
and the same as the driving style factor Fstyle.
Driver Characteristics Based on Driver Behavior 393
As for the group divided by accidental history, the
average values of the driving skill factor Fskill with the
no accidental history group are lower than those of
the accidental history group and the significant level
is lower than 0.1, which is an obvious difference. The
above analysis indicates that the driver characteristics
and the difference of the accident history can be
reflected by the common factor. And the driver with
the high driving skill factor Fskill or driving style factor
Fstyle is much easier to have an accident.
Based on the comparisons between quantified
factor scores and driver self-evaluation, the group anal-
ysis, the reasonability of the quantification method is
validated that the quantified factors could embody the
driver’s characteristics tendencies on skill and style.
Future Directions
In this paper, a quantification method is explored to
investigate the human driver behavior characteristics.
Thirty-three Chinese drivers participate in a self-reported
survey based on Driver Behavior Questionnaire (DBQ)
and a two-dimensional factor structure is established
from the original data with factor analysis theory.
The extracted Fskill and Fstyle factors can be calcu-
lated by the data of driver self-reported survey expedi-
ently, which describe the driver’s unintentional driving
skill and intentional driving style characteristics,
respectively. Some comparisons and discussions are
also carried out to validate this method. The quantifi-
cation method of driving skill and driving style could
be applied to the development of adaptive Driver Assis-
tance Systems, e.g., providing more assistance to
drivers with less driving skill and effective warning to
the drivers with aggressive driving style. The quantified
evaluation of drivers could also be utilized to train and
educate drivers and help to improve the driving skill
and correct the driving style. Meanwhile, the compar-
ative analysis of the DBQ and real road experimental
data can be carried out in the future work.
Acknowledgments
The project was supported by NSFC (No. 50705047).
The authors especially thank Mr. Qing Xiao
(China Agricultural University) and Mrs. Furui Yang
(China Agricultural University) for their contribution
to the preparation of the chapter.
394 Driver Characteristics Based on Driver Behavior
Bibliography
1. Elander J, West R, French D (1993) Behavioral correlates of
individual differences in road traffic crash risk: an examination
of methods and findings. Psychol Bull 113:279–294
2. Aufrere R, Gowdy J, Mertz C, Thorpe C, Wang C, Yata T (2003)
Perception for collision avoidance and autonomous driving.
Mechatronics 13(10):1149–1161
3. Deffenbacher JL, Oetting ER, Lynch RS (1994) Development of
a driving anger scale. Psychol Rep 74:83–91
4. Lajunen T, Parker D, Summala H (2003) The Manchester driver
behaviour questionnaire: a cross-cultural study. Accid Anal
Prev 942:1–8
5. Reason JT, Manstead ASR, Stradling SG, Baxter JS, Campbell
K (1990) Errors and violations on the road: a real distinction?
Ergonomics 33:1315–1332
6. Xie C, Parker D (2002) A social psychological approach to
driving violations in two Chinese cities. Transport Res Part F:
Traffic Psychol Behav 5(4):293–308
7. Ozkana T, Lajunen T, Summala H (2006) Driver behaviour
questionnaire: a follow-up study. Accid Anal Prev 38:386–395
8. Lajunen T, Summala H (2003) Can we trust self-reports of
driving? Effects of impression management on driver behavior
questionnaire responses. Transp Res F 6:97–107
9. Davey J, Wishart D, Freeman J, Watson B (2007) An application
of the driver behaviour questionnaire in an Australian
organisational fleet setting. Transp Res F 10:11–21
10. Huixin K, Shen H (2005) The statistical analysis of the research,
2nd edn. The Press of China Media University, Beijing
11. Cronbach LJ (1951) Coefficient alpha and the internal
structure of tests. Psychometrika 16(3):297–334

Driver Inattention Monitoring
System for Intelligent Vehicles
YANCHAO DONG1, ZHENCHENG HU2
1
Graduate School of Science and Technology,
Kumamoto University, Kumamoto, Japan
2
Graduate School of Science and Technology,
Computer Science Department, Kumamoto
University, Kumamoto, Japan
Article Outline
Glossary
Definition of the Subject
Introduction
Distraction and Fatigue Effects on Driving Behavioral
Performance
Commercial Products and Activities for Driver
Inattention Detection
Current Methods to Detect Driver Inattention
Discussion: Future Directions
Conclusion
Bibliography
Glossary
DIMS Driver inattention monitoring system, to
monitor the attention status of the driver.
Distraction Driver distraction is a diversion of
attention away from activities critical for safe
driving toward a competing activity.
Driver biological Utilize driver biological signals, e.g.,
electroencephalography (EEG), electrocardiogram
(ECG), electro-oculography (EOG), surface electro-
myogram (sEMG), to estimate driver attention status.
Driver physical Utilize driver’s physical signals,
e.g., eye closure duration, blink frequency, nodding
frequency, fixed gaze, and frontal face pose, to
estimate driver attention status.
Driving performance Utilize driving performance,
e.g., pressure distribution on the seat, car-following,
steering wheel angle, accelerator pedal position, lane
boundaries, and upcoming road curvature, to
estimate driver attention status.
Fatigue Driver fatigue refers to a combination of
symptoms such as impaired performance and
a subjective feeling of drowsiness.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Driver Inattention Monitoring System for Intelligent Vehicles 395
Hybrid Combining driver physical measures with
driving performance measures to estimate driver
attention status.
Inattention Driver inattention represents diminished
attention to activities that are critical for safe
driving in the absence of a competing activity.
Measures The way to estimate the driver’s attention
status.
Physical signal extraction The approaches for
extracting driver physical signals.
Subjective report Subjective self-assessment of
attention status.
Definition of the Subject
Driver inattention is a major factor in highway crashes.
The National Highway Traffic Safety Administration
(NHTSA) estimates that approximately 25% of
police-reported crashes involve some forms of driver
inattention – the driver is distracted, asleep or fatigued,
or otherwise “lost in thought” [1]. This entry reviews
the state-of-the-art technologies for monitoring driver
inattention, which can be classified into two main
categories: distraction and fatigue. Driver inattention
is a major factor in most traffic accidents. Research and
development has been actively carried out for decades
with the goal of precisely determining the drivers’ state
of mind. This entry summarizes these approaches by
dividing them into five different types of measures:
1. Subjective report measures
2. Driver biological measures
3. Driver physical measures
4. Driving performance measures
5. Hybrid measures
Among these approaches, subjective report measures
and driver biological measures are not suitable under
real driving conditions, but could serve as some rough
ground truth indicators. The hybrid measures are
believed to give more reliable solutions compared with
single driver physical measures or driving performance
measures, because the hybrid measures minimize the
number of false alarms and maintain a high recognition
rate, which promote acceptance of the system. Also,
a discussion on some nonlinear modeling techniques
commonly used in the literature is made.
# 2012, DOI 10.1007/978-1-4419-0851-3
396 Driver Inattention Monitoring System for Intelligent Vehicles

Introduction The phenomenon of fatigue is different from that of

distraction. The term “fatigue” refers to a combination
Driver inattention is a major factor in highway crashes.
of symptoms such as impaired performance and
The National Highway Traffic Safety Administration
a subjective feeling of drowsiness [4]. Even with the
(NHTSA) estimates that approximately 25% of
intensive research that has been performed, the term
police-reported crashes involve some forms of driver
“fatigue” still does not have a universally accepted
inattention – the driver is distracted, asleep or fatigued,
definition [5]. Thus, it is difficult to determine the
or otherwise “lost in thought” [1]. A common
level of fatigue-related accidents. However, studies
definition of driver inattention is given in [2]:
show that 25–30% of driving accidents are fatigue
“Driver inattention represents diminished attention
related [6]. In their definition, the European Transport
to activities that are critical for safe driving in the
Safety Council (ETSC) states that fatigue “concerns the
absence of a competing activity.”
inability or disinclination to continue an activity,
A study by the AAA FTS (American Automobile
generally because the activity has been going on for
Association Foundation for Traffic Safety) utilized five
too long” [7]. From the viewpoint of individual organ
categories for the driver attention status: attentive, dis-
functionality, there are different kinds of fatigue such as
tracted, looked but did not see, sleepy, and unknown
local physical fatigue (e.g., in a skeletal or ocular mus-
[3]. The category of looked but did not see can be
cle), general physical fatigue (following heavy manual
considered to be a kind of cognitive distraction, and
labor), central nervous fatigue (sleepiness), and mental
the word “sleepy” could be replaced by the more com-
fatigue (not having the energy to do anything). Central
prehensive word “fatigued.” In this entry, two catego-
nervous fatigue and mental fatigue are the most
ries of inattention are proposed: distraction and fatigue.
dangerous types for driving, because these will
The causes of driver distraction are diverse and pose
eventually lead to sleepiness, increasing the probability
large risk factors – over half of the crashes involving
of an accident.
inattention were caused by driver distraction [1, 2].
The ETSC defines four levels of sleepiness based on
After an intensive study on the various definitions of
behavioral terms [7]: (a) completely awake, (b) moder-
driver distraction appeared in the literature, a more
ate sleepiness, (c) severe sleepiness, and (d) sleep. In an
general definition is proposed by Lee et al. [2]: “Driver
attempt to avoid having an accident, most sleepy
distraction is a diversion of attention away from
drivers will try to fight against sleep with different
activities critical for safe driving toward a competing
durations and sequences of the physiological events
activity.”
that precede the onset of sleep [8]. When a driver
Thirteen types of potentially distracting activities
becomes fatigued and begins to fall asleep, the follow-
are listed in [3]: eating or drinking, outside person,
ing symptoms can be observed: repeated yawning, con-
object or event, talking or listening on cellular phone,
fusion and thinking seems foggy, feeling depressed and
dialing cellular phone, using in-vehicle-technologies,
irritable, slower reaction and responses, daydreaming,
etc. Because the distracting activities take many forms,
difficulty keeping eyes open and burning sensation in
the NHTSA classifies distractions into four categories
the eyes, lazy steering, difficulty maintaining concen-
from the viewpoint of the driver’s functionality: visual
tration, swaying of head or body from nodding off,
distraction (e.g., looking away from the roadway), cog-
vehicle wanders from the road or into another lane,
nitive distraction (e.g., being lost in thought), auditory
nodding off at the wheel, breathing becomes shallow,
distraction (e.g., responding to a ringing cell phone),
heart races, etc. Different individuals show different
and biomechanical distraction (e.g., manually adjusting
symptoms and the degrees vary. Thus, there is no con-
the radio volume) [1]. Two more categories are added
crete method to measure the level of fatigue. The ETSC’s
by Lee et al. [2]: olfactory distraction and gustatory
study [7] showed that the level of fatigue or sleepiness
distraction. Many distracting activities can involve
(sleepiness is the outside exhibition of fatigue) is
more than one of these components (e.g., talking to
a function of the amount of activity in relation to the
a phone while driving creates a biomechanical, audi-
brain’s physiological waking capacity. Several factors
tory, and cognitive distraction).
can influence this physiological waking capacity, and

hence lower the fatigue threshold [4, 5, 7, 9] such as
disturbed sleep, the low point in the circadian rhythm,
hard work prior to driving, etc. These factors are inde-
pendent of the activity being undertaken, but result
in the fatigue effect of that activity appearing more
quickly. Thus, fatigue cannot be seen simply as a func-
tion of the duration of time engaged in driving.
Driving inattention includes two main categories and
each of them includes a few subcategories, as shown in
Fig. 1. The purpose of driving inattention monitoring
system (DIMS) is to monitor the attention status of the
driver and if driving inattention is detected some
measures should be done to make the driving safe,
depending on the inattention type and level.
Driving is a process involving situation awareness
of the environment, decision making, and the
performance of actions, as shown in Fig. 2 [10]. In
this process, the most complicated stage is the situation
awareness. In [10], a three-level situation awareness
model is defined as “The perception of the elements
in the environment within a volume of time and space,
the comprehension of their meaning, and the projec-
tion of their status in the near future.” The deployment
of attention in the perception process acts to present
certain constraints on a person’s ability to accurately
perceive multiple items in parallel, and is a major
Distraction
Cognitive Visual Olfactory
Auditory Bio.-Mech. Gustatory
Driver Inattention Monitoring System for Intelligent Vehicles. Figure 1
Inattention categories
Driver Inattention Monitoring System for Intelligent Vehicles 397
limitation on situation awareness. Direct attention is
needed, not only for perceiving and processing the
available cues, but also in the later stages of decision
making and response execution. In a complex and
dynamic driving environment, attention demands
result from information overload, complex decision
making, and the performance of multiple tasks. Thus,
monitoring the attention status is vital for maintaining
safe driving.
The purpose of the driver inattention monitoring
system (DIMS) is to monitor the attention status of the
driver. If driver inattention is detected, different coun-
termeasures should be taken to maintain driving safety,
depending on the types and levels of inattention. DIMS
has been an active research field for decades. The first
international conference on driver distraction and inat-
tention was held in 2009 [11]. Some auto companies
have already installed some simple function driver
fatigue monitoring systems in their high-end vehicles.
Yet, there is still a great need to develop a more reliable
and fully functional DIMS using cost-efficient methods
for a real driving context. It is believed that the
development of signal processing and computer vision
techniques will attract more attention to the study of
this field in the coming years. With the intention
of benefiting those interested in, or about to enter,
Inattention
Fatigue
Local-Phy Gen.-Phy.
Central-Ner. Mental
398 Driver Inattention Monitoring System for Intelligent Vehicles

Feedback

system Capability
ATTENTION Interface Design
DRIVER Stress & Work load
Complexity
Automation

Tasks
Situation Awareness

ENVIRONMENT Performance
Perception Comprehension Projection Decision VEHICLE
of Actions
Level 1 Level 2 Level 3

Individual Factors
Information Processing
Goals & Objective Mechanisms
Preconception
(Expectations) Long Term
Automaticity
Memory Stores

Abilities
Experience
Training

Driver Inattention Monitoring System for Intelligent Vehicles. Figure 2

Information processing and attention [10]

this field, this entry gives a comprehensive review of the
state of the knowledge on driver inattention. It thus
provides a clear view of the previous achievements and
the issues that still need to be considered.
The arrangement of this entry is as follows. We
introduced the driver inattention concept in section
“Introduction”. Next, the effects of driver distraction
and fatigue on driving performance are presented in
section “Distraction and Fatigue Effects on Driving
Behavioral Performance”. Because some commercial
products relative to inattention detection have emerged
on the market in recent years, section “Commercial
Products and Activities for Driver Inattention
Detection” is devoted to reviewing these products. sec-
tion “Current Methods to Detect Driver Inattention”
presents a detailed review of the scientific researches
on inattention detection. Five types of measures for
inattention detection are presented in this section:
(a) subjective report measures, (b) driver biological
measures, (c) driver physical measures, (d) driving
performance measures, and (e) hybrid measures.
After a discussion of future directions in section
“Discussion: Future Directions”, a conclusion is given
in section “Conclusion”.
Distraction and Fatigue Effects on Driving
Behavioral Performance
This section concentrates on how distraction and
fatigue affect a driver’s behavior and driving
performance. Exploring these effects could provide
useful information for the development of real-time
distraction and fatigue detection algorithms.
Effects of Distraction
Performing a cognitively demanding task while driving
would influence both the driver’s visual behavior and
driving performance (as indicated by braking
behavior).
Driver Behavior Patterns With an increase in the
cognitive demand, many drivers changed their

inspection patterns on the forward view. Angell et al.
[12] indicated that the eye-glance pattern could be used
to discriminate driving while performing a secondary
task from driving alone, and could be used to discrim-
inate high- from low-workload secondary tasks. More
facts associated with cognitive distraction driving can
be found in [13, 14]: Drivers narrowed their inspection
of the outward view and spent more time looking
directly ahead. They reduced their inspection of the
instruments and mirrors, and reduced their glances at
traffic signals and the area around an intersection.
Rantanen and Goldberg [14] found that the visual
field shrank by 7.8% during a moderate-workload
counting task and by 13.6% during a cognitively
demanding counting task. Drivers had fewer saccades
per unit time, which was consistent with a reduction in
glance frequency and less exploration of the driving
environment, and in some cases drivers shed these
tasks completely and did not inspect these areas at all
[15]. Hayhoe [16] showed links between eye movement
(fixation, saccade, and smooth pursuit), cognitive
workload, and distraction. Fixations occur when an
observer’s eyes are nearly stationary. Saccades are very
fast movements that occur when visual attention shifts
from one location to another. Smooth pursuits occur
when an observer tracks a moving object such as
a passing vehicle. Saccade distance decreases as task
complexity increases, which indicates that saccades
may be a valuable index of mental workload [17].
In contrast, the amount of head movement increased
when cognitive loads were imposed. It is believed that
this is a compensatory action by which a driver
attempts to obtain a wider field of view [18]. Miyaji
et al. [18] proposed that the standard deviations of eye
movement and head movement could be suitable for
detecting the states of cognitive distraction in subjects.
Both cognitive and visual distractions caused gaze
concentration and slow saccades when drivers looked
at the roadway, and cognitive distraction increased
blink frequency [19]. Liang and Lee [19] found that
visual distraction resulted in frequent, long off-road
glances. A report from the Safety Vehicle Using
Adaptive Interface Technology (SAVE-IT) program
showed that eyes-off-road glance duration, head-off-
road glance time, and standard deviation of lane
position (SDLP) are good measures of visual
distraction [20].
Driver Inattention Monitoring System for Intelligent Vehicles 399
Other Physiological Responses When cognitive
loads (conversation or arithmetic) were imposed on
subjects, pupil dilation occurred by the acceleration of
the sympathetic nerve [18]. The average heart rate also
increased by approximately 8 beats per minute.
However, the average value of the heart rate RRI
decreased under the same situation [18]. Itoh [21]
pointed out that performing a cognitively distracting
secondary task (e.g., talking, thinking about
something, etc.) during driving caused a decrease in
the driver’s temperature at the tip of the nose, and this
effect was reproducible. It was reported in [22] that
a considerable and consistent skin temperature
increase in the supraorbital region could be observed
during cognitive and visual distractions. Berka [23]
found that the electroencephalography (EEG) signal
also contained information about the task engagement
level and mental workload.
Driving Performance Significant changes were
observed in a driver’s vehicle control as a consequence
of performing the additional cognitive tasks while
driving. Ranney [24] found that distraction may be
associated with lapses in vehicle control, resulting in
unintended speed changes or allowing the vehicle to
drift outside the lane boundaries. Zhou et al. [25]
found the influences on the lane-changing behavior
when a secondary task was being performed, which
included a reduction in the frequency of the checking
behavior (check a side mirror or speedometer), a delay
in the checking behavior, and a longer time for the
checking behavior. Carsten and Brookhuis [26] found
that the effects of cognitive distraction on driving
performance differed considerably from those of visual
distraction. Visual distraction affects a driver’s steering
ability and lateral vehicle control, while cognitive
distraction affects longitudinal vehicle control,
particularly car-following. Liang and Lee [19] also
found that cognitive distraction made steering less
smooth, but improved lane maintenance. Liang and
Lee [19] found that steering neglect and overcompen-
sation are associated with visual distraction, while
under-compensation is associated with cognitive
distraction. Overall, visual distraction interferes with
driving performance more than cognitive distraction.
An apparently anomalous finding is that when second-
ary task cognitive demands increased, a driver’s lateral
400 Driver Inattention Monitoring System for Intelligent Vehicles
control ability was found to improve [26]. Harbluk
et al. [13, 15] found an increased incidence of hard
braking associated with cognitive distraction driving.
Effects of Fatigue
When a driver is fatigued, certain physical and physio-
logical phenomena can be observed. These include
changes in brain waves or EEG, eye activity, facial expres-
sions, head nodding, body sagging posture, heart rate,
pulse, skin electric potential, gripping force on the
steering wheel, and other changes in body activities.
Driver Behavior Patterns Eskandarian et al. [27]
found that the follow actions were correlated with
fatigue: Drivers exhibited a reflexive head nod after
checking the side mirrors; the head motions were
significantly less frequent; the number of times drivers
touched or scratched their chin, face, head, ears, eyes,
and legs significantly increased; drivers were inclined to
turn their head to the left to relieve muscular tension in
the neck; eye blinking activity radically increased;
episodes of yawning were more frequent; and they
tended to adopt more relaxed hand positions on the
steering wheel. Particularly for eye blinking patterns,
PERCLOS [28], the percentage of time the eye is more
than 80% closed, is one of the most widely accepted
measures in the scientific literature for drowsiness
detection. It has been validated using both EEG data
and subjective evaluation.
Other Physiological Responses The activity of a low
frequency EEG ranging from 0 to 20 Hertz has a signif-
icant relationship with sleepiness. The spectral analysis
of an EEG shows the transition from wakefulness to
sleep can be described as a shift toward slower EEG
frequencies. In the alert condition, the appearance of b
activity is common in the EEG. a activity is also
normally found in the occipital regions (O1 and O2)
in the awake and relaxed condition. When a driver
gets drowsy, a burst of a activity can often be seen
in the central regions of the brain (C3 and C4).
However, some people do not show any a activity.
As the driver gets drowsier, the a activity is replaced
by y activity. When d activity occurs in the EEG,
the driver is no longer awake; this is an indicator of
deep sleep [29].
Driving Performance It has been reported that
sleep-deprived drivers have a lower frequency of
steering reversals (every time the steering angle crosses
zero degrees) [30], a deterioration of steering
performance [31], a decrease in the steering wheel
reversing rate [32], more frequent steering maneuvers
during wakeful periods, no steering correction for
a prolonged period of time followed by jerky motion
during drowsy periods [33], low velocity steering [34],
large amplitude steering wheel movements, and large
standard deviations in the steering wheel angle [35].
Zhong et al. [36] found that when drivers had
a fatigued status, the steering wheel angle and vehicle
tracking became irregular, and the range of deviation
greatly increased. Several researchers found that the
lane-tracking ability decreased as the time on the task
increased [31]. Variables such as the times of lane
departures, SDLP, and maximum lane deviation were
found to be highly correlated with eye closures [37].
The mean square of lane deviation, mean square of
high pass lateral position, and SDLP showed good
potential as drowsiness indicators [38].
Dingus et al. [34] found that the yaw deviation
variance and the mean yaw deviation (calculated over
a 3 min period) showed some promise as drowsiness
indicators. However, no strong correlations between
drowsiness and braking or acceleration were found in
[34, 39]. Generally, vehicle speed variability has not
shown any strong correlation with drowsiness [39].
However, some reports found that the standard
deviation of speed increased from the third driving
hour with a time interval of 45 min [40].
Commercial Products and Activities for Driver
Inattention Detection
Auto Companies
Many famous auto companies are currently conducting
researches on driver inattention monitoring systems,
including Toyota, Nissan, Volvo, Mercedes-Benz,
and Saab.
Saab’s Driver Attention Warning System [41, 42] is
a project designed to counter two of the most common
causes of road accidents: driver drowsiness and
distraction. The system utilizes two miniature infrared
cameras, one installed at the base of the driver’s A-pillar
and the other in the center of the main fascia, which are

focused on the driver’s eyes. It also utilizes SmartEye
[43] software to get accurate eyelid, gaze, and
head orientation information. In their algorithm, the
driver’s eye blinking frequency is measured. If a pattern
of long duration eyelid closures is detected, it indicates
the potential onset of drowsiness. A three-level warning
interface was designed for drowsiness detection. This
starts with a chime sound and text message, moves on
to a spoken message, and finally a stronger warning
tone audio message is persistently delivered until the
driver presses the reset button. As soon as the driver’s
gaze moves away from what is defined as the “primary
attention zone”-the central part of the windshield in
front of the driver-a timer starts counting. If within 2 s
of the timer being triggered the driver’s eyes and head
do not return to the “straight ahead” position, it is
considered to be a distraction. In a case involving
peripheral tasks such as looking in the rearview mirror,
a side mirror, or turning a corner, the timer’s elapse
time becomes longer. Once the driver distraction has
been detected, a seat vibration signal will be issued to
warn the driver. However, there is no report about
the robustness of this system during daytime and
nighttime driving under different kinds of weather
conditions, providing no driver status ground truth as
a reference.
Toyota developed their Driver Monitoring System
in 2006 for its latest Lexus models. This system features
a camera, using near-infrared technology, mounted on
top of the steering column cover. It monitors the exact
position and angle of the driver’s head while the vehicle
is in motion. If the Advanced Pre-Crash Safety system
detects an obstacle ahead, and at the same time the
Driver Monitoring System establishes that the driver’s
head has been turned away from the road for too long,
the system automatically activates pre-crash warnings.
If the situation persists, the system can briefly apply the
brakes to alert the driver [44]. In 2008, the Toyota
Crown system went further. It can detect if the
driver is becoming sleepy by monitoring their eyelids.
Toyota’s solution combines driver face orientation and
environmental obstacle detection to determine
accident potential, and utilizes eyelid activity to
identify drowsiness.
In the spring of 2009, Mercedes-Benz introduced
Attention Assist into its series production [45].
Attention Assist works by first observing a driver’s
Driver Inattention Monitoring System for Intelligent Vehicles 401
behavior, and then using this information to create
a unique driver profile. During operation, a series of
tests continually monitor the driver input in relation to
this profile, and in the event that a deviation is encoun-
tered, the system then determines whether or not the
deviation is a result of fatigue. If it is, Attention Assist
alerts the driver both visually and audibly that it is time
to take a break. The factors taken into account to
determine a driver’s profile include the speed,
longitudinal and lateral acceleration, angle of the
steering wheel, way that the indicators and pedals are
used, certain driver control actions, and even various
external influences such as a side wind or an uneven
road surface. The Attention Assist system only uses
vehicle parameters to determine driver drowsiness,
which requires no additional hardware setup. However,
this system needs to establish individual profiles for
different drivers, which would affect the acceptance of
the system in real life.
In 2007, Volvo Cars introduced Driver Alert
Control to alert tired and nonconcentrating drivers
[46]. With the idea that the technology for monitoring
a driver’s eyes is not yet sufficiently mature and human
behavior varies from one person to another, they
developed their system based on the cars progress on
the road. It is reported that Driver Alert Control
monitors the car’s movements and assesses whether
the vehicle is being driven in a controlled or
uncontrolled way. It can also cover situations where
the driver is focusing too much on his or her cell
phone or children in the car, thereby not having full
control of the vehicle.
Other Commercial Products and Activities
Technological approaches have continued to emerge in
recent years, and hold promise for detecting and
monitoring dangerous levels of driver inattention.
While many of these projects are now in the develop-
ment, validation testing, or early implementation
stages, some companies can provide usable devices or
prototype to give information about driver behavior.
For the purpose of nonintrusive measurement, these
devices mainly utilize video cameras and computer
vision technologies.
Attention Technology, Inc. has designed and
developed the DD850 Driver Fatigue Monitor (DFM),
402 Driver Inattention Monitoring System for Intelligent Vehicles
the only real-time, on-board drowsiness monitor that
is currently being tested in an extensive field
operational test. The DFM is a video-based drowsiness
detection system that works by measuring slow eyelid
closure. It is designed to mount on a vehicle’s
dashboard just to the right of the steering wheel, and
provides a continuous real-time measurement of eye
position and eyelid closure [47]. Specifically, the DFM
estimates PERCLOS to determine drowsiness, which is
the proportion of time the eyes are closed 80% or more
over a specified time interval. The DFM uses
a structured illumination approach to identify the
driver’s eyes. This approach obtains two consecutive
images of the driver using a single camera. The first
image is acquired using an infrared (IR) illumination
source that produces a bright-pupil image. The second
image uses an IR illumination source at a different
wavelength to produce an image with dark pupils.
These two images are essentially identical except for
the brightness of the pupils in the images. A third image
calculates the difference between these two images,
enhancing the bright eyes and eliminating all image
features except for the bright pupils. The driver’s eyes
are identified in this third image by applying
a threshold to the pixel brightness. The bright pupil
effect utilized by the DFM is a simple and effective
eye-tracking approach for pupil detection based on
a differential lighting scheme. However, the success of
the bright pupil technique strongly depends on the
brightness and size of the pupils, which are often
functions of face orientation, external illumination
interference, the distance of the subject from the cam-
era, and race. For real-world in-vehicle applications,
sunlight can interfere with IR illumination, reflections
from eyeglasses can create confounding bright spots
near the eyes, and sunglasses tend to disturb the IR
light and make the bright-pupil phenomenon appear
very weak.
Delphi believes that computer vision offers the
most direct method for detecting the early onset of
sleepiness and distraction, and it is also seen as an
excellent platform to be shared with other vision-
based driver assistance applications in the future.
They integrated two products, the ForeWarn Drowsy
Driver Alert and the ForeWarn Driver Distraction
Alert, into a comprehensive Driver State Monitor
(DSM) [47]. The DSM is a computer vision system
that uses a single camera mounted on the dashboard
directly in front of the driver and two IR illumination
sources. Upon detecting and tracking the driver’s facial
features, the system analyzes eye closures and head pose
over time to infer fatigue or distraction level. It
provides an extended eye-closure warning for closures
longer than 2.5 s, and also provides an extended
distraction warning for nonforward gaze states in
excess of 2.5 s. The fatigue detection algorithm predicts
AVECLOS, the percentage of time the eyes are
estimated to be fully closed over a 1-min interval.
Because this is a less complex measure of drowsiness
than PERCLOS, it permits the use of an automotive-
grade data processor, in contrast to the high-grade PC
processor required for PERCLOS.
Seeing Machines is engaged in the research,
development, and production of advanced computer
vision systems for research on human performance
measurement, advanced driver assistance systems,
transportation, etc. [48]. Their signature product,
faceLAB, provides head and face tracking, as well as
eye, eyelid, and gaze tracking for human subjects, using
a noncontact, video-based sensor. faceLAB provides
comprehensive blink analysis and PERCLOS assess-
ment, including the delivery of raw data on the details
of eyelid behavior. Instead of using traditional corneal
reflection techniques, input is obtained using a stereo
camera pair. Seeing Machines’s faceLAB has been
employed extensively as a PC-based research tool.
Although the device reportedly works very well in
a simulator environment, the numerous challenges
faced in a real driving environment prevent it from
working robustly. Seeing Machines also provides
another product, Driver State Sensor (DSS). It consists
of one camera, two IR LED illuminators, and one
special computing and communication unit. The goal
of DSS is to detect driver fatigue by analyzing eyelid
activity.
Smart Eye AB is another company that provides
computer vision–based software for detecting human
face/head movement, eye movement, and gaze
direction [43]. Their product, Smart Eye Pro 3.0, is
a machine vision system that estimates head pose
using a simple and robust method based on tracking
individual facial features and a three-dimensional (3D)
head model. While the face is being tracked, the gaze
direction and eyelid positions are determined by

combining image edge information with 3D models of
the eye and eyelids. A major advantage is that eye and
head tracking can continue even if one camera is fully
occluded or otherwise nonoperational. This also allows
for large head motions (translation and rotation).
Smart Eye has not developed an algorithm that moni-
tors drowsiness.
SensoMotoric Instruments GmbH (SMI) [49] is
a German company. Their product, InSight, can
measure head position and orientation, gaze direction,
eyelid opening, and pupil position and diameter. It uses
a sampling rate of 120 Hz for head pose and gaze
measurement, 120 Hz for eyelid closure and blink
measurement, and 60 Hz for combined gaze, head
pose, and eyelid measurement. It also provides
PERCLOS information for drowsiness detection. It is
a computer based system and needs user calibration.
Current Methods to Detect Driver Inattention
In the scientific literature, five main types of
measures for inattention detection are commonly
used: (a) subjective report measures, (b) driver
biological measures, (c) driver physical measures,
(d) driving performance measures, and (e) hybrid
measures. With the exception of subjective report mea-
sures, these measures are based on nonlinear modeling
techniques. This section briefly reviews the most
common nonlinear modeling techniques. Then, the
researches on the five main types of measures will be
explored. Finally, the extraction of physical signals
from a driver by image processing will be discussed at
the end of this section, since driver physical measures
offer distraction detection through eye gaze
monitoring and fatigue detection through eye gaze,
blink, head, and mouth tracking.
Nonlinear Modeling Techniques
Human cognition can hardly be represented by a linear
model. Hence, nonlinear modeling techniques are
greatly adopted in the driver inattention detection
area. Nonlinear modeling with machine learning
techniques can extract information from noisy data,
and do not require prior knowledge before training.
There are also some mechanisms in machine learning
that can avoid over-fitting for nonlinear modeling,
Driver Inattention Monitoring System for Intelligent Vehicles 403
producing more robust and general models than
traditional learning methods (e.g., logistic regression),
which only minimize training error.
Artificial neural networks (ANNs) have been stud-
ied and utilized in numerous scientific and engineering
fields. One of the main advantages of ANNs is that they
infer solutions from data without any prior knowledge
of the patterns in the data, that is to say, they extract
the patterns empirically even if the equation between
the inputs and outputs does not exist. This character-
istic is very important because in most practical
cases the exact input-output relationship is difficult to
establish. ANNs also have the ability to generalize (i.e.,
they respond with a reasonable accuracy to patterns
that are broadly similar to the original training
patterns), which is very useful because real-world data
is noisy, distorted, and often incomplete. ANNs
are nonlinear, which allows them to solve some
complex problems more accurately than linear tech-
niques [27].
The fuzzy inference system (FIS) is famous for its
well-known linguistic concept modeling ability. The
fuzzy rule expression is close to an expert natural
language. A fuzzy system then manages the uncertain
knowledge and infers high-level behaviors from the
observed data. On the other hand, as it is a universal
approximator, the FIS can be used for knowledge
induction processes [50].
Support vector machine (SVM) is based on the
statistical learning technique and can be used for
pattern classification and the inference of nonlinear
relationships between variables. This method has been
successfully applied to the detection, verification, and
recognition of faces, objects, handwritten characters
and digits, text, speech, and speakers, along with the
retrieval of information and images [51]. The learning
technique of the SVM method makes it suitable for
measuring the cognitive states of humans. SVMs can
generate both linear and nonlinear models and are able
to compute the nonlinear models as efficiently as the
linear ones. Given a set of input data, this method first
transforms the input domain through a kernel, and
then looks for a hyperplane in the transformed domain
that separates the data with minimum error and
maximum gain. Finally, the hyperplane is transformed
back to the input domain to obtain the decision
boundaries, which may potentially be nonlinear.
404 Driver Inattention Monitoring System for Intelligent Vehicles
AdaBoost is a learning algorithm that uses the
pattern recognition algorithm called Boosting [52]. Its
advantages include a high classification performance,
fast recognition process time, and the potential
extension of recognition features. In AdaBoost, learn-
ing involves the creation of different classifiers while
successively changing the weighting of the learning
data. A weighted majority decision is then made for
these multiple classifiers in order to obtain the final
classifier function. Individual classifiers are referred
to as “weak classifiers,” while the combination of
classifiers is a “strong classifier.”
Bayesian networks (BNs) have several advantages
that make them well suited for describing human
behavior. First, the hierarchical structure of BNs can
systematically present information from different
sources and at different levels of abstraction, and can
also capture probabilistic relationships. Second, a BN is
not only a computational model but also a form of
knowledge representation. Unlike other data mining
approaches such as the SVM, BNs reveal the
relationships that generate the model predictions.
Third, BNs can handle situations with missing data.
The certainty of the hypothesis will change according to
the BN’s reasoning, which incorporates new data using
a probabilistic dependence network when new evidence
is added. Because of these advantages, BNs are
applicable to human-behavior modeling and have
been used to detect inattention [53]. Despite these
advantages, creating a correct and stable BN model
requires extensive computational capability and
a large amount of training data.
Another emerging trend has been to borrow
techniques based on hidden Markov models (HMMs)
from the speech processing and language technology
field and apply these to driver behavior modeling for
route recognition, driver identification, and distraction
detection in a manner analogous to speech recognition,
speaker identification, and stress detection in speech
[54]. The foundation of HMM is a stochastic Markov
process consisting of a number of states with
corresponding transitions. At discrete time intervals,
the Markov process moves from one state to another
according to a set of transition probabilities. State
changes in the Markov process are hidden from the
user. Sathyanarayana et al. [54] constructed a hidden
Markov model using vehicle speed, steering wheel
angle, and braking force to predict route maneuvers
(left turn, right turn, and lane change).
Subjective Report Measures
The Karolinska sleepiness scale (KSS) is the most
commonly used tool for the subjective self-assessment
of sleepiness; the values used in the KSS are shown in
Table 1. Kaida et al. [55] investigated the validity and
reliability of the KSS using EEG, behavioral, and other
subjective indicators of sleepiness. Their study showed
that the KSS was closely related to EEG and behavioral
variables, which indicates that the KSS has a high
validity for measuring sleepiness.
Ingre et al. [56] verified the close relationship
between subjective sleepiness measured with the KSS
and blink duration (BLINKD) and lane drifting,
calculated as the standard deviation of the lateral
position (SDLP) in a high-fidelity moving base driving
simulator. Their experiments showed a significant
effect of the KSS on both BLINKD and SDLP. A test
for a quadratic trend suggested a curvilinear effect with
a steeper increase at high KSS levels for both SDLP and
BLINKD. Craig et al. [57] also found that psychological
factors correlated consistently with self-reported
fatigue. However, the KSS is recorded over relatively
long time intervals, say every 15 min, as a trade-off
between high temporal resolution and avoiding
intrusive feedback. As a consequence, the KSS was not
Driver Inattention Monitoring System for Intelligent
Vehicles. Table 1 Karolinska sleepiness scale (KSS)
KSS Meaning
1 Extremely alert
2 Very alert
3 Alert
4 Rather alert
5 Neither alert nor sleepy
6 Some signs of sleepiness
7 Sleepy, no effort to stay awake
8 Sleepy, some effort to stay awake
9 Very sleepy, great effort to keep awake, fighting
sleep
 Driver Inattention Monitoring System for Intelligent Vehicles 405

capable of recording sudden drowsiness variations
caused by different situations.
Driver Biological Measures
Biological signals include EEG, electrocardiogram
(ECG), electro-oculography (EOG), surface electro-
myogram (sEMG), etc. These signals are collected
through electrodes in contact with the skin of the
human body. Table 2 summarizes some typical
methods used in this field. EEG has a spatial resolution
of 20 mm and a temporal resolution of 0.001 s. It is
widely used in the brain activity research field. Recent
research has proposed various methods to extract fea-
tures from a segment of raw EEG data for fatigue detec-
tion. In the time domain, the average value, standard
deviation, and sum of the squares of EEG amplitude are
the most commonly used features. In the frequency
domain, the energy content of each band (b, a, y, d),
mean frequency, and center of gravity of the EEG
spectrum are commonly used. Other models such as
the auto-regressive moving average (ARMA) and
power spectrum estimation are also used by some
researchers to extract EEG features. The most reliable
patterns in terms of their consistency and occurrence
for fatigue are the b, a, y, and d waves (see Fig. 3).
EEG is widely accepted as a good indicator of the
transition between wakefulness and sleep, as well as
between the different sleep stages. It is often referred
to as the gold standard. Svensson [29] proposed
objective sleepiness scoring (OSS), derived from EEG
signals, as the ground truth for validating other drows-
iness detection algorithms. The five-level OSS scores
are described in [29], and are shown in Table 3.
The four EEG activities (b, a, y, and d) were
assessed in [58] for 52 subjects during a monotonous

Driver Inattention Monitoring System for Intelligent Vehicles. Table 2 Summary of biological measures
Bio-signal
References type Objective Analysis methods Result
[58] EEG Fatigue Assess d, y, a, b (y + a)/b ↑
[59] Sample entropy, phase Sample entropy, phase
synchronization synchronization ↑
[60] SVM Predict alert ! drowsy
[61] Probabilistic-SVM Better than standard SVM
[62, 63] ICA, FFT, correlation analysis, LRM Est. drowsy level with 87% accuracy
[64] KPCA algorithm Complexity decreases as fatigue
increases
[23] Mental Inspection with second timescale Different mental task can be
engagement detected
[65] XCS Different mental task can be
detected
[66] EEG, ECG Fatigue Dynamic Bayesian network More features are favorable
[67] ECG, PPG PVT Multi linear regression model ECG, PPG is useful for est. PVT
[68] EOG Hypovigilance Eight eye actives, fuzzy expert sys Pre. sleep-related accidents with
high acc
[69] EOG Drowsiness Eleven eye actives, SVM Accuracy is quite high for “very
sleepy”
[70] sEMG Fatigue Statistic analysis Statistic trends were given
[71] Frequency and statistic analysis MDF, MNF, RMS show large change
406 Driver Inattention Monitoring System for Intelligent Vehicles

Beta (β) 13–30 Hz

Parietally and
frontally

Alpha (α) 8–13 Hz

Occipitally

Theta (Θ) 4–8 Hz

Children,
sleeping adults

Delta (δ) 0.5–4 Hz

Infants,
sleeping adults

Driver Inattention Monitoring System for Intelligent Vehicles. Figure 3

Four types of EEG waves [29]

Driver Inattention Monitoring System for Intelligent
Vehicles. Table 3 Objective sleepiness scoring (OSS)
OSS
score EEG content
0 Background of continuous b waves, no a, no y
waves
1 Occurrence of a and/or y waves, in at least two
regions of the brain, for less than a cumulative
length of 5 s
2 Occurrence of a and/or y waves, in at least two
regions of the brain, for less than a cumulative
length of 5 s
or
Occurrence of a and/or y waves, in at least two
regions of the brain, for less than a cumulative
length of 5 s
3 Occurrence of a and/or y waves, in at least two
regions of the brain, for less than a cumulative
length of 5 s
4 Continuous a and/or y waves
Source: Derived from EEG data [29]
driving session. The results showed an increasing trend
for the ratio of slow wave to fast wave EEG activities
over time. In [59], sample entropy and phase
synchronization were adopted to detect fatigue from
EEG signals, with the results showing that phase syn-
chronization among the hemispheres gradually
increased and sample entropy decreased, both pointing
to a gradual increase in sleepiness, which is related to
a decrease in EEG complexity. Yeo et al. [60] trained the
SVM to classify EEG signals into four principal fre-
quency bands, and then to predict the transition from
alertness to drowsiness. Shen et al. [61] compared
a probabilistic-based multi-class SVM and standard
multi-class SVM as classifiers for distinguishing mental
fatigue into five mental-fatigue levels, and showed that
the accuracy of the probabilistic-based multi-class
SVM was better. Lin et al. [62] established a linear
regression model to estimate the drowsiness level
from the ICA of 33-channel EEG signals, and could
estimate the drowsiness level with 87% accuracy. They
then implemented a real-time embedded EEG-based
driver drowsiness estimate system in [63], which
adopted only four channels of EEG data.
Although not apparently related, some researchers
have shown that it is also possible to estimate the
distraction level from EEG data. Berka et al. [23] tried
to use EEG data to continuously and unobtrusively
monitor the levels of task engagement and
mental workload in an operational environment. An
inspection on the EEG data using a second-by-second

timescale revealed associations between the workload
and engagement levels when aligned with specific task
events, which provided preliminary evidence that
second-by-second classifications reflect parameters of
task performance. Skinner et al. [65] investigated the
efficacy of the genetic-based learning classifier system
XCS in classifying artifact-inclusive EEG signals into
four mental tasks designed to elicit hemispheric
responses. In Liu et al. [64], the KPCA algorithm was
employed to extract nonlinear features from the
complexity parameters of EEG (approximate entropy
(ApEn) and Kolmogorov complexity (Kc)) and
improve the generalization performance of an HMM.
The result showed that both complexity parameters
decreased significantly as the mental fatigue level
increased, and the classification accuracy reached 84%.
In addition to EEG, other biological signals such as
ECG, EOG, and sEMG have also been tested to estimate
driver mental state. Yang et al. [66] employed
a dynamic Bayesian network with EEG and ECG to
estimate fatigue. A first-order hidden Markov model
(HMM) was employed to compute the dynamics of
a Bayesian network at two different time slices. The
results showed that more features are favorable
for inferring the driver fatigue more reliably and accu-
rately. In Chua et al. [67], a multiple linear regression
model was established to estimate PVT (psychomotor
vigilance test) values from a combination of
ECG and photoplethysmogram (PPG) data. Damousis
and Tzovaras [68] selected eight eye activity
features, extracted from EOG, to develop a fuzzy expert
system (FES) for the detection of hypovigilance.
Shuyan and Gangtie [69] employed an SVM to perform
drowsiness prediction with 11 eyelid-related features
extracted from EOG. These eyelid features included
blink duration, blink duration 50–50, amplitude, lid
closure speed, peak closing velocity, lid opening speed,
peak opening velocity, delay of eyelid reopening, dura-
tion at 80%, closing time, and opening time. It was
reported that the drowsiness detection accuracy was
86% for “sleepy.” In Balasubramanian and Adalarasu
[70], the surface sEMG of the shoulder and neck was
analyzed while the participant was driving to deter-
mine the onset of fatigue and prove that the develop-
ment of muscular fatigue is a consequence of driving.
In Katsis et al. [71], frequency analysis and statistical
analysis were performed on sEMG signals from the left
Driver Inattention Monitoring System for Intelligent Vehicles 407
bicep, right bicep, left forearm flexor, right forearm
flexor, and frontal muscles. The results showed that
the middle frequency decreased by about 9.5–18.9%,
the mean frequency decreased by about 11.3–18.4%,
and the root mean square amplitude increased by
about 25.1–47.7% from their initial values for
a predefined driving route.
Driver Physical Measures
Fatigue Detection In Bergasa et al. [50], PERCLOS,
eye closure duration, blink frequency, nodding
frequency, fixed gaze, and frontal face pose were
normalized and used as inputs to the fuzzy inference
system for fatigue detection. Different linguistic terms
and their corresponding fuzzy sets were distributed in
each of the inputs using induced knowledge based on
the hierarchical fuzzy partitioning (HFP) method.
They then chose the fast prototyping algorithm with
the pruned method (FDT + P) to automatically
generate fuzzy rules that were consistent, lacked
redundancy, and were interpretable. Afterward,
a simplification process was applied to achieve a more
compact knowledge base to improve the interpretabil-
ity and maintain the accuracy. Finally, three variables
(fixed gaze, PERCLOS, and eye closure duration) were
determined to be crucial cues for detecting a driver’s
fatigue. By fusing them with a fuzzy system, a final
fatigue detection accuracy of 98% was achieved.
Fan et al. [72] utilized a Gabor-features representa-
tion of the face for fatigue detection. After the face was
located, Gabor wavelets were applied to the face area to
obtain different scale and orientation features of the
face. Then, features on the same scale were fused into
a single one to reduce the dimension. Finally, the
AdaBoost algorithm was used to extract the most crit-
ical features from the dynamic feature set and construct
a strong classifier for fatigue detection. It was reported
that this method worked well on a wide range of
human subjects with different genders, poses, and
illuminations.
Friedrichs and Yang [73] explore 18 features of eye
movement for drowsiness detect. The features are listed
in Table 4. Rather than using principal component
analysis (PCA) or linear discriminate analysis (LDA)
to reduce the dimension of the features, they chose the
sequential floating forward selection (SFFS) [74]
408 Driver Inattention Monitoring System for Intelligent Vehicles
Driver Inattention Monitoring System for Intelligent
Vehicles. Table 4 Eighteen features of eye movement
[73]
Features
Average eye closure speed
Amplitude/velocity ratio (APVC)
APCV with regression
Blink amplitude
Blink duration
BLINKDUR baselined
Blinking frequency
Energy of blinking (EC)
EC baselined
Microsleep event 0.5 s rate
Microsleep event 1.0 s rate
Mean square eye closure
Mean eye closure
Percentage eyes >70% close (PERCLOS70)
Percentage eyes >80% closed (PERCLOS80)
PERCLOS70 baselined
PERCLOS80 EWMA baselined
Head nodding
algorithm to select the most promising features to
construct a classifier. The advantage of SFFS over fea-
ture transform techniques is its high transparency,
because the selected features remain unchanged. An
ANN classifier was trained to detect the drowsiness,
and the results showed that as long as the blinking
signals were correctly detected (high confidence), the
drowsiness detection accuracy could reach 82.5%.
Some other methods have also been used for fatigue
detection. In Sun et al. [75], a Bayesian network was
employed to infer fatigue from gaze information.
Orazio et al. [76] used a mixture Gaussian model to
model the “normal behavior” statistics from the eye
closure duration (ECD) and frequency of eye closure
(FEC) for each person, in order to identify anomalous
behaviors. Suzuki et al. [77] derived three factors from
the blinking waveform: the length of a blink, the closure
rate, and the blink rate. These factors were then
weighted using a multiple regression analysis for each
individual to calculate the drowsiness level. In
Senaratne et al. [78], four cues were fused using fuzzy
logic to detect driver fatigue: PERCLOS, head nodding
frequency, slouching frequency, and posture adjust-
ment frequency. In addition to analyzing eye activities,
some researches also analyzed mouth activities [79–81]
to estimate the level of driver inattention. Fan et al. [80]
used an LDA to classify the mouth into two states:
normal and yawning. In Vural et al. [81], they used
a BP ANN to estimate three mouth states from lip
features: normal, yawning, and talking. Vural et al.
[81] used a facial action coding system (FACS) to code
facial expressions, and then employed machine learn-
ing to discover which facial configurations were suit-
able for fatigue detection, with 31 facial actions
employed to predict drowsiness. This system claimed
to be able to predict sleep-and-crash episodes with
a 96% accuracy within subjects, and an accuracy
above 90% across subjects.
Distraction Detection Kircher et al. [82] described
and compared two different algorithms for gaze-based
driver distraction detection, based on the eye-tracking
data obtained in a field study. One algorithm relied on
the metric “percent road center” (PRC) of gaze
direction, where a PRC of more than 92% was consid-
ered to be indicative of a gaze concentration resulting
from cognitive distraction, while a PRC below 58%,
computed over 1 min, was a indicator of visual distrac-
tion. Fixations were used for the computation of PRC.
The second algorithm was based on a 3D world model
with different interior zones such as the windshield,
speedometer, mirrors, dashboard, etc., and on the time
the driver spends glancing at those zones. A time-based
“attention buffer” with a maximum value of 2 s was
decreased over time when the driver looked away from
the “field relevant for driving” (FRD), while it was
increased when the driver’s glance was inside the FRD,
until the maximum value was reached. When the buffer
reached zero the driver was considered to be distracted,
and when further conditions were met (direction indica-
tor not activated, speed above 50 km/h, no brake activa-
tion, and no extreme steering maneuvers), a warning was
issued. The results showed that both algorithms have

potential for detecting driver distraction, and that fully
attentive drivers had a PRC of about 70–80%.
Pohl et al. [83] used head pose and eye gaze
information to model the visual distraction level, which
was time dependent on the visual focus, with the assump-
tion that the visual distraction level was nonlinear: Visual
distraction increased with time (the driver looked away
from the road scene) but decreased nearly instanta-
neously (the driver refocused on the road scene).
Based on the pose/eye signals, they established their
algorithm for visual distraction detection. First, they
used a distraction calculation to compute the instanta-
neous distraction level. Then, a distraction decision-
maker determined whether the current distraction level
represented a potentially distracted driver.
Bergasa et al. [84] tried to detect visual distraction
with head pose and fatigue with yawning, eyebrow
raising, and PERCLOS. Although they developed
an algorithm for extracting the required cues, the
algorithm for fusing them was unclear.
Driving Performance Measures
A change in the mental state can induce a change in
driving performance. In Furugori et al. [85], the
pressure distribution on the seat of male subjects was
measured during simulated long-term driving, and the
results showed there was a relationship between
changes in the load center position (LCP) and driver
reported subjective fatigue. Their algorithm to derive
a fatigue index was calculated on a time interval of
10 min, which was a considerable delay.
Farid et al. [86] tried to distinguish between atten-
tive and inattentive driving in car-following situations
by analyzing the vehicle following distance and steering
angle. They built up a real-time model using hidden
Markov models with Gaussian mixtures to infer the
intentions of the driver, and this model was able to
detect a lane change half a second earlier than
conventional approaches. Zhong et al. [36] performed
a localized energy analysis of the steering wheel angle
dynamics and vehicle tracking to detect driver fatigue,
and found a trend of localized energy increase with
driving time. In Takei and Furukawa [87], chaos theory
was employed to explain the dynamics of steering
wheel motion, and estimate driver fatigue. Using
a proper time delay, they found the attractors, which
Driver Inattention Monitoring System for Intelligent Vehicles 409
involved the Chaos characteristics. They stated that they
will study the Lyapunov exponent of this chaos to esti-
mate the driver fatigue. In addition to an energy analysis,
in [88], a Gaussian mixture model was adopted to
identify the driver based on the driving behavior sig-
nals: forces on the pedals and vehicle velocity.
Kari Torkkola and Wood [89] adopted the steering
wheel position, accelerator pedal position, lane bound-
aries, and upcoming road curvature to infer driver
status. First, the original signals were preprocessed
(averaging, entropy, etc.), which yielded a huge set of
features. Then, random forest (RF) [90], a technique
based on ensembles of learners, was employed to select
the optimal parameters from the derived features. The
classifier was also constructed using RF, and the final
accuracy reached 80%.
In Ersal et al. [91], a radial-basis neural-network-
based modeling framework was developed to character-
ize normal driving behavior. Then, in conjunction with
an SVM, it was able to classify normal and distracted
driving. Vehicle dynamics and driving performance data
such as vehicle position, velocity, and acceleration, as
well as throttle and brake pedal positions, were adopted
to model normal driving. The average and standard
deviations of the residuals (the differences between the
actual and model-predicted driver actions) were chosen
as the inputs for the SVM. The results showed that the
accuracy varied between individuals.
Hybrid
Combining driver physical measures with driving
performance measures could intuitively increase the
inattention detection confidence. On the other hand,
road scene analysis and observations of the driver’s face
would make it possible to estimate what the driver
knows, what the driver needs to know, and when the
driver should know it. Combining driver gaze
information with road scene information offers several
potential benefits: context-relevant information
selection, unnecessary information suppression, and
anticipatory information selection. Table 5 shows
a summary of some researches that utilized a hybrid
method for detecting driver inattention.
Fatigue Detection Eskandarian et al. [27] utilized
artificial neural networks (ANNs) to analyze vehicle
410 Driver Inattention Monitoring System for Intelligent Vehicles

Driver Inattention Monitoring System for Intelligent Vehicles. Table 5 Summary of hybrid measures
References Raw signals Fusion technique Object
[27] Vehicle parameter data and eye closure data ANN Fatigue detection
[18] Eye gaze, head orientation, diameter of pupils, heart SVM and Adaboost Cognitive distraction
rate (RRI) detection
[92] Leg and head motions, CAN signals K-nearest neighbors Distraction detection
classifier
[93] Audio signal and CAN signals GMM/UBM Distraction detection
[94] Head orientation and the surround salience map Direct matching Visual distraction
detection
[95] Gaze variables, driving data and road geometry ANOVA and binary Distraction detection
logistic regression.
[51] Eye movement and vehicle parameters SVM Cognitive distraction
detection
[53] Eye movement and vehicle parameters Bayesian network Cognitive distraction
detection
[96] Head/eye and vehicle parameters SVM Visual and cognitive
distraction detection
[97] Vehicle and environment parameters ANFIS Distraction estimation
[98] Eye gaze, blink, head pose, and environment Region matching Visual and cognitive
parameters distraction detection
[99] Head dynamics, facial features, upper body posture In developing Driver assistance
information and vehicle dynamics

parameter data and eye closure data to infer driver
fatigue. The vehicle parameter data included speed,
acceleration, vehicle lane position, steering angle, brak-
ing, and heading angle, which was recorded at
a frequency of 20 Hz. The eye closure data was recorded
at 60 Hz using PC-based equipment by Applied Science
Laboratory (ASL), which recorded pupil diameter by
capturing reflections from the pupils (bright pupil).
Then they analyzed the data to identify the potential
variables that were correlated with drowsiness. This
analysis found four variables highly correlated with
fatigue: PERCLOS, vehicle crash, vehicle lateral dis-
placement, and steering wheel angle. Out of consider-
ation for the simplicity and robustness of the data
acquisition, Eskandarian et al. [27] implemented two
ANN fatigue detectors: one utilized the steering wheel
angle signal as an input and the other utilized both
the steering wheel angle signal and the eyelid signal
as inputs. The steering angle was preprocessed
before being input in the ANN. The preprocessing
scheme involved normalization for road curvature,
discretization at different ranges, coding, and 15-s
accumulation. For the eyelid signal, the preprocessing
scheme was the same as that for the steering angle,
except for eliminating the normalization stage. It was
reported that after proper training and cross validation,
the steering-eye ANN had an accuracy of 88%, with
a false alarm rate of 9%, while the steering-only ANN
had an accuracy of 85%, with a false alarm rate of 14%.
Distraction Detection In Miyaji et al. [18], the stan-
dard deviations of eye gaze, head orientation, pupil
diameter, and average heart rate (RRI) were combined
to improve the accuracy of the driver cognitive distrac-
tion detection. The eye and head parameters were
obtained using faceLAB, while the RRI data came
from ECG. In Miyaji et al. [18], two machine learning
techniques, SVM and Adaboost, were implemented

under the same conditions. The results showed that the
classification performance of Adaboost was slightly
better than that of SVM, while the recognition time of
AdaBoost was approximately 1/26 that of the SVM.
Sathyanarayana et al. [92] tried to detect distraction
by combining motion signals from the leg and head
with vehicle signals. The motion signals included the
three-axis acceleration of the right leg and two-axis
orientation of the head. The vehicle signals adopted
included vehicle speed, braking, acceleration, and
steering angle. Then, a group of features were derived
from these signals based on the nature of the signals.
The feature types are listed in Table 6. Next, these
derived features were analyzed using LDA to reduce
the dimension. Then, a K-nearest neighbors classifier
was trained and verified.
In order to cope with the variability between drivers
and maneuvers (context), [93] utilized a GMM/univer-
sal background model (UBM) and likelihood maximi-
zation learning scheme to first identify the driver
through an audio signal and then recognize the maneu-
vers (right/left turn and lane change) through
CAN-bus signals. Finally, the CAN-bus signals were
also used to detect distraction for a particular driver
Driver Inattention Monitoring System for Intelligent
Vehicles. Table 6 Candidate signal features [92]
Signal features
Maximum value of the signal
Minimum value of the signal
Amplitude of the difference between the first value and
the last value
Duration of the signal
Maximum difference between any two consecutive
values
Median of the signal
Mean of the signal
Difference between the maximum and minimum value
of the signal
Standard deviation of the signal
Root mean square value of the signal
Difference between the maximum and minimum value
of the differential of signal
Driver Inattention Monitoring System for Intelligent Vehicles 411
and particular maneuver. It was reported that this system
could reach an accuracy of 70% for distraction detection.
Doshi and Trivedi [94] fused head orientation detec-
tion and a saliency map of the surroundings to deter-
mine whether there was a salient object in the driver’s
view, which gave an indication of whether a driver’s head
turn was motivated by the goal in his or her mind or
some distracting object/event in the environment.
It is known that road geometry influences gaze
behavior [100], and this aspect was taken into account
by including road geometry as an additional factor
when detecting driver distraction in [95]. They utilized
an analysis of variance (ANOVA) and binary logistic
regression to analyze and establish a model for distrac-
tion detection based on gaze variables and driving data:
fixations (number and duration), scan path, standard
deviation of gaze location, speed (minimum, maxi-
mum, average and percentage change in speed), lateral
acceleration (maximum), and longitudinal decelera-
tion (maximum). The results showed that the road
geometry does influence the accuracy of distraction
detection based on driving data, but gaze behavior is
mainly influenced by distraction, with little or no influ-
ence by road geometry.
Liang et al. [51] tried to detect the driver cognitive
distraction caused by interacting with In-Vehicle Infor-
mation Systems (IVISs) in real time by fusing eye
movement and driving performance using an SVM.
The measured signals included fixation, saccade,
smooth pursuit of eye (calculated from raw gaze vector
obtained using faceLAB [48]), steering wheel angle,
lane position, and steering error. These measures were
summarized over various windows to create instances
that became the SVM model inputs. After training, the
SVM model could detect driver distraction with an
average accuracy of 81.1% (sd = 9.05%). Lee et al. [53]
utilized the same conditions as [51] but adopted a
Bayesian network to detect cognitive distraction, show-
ing that compared to an SVM model, the dynamic
Bayesian network produced better accuracy.
Markkula and Kutila [96] concentrated on
processing head/eye and vehicle performance informa-
tion to estimate both visual and cognitive distractions;
their algorithm is shown in Fig. 4. The head/eye infor-
mation derived from stereo cameras included head
position, head orientation, gaze orientation, saccade,
and blink identification, as well as confidence values.
412 Driver Inattention Monitoring System for Intelligent Vehicles
Sensor output
Sensors
post processing
Head/eye Head/eye
Gaze-world
sensor data post
mapping
processing
Lane pos.
sensor
Vehicle
Cognitive distraction
Vehicle data post
indicator calculation
sensor processing
Driver Inattention Monitoring System for Intelligent Vehicles. Figure 4
Overview of the distraction detection algorithms [96]
The vehicle performance information included lane
position, vehicle speed, etc. Based on the head/eye
information, they developed Gaze-World Mapping
and Eyes-Off-Road Detection, which could detect
momentary visual distraction. Another algorithm,
visual time-sharing detection, was developed to mea-
sure longer term visual distractions. For cognitive dis-
traction, they used three indicators to classify the
cognitive tasks with an SVM: the standard deviation
of gaze angle, standard deviation of head angle, and
standard deviation of lane position. However, in the
Gaze-World Mapping phase, which mapped gaze and
head angles onto actual real-world targets of visual
attention, the road-ahead target was static and deter-
mined off-line by inspecting the distribution of gaze
angles for road-ahead data, and then manually
enclosing the distribution in a rectangle.
Tango et al. [97] proposed a method to derive the
distraction level from relevant vehicle and environment
data using the adaptive neuro-fuzzy inference system
(ANFIS). Rather than a binary “yes” or “no,” they chose
reaction time as the output to train, validate, and test
their ANFIS model. The candidates to be selected as
input for the ANFIS included the environment visibil-
ity, traffic density, and the standard deviations in
speed, steering angle, lateral position, lateral accelera-
tion, and deceleration jerk. After preprocessing, the
level of difficulty of an IVIS and the standard deviation
of steering angle were found to have the highest
correlations with the reaction time. Thus, they were
selected as the input. No accuracy information was
provided in [97].
Distraction detection
Eyes-off-road
detection
Visual time sharing
detection
Cognitive distraction
detection
Fletcher and Zelinsky [98] utilized faceLAB to
obtain information such as eye gaze direction, eye
closure, and blink detection, as well as head position
information. In this system, upper and lower bounds
were placed on the percentage of time the driver spent
observing the road ahead, called the percentage road
center (PRC). A percentage that was too high (>90%)
could indicate a fatigued state (e.g., vacant staring).
A percentage that was too low (<20%) might indicate
a distracted state (e.g., tuning radio). Similar to the
PRC metric, they analyzed driver gaze to detect even
shorter periods of driver distraction. They used gaze
direction to reset a counter. When the driver looked
forward at the road scene, the counter was reset. If the
driver’s gaze diverged, the counter began timing. When
the gaze had been diverted for more than a specified
time period, a warning was given. The time period for
the permitted distraction was a function of the vehicle
velocity. As the speed increased, the permitted time
period would decrease, either as the inverse (reflecting
time to impact) or the inverse squared (reflecting the
stopping distance). They tried to integrate driver gaze
information into other driver assist systems to make
the system more acceptable and safer. The framework is
shown in Fig. 5. They also spent a significant amount of
effort on integrating driver gaze information into lane
tracking and sign reading systems. The lane-tracking
system was used to orient the driver gaze information.
A strong correlation was found to exist between the eye
gaze direction and the curvature of the road during
normal driving [101], with a slight correlation being
a potential indicator of inattention. Fletcher and

Signs DAS
Platooning DAS
Avoidance DAS
Fatigue DAS
Driver
State
Vehicle
State
Driver Inattention Monitoring System for Intelligent Vehicles. Figure 5
The distributed modular software architecture [98]
Zelinsky [98] integrated driver visual information with
sign detection to implement a Sign Driver Assist Sys-
tem. This system recognized critical signs in the envi-
ronment. At the same time, the driver monitoring
system verified whether the driver looked in the direc-
tion of the sign. If it appeared that the driver was aware
of the sign, the information could be made available
passively to the driver. In contrast, if it appeared that
the driver was unaware of the information, it could be
highlighted.
A driver’s body posture information is potentially
related to driver intent, driver affective state, and driver
distraction. Tran and Trivedi [99] explored the role of
3D driver posture dynamics in relation to other
contextual information (e.g., head dynamics, facial
features, and vehicle dynamics) for driver assistance.
It focused on head pose and upper body posture
extraction, but no significant results on driver
assistance were found.
Driver Physical Signal Extraction
Numerous researches have adopted commercial eye
trackers to obtain the physical signals related to the
face/eye, which have allowed them to concentrate on
exploring the inattention detection algorithm rather
than image processing. However, these commercial
Driver Inattention Monitoring System for Intelligent Vehicles 413
Sign reader
Vehicle tracker
Pedestrian detector
Obstacle detector
Monotony detector
Lane tracker
eye trackers can only work well under specific
constrained environments. They do not normally
work well for real road conditions. For example, [73]
adopted the driver state sensor from the seeing
machines [48] to obtain the eye signal. However, even
after many improvements, there were still some issues:
Reflections from glasses led to bad signal quality and
varying light conditions during daytime driving posed
problems for the eye signal tracking (see Fig. 6). There-
fore, much research has been conducted to make the
physical signals extracted using image processing more
accurate and robust. The methods for physical signal
extraction are summarized in Table 7.
For face segmentation in driver inattention detec-
tion, the commonly used methods in the literature
include a boosted cascade of Haar-like features [102],
adaptive boosting [75], landmark model matching [78],
skin color [79], and gravity center template [80].
Eren et al. [103] adopted stereo cameras and extracted
a face from a disparity map on the assumption that the
driver’s face had a smaller depth than the background.
They then used an embedded HMM to recognize the
forehead, eyes, nose, mouth, and chin.
After the face area has been segmented, it is neces-
sary to extract the eye area and mouth area for further
processing to obtain physical signals. In the literature,
the following methods have been employed to extract
414 Driver Inattention Monitoring System for Intelligent Vehicles

a b
Reflections on glasses Bad light due to sun backlight

Driver Inattention Monitoring System for Intelligent Vehicles. Figure 6

Image processing problems [73]

Driver Inattention Monitoring System for Intelligent Vehicles. Table 7 Summary of physical signal extraction

Mouth Mouth
Face segment Eye segment Blink feature Gaze segment feature Head pose
Boosted cascade Darkest regions Optical flow Hough transform Fisher Connected Two eyes and
of Haar-like search [102], [102], and gradient classifier component the center of
features [102], template derivative [77], direction [75], [79], analysis to face [104],
adaptive matching method SVM [78], finite Kalman filters and gravity determine lip pupil and the
boosting [75], [75], neural state machine the FSM [50], center [79], Gabor nostril
landmark model network [77], edge (FSM) [50], relative position template wavelet to get position [50],
matching [78], map [78], p-tile Kalman between pupil [80] mouth corners headband
skin color [79], and k-means filtering [105] and glint [105], [80] [106]
gravity center algorithm [104], pupil features +
template [80], Hough transform eigenspace [105],
disparity map + neural classifier headband + 3D
[103] [76], “bright pupil” eyeball model
[50] [105, 106], [106]
Sobel edges +
SVM [107],
templates
matching [108]

the eye area. In Brandt et al. [102], the eyes were
extracted by assuming that they were the darkest
regions in the face; Sun et al. [75] located the eyes
using a template matching method; Suzuki et al. [77]
used a neural network to detect the eyes; Senaratne
et al. [78] used an edge map to locate the irises of the
eyes; and Su et al. [104] used a p-tile algorithm and
k-means algorithm to locate the eyes. In [76], candidate
eye regions were first extracted using a modified Hough
transform, then symmetric regions in the candidates
were chosen as further candidates, and finally a neural
classifier was used to validate the presence of the eye in
the image.
Another popularly adopted method for locating
eyes involves the use of the “bright pupil” effect
produced by near-infrared light. In [50, 105, 106],
a camera equipped with a two-ring IR illuminator
was adopted to acquire a driver image. The ring sizes
were empirically calculated to obtain a dark pupil
image when the outer ring was turned on, and
a bright pupil image when the inner ring was turned
on. A controller was designed to synchronize the IR
illuminator with the image frame rate: to ensure that
the images with and without bright pupils were inter-
laced. Digitally subtracting the dark pupil image from
the bright pupil image produced a difference image,
where the pupils appear to be the brightest regions in
the image, as can be seen in Fig. 7. The pupils were
detected on the resulting image by searching the entire
image to locate two bright blobs that satisfied certain
constraints. After locating the eyes in the initial frames,
Bergasa et al. [50] used two Kalman filters, one for each

Driver Inattention Monitoring System for Intelligent Vehicles. Figure 7
Fields captured and subtraction. (a) Image obtained with inner IR ring. (b) Image obtained with outer IR ring. (c) Difference
image [50]
pupil, to continuously and robustly monitor a driver
with eye closure or oblique face orientation. Huang
et al. [107] eliminated the need for the synchronizer
by acquiring the pupil location from a single image:
First, pupil candidates were obtained through Sobel
edges, and then they were identified using an SVM
with a Gaussian kernel. In Zhu et al. [108], a round
template two values matching algorithm was proposed
for locating bright pupils, which had an accuracy of
96.4% but consumed 1,011 ms on a PIII 800 MHz
computer.
After the location of the eye is extracted, the
blinking and gaze parameters should be calculated.
In Brandt et al. [102], blinks were measured by analyz-
ing the optical flow of the eye region. Suzuki et al. [77]
used a derivative method to detect the eyelids and
produce a blinking waveform. Senaratne et al. [78]
used an SVM to classify the state of the eye as open
or closed to get the PERCLOS value. In the “bright
pupil” condition, Bergasa et al. [50] implemented a
Driver Inattention Monitoring System for Intelligent Vehicles 415
finite state machine (FSM), with five states
defined: tracking_ok, closing, closed, opening, and
tracking_lost. The transitions between states were
achieved from frame to frame as a function of the
width/height ratio of the pupils. The ocular parameters
such as eye closure duration, PERCLOS, eye closure/
opening speed, and blink frequency were calculated
as functions of the FSM. Ji and Yang [105] used
Kalman filtering to track eyelid movements and
compute the PERCLOS and average eye closure
speed (AECS).
The gaze was estimated by combining the Hough
transform and gradient direction in [75], while Bergasa
et al. [50] calculated the gaze based on the position and
speed data using Kalman filters and the FSM. Ji and
Yang [105] estimated gaze direction using information
about the head movement and relative position
between pupil and glint, with the gaze direction
quantized into nine zones: left, front, right, up, down,
upper left, upper right, lower left, and lower right.
416 Driver Inattention Monitoring System for Intelligent Vehicles
Cudalbu et al. [106] utilized a headband and
a simplified 3D eyeball model to estimate the gaze
orientation with an accuracy that varied from 1 to 3 .
Besides the eye, estimating the position of the
mouth is also useful in fatigue detection. Rongben
et al. [79] used a fisher classifier to extract the mouth
area from the face region, while Fan et al. [80] used
a gravity center template to extract the mouth area.
Then, Rongben et al. [79] used connected component
analysis to find the lips and [80] used a Gabor wavelet
to get the corners of the mouth.
The head nodding frequency, slouching frequency,
and posture adjustment frequency were derived from
changes in the head position in [78]. Su et al. [104]
clustered facial orientations into five clusters: frontal,
left, right, up, and down, depending on the position of
the eyes and the center of the face. Similarly, based on
the pupil and nostril positions, Bergasa et al. [50] made
a coarse 3-D face pose estimation. Ji and Yang [105]
used an eigenspace algorithm to map seven pupil
features (inter-pupil distance, sizes of left and right
pupils, intensities of left and right pupils, and
ellipse ratios of left and right pupils) to determine
face orientation, which was quantized into seven
angles: 45 , 30 , 15 , 0 , 15 , 30 , and 45 .
Cudalbu et al. [106] employed a headband with IR
reflective markers to estimate the 6 DOF head pose
with an average error of 0.2 .
Discussion: Future Directions
Issues with Detection
The subjective report measures can produce some rea-
sonable results in quantifying fatigue level. Because this
kind of approach requires that the driver report his or
her state frequently, both the fatigue level result and the
driver himself or herself could cause interference. Large
individual differences have been observed with the
overall driving performance and blink duration inde-
pendent of the KSS values [56]. In addition, [109]
demonstrated that drivers have difficulty judging
their fitness, especially after about 3 h of continuous
monotonous daytime driving with increasing drowsi-
ness. For these reasons, it is not sufficient to solely
record the KSS. However, if only a rough fatigue level
is needed and the lowest cost is required, this kind of
approach may be the best choice. The driver biological
measures directly measure biological signals from
a driver’s body, and have been found to be highly
accurate when used to detect a driver’s fatigue level.
Svensson [29] even proposed an objective sleepiness
scoring (OSS) method that relied on EEG. However,
most of the driver biological measures are intraper-
sonal. The results of [71] showed that intrapersonal
data had a good linear trend, while interpersonal data
showed a different threshold. Bouchner [110] also
showed that the EEG was very dynamic and very sen-
sitive to outside factors. In addition, EEG patterns vary
between individuals. Therefore, these two kinds of
measures should be treated as rough ground truth
indicators for other methods. Driver physical measures
and driving performance measures are the most
promising methods in the real driving context, because
neither rely on intrusive measurements that might
affect the driver.
For fatigue detection, the most popularly used
parameter in driver physical measures is PERCLOS.
However, one of the limitations of PERCLOS is that
its prediction is good only when using large time inter-
vals. Moreover, PERCLOS does not take into account
the variability in human behavior, because blinking
activity can significantly differ between individuals.
Another challenge for driver physical measures is the
robustness of the algorithm under all driving condi-
tions (day and night, sunny and cloudy, etc.), because
this type of method mainly relies on image processing.
Many researchers have adopted infrared illumination
techniques in image acquisition systems for three pur-
poses. First, they minimize the impact of different
ambient lighting conditions. Second, they allow the
bright pupil effect to be produced, which makes eye
detection easier. Third, because near-infrared is barely
visible to the driver, it minimizes any interference with
their driving. The “bright pupil” effect does benefit the
eye extraction process, but it only works well under
some constrained lighting conditions. Moreover, in
real driving scenarios, these constraints cannot be sat-
isfied most of the time. In Bergasa et al. [50], three
main illumination challenges were encountered: artifi-
cial light from elements outside the road (such as
streetlights), vehicle lights, and sunlight, as shown in
Fig. 8. The “bright pupil” effect will disappear under
these conditions, which causes the eye detection to fail,
and consequently influences the inattention detection.

Driver Inattention Monitoring System for Intelligent Vehicles. Figure 8
Effects of external lights on the acquisition system. (a) Out-of-the-road lights effect. (b) Vehicle lights effect. (c) Sunlight
effect. (d) Sunlight effect with filter [50]
For example, sunlight and reflections from glasses
could cause the performance to drop considerably, to
30% [50]. No matter how the hardware is adjusted, the
“bright pupil” effect is not robust, especially in daytime
[50] or when wearing glasses [27]. Even under
constrained conditions, the reflection of the IR in the
pupils varies by individual. Even with the same driver,
the intensity depends on the gaze point, head position,
and opening of the eye. Therefore, more reliable real-
time eye detection algorithms are preferred over the
“bright pupil” effect. As described in subsection
“Driver Physical Signal Extraction”, most studies have
concentrated on image processing, and have estimated
the driver physical parameters (e.g., gaze, face pose,
mouth activity) quite roughly. Combining the image
processing with some face mathematical models leads
to more accurate estimation. Dong et al. [111] devel-
oped a real-time tracking kernel for stereo cameras to
estimate face pose and face animation, including the
movement of the eyelid, eyeball, eyebrow, and mouth,
for driver inattention detection.
The advantage of driving performance measures is
that the signals are meaningful and readily available.
Moreover, the literature shows that they are useful for
estimating driver fatigue, and show good promise in
a real driving context. Although in subsection “Effects
of Distraction”, many researchers have found that
driver distraction influences driving performance, few
studies have utilized driving performance measures to
detect distraction, with most concentrating instead on
fatigue or abnormal detection.
One more issue that should be pointed out is that
many of the researches claimed very high detection
accuracies, which were true only for their particular
Driver Inattention Monitoring System for Intelligent Vehicles 417
hypothetical fatigue/distraction definitions. These
hypothetical definitions usually covered a limited
region of the whole fatigue/distraction definition.
Without this condition, the accuracy rates had no
meaning.
Because of the significant difficulties inherent in
measuring driver attention, the magnitude, and partic-
ularly the safety implications, of driver distraction have
been very difficult to determine. Indeed, unlike seat belt
use, the driver’s attention status cannot be categorized
as “yes” or “no,” but should be quantified in the same
manner as blood alcohol level [24].
The factors that influence driver fatigue/drowsiness
include greater daytime sleepiness, less sleep, a more
difficult schedule, more hours of work, age, driving
experience, cumulative sleep debt, the presence of
a sleep disorder, and the time of day. This entry focused
on real-time inattention detection technology, rather
than on long-term sleep/wake regulation prediction
technology. Biomathematical models that quantify
the effects of the circadian and sleep/wake processes
on the regulation of alertness and performance have
been developed in an effort to predict the magnitude
and timing of fatigue-related responses in transporta-
tion operations. These models of fatigue and perfor-
mance typically use input information about sleep
history, duration of wakefulness, work and rest pat-
terns, and circadian phase to predict sleepiness, perfor-
mance capability, and/or fatigue risk [47]. Researches
on biomathematical models could enhance the confi-
dence of the real-time estimation, because the former
could be used to predict when the probability of sleep-
iness will become higher. For descriptions of these
biomathematical models, see [47].
418 Driver Inattention Monitoring System for Intelligent Vehicles
Systematical Design Consideration
This entry indicated that no single measure can be used
to reliably detect inattentive driving. A combination of
different measures is recommended, e.g., analyses of
lateral control performance and eye blink patterns.
According to the definition of driver distraction [2],
when driver distraction occurs there should be some
kind of distraction source that exists inside or outside
of the vehicle. Therefore, fusing driver gaze informa-
tion, vehiclears’ ego state (steering, lane position,
speed, and state of IVIS) with current road scenario
(e.g., the type of road, weather conditions, and traffic
density) will lead to a more comprehensive under-
standing and recognition of driver distraction.
The level of distraction associated with a given sec-
ondary task depends on the extent to which a driver is
engaged in the task. Different secondary tasks have
different requirements for concentration. Thereby,
they have different levels of distraction. Based on the
number of button presses and/or glances away from the
forward road, Klauer et al. [112] defined three catego-
ries of secondary tasks: complex tasks, moderate tasks,
and simple tasks. It was found that complex tasks
carried three times the risk of involvement in a crash
or near-crash as simple tasks or no secondary tasks.
Specifically, for drivers performing complex secondary
tasks, elevated likelihood ratios were found for the
following conditions: dusk and unlighted darkness,
rain, divided roads, and roads with grades (straight or
curved). Therefore, identifying the environmental con-
ditions is important to correctly assess the risk of the
distraction.
Different environments contribute different risk
levels for the same inattention state. Different environ-
ments could also induce the occurrences of different
distractions. Specifically, Stutts et al. [3] reported that
crashes associated with adjusting audio devices were
more likely at night, those associated with moving
objects inside the vehicle were more likely on nonlevel
grades, and distractions involving communication with
other occupants were more likely at intersections.
Horne and Reyner [113] found various criteria associ-
ated with drowsiness-related accidents: the vehicle run-
ning off the road, no sign of braking, no mechanical
defect, good weather, and the elimination of speeding.
The NHTSA [27] reported some indirect cues:
Accidents were more likely to be associated with the
period from midnight to early morning, rural highways
with a 55–65 mph speed limit, and fixed objects (trees,
guardrail, highway sign, etc.). This, if these contextual
cues could be taken into account when determining the
risk level of an inattention occurrence and determining
which countermeasure should be adopted, it would
make the driver inattention monitoring system more
reliable and acceptable.
Conclusion
This entry reviews the current state of the knowledge
about driver inattention monitoring. Driver inatten-
tion increases driving risk, and has become a major
factor in a considerable percentage of traffic accidents.
Driver inattention has no universally accepted defini-
tion. However, based on a review of the literature,
driver inattention is classified into two main categories,
distraction and fatigue, each of which also contains
a few subcategories. Summarily, distraction means the
driver has the capability of paying attention, but their
attention is shifted away from the primary driving task
to some secondary task or attracted by some attractive
object/event. Fatigue means the driver has exhausted
his or her attention energy and cannot maintain suffi-
cient attention for driving. The causes of distraction
and fatigue are different, and they impose different
influences on the driver and driving performance.
Revealing these influences could help when selecting
appropriate measures to develop a real-time inatten-
tion monitoring system. Recently, many commercial
products relative to driver inattention monitoring
have emerged. Auto companies such as Toyota, Nissan,
Volvo, Mercedes-Benz, and Saab have installed driver
inattention monitoring systems on their top-brand
vehicles and/or are conducting researches on such sys-
tems. A few third parties, like Seeing Machines and
SmartEye, are providing camera-based nonintrusive
tools for measuring driver physical signals such as
gaze, head pose, and mouth activity. It should be
pointed out that in most of the cases, neither the
scientific and technological method behind, nor the
exhaustive results of the performance can be provided
for these commercial products.
Many articles have reported that these tools work
well under constrained conditions, but are not robust

under real driving conditions. Thus, there is still much
progress to be made to improve the robustness and
accuracy of the physical measuring tools. In the scien-
tific literature, five different types of measures could be
found for detecting driver inattention: (a) subjective
report measures, (b) driver biological measures,
(c) driver physical measures, (d) driving performance
measures, and (e) hybrid measures. Although not suit-
able for a real life context, subjective report measures
and driver biological measures could serve as some
rough ground truth indicators. Because driver physical
measures and driving performance measures have
advantages and disadvantages, hybrid measures are
believed to provide more reliable solutions, which will
both accurately detect driver inattention and minimize
the number of false alarms to promote acceptance of
the system. After all, the goal of a driver inattention
monitoring system is to reduce the driving risk.
To obtain this goal, three distinct sources of data
should be combined: driver physical variables, driving
performance variables, and information from the IVIS.
In addition to these variables, it is important to con-
sider the characteristics of the driving environment
(e.g., the type of road, weather conditions, and traffic
density).
Bibliography
1. Ranney TA, Mazzae E, Garrott R, Goodman MJ (2000) Nhtsa
driver distraction research: past, present, and future. Technical
report, Transportation Research Center Inc
2. Lee JD, Young KL, Regon MA (2009) Driver distraction:. CRC
Press/Taylor & Francis Group, London
3. Stutts JC, Reinfurt DW, Staplin L, Rodgman EA (2001) The role of
driver distraction in traffic crashes. Technical report, American
Automobile Association AAA Foundation for Traffic Safety
4. S.S.G. on Fatigue (2006) Canadian operational definition of
driver fatigue. Technical report, STRID Sub-Group on Fatigue
5. Brill JC, Hancock PA, Gilson RD (2003) Driver fatigue: is some-
thing missing? In: Proceedings of the second international
driving symposium on human factors in driver assessment,
training and vehicle design, Park City, pp 138–142
6. T.R.S. for the Prevention of Accidents (2001) Driver fatigue and
road accidents: a literature review and position paper. Tech-
nical report, The Royal Society for the Prevention of Accidents
7. Croo HD, Bandmann M, Mackay GM, Rumar K, Vollenhoven P
(2001) The role of driver fatigue in commercial road transport
crashes. Technical report, European Transport Safety Council
8. Horne J, Reyner L (1999) Vehicle accidents related to sleep: a
review. Occup Environ Med 56(5):289–294
Driver Inattention Monitoring System for Intelligent Vehicles 419
9. Nilson T, Nelson TM, Carlson D (1997) Accid Anal Prev
29(4):479
10. Endsley MR (1995) Hum Factors 37(1):32
11. http://www.chalmers.se/safer/driverdistraction-en/programme
12. Angell L, Auflick J, Austria A, Kochhar D, Tijerina L, Biever W,
Diptiman T, Hogsett J, Kiger S (2006) Driver workload metrics
project-task 2 final report. Technical report, U.S. Dept. of
Transportation National Highway Traffic Safety Administration
13. Harbluk JL, Noy YI, Trbovich PL, Eizenman M (2007) Accid Anal
Prev 39:372
14. Rantanen EM, Goldberg JH (1999) The effect of mental work-
load on the visual field size and shape. Ergonom 42(6):
816–834
15. Harbluk JL, Noy YI (2002) The impact of cognitive distraction
on driver visual behaviour and vehicle control. Technical
report, Ergonomics Division, Road Safety Directorate and
Motor Vehicle Regulation Directorate
16. Hayhoe MM (2004) Infancy 6(2):267
17. May JG, Kennedy RS, Williams MC, Dunlap WP, Brannan JR
(1990) Acta Psychol 75(1):75
18. Miyaji M, Kawanaka H, Oguri K (2009) Driver’s cognitive dis-
traction detection using physiological features by the
adaboost. In: Proceedings of the 12th international IEEE con-
ference on intelligent transportation systems, St. Louis, 2009
19. Liang Y, Lee JD (2010) Accid Anal Prev 42:881
20. Zhang H, Smith M, Dufour R (2008) A final report of safety
vehicles using adaptive interface technology (phase ii: Task 7c):
visual distraction. Technical report, Delphi Electronics and Safety
21. Itoh M (2009) Individual differences in effects of secondary
cognitive activity during driving on temperature at the nose
tip. In: Proceedings of the 2009 IEEE International conference
on mechatronics and automation, Changchun, China, 2009
22. Wesley A, Shastri D, Pavlidis I (2010) A novel method to
monitor driver’s distractions. In: Proceedings of the 28th of
the international conference extended abstracts on human
factors in computing systems CHI EA 10 (2010), Atlanta, 2010
23. Berka C, Levendowski DJ, Lumicao MN, Yau A, Davis G,
Zivkovic VT, Olmstead RE, Tremoulet PD, Craven PL
(2007) Aviat Space Environ Med 78(5):B231
24. Ranney TA (2008) Driver distraction: a review of the current
state-of-knowledge. Technical report, NHTSA
25. Zhou H, Itoh M, Inagaki T (2008) Influence of cognitively
distracting activity on driver’s eye movement during prepara-
tion of changing lanes. In: SICE annual conference, Tokyo,
2008
26. Carsten OMJ, Brookhuis K (2005) Transp Res Part F: Traffic
Psychol Behav 8:191
27. Eskandarian A, Sayed R, Delaigue P, Blum J, Mortazavi A (2007)
Advanced driver fatigue research. Technical report, Report No.
FMCSA-RRR-07-001. U.S. Department of Transportation Fed-
eral Motor Carrier Safety Administration
28. Dinges DF, Richard G (1998) Perclos: a valid psychophysiolog-
ical measure of alertness as assessed by psychomotor
vigilance. Technical report, Federal Highway Administration,
Office of Motor Carriers
420 Driver Inattention Monitoring System for Intelligent Vehicles
29. Svensson U, (2004) Electrooculogram analysis and develop-
ment of a system for defining stages of drowsiness.
Master’s thesis, Department of Biomedical Engineering,
Linkoping University, Sweden
30. Hulbert S (1972) Effects of driver fatigue. In: Forbes TW (ed)
Human factors in highway traffic safety research. Wiley-
Interscience, New York
31. Mast T, Jones H, Heimstra N (1989) Effects of fatigue on
performance in a driving device. Highway research record.
Technical report, Driver Fatigue Research: Development of
Methodology, Haworth, Vulcan, Triggs, and Fildes (eds)
Accident Research Center, Monash University Australia
32. Kahneman D (1973) Attention and effort. Technical report,
Prentice Hall, Englewood Cliffs
33. Yabuta K, Iizuka H, Yanagishima T, Kataoka Y, Seno T (1985)
The development of drowsiness warning devices. In: Proceed-
ings of the 10th International technical Conference on Exper-
imental Safety Vehicles, Washington, DC
34. Dingus TA, Hardee L, Wierwille WW (1985) Development of
impaired driver detection measures. Technical report, Depart-
ment of Industrial Engineering and Operations Research, Vir-
ginia Polytechnic Institute and State University, (Departmental
Report 8504), Blacksburg
35. Elling M, Sherman P (1994) Evaluation of steering wheel mea-
sures for drowsy drivers. In 27th ISATA, Aachen, Germany
36. Zhong YJ, Du LP, Zhang K, Sun XH (2007) Wavelet Anal Pattern
Recognit 4:1843
37. Skipper JH, Wierwille W, Hardee L (1984) An investigation
of low level stimulus induced measures of driver
drowsiness. Technical report, Virginia Polytechnic Institute
and State University IEOR Department Report No.8402,
Blacksburg
38. Stein AC (1995) Detecting fatigued drivers with vehicle simu-
lators. Technical report, Drive r Impairment, Driver Fatigue and
Driving Simulation, Hartley L (ed) Taylor & Francis, Bristol,
pp 133–150
39. Safford R, Rockwell TH (1967) Highway Res Record 163:68
40. Riemersama JB, Sanders AF, Wildervack C, Gaillard AW
(1977) Performance decrement during prolonged night
driving. Technical report, Vigilance: Theory, Operational
Performance and Physiological Correlates, Makie RR (ed)
Plenum Press, New York
41. www.saabnet.com/tsn/press/071102.html
42. Nabo A (2009) Driver attention – dealing with drowsiness and
distraction. Technical report, Saab Automobile AB
43. www.smarteye.se
44. www.testdriven.co.uk/lexus-ls-600h/
45. www.emercedesbenz.com
46. www.media.volvocars.com
47. Barr L, Popkin S, Howarth H (2009) An evaluation of
emerging driver fatigue detection measures and technolo-
gies. Technical report, Volpe National Transportation Systems
Center
48. www.seeingmachines.com
49. www.smivision.com
50. Bergasa L, Nuevo J, Sotelo M, Barea R, Lopez E (2006) IEEE
Trans Intell Transp Syst 7(1):63
51. Liang Y, Reyes ML, Lee JD (2007) IEEE Trans Intell Transp Syst
8(2):340
52. Freund Y, Schapire RE (1997) J Comput Syst Sci 55:119
53. Lee J, Reyes M, Liang Y, Lee YC (2007) Safety vehicles using
adaptive interface technology (task 5): algorithms to assess
cognitive distraction. Technical report, The University of
Iowa
54. Sathyanarayana A, Boyraz P, Hansen JHL (2008) Driver
behavior analysis and route recognition by hidden markov
models. In: Proceedings of the 2008 IEEE international
conference on vehicular electronics and safety, Columbus,
2008
55. Kaida K, Takahashi M, Akerstedt T, Nakata A, Otsuka Y, Haratani T,
Fukasawa K (2006) Clin Neurophysiol 117(7):1574
56. Ingre M, Akerstedt T, Peters B, Anund A, Kecklund G (2006)
J Sleep Res 15(1):47
57. Craig A, Tran Y, Wijesuriya N, Boord P (2006) Biol Psychol
72:78
58. Jap BT, Lal S, Fischer P, Bekiaris E (2009) Expert SystAppl
36(2):2352
59. Chouvarda I, Papadelis C, Papadeli CK, Bamidis PD,
Koufogiannis D, Bekiaris E, Maglaveras N (2007) Medinfo
12(2):1294
60. Yeo MVM, Li XP, Shen K, Smith EPW (2009) Safety Sci 47(1):115
61. Shen KQ, Li XP, Ong CJ, Shao SY, Smith EPW (2008) Clin
Neurophysiol 119:1524
62. Lin CT, Wu RC, Liang SF, Chao WH, Chen YJ, Jung TP
(2005) Circuits and systems I: regular papers, IEEE Trans
52(12):2726
63. Lin CT, Chen YC, Huang TY, Chiu TT, Ko LW, Liang SF,
Hsieh HY, Hsu SH, Duann JR (2008) Biomed Eng IEEE Trans
55((5):1582
64. Liu J, Zhang C, Zheng C (2010) Biomed Signal Process Control
5:124
65. Skinner BT, Nguyen HT, Liu DK (2007) Classification
of eeg signals using a genetic-based machine learning
classifier. In: 29th annual international conference of the
IEEE, Lyon. Engineering in Medicine and Biology Society,
pp 3120–3123
66. Yang G, Lin Y, Bhattacharya P (2010) Inf Sci 180:1942
67. Chua CP, McDarby G, Heneghan C (2008) Physiol Meas
29(8):857
68. Damousis IG, Tzovaras D (2008) Intell Transp Syst IEEE Trans
9(3):491
69. Shuyan H, Gangtie Z (2009) Expert Syst Appl 36:7651
70. Balasubramanian V, Adalarasu K (2007) J Bodyw Mov Ther
11:151
71. Katsis CD, Ntouvas NE, Bafas CG, Fotiadis DI (2004) Assessment
of muscle fatigue during driving using surface emg. In: Pro-
ceedings of the IASTED international conference on biomed-
ical engineering, Article, Innsbruck, pp 259–262
72. Fan X, Sun Y, Yin B, Guo X (2010) Pattern Recognit Lett
31:234

73. Friedrichs F, Yang B (2010) Camera-based drowsiness refer-
ence for driver state classification under real driving condi-
tions. In: 2010 IEEE intelligent vehicles symposium, San
Diego, 2010
74. Pudil P, Ferrii FJ (1994) Floating search methods for feature
selection with nonmonotonic criterion functions. In: Pro-
ceedings of the 12th international conference on pattern
recognition, IAPR, Jerusalem, 1994
75. Sun XH, Xu L, Yang JY (2007) In: MIPPR 2007: automatic target
recognition and image analysis; and multispectral image
acquisition
76. Orazio TD, Leo M, Guaragnella C, Distante A (2007) Pattern
Recognit 40(8):2341
77. Suzuki M, Yamamoto N, Yamamoto O, Nakano T, Yamamoto
S (2006) Systems, Man Cybern 4:2891
78. Senaratne R, Hardy D, Vander B, Halgamuge S (2007) Lect
Notes Comput Sci 4492:801
79. Rongben W, Lie G, Bingliang T, Lisheng J (2004) Monitoring
mouth movement for driver fatigue or distraction with one
camera. In: IEEE 7th conference on intelligent transportation
systems, pp 314–319
80. Fan X, Yin BC, Sun YF (2007) Mach Learn Cybern 2:664
81. Vural E, Cetin M, Ercil A, Littlewort G, Bartlett M, Movellan J
(2007) Drowsy driver detection through facial movement
analysis. Springer, Berlin/Heidelberg
82. Kircher K, Ahlstrom C, Kircher A (2009) Comparison of two
eye gaze based real-time driver distraction detection
algorithms in a small-scale field operational test. In: Proceed-
ings of the fifth international driving symposium on human
factors in driver assessment, training and vehicle design, Big
Sky, 2009
83. Pohl J, Birk W, Westervall L (2007) J Syst Control Eng
221(14)):541
84. Bergasa LM, Buenaposada JM, Nuevo J, Jimenez P, Baumela L
(2008) Analysing driver’s attention level using computer
vision. In: 11th international IEEE conference on intelligent
transportation systems, Beijing
85. Furugori S, Yoshizawa N, Iname C, Miura Y (2005) Rev
Automot Eng 26(1):53
86. Farid M, Kopf M, Bubb H, Essaili A (2006) Safety Lit 1960:639
87. Takei Y, Furukawa Y (2005) IEEE international conference on
systems, Man Cybern 2:1765
88. Wakita T, Ozawa K, Miyajima C, Igarashi K, Itou K, Takeda K,
Itakura F (2006) IEICE Trans Inf Syst E89-D(3):1188
89. Torkkola K, Massey N, Wood C (2004) Driver inattention
detection through intelligent analysis of readily
available sensors. In: The 7th international IEEE conference
on intelligent transportation systems, Washington, DC,
2004
90. Breiman L (2001) Random forests. Technical report, Univer-
sity of Califomia, Berkeley
91. Ersal T, Fuller HJA, Tsimhoni O, Stein JL, Fathy HK (2010)
Model-based analysis and classification of driver distraction
under secondary tasks. IEEE transactions on intelligent trans-
portation systems 11(3):692–701
Driver Inattention Monitoring System for Intelligent Vehicles 421
92. Sathyanarayana A, Nageswaren S, Ghasemzadeh H, Jafari R,
Hansen JHL (2008) Body sensor networks for driver
distraction identification. In: IEEE international conference
on vehicular electronics and safety (ICVES), Columbus, 2008
93. Sathyanarayana A, Boyraz P, Hansen JHL (2011) Information
fusion for robust ‘context and driver aware’ active vehicle
safety systems. Inf Fusion 12(4):293–303
94. Doshi A, Trivedi M (2009) Investigating the relationships
between gaze patterns, dynamic vehicle surround analysis, and
driver intentions. In: Intelligent vehicles symposium. IEEE, Xi’an
95. Weller G, Schlag B (2009) In: European conference on human
centred design for intelligent transport systems
96. Markkula G, Kutila M (2005) Online detection of driver
distraction – preliminary results from the aide project. In:
Proceedings of the 2005 international truck and bus safety
and security symposium, Virginia, pp 86–96
97. Tango F, Calefato C, Minin L, Canovi L (2009) Moving
attention from the road: a new methodology for the driver
distraction evaluation using machine learning approaches.
In: HSI 2009, Catania
98. Fletcher L, Zelinsky A (2007) Driver state monitoring to
mitigate distraction. Technical report. In: Faulks IJ, Regan M,
Stevenson M, Brown J, Porter A, Irwin JD (eds) Distracted
driving. Sydney, NSW: Australasian College of Road Safety
pp 487–523
99. Tran C, Trivedi MM (2010) Towards a vision-based system
exploring 3d driver posture dynamics for driver assistance:
issues and possibilities. In: 2010 IEEE intelligent vehicles
symposium, San Diego, 2010
100. Land MF, Lee DN (1994) Nature 369:742
101. Apostoloff N, Zelinsky A (2003) Int J Robot Res 5:28
102. Brandt T, Stemmer R, Rakotonirainy A (2004) IEEE Int Conf
Syst Man Cybern 17:6451
103. Eren H, Celik U, Poyraz M (2007) Stereo vision and statistical
based behavior prediction of driver. In: Proceedings of IEEE
intelligent vehicles symposium, Istanbul, 2007
104. Su MC, Hsiung CY, Huang DY (2006) Syst Man Cybern 1:429
105. Ji Q, Yang XJ (2002) Real-Time Imaging 8:357
106. Cudalbu C, Anastasiu B, Radu R, Cruceanu R, Schmidt E, Barth E
(2005) Signals Circuits Syst 1:219
107. Huang H, Zhou YS, Zhang F, Liu FC (2007) Wavelet Anal
Pattern Recognit 3:1144
108. Zhu ZW, Fujimura K, Ji Q (2007) Data Sci J 6:636
109. Schmidt EA, Schrauf M, Simona M, Fritzsche M, Buchnerb A,
Kincsesa WE (2009) Accid Anal Prev 41:1087
110. Bouchner P (2006) WSEAS Trans Syst 5(1):84
111. Dong Y, Hu Z, Uchimura K, Murayama N (2010) A robust and
efficient face tracking kernel for driver inattention monitoring
system. In: Intelligent vehicles symposium (IV), 2010 IEE
112. Klauer SG, Dingus TA, Neale VL, Sudweeks JD, Ramsey DJ
(2006) The impact of driver inattention on near-crash/crash
risk: An analysis using the 100-car naturalistic driving study
data. Technical report, National Highway Traffic Safety
Administration, Washington, DC
113. Horne JA, Reyner LA (1995) Br Med J 310(6979):565
422 Driving Under Reduced Visibility Conditions for Older Adults
Driving Under Reduced Visibility
Conditions for Older Adults
RUI NI
Department of Psychology, Wichita State University,
Wichita, KS, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Car Following in Fog
Collision Detection in Fog
Steering Control Under Reduced Contrast Conditions
Spatial Integration in Processing Optical Flow
Information
Future Directions
Bibliography
Glossary
Contrast sensitivity The ability to distinguish differ-
ent levels of luminance for a given display.
Definition of the Subject and Its Importance
An important and consistent finding regarding driving
safety is that accident risk increases for older driver
populations [1–3]. Evans [2] systematically examined
data from the FARS (Fatality Analysis Reporting
System) and found steady increases in accident fatalities
and rate of severe crashes for older drivers who were
older than 60. This increased accident rate occurred for
both men and women and was independent of miles
driven.
For many people, driving gives us a sense of inde-
pendence. This becomes even more important for older
population to maintain the quality of life. Unfortu-
nately, natural age-related functional deteriorations
cause decreased driving performance and increased
crash risks. There are an increasing number of older
adults on the road today compared to 10 years ago [4],
and the older driver population will continue to
increase in the next few decades. Thus, it is important
to understand older adults’ driving behaviors and their
limitations so that we can ultimately help them to
sustain a healthy and active lifestyle.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Owsley [5] identified four main functional impair-
ments that contribute to decreased driving perfor-
mance: (1) visual, (2) visual-cognitive, (3) cognitive,
and (4) physical. These impairments can come at any
age but are more common among the older popula-
tion. Research on visual impairments and driving has
mainly focused on the decline of visual acuity for older
drivers and visual acuity assessment has been
a standard screening test for the driver’s license test.
However, it has been shown that visual acuity is not
a good predictor for accident risk [6]. Instead, dynamic
acuity, contrast sensitivity, and useful field of view [7]
were proved to be much better candidates to predict
accident risks. Wood and Owens [6] found that real-
world recognition performance of all age groups was
independent of visual acuity, which however was
seriously degraded during night driving, and this
impairment was greater for the older participants.
McGwin, Chapman, and Owsley [8] found that con-
trast sensitivity was significantly associated with the
frequency of making left-hand turns, driving on high-
traffic roads, driving during rush hours, driving alone,
and parallel parking. Glass [9] showed that contrast
sensitivity accounted for more age-related variability
and was highly correlated with high sensory-demand
tasks as compared with low sensory-demand tasks.
Thus, it is critical to systematically examine older
drivers’ driving behaviors under different contrast
conditions.
Introduction
The increased crash risk for older drivers could result
from a number of age-related factors. These factors
range from sensory processing and perceptual
processing to attention and cognitive ability.
Age-related declines in sensory processing have been
found in accommodation [10], contrast sensitivity
[11–13], dark adaptation [14, 15], visual acuity
[16, 17], spatial vision [18], and dynamic visual acuity
[19]. Age-related changes in perceptual processing have
been found in motion perception [20–24], optical flow
[22, 25, 26] and depth perception [27, 28].
These types of age-related changes have important
implications for driving safety. For example, declines in
motion and depth perception can result in
# 2012, DOI 10.1007/978-1-4419-0851-3

performance decrements in detecting impending colli-
sions during decelerations [25] and of approaching
objects [26]. Age-related declines in attention include
performance decrements for both focused [29, 30] and
divided attention tasks [31]. One issue that has been
extensively studied is the decline in the useful field of
view [7, 32–35]. Finally, age-related declines in cogni-
tive ability include a consistent result in generalized
slowing of cognitive processing [36, 37].
Safely traveling in a vehicle relies on proper skills
and capabilities performing complex tasks which can
be categorized into three types of activities – strategic,
tactical, and control [38]. Strategic tasks involve the
planning of the trip and the overall goals for the driv-
ing. Before we even begin to drive, we must plan the
route to our new destination. Choosing a route can be
influenced by many factors. For example, we may
choose a particular route because it is the shortest
route in distance. Other times, we may decide on
a route which is the shortest in travel time, the most
familiar one, or the most fuel efficient one. Once
a route is determined, the driver attempts to follow
that route, experiencing a phenomenon called
wayfinding. Wayfinding is the ability for a person to
navigate through and around an environment. Sub-
tasks of wayfinding that are equally important to safe
driving include maintaining adequate heading, vehicle
velocity [39], and developing a cognitive representa-
tion of the environment [40].
Tactical tasks of driving focus on the decisions
made for the on-hand goals of arriving at a destination.
These tasks include speed selection, decisions to pass
other vehicles, and lane selection. For example, detec-
tion of an imminent collision is crucial in safe driving.
Imagine a driver waiting to make a left turn at an
intersection. If there are oncoming vehicles
approaching the intersection, the driver needs to decide
whether or not it is safe to make the turn. In other
words, the driver has to judge whether there will be
a collision before the turn is finished, by estimating the
time-to-contact (TTC) from a stationary standpoint. It
will be safe to make the left turn if the TTC is greater
than the time that it would take for the driver to finish
the turn. Now imagine you are sitting in the upcoming
vehicle approaching this intersection. You need to esti-
mate the TTC of the left-turning vehicle from a moving
standpoint. If a collision is imminent, you should make
Driving Under Reduced Visibility Conditions for Older Adults 423
a quick judgment and execute the proper reactions in
attempt to avoid the collision, by either swerving, brak-
ing, or even speeding up.
Control tasks focus on the operation of the vehicle.
These tasks include maintaining safe distances between
cars, maintaining speeds, maintaining lane positions,
and steering the vehicle. For example, proper car
following entails maintaining a safe distance between
a driver’s vehicle and the lead vehicle in front. A safe
following distance provides an ample amount of time
for the driver to react appropriately to sudden hazards.
It is very often that drivers have to gather information
about the surroundings with just one glance.
DeLucia [41] found that young drivers were able to
derive motion information in car following scenes fairly
well, even when visual information was partially absent.
However, older drivers extrapolated this information
much slower than their younger counterparts.
Lane keeping and lane changing are similar driving
tasks. Lane keeping is the act of keeping the vehicle
within the lane boundaries as the road and environ-
ment change, e.g., on a curved road. Lane changing is
also necessary to avoid certain situations (e.g., avoid
colliding into a car merging into the freeway) as well as
maintaining safe, efficient wayfinding. Recent study by
Macuga and colleagues [42] investigated whether peo-
ple could perform a proper lane change with little or no
visual information. When the needed heading direc-
tion was made implicit, participants were able to
change lanes. Further, drivers were able to navigate
properly even when all visual information was
removed, by using inertial cues in a portable virtual
reality vehicle. Again, research has demonstrated that
drivers are able to perform certain driving tasks, such as
lane changing, with little or no visual information
within a certain amount of time.
Driving performance can also be assessed in terms
of steering control error. Steering control is a basic task
in driving that is to control the heading direction of the
automobile. Steering control is also essential for many
driving tasks that are previously discussed, e.g., car
following, lane changing and keeping, and collision
avoidance. Perceiving correct heading direction is nec-
essary to maintain or adjust locomotion and may be
the lowest-level steering control task. Therefore, it has
been the primary driving performance measure for
a large body of driving-related literature.
424 Driving Under Reduced Visibility Conditions for Older Adults
In the current article, a series of driving studies are
reviewed that investigated age-related decrements in
driving performance, especially in low visibility condi-
tion (i.e., in low contrast driving scenes). Several essen-
tial driving tasks are examined, including car following,
collision detection, and steering control. In addition,
the age-related spatial integration mechanism is pro-
posed to account for age-related deficiency in driving
in optical flow field. In the end, future research direc-
tions are considered briefly on both the advances of
new technology and improving driving performance
through training.
Car following in Fog
In this part age-related differences are examined in
sensory and perceptual processing for a task important
for driving safety – car following. Effective car follow-
ing allows drivers to maintain safe distances that are
necessary to take proper executions in case of emer-
gency. Failure to correctly respond to lead vehicle (LV)
speed changes can have serious consequences for the
safety of the driver. For example, if a driver in
a following vehicle fails to respond to a reduction in
LV speed then the headway distance between the fol-
lowing and lead vehicle is reduced. This can result in
a following distance that is too close (i.e., does not
allow for sufficient response time should the LV sud-
denly decelerate), leading to an increased risk of
a crash.
Previous research on car following [43, 44] has
assumed that drivers have precise information regard-
ing headway distance and speed of the lead and follow-
ing vehicle. A limitation of this research is that drivers
do not have access to this precise information. Instead,
drivers can estimate distance and speed based on avail-
able sources of visual information. Recently, Andersen
and Sauer [45] proposed a new model for car follow-
ing, referred to as the DVA (driving by visual angle)
model, based on the visual angle and speed informa-
tion available to a driver. The DVA model consists of
two components – one component provides informa-
tion for distance perception (based on visual angle of
the LV) and the second component provides informa-
tion useful for speed perception (based on instanta-
neous changes in LV visual angle). The results of their
study indicate that the DVA model, as compared to
other car following models based on precise headway
distance and LV speed, could better predict driver per-
formance in both simulator and real-world driving
conditions. In a related study, Andersen and Sauer
[46] showed the use of the driving scene information
in car following. The results indicated greater accuracy
in driving performance when the surrounding scene
was visible as compared to conditions when it was not
visible. They argued that the surrounding scene is use-
ful for specifying edge rate information, which is used
to estimate the speed of the driver’s vehicle.
Recent study by Ni, Kang, and Andersen [47] exam-
ined two hypotheses concerning age-related changes in
car following performance. Previous studies have
shown age-related differences in the use of scene infor-
mation to judge distance and layout of a scene [48].
This finding suggests that older drivers, as compared to
younger drivers, will have decrements in perceived
distance to the LV. This hypothesis is referred to as the
aging and distance perception hypothesis. Previous
studies have also found age-related declines in judging
speed of the driver’s vehicle [25] and of the
approaching objects [32, 49]. These findings suggest
that older drivers, as compared to younger drivers, will
have decrements in perceiving self-speed and relative
speed change between the lead and following vehicle.
This hypothesis is referred to as the aging and speed
perception hypothesis.
In addition to examining age-related differences in
car following performance, Ni’s study [47] examined
a set of environmental conditions, the presence of fog,
that are likely to be problematic for older drivers. Fog
reduces the overall contrast and visibility of the driving
scene, with the magnitude of reduced visibility increas-
ing as a function of distance. As a result, the ability to
see detail of the driving scene is reduced as a function of
the distance between the driver and objects in the scene.
Epidemiological studies of older driver crash rates have
found increased crash risk for older drivers under
reduced visibility conditions due to weather or dusk/
nighttime conditions [1, 50–52]. A more recent study
[53] found reduced car following performance for col-
lege-age drives under simulated fog conditions.
As discussed earlier, it is well documented in the
literature that contrast sensitivity is reduced with
increased age [11–13]. Indeed, studies have found
that for photopic vision (daylight conditions) there is

a 41% reduction of contrast sensitivity for 70 year old
subjects as compared to 20 year old subjects in
detecting mid- to high level spatial frequency targets
[13]. These results suggest that older drivers are likely
to have poorer car following performance than younger
drivers because of reduced visibility of the LV under
dense fog conditions.
Previous research [53, 54] has found decreased car
following performance (failure to maintain following
distance) under simulated fog conditions. Two factors
may impact car following performance under foggy
conditions. First, the reduced visibility of the scene
may result in a compression of the overall perceived
depth of the driving scene. This situation is likely to
occur as the reduction in contrast of the surrounding
scene will remove information important for perceiv-
ing scene depth such as texture gradients and linear
perspective [55]. If the perceived depth is compressed
then smaller headway distance at high fog density con-
ditions is expected. Second, reduced visibility of the
scene may result in increased difficulty in estimating
speed. Previous research has found that edge rate infor-
mation (the rate at which local edges cross a fixed
reference point in the visual field) is important for
determining the perceived speed of vehicle motion
[56] and is used by drivers for tasks such as braking
[57, 58]. If the reduced visibility of the scene from fog
decreases the visibility of edge rate information in the
driving scene then greater error in tracking changes in
LV speed is expected.
The rest of this section then focuses on the study by
Ni and colleagues [47]. In this study, Eight college
students (age mean and standard deviation of 21.0
and 2.6, respectively) and eight older subjects (age
mean and standard deviation of 72.6 and 4.6, respec-
tively) participated in the study and all reported nor-
mal or corrected-to-normal vision and were currently
licensed drivers. All drivers had experience driving in
fog and reported driving at least 3 days per week.
The study was conducted in a fix-based driving
simulator, in which the roadway in an urban setting
consisted of three traffic lanes (representing a three lane
one way road) with the driver and LV located in the
center lane. Drivers were presented with a car following
scenario in which the LV varied its speed according to
a sum of three equal-energy sinusoids (i.e., the peak
accelerations and decelerations of each sine wave in the
Driving Under Reduced Visibility Conditions for Older Adults 425
signal were equivalent). At the beginning of each trial
run, drivers were given 5 s of driving at a constant speed
18 m behind the constant speed LV. Drivers were
instructed that the headway distance during this
phase of the trial was the desired headway distance.
Following 5 s a tone sounded to indicate to the driver
that the LV speed would vary, which was based on the
sum of three non-harmonic sine-wave frequencies.
This signal does not repeat and thus prevents the driver
from anticipating changes in LV speed.
To simulate realistic effects of fog, specific fog den-
sity values were selected to represent a range of condi-
tions from high visibility (0.0 fog condition) to low
visibility (0.16 fog condition). The low visibility con-
dition was selected based on informal observations
indicating that under this condition the visibility of
the LV was considerably reduced at the desired follow-
ing distance of 18 m. The LV was visible to all drivers
under the simulated fog conditions at the
predetermined driving distance.
Car following performance was assessed using
a variety of measures that examine overall performance
changes based on distance information and speed
information. Specifically, mean and variance of dis-
tance headway are based on distance perception
whereas RMS speed error is based on speed perception.
If the age and distance perception hypothesis is correct,
then age-related declines in measures based on distance
perception are expected. If the age and speed percep-
tion hypothesis is correct, then age-related declines in
measures based on speed perception are expected. In
addition, if older drivers have greater difficulty in
detecting and responding to speed changes then age-
related declines should be greater as the overall speed of
the LV is increased.
An important issue in car following is how to quan-
tify a safe following distance. Time headway (THW) is
a measure that is derived by the ratio of headway
distance to the velocity of the driver vehicle. This mea-
sure indicates the time between two vehicles passing the
same point traveling in the same direction and thus has
been used as an indication of a safe margin between the
driver and lead vehicle. In Ni’s study they required
drivers to maintain a fixed distance of 18 m across the
three speed conditions. Thus, it is natural that THW
would vary as a function of speed. The use of THW
allows one to determine changes in safe driving
426 Driving Under Reduced Visibility Conditions for Older Adults
performance as a function of fog density. For example,
consider car following at a specific constant speed. If
drivers have difficulty in maintaining a safe margin as
a function of fog then increased fog density should
result in a decrease in THW. If older drivers have
greater difficulty than younger drivers under foggy
conditions in determining distance and speed then
they may follow at a closer distance resulting in
a decrease in THW – an indication of increased
crash risk.
The results showed that the mean distance headway
(distance between driver and lead vehicle in meters) for
the 0.0, 0.04, 0.08, 0.12, and 0.16 simulated fog density
levels were 19.3, 20.1, 19.3, 18.2, and 17.9 m, respec-
tively. Mean distance headway for the 40, 60, and 80
km/h speeds were 17.7, 19.2, and 20.0 m, respectively.
The result also showed that older drivers maintained
a slightly greater headway distance for the no fog
(0.0 fog density) condition. However, at higher fog
density levels and at increased speed older drivers
maintained a closer headway distance than younger
drivers. The largest age-related difference in distance
headway occurred at the intermediate speed (60 km/h)
and highest fog density condition. However, a notable
exception was the highest fog density condition and
highest speed. For this condition older drivers had
a greater following distance than younger drivers.
This result is likely due to a change in strategy
employed by older drivers. Specifically, older drivers
increased following distance at high speeds to minimize
the likelihood of a collision because of difficulty in
perceiving changes in the visual angle of the LV due
to reduced contrast. In addition, the result indicated
that older drivers had greater headway variance (mean
variance of 25.6) compared to younger drivers (mean
variance of 10.1). Analysis of RMS speed error showed
greater RMS error for older drivers (mean RMS error of
6.17 km/h) as compared to younger drivers (mean
RMS error of 4.83 km/h).
To examine the effects of fog on the safety margin
for car following time headway (THW) data was
derived. THW varied as a function of the average
speed of the LV with THW decreasing at increased
average speed of the LV. The result showed decreased
THW as a function of fog density. These results showed
that at the intermediate speed (60 km/h) THW
decreased for both older and younger drivers with an
increase in fog density. However, older drivers had the
largest decrease in THW at the highest fog condition.
This result indicates that older drivers, as compared to
younger drivers, had a significant reduction in the
safety margin at the highest fog density level. At the
highest speed (80 km/h) both older and younger
drivers had a decrease in THW with an increase in fog
density with the exception of the highest fog density
condition. For younger drivers THW continued to
decrease at the highest fog density condition. However,
for older drivers THW increased at the highest fog
density condition. This result suggests that older
drivers increased the following distance to increase
the safety margin.
As discussed earlier, two hypotheses were examined
in this study concerning age-related differences in car
following performance under foggy conditions.
According to the aging and distance perception
hypothesis, older drivers, as compared to younger
drivers, will have decrements in perceived distance to
the LV. According to the aging and speed perception
hypothesis, older drivers, as compared to younger
drivers, will have decrements in perceived speed and
relative speed change between the lead and following
vehicle. The results of this study provided evidence in
support of both of these hypotheses. With regard to
distance information, they found that older drivers, as
compared to younger drivers, followed at a closer head-
way distance with an increase in fog density. In addi-
tion, although mean headway distance varied for both
younger and older drivers as a function of speed, older
drivers showed a greater change as a result of speed.
This effect was especially pronounced for the highest
fog density conditions examined. They also found
greater variance in distance headway for older drivers,
as compared to younger drivers. These results provide
evidence in support of the age and distance perception
hypothesis.
With regard to speed perception the results indi-
cated that older drivers had greater RMS speed error
than younger drivers. In addition, older drivers had
lower squared coherence scores than younger drivers,
indicating that older drivers had poorer performance in
tracking local variations in speed at specific frequencies
of LV speed. These results provide evidence in support
of the age and speed perception hypothesis. The results,
considered together, suggest that older drivers have

decreased car following performance as a result of
difficulty in judging both speed and distance.
An important finding in this study was the interac-
tion of age, speed, and fog density for the distance
headway measure. Of particular interest were the age-
related performance differences for the 60 km/h speed
condition under the highest fog density condition.
Under this condition, older drivers followed at a very
close distance and shorter THW. The authors believed
that older drivers followed at a close distance because
reduced visibility due to fog increased the difficulty to
perceive LV speed changes. Collision risk can be
assessed by deriving the THW. The shorter the THW,
the less is the time available to the driver to avoid
a collision. For the 60 km/h/0.16 fog density condition
younger drivers had an average time gap of 1.1 s.
However, older drivers had a time gap of 0.87 s,
which represents a 21% reduction in available response
time to avoid a collision. Given the well-documented
finding of slower reaction time with age, this result
suggests that older drivers may be at considerable risk
of a collision under high fog density conditions at
moderate speeds.
Collision Detection in Fog
In addition to car following, there is another important
perceptual task for driving safety – collision detection.
Failure to correctly detect an impending collision event
can greatly increase accident risk for drivers. For exam-
ple, if a driver in a vehicle fails to see an upcoming
vehicle early enough or does not see it at all when
making a left turn, then a crash might occur resulting
in a severe injury or even death for the driver. Accurate
detection of an impending collision is the basis of
making appropriate responses, such as steering or
braking to avoid a collision. DeLucia et al. [59] studied
the age-related differences in detecting collision events
when an object was moving toward the observer at
a constant speed. They found an age-related decrement
in sensitivity for older females but not older males.
Andersen and Enriquez [26] found that older
observers, as compared to younger observers, were
less sensitive to collision events and required more
time to process the visual information.
As reviewed earlier, previous studies of older driver
have shown increased crash rates under reduced
Driving Under Reduced Visibility Conditions for Older Adults 427
visibility conditions due to weather or dusk/nighttime
conditions. Specifically, the presence of fog reduces the
overall contrast and visibility of the driving scene,
which results in reduced visible details as a function
of increasing distance. Such reduced visibility of driv-
ing scene may lead to higher injury and death rates in
accidents [60]. Driving simulation studies have found
evidence of decreased driving performance under sim-
ulated fog conditions. Recent study [47] reviewed
above on car following task has shown reduced perfor-
mance under simulated fog conditions for older drivers
compared with younger drivers. As discussed earlier,
contrast sensitivity is reduced with increased age. These
findings suggest that older observers may have more
difficulty than younger observers in recovering visual
information needed to correctly detect impending col-
lision events under dense fog conditions.
In this section, I will review a most recent research
[61] which investigated detecting collision events when
an object was approaching the observer at a constant
speed on a linear trajectory under foggy conditions.
Two factors might affect collision detection perfor-
mance in fog. First, the reduced visibility of the scene
may result in a reduced visibility of bearing angle of the
approaching object, which is crucial to judge the colli-
sion event on linear trajectory [62]. This situation is
likely to occur as the reduction in contrast of the
surrounding scene will remove information important
for perceiving object distance such as the horizon,
texture gradients, and linear perspective. If the infor-
mation on object position in the visual field is
compromised, then reduced collision detection perfor-
mance at dense fog condition is expected. Second,
reduced visibility of the scene may result in increased
difficulty in estimating speed. Previous research has
found that edge rate information (the rate at which
local edges cross a fixed reference point in the visual
field) is important for determining the perceived speed
of vehicle motion [56] and is used by drivers for tasks
such as braking [25, 58]. When moving in the driving
scene, drivers need to predict not only the motion
path of the approaching object but also the vehicle
position after a certain amount of time, which is
dependent of estimated speed. Thus, it is expected
that the decreased visibility of edges in the driving
scene under dense fog condition will decrease collision
detection performance.
428 Driving Under Reduced Visibility Conditions for Older Adults
In this study, they investigated the effects of fog on
collision detection performance when optical variables
(i.e., visual angle and change in visual angle of the
object, and the relative speed between the object and
the vehicle) were constant. Drivers were presented
with a driving simulation scene of a straight roadway
in a suburban setting. A spherical object traveled on
a linear trajectory toward the vehicle for 9 s before it
either collided with or passed the vehicle. The vehicle
was either stationary in Experiment one or moved
straight ahead at a constant speed in Experiment two
while the relative speed between the object and the
vehicle remained constant. Only a part of the travel
path was presented and drivers were instructed to
make a judgment whether the object would collide
with the vehicle or not by pressing one of the two
keys on the keyboard.
In the first experiment, three independent variables
were examined: Age (younger and older drivers), sim-
ulated fog density (no fog, mild fog, medium fog, or
dense fog, corresponding to 0, 0.08, 0.16, or 0.24 in
density, respectively), and the time-to-contact (TTC)
on the last frame of the display (2, 4, or 6 s).
In the second experiment, the stimuli were the same
as in first experiment except that the vehicle moved at
one of the three speeds (20, 50, and 80 km/h). Since the
relative speed between object and observer was kept
constant, which is 90 km/h, the object moved at
a speed of 70, 40, or 10 km/h, respectively, depending
on the vehicle moving speed. Four independent vari-
ables were examined: Age (younger and older
drivers), simulated fog density (no fog or dense fog,
corresponding to 0 or 0.24 in fog density, respectively),
moving speed of the vehicle (20, 50, or 80 km/h), and
TTC (2, 4, or 6 s).
The displays simulated a 3D scene consisting of
a roadway, roadway strip, textured ground, and an
approaching bright red spherical object (See Fig. 1).
The object motion trajectories were generated using the
same method employed by Ni and Andersen [62]. The
initial position of the object was chosen randomly from
an arc (+/
20 from the center of the display) at
a fixed distance of 223 m from the viewpoint. The
end position of the object was 50, 100, or 150 m from
the view point for TTC of 2, 4, and 6 s, respectively. The
3D velocity for collision and non-collision events was
constant. The simulated foggy effect was achieved using
Driving Under Reduced Visibility Conditions for Older
Adults. Figure 1
Example of the display in dense fog condition. A bright red
spherical object moves along a linear trajectory toward the
observer, which might or might not result in colliding with
the observer
the same method described above. The low visibility
condition was selected based on informal observations
indicating that under this condition the visibility of the
object was considerably reduced while still visible at the
original position. All observers reported during
debriefing that they were able to see the sphere in all
conditions.
The two experiments were conducted in a darkened
room. Observers viewed the monitor binocularly
through a collimating lens with their head in a chin
rest. They were informed that they would be shown
a series of displays consisting of a 3D scene with a single
object in a distance that was moving toward the
observer. Their task was to determine whether the
moving object would hit the observer. Observers were
next shown a series of demonstration displays of colli-
sion and non-collision stimuli in which the complete
motion trajectory was displayed. After observers
understood the task, they were presented with eight
practice trials with complete 9-second events, half of
which simulated a collision, and were asked to indicate
whether or not the object was on a collision trajectory
by pressing one of the two designated keys on
a standard keyboard. Subjects pressed the number “6”

key for non-collision responses and the number “4” key
for collision responses. No feedback was given during
the practice trials and experimental trials. Subjects were
required to correctly identify seven out of the eight
practice trials before proceeding to the experimental
trials. Following the practice trials subjects were
instructed that the experiment would include brief pre-
sentations of the collision and non-collision events. If
they were uncertain of whether the display was
a collision or non-collision event, they should make
the best judgment possible.
The average proportion of hits (collision response
for trials that simulated a collision) and false alarms
(collision response for trials that did not simulate
a collision) was calculated for each subject in each
condition and used to derive sensitivity (d’) and
response bias (b) measures[63]. In conditions where
observers performed perfectly (100% hit rate or 0%
false alarm rate), d’ and b were calculated based on
the assumption that one error was made.
The analysis of d’ values indicates that younger
observers (mean d0 = 3.43) had greater sensitivity
than older observers (mean d0 = 2.78). The increase of
fog density resulted in decreased sensitivity for older
observers but not for younger observers. When the
TTC was 2 s, older observers performed as well as
younger observers. As the TTC increased, the sensitiv-
ity for both age groups decreased. However, older
observers’ sensitivity decreased much faster than that
of younger observers.
The analysis of b values suggests that both age
groups were equally likely to make a collision judgment
and the fog density did not bias the observers’ likeli-
hood to report a collision event. It was also found that
observers were more likely to make a collision judg-
ment with increased TTC.
The results of the second experiment are consistent
with those in the first experiment, indicating that
younger observers (mean d’ = 2.65) had greater sensi-
tivity than older observers (mean d’ = 2.13). It was
shown that with increased TTC the sensitivity for
both age groups decreased. However, older observers’
sensitivity decreased faster than that of younger
observers. The analysis of b values suggests consistent
results as in the first experiment that when fog was
simulated or when the TTC increased, observers were
more likely to respond that a collision would occur.
Driving Under Reduced Visibility Conditions for Older Adults 429
According to the contrast and bearing angle
hypothesis, older drivers, as compared to younger
drivers, will have more difficulty in perceiving constant
bearing information under low contrast condition
caused by fog. The results in Experiment 1 provide
evidence supporting this hypothesis. It was found that
older drivers, as compared to younger drivers, had
significant decrements in sensitivity to detect an immi-
nent collision event in dense foggy condition. In addi-
tion, although a decreased performance was found for
both younger and older drivers with decreased display
duration, older drivers showed greater decrements as
a result of shorter display duration. This is a finding
consistent with previous research (Andersen &
Enriquez, 26). This result is probably due to the
increased difficulty in discriminating bearing angle in
driving scene with reduced contrast in distance.
Since reduced contrast in driving scene in dense fog
condition caused decreased sensitivity of detecting col-
lision events, one may ask whether this decrease can be
predicted by contrast sensitivity. To address this con-
cern, a linear regression between contrast sensitivity
and the sensitivity of collision detection in fog was
conducted for both age groups. The results indicated
that contrast sensitivity was not a significant predictor
for sensitivity of collision detection for either you-
nger observers (r2 = 0.09, p = 0.37) or older observers
(r2 = 0.06, p = 0.46). There are two explanations which
might account for this result. First, the contrast sensi-
tivity for each observer was measured using Pelli
Robson test [64], which may not have sufficient preci-
sion to assess individual differences. For example,
10 out of 11 younger observers in this study had the
same contrast sensitivity whereas their d’ ranged from
2.38 to 3.24. An important issue for future research
would be to precisely derive contrast sensitivity thresh-
olds. Second, the Pelli Robson test only measured con-
trast sensitivity in static scenes whereas our stimuli
were in motion. Observers might have different con-
trast sensitivity when viewing motion stimuli as com-
pared to when viewing static stimuli. A similar result
has been found for studies examining sensitivity to
binocular disparity [65]. Specifically this research has
found that observers who were stereo-blind using static
stimuli had sensitivity to disparity when the stimuli
were moving. If similar conditions exist for low con-
trast stimuli, then the assessment of contrast sensitivity
430 Driving Under Reduced Visibility Conditions for Older Adults
with static stimuli would not be predictive of sensitivity
to low contrast moving stimuli.
With regard to edge rate information, it was found
that both younger and older drivers had considerable
decrease in their sensitivity to collision events when the
vehicle is moving. The results in Experiment 2 were
consistent with those in Experiment 1 in that decreased
performance was found with decreased display dura-
tion (i.e., greater TTC). Considered together, these
results suggest that under inclement weather condi-
tions, such as fog, older drivers may be subject to an
increased crash risk due to a decreased ability to detect
impending collision events. One important finding in
this study was the interaction of age and display dura-
tion, specifically the age-related difference for the
3 s condition. Under this condition, older drivers, as
compared to younger drivers, had significant decre-
ments in detecting the collision event which suggests
a serious collision risk for older drivers. The longer it
takes a driver to detect imminent collision events, the
less time the driver has to take appropriate response,
e.g., to decelerate or steer the vehicle. Given the well-
documented finding of slower reaction time and
reduced divided attention capacity with age this result
suggests that older drivers may be at considerable risk
of a collision under high fog density conditions.
Steering Control Under Reduced Contrast
Conditions
Besides accelerating and decelerating the vehicle, the
other control task that a driver performs during driving
is steering control, which is to steer the vehicle to move
in desired direction. To successfully steer the vehicle,
one needs to first perceive the current status and mov-
ing direction from the visual information in the driving
scene. Failure to accurately detect changes in the path
of motion and make corrections to the vehicle’s path of
motion could have serious consequences for driver
safety. One source of information that has been exten-
sively studied for the perception and control of steering
is optical flow – the perspective transformation of the
optic array [66, 67]. Research on optical flow has dem-
onstrated the usefulness of this information for the
perception of heading [68, 69], the perception of self-
motion [70], and the perception of egospeed [56].
A critical assumption of optical flow research is that
the spatial and temporal characteristics of the stimuli
result in a perception of apparent motion.
Previous research has demonstrated age-related
decrements in the perception of motion and in the
use of optical flow information for the perception of
3D shape [22]. In this part, I will introduce a most
recent study on age-related differences in the use of
optical flow for steering control under low contrast
conditions. In the study by Nguyen et al. [71], drivers
were presented with computer generated displays sim-
ulating forward vehicle motion through a 3D scene of
random dots on a ground plane. The horizontal posi-
tion of the vehicle was perturbed according to a sum of
sinusoidal functions, and drivers were asked to steer the
vehicle to maintain the initial straight path of motion.
Previous research has shown that the use of optical
flow information for perceiving the path of observer
motion requires the spatial integration of velocity
information. Recently, it was found that older
observers had a decreased ability to spatially integrate
velocity information to perceive motion defined edge
boundaries ([72], will be reviewed in details below).
Nguyen et al. [71] proposed that older drivers, as com-
pared to younger drivers, would show a decreased sen-
sitivity to spatially integrate velocity information for
steering control from optical flow, especially in low
visibility conditions. This was referred to as the spatial
integration hypothesis. To test this hypothesis they
manipulated the dot density display (resulting in
a greater spatial separation of the velocity information)
and contrast of the dots display. If the spatial integra-
tion hypothesis is correct, then age-related differences
in steering performance should be more pronounced at
low as compared to high dot densities and at low-
contrast conditions as compared to high-contrast
conditions.
In their study, Nguyen et al. manipulated the num-
ber of dots in the optic flow field (25 or 325), and
contrast level (low, medium, and high level,
representing 0.15, 0.2, and 0.25 in Michelson contrast,
respectively). The displays simulated driving through
a 3D array of dots located on a ground plane, as illus-
trated in Fig. 2. The simulated speed was 72 km/h. The
horizontal position of the dots pattern was perturbed
according to a sum of three prime sine-wave functions.
The frequencies, amplitudes, and phases of the
sinusoidal functions were chosen in a similar way as

Driving Under Reduced Visibility Conditions for Older
Adults. Figure 2
Low contrast optic flow showing straight ahead moving
direction. In the actual experiment, the contrast was
reversed so that white dots were displayed on a black
background
in Ni et al [47] so that the output of the sum of three
sine-wave functions was always zero at the beginning of
each trial and unpredictable. The scene consisted either
25 or 325 randomly positioned white dots on a black
background to form the ground plane. The back-
ground had an average luminance of 16 cd/m2. The
dots were randomly positioned at the beginning of each
trial, and then moved to the observer. Any dot that
moved out of view port would be randomly
repositioned at the far end of the ground plane.
Drivers were instructed that their task was to main-
tain a fixed heading direction relative to the moving
dots as if driving down a straight roadway with lateral
wind gusts perturbing their positions on the roadway.
Steering control performance was assessed by calculat-
ing RMS (root means square) steering error for each
driver on each trial in each condition. The average
results for RMS error in their study revealed that
although both older and younger drivers had less
steering error with a greater number of dots in the
flow field, younger drivers had significantly less
steering error (9.9) than older drivers (12.4). Under
low dot density condition, in which less optical flow
information was presented, both younger and older
drivers showed decreased steering performance with
decreased contrast of the scene. However, older drivers’
Driving Under Reduced Visibility Conditions for Older Adults 431
performance deceased much faster with decreased con-
trast, as compared with younger drivers. This result
suggests that older drivers are more prone to accident
risk than younger drivers under low contrast condition,
especially with less optical flow information.
Previous study by Ni and Andersen [73] found
older drivers, as compared to younger drivers, relied
more on optical flow than landmark information for
their steering control task. One possible explanation for
such result is that in their displays, only one color coded
dot was present at a time which provided landmark
information. In order to use landmarks for steering
control, a driver must encode the spatial location of
landmark positions and use this positional information
as reference information to maintain the initial path of
motion. Older drivers may have difficulty encoding and
using landmark position information, which results in
reduced accuracy in steering control, especially when
small amount of landmark information is present. If
this hypothesis is true, with increased landmark infor-
mation there should be an increase in the steering per-
formance for older drivers. However, older drivers might
show less improved driving performance, which results
from the extra landmark information added to optical
flow field, as compared to younger drivers.
In a related study by Ni et al. [74], they used similar
experimental setting as in the study by Nguyen et al.
[71], except that different numbers of landmark dots
were presented in the optic flow field (0, 1, 4, 7, and 10).
Their result showed that younger drivers had signifi-
cantly less steering error (5.8) than older drivers (9.9).
Both older and younger drivers had less steering error
with a greater number of landmark dots in the flow
field. However, younger drivers, as compared to older
drivers, were more efficient at using landmark infor-
mation to improve steering control.
Overall, these findings indicate that older drivers
may be more reliant on optical flow information for
controlling a vehicle and may have a reduced ability to
use alternative sources of information for steering con-
trol. The decreased reliance of landmark information
for older drivers, when optical flow information is
reduced, may result from age-related deficits in atten-
tion. Previous research [75] found that older subjects
had greater difficulty than younger subjects in ignoring
distracting items when scanning a display, presumably
a result of difficulty in disengaging attention to
432 Driving Under Reduced Visibility Conditions for Older Adults
irrelevant information. In Ni et al’s study [74], older
drivers may, under reduced optical flow conditions,
focus their attention on the landmarks and have diffi-
culty disengaging their attention from the landmarks to
scan other parts of the scene. However, when more
landmark information was presented on a dense optical
flow ground plane, less steering control error occurred
for older drivers since they can rely more on optical
flow information in this condition.
Spatial Integration in Processing Optical Flow
Information
Visual processing of optical flow information requires
the integration of information over space and time.
This section attempts to explain age-related difference
in using optical flow information in terms of spatial
integration. Now consider how the visual system per-
ceives a 2D form from fragmented image information.
In order to perceive the form, the visual system must
combine local fragments (e.g., line segments) across
space to recover the shape of the form. Information
for perception can also be available from different time
intervals. For example, consider the perception of
a moving object partly obscured by nearer objects. In
order to perceive the form, the visual system must
integrate the different parts of the object visible
at different time intervals. These examples highlight
the role of spatial and temporal integration in vision.
Previous studies of neuronal activity in the visual
system have shown evidence of spatial and temporal
integration. For example, retina cells respond to informa-
tion present over a limited region of the visual field [76].
The visual system combines or spatially integrates infor-
mation from local receptive field regions via intracortical
connections from other cells within a cortical region
[77–79] and from cells in higher cortical regions
[80–82]. Spatial integration has been demonstrated to
be important for visual processing including motion
perception [83] and form perception [84–86].
Neurophysiological studies have also examined the
role of temporal integration by the visual system. For
example, studies have examined the role of temporal
integration for motion perception in primary visual
cortex (V1) and extrastriate motion area (MT/V5)
[87]. It has been argued that temporal integration at
a neural level is due to the synchronous activity of
Frame 1 Frame 2
Driving Under Reduced Visibility Conditions for Older
Adults. Figure 3
Illustration of 2D form perception results from occlusion
when the form moves from frame 1 to frame 2. Note the
stars in low contrast represent the dots that are occluded
by 2D form above (a diamond in this illustration)
groups of neurons rather than the firing rate of
a single neuron [88]. Specifically, temporal integration
has been shown to be important in visual processing in
motion perception [89] and form perception [90–93].
Recently, Andersen and Ni [72] used a novel per-
ceptual task to examine whether age-related declines in
visual processing resulted from changes in spatial or
temporal integration. To assess this issue, they used
a visual task that requires both spatial and temporal
integration – the recovery of 2D shape from kinetic
occlusion. Kinetic occlusion can specify 2D shape by
providing information regarding edge boundaries of an
object based on the accretion and deletion of texture
over time. Consider an opaque object positioned in
front of a background (see Fig. 3). If the object and
background have identical random texture (randomly
positioned dots) then the object boundaries cannot be
seen. However, when the object starts moving (or the
background starts moving) the accretion and deletion
of the background texture due to occlusion can be used
to reveal the boundaries of the object. In order to
recover the object boundary, the visual system must
integrate the accretion and deletion of texture over
space (discrete samples of the edge boundary are
specified by the local disappearance/reappearance of
texture) and time (the accretion and deletion of texture
occur in local intervals of time).
Previous research [90] examined the importance of
kinetic occlusion for specifying the shape of 2D objects.
College-age subjects were presented with displays of
a moving opaque object with random dot texture

against a random dot texture background. The velocity
and texture of the display were varied and subjects were
asked to identify which of four shapes were present.
The results indicated that an increase in velocity and
density resulted in greater accuracy in shape identifica-
tion. The results also indicated that density was the
primary factor in determining performance. An impor-
tant issue examined in their study was whether perfor-
mance was based on the rate of occlusion events (the
number of occlusion events per unit time). To examine
this issue, they compared performance across condi-
tions in which velocity and density changed propor-
tionally resulting in a constant rate of occlusion events.
For example, an occlusion stimulus with a velocity of
0.8 deg/s and a density of 1.2 dots/deg2 was compared
to an occlusion stimulus with a velocity of 1.6 deg/s and
a density of 0.6 dots/deg2. These two conditions have
identical rates of occlusion events (i.e., disappearance/
reappearance of texture). The results indicated that
performance varied across conditions with a constant
rate of occlusion events, with performance increasing
due to an increase in density. These findings led Ander-
sen and Cortese to conclude that spatial integration
from texture density was the primary factor in shape
perception from kinetic occlusion.
In Andersen and Ni’s study [72], they examined
conditions similar to those examined by Andersen
and Cortese [91]. Subjects were shown kinetic occlu-
sion displays and were asked to identify the 2D shape of
a moving form. Information for spatial and temporal
integration was manipulated by changing velocity and
texture density. Information for spatial and temporal
integration was manipulated by changing the texture
density of the display. Specifically, an increase in den-
sity will result in an increase in spatial and temporal
information for the edge boundary. Information for
temporal integration was manipulated by varying the
speed of motion. Greater velocity of the object
increases the rate of accretion and deletion, resulting
in an increase in information for temporal integration.
If age-related decrements in visual processing are
due to changes in temporal integration, then it should
be found that object identification performance will be
significantly worse for older observers as compared to
younger observers at slower speeds.
To isolate the role of spatial integration, they
examined variations in performance when the rate of
Driving Under Reduced Visibility Conditions for Older Adults 433
occlusion events remained constant. Specifically, if
the velocity of the moving object is decreased and the
texture density is increased proportionately, then the
spatial separation of accretion and deletion of texture
that defines the object boundary is decreased. If age-
related decrements in visual processing are due to
changes in spatial integration, then object identifica-
tion performance will be significantly worse for older
observers as compared to younger observers at lower
texture densities under constant rates of occlusion.
Two experiments were conducted by Ander and Ni
[72] to test their hypothesis. In the first experiment,
they examined age-related decrements in the percep-
tion of 2D shape from kinetic occlusion. The displays
consisted of random dots projected onto either a 2D
shape (circle, square, diamond, or 5 point star) or the
background. The primary variables of interest were dot
density, velocity of object, and the absence/presence of
background texture (i.e., providing kinetic occlusion
information when background texture was present). If
older observers have a decreased capacity for spatially
integrating kinetic occlusion information, then poorer
performance is expected for older observers with
a decrease in dot density when background texture is
present, particularly when the rate of occlusion is
constant.
Older and younger observers showed similar levels
of performance for the kinetic occlusion absent condi-
tion. When kinetic occlusion was added, both groups
showed greater shape identification performance, but
a greater increase in performance was found for the
younger as compared to older observers. This result
suggests that younger observers, as compared to older
observers, had a greater capacity for integrating
spatial and temporal information for identifying the
boundaries of the object.
Based on these results, they conducted additional
analyses of the data specifically for the condition when
kinetic occlusion was present. No significant interac-
tions were obtained between age and velocity, age and
size, or any other higher order interactions with age. It
was shown that age-related decrements in shape iden-
tification were approximately constant across varia-
tions in velocity. The lack of a significant interaction
of age and velocity indicates that the age-related decre-
ments are not due to a result of an age-related decline in
temporal integration.
434 Driving Under Reduced Visibility Conditions for Older Adults
Andersen and Ni further examined the conditions
in which two pairs of velocity/density combinations
have identical spatial/temporal information. They
found significant age effects between those conditions,
such as between the combination of 1.2 deg/s velocity
with 1.2 dots/deg2 density and the combination of
0.6 deg/s velocity with 2.44 dots/deg2 density. The
interaction of age and performance for these two con-
ditions was significant, indicating significant age effects
when spatial/temporal information is constant. The
greater age-related differences at lower density condi-
tions indicate age-related decrements in spatial inte-
gration. Younger observers had a 32% increase in
detection performance when kinetic occlusion infor-
mation was present, as compared to an 11.5% increase
for older observers.
To provide converging evidence in supporting spa-
tial integration hypothesis, they conducted a second
experiment. In this experiment, they managed to iso-
late the effects of spatial integration by manipulating
limited lifetime of the stimuli, indicating the duration
of individual dots was restricted. When limited lifetime
displays are used, the object and background dots in
the display would appear for a limited period of time
before being randomly repositioned in the display. As
a result, the salience of the edge boundary is consider-
ably reduced because the disappearance/appearance of
the dots can be either caused by occlusion or simply by
the limited lifetime. In order to recover the boundary,
the visual system must determine regions of local
motion of limited duration and regions of no motion
of limited duration. As a result, variations in the life-
time of the dots provide an opportunity to more
directly assess the integration of information over time.
If the age-related decrements in using kinetic occlu-
sion information were due to a loss in temporal inte-
gration, then decreasing the lifetime of local dot
motion should greatly limit the temporal integration,
resulting in a subsequent decrease in performance for
older observers. Thus in their second experiment, the
presentations of individual dots were shown for dura-
tions of 16, 33, 66, or 100 ms based on presentations of
2, 3, 4, or 6 frames. The proportion of dots repositioned
from frame to frame varied as a function of lifetime.
For example, for the two frame lifetime stimuli, 50% of
the dots were repositioned on each frame; for the three
frame lifetime stimuli, 33% were repositioned, etc.
The result showed a significant main effect of dura-
tion, indicating that an increase in lifetime result in
increased detection performance. Surprisingly, planned
comparisons of the effect of age at each duration con-
dition showed no significant effects. These results
suggest that age-related changes in visual processing
are the result of changes in spatial integration.
An important question is that whether other
aspects of visual processing, which are known to
change with age (e.g., motion perception, pattern rec-
ognition, and face recognition), may be due to
decreased spatial integration. Age-related changes
have also been found in visual attention tasks that
involve spatial information [94, 95]. Spatial integration
may not only explain age-related effects in visual per-
ception, but also be an important factor in age-related
changes in performing driving-related visual tasks.
Future Directions
The results of the preset study suggest two practical
applications in improving driving safety for older
drivers. The first application concerns driving assisting
system. Major automobile manufacturers have devel-
oped various intelligent safety systems, e.g., “Crash
Mitigation Braking System” from Honda, Volvo’s
“City Safety,” Toyota’s “Pre-crash System,” and
Mercedes Benz’s “Distronic Plus” system. However,
there are two major concerns regarding the current
assisting system prototypes. First, any system utilizing
optical information, e.g., laser radar, is prone to
weather change such as rain and fog. Second, most of
these systems were tested on younger drivers which do
not accommodate the specific characteristics of older
driver population. A successful driving assisting system
should take age-related differences in driving perfor-
mance into account, as well as the environmental
changes in inclement weather conditions.
A second application concerns the potential
improvement in driving performance through training
for older drivers. Andersen et al. [96] found that fol-
lowing training with subthreshold stimuli, older
observers were able to improve performance in
a texture discrimination task, and that this improve-
ment was retained for up to 3 months. Indeed, the
training, which occurred over a 2 day period, made
the older observers to perform equally well as

compared to the pre-training performance of college
age. In another study, Richards et al. [97] found that
older observers could improve their performance in
divided attention through repeated training with the
UFOV (useful field of view) task. An interesting topic
for future research would be to investigate whether older
observers can improve driving performance under low
visibility conditions using different training techniques.
Bibliography
1. Langford J, Koppel S (2006) Epidemiology of older driver
crashes – identifying older driver risk factors and exposure
patterns. Transp Res F Traffic Psychol Behav 9:309–321
2. Evans L (2004) Traffic safety. Science Serving Society, Bloom-
field Hills, p 179
3. Owsley C, Ball K, Sloane ME, Roenker DL, Bruni JR (1991) Visual/
cognitive correlates of vehicle accidents in older drivers.
Psychol Aging 6:403–415
4. National Highway Traffic Safety Administration (2008) Traffic
safety fact – 2007 data (NHTSA’s national center for statistics
and analysis. Publication No.DOT HS 810 992). Washington, DC
5. Owsley C, McGwin G Jr, Phillips JM, McNeal SF, Stalvey BT
(2004) Impact of an educational program on the safety of
high-risk, visually impaired, older drivers. Am J Prev Med
26:222–229
6. Wood JM, Owens DA (2005) Standard measures of visual
acuity do not predict drivers’ recognition performance under
day or night conditions. Optom Vis Sci 82:698–705
7. Ball K, Owsley C, Beard B (1990) Clinical visual perimetry
underestimates peripheral field problems in older adults.
Clin Vis Sci 5:113–125
8. McGwin G, Chapman V, Owsley C (2000) Visual risk factors for
driving difficulty among older driver. Accid Anal Prev
32(6):735–744
9. Glass JM (2007) Visual function and cognitive aging: differen-
tial role of contrast sensitivity in verbal versus spatial tasks.
Psychol Aging 22:233–238
10. Schachar RA (2006) Effect of change in central lens thickness
and lens shape on age-related decline in accommodation.
J Cataract Refract Surg 32:1897–1898
11. Richards OW (1977) Effects of luminance and contrast on
visual acuity, ages 16 to 90 years. Am J Optom Physiol Opt
54:178–184
12. Derefeldt G, Lennerstrand G, Lundh B (1979) Age variations in
normal human contrast sensitivity. Acta Ophthalmol 57:679–690
13. Owsley C, Sekuer R, Siemsen D (1983) Contrast sensitivity
throughout adulthood. Vis Res 23:689–699
14. McFarland RA, Domey RG, Warren AB, Ward DC (1960) Dark
adaptation as a function of age: I. A statistical analysis.
J Gerontol 15:149–154
15. Domey RG, McFarland RA, Chadwick E (1960) Dark adaptation
as a function of age and time. II. A derivation. J Gerontol
15:267–279
Driving Under Reduced Visibility Conditions for Older Adults 435
16. Chapanis A (1950) Relationships between age, visual acuity
and color vision. Hum Biol 22:1–33
17. Kahn HA, Leibowitz HM, Ganley JP, Kini MM, Colton T,
Nickerson RS et al (1977) The Framingham Eye Study I. Outline
and major prevalence findings. Am J Epidemiol 106:17–32
18. Sekuler R, Hutman LP, Owsley CJ (1980) Human aging and
spatial vision. Science 209:1255–1256
19. Long GM, Crambert RF (1990) The nature and basis of
age-related changes in dynamic visual acuity. Psychol Aging
5:138–143
20. Trick GL, Silverman SE (1991) Visual sensitivity to motion:
age-related changes and deficits in senile dementia of the
Alzheimer type. Neurology 41:1437–1440
21. Gilmore GC, Wenk HE, Naylor LA, Stuve TA (1992) Motion
perception and aging. Psychol Aging 7:654–660
22. Andersen GJ, Atchley P (1995) Age-related differences in the
detection of three dimensional surfaces from optic flow.
Psychol Aging 10:650–658
23. Betts LR, Taylor CP, Sekuler AB, Bennett PJ (2005) Aging
reduces center-surround antagonism in visual motion
processing. Neuron 45:361–366
24. Bennett PJ, Sekuler R, Sekuler AB (2007) The effects of aging
on motion detection and direction identification. Vis Res
47:799–809
25. Andersen GJ, Cisneros J, Atchley P, Saidpour A (1999) Speed,
size, and edge-rate information for the detection of collision
events. J Exp Psychol Hum Percept Perform 25:256–269
26. Andersen GJ, Enriquez A (2006) Aging and the detection of
observer and moving object collisions. Psychol Aging 21:
74–85
27. Norman FJ, Clayton AM, Shular CF, Thompson SR (2004) Aging
and the perception of depth and shape from motion parallax.
Psychol Aging 19:506–514
28. Norman FJ, Crabtree CE, Hermann D (2006) Aging and the
perception of 3-D shape from kinetic patterns of binocular
disparity. Percept Psychophys 68:94–101
29. Folk CL, Hoyer WJ (1992) Aging and shifts of visual spatial
attention. Psychol Aging 7:453–465
30. Kramer AF, Hahn S, Irwin DE, Theeuwes J (1999) Attentional
capture and aging: Implications for visual search performance
and oculomotor control. Psychol Aging 14:135–154
31. Hartley AA, Little DM (1999) Age-related differences and
similarities in dual-task interference. J Exp Psychol
128:416–449
32. Scialfa CT, Thomas DM, Joffe KM (1994) Age differences in the
useful field of view: An eye movement analysis. Optom Vis Sci
71:736–742
33. Sekuler AB, Bennett PJ, Mamelak M (2000) Effects of aging on
the useful field of view. Exp Aging Res 26:103–120
34. Owsley C, Sloane ME (1990) Vision and aging. In: Nebes RD,
Corkin S (eds) Handbook of neuropsychology, vol 4. Elsevier
Science, New York, pp 229–249
35. Hoffman L, McDowd JM, Atchely P, Dubinsky R (2005) The role
of visual attention in predicting driving impairment in older
adults. Psychol Aging 20:610–622
436 Driving Under Reduced Visibility Conditions for Older Adults
36. Salthouse TA, Somberg BL (1982) Time-accuracy relationships
in young and old adults. J Gerontol 37:49–53
37. Salthouse TA, Somberg BL (1982) Isolating the age deficit in
speeded performance. J Gerontol 37:59–63
38. Michon JA (1989) Explanatory pitfalls and rule-based driver
models. Accid Anal Prev 21(4):341–353
39. Cutting JE (1996) Wayfinding from multiple sources of local
information in retinal flow. J Exp Psychol Hum Percept Perform
22(5):1299–1313
40. Wilkniss SM, Jones MG, Korol DL, Gold PE, Manning CA
(1997) Age-related differences in an ecologically based study
of route learning. Psychol Aging 12(2):372–375
41. DeLucia PR, Mather RD (2006) Motion extrapolation of car-
following scenes in younger and older drivers. J Hum Factors
Ergon Soc 48(4):666–674
42. Macuga KL, Beall AC, Kelly JW, Smith RS, Loomis JM
(2007) Changing lanes: inertial cues and explicit path informa-
tion facilitate steering performance when visual feedback is
removed. Exp Brain Res 178:141–150
43. Chandler RE, Herman R, Montroll EW (1958) Traffic dynamics:
studies in car following. Oper Res 6:165–184
44. Brackstone M, McDonald M (1999) Car-following: a historical
review. Transp Res Rec F 2:181–196
45. Andersen GJ, Sauer CW (2007) Optical information for car
following: the driving by visual angle (DVA) model. Hum Fac-
tors 49:878–896
46. Andersen GJ, Sauer CW (2005) Visual information for car fol-
lowing by drivers: the role of scene information. Transp Res
Rec F 1899:104–109
47. Ni R, Kang J, Andersen GJ (2010) Age-related declines in car
following performance under simulated fog conditions. Accid
Anal Prev 42(3):818–826
48. Bian Z, Andersen GJ (2008) Aging and the perceptual organi-
zation of 3-D scenes. Psychol Aging 23:342–352
49. Scialfa CT, Guzy LT, Leibowitz HW, Garvey PM et al (1991) Age
differences in estimating vehicle velocity. Psychol Aging
6(1):60–66
50. Massie DL, Campbell KL, Williams AF (1995) Traffic accident
involvement rates by driver age and gender. Accid Anal Prev
27:73–87
51. Stutts JC, Martell C (1992) Older driver population and crash
involvement trends 1974–1988. Accid Anal Prev 24:317–327
52. McGwin G, Brown DB (1999) Characteristics of traffic crashes
among young, middle-aged and older drivers. Accid Anal Prev
31:181–198
53. Kang JJ, Ni R, Andersen GJ (2008) The effects of reduced
visibility from fog on car following performance. J Transp Res
Board 2069:9–15
54. Broughton KLM, Switzer F, Scott D (2007) Car following
decisions under three visibility conditions and two
speeds tested with a driving simulator. Accid Anal Prev
39:106–116
55. Andersen GJ, Braunstein ML (1998) The perception of
depth and slant from texture in 3D scenes. Perception
27:1087–1106
56. Larish JF, Flach JM (1990) Sources of information useful for
perception of speed of rectilinear motion. J Exp Psychol Hum
Percept Perform 16:295–302
57. Andersen GJ, Cisneros J, Saidpour A, Atchley P (2000) Age-
related differences in collision detection during deceleration.
Psychol Aging 15:241–252
58. Andersen GJ, Sauer CW (2004) Optical information for collision
detection during deceleration. In: Hecht H, Kaiser M (eds)
Theories of time to contact, advances in psychology series,
North Holland. Elsevier, Amsterdam, pp 93–108
59. DeLucia PR, Bleckley MK, Meyer LE, Bush JM (2003) Judgments
about collisions in younger and older drivers. Transp Res part
F Traffic Psychol Behav 6:63–80
60. Al-Ghamdi AS (2007) Experimental evaluation of fog warning
system. Accid Anal Prev 39:1065–1072
61. Bian Z, Ni R, Guindon A, Andersen GJ (2009) Aging and the
detection of collision events in fog. In: Proceedings of the fifth
international driving symposium on human factors in driver
assessment, training and vehicle design, pp 69–75
62. Ni R, Andersen GJ (2008) Detection of collision events on
curved trajectories. Percept Psychophys 70(7):1314–1324
63. Green DM, Swets JA (1966) Signal detection theory and psy-
chophysics. Wiley, New York
64. Pelli DG, Robson JG, Wilkins AJ (1988) Designing a new letter
chart for measuring contrast sensitivity. Clin Vis Neurosci
2:187–199
65. Braunstein ML, Andersen GJ, Rouse MW, Tittle JS (1986) Recov-
ering viewer-centered depth from disparity, occlusion, and
velocity gradients. Percept Psychophys 40:216–224
66. Gibson T (1979) The ecological approach to visual perception.
Houghton Mifflin, Boston
67. Gibson T (1966) The senses considered as perceptual systems.
Houghton Mifflin, Boston
68. Warren WH, Morris M, Kalish M (1988) Perception of transla-
tional heading from optical flow. J Exp Psychol Hum Percept
Perform 14:646–660
69. Warren WH, Mestre DR, Blackwell AW, Morris MW (1991) Per-
ception of circular heading from optical flow. J Exp Psychol
Hum Percept Perform 17:28–43
70. Andersen GJ, Braunstein ML (1985) Induced self-motion
in central vision. J Exp Psychol Hum Percept Perform
11:122–132
71. Nguyen B, Zhuo Y, Ni R (2011) Aging and the steering control
under reduced visibility conditions. In: Proceedings of the
sixth international driving symposium on human factors in
driver assessment, training and vehicle design (in press)
72. Andersen GJ, Ni R (2008) Aging and visual processing: declines
in spatial not temporal integration. Vis Res 48:109–118
73. Ni R, Andersen GJ, Rizzo M (2005) Age related decrements in
steering control: the effects of landmark and optical flow
information. In: Proceedings of the third international driving
symposium on human factors in driver assessment, training
and vehicle design, pp 144–150
74. Ni R, Bian Z, Andersen GJ (2009) Age-related differences in
using optical flow and landmark information for steering

control. In: Proceedings of the IEEE intelligent vehicles
symposium, 2009, pp 691–694
75. Ni R, Andersen GJ (2006) Age related performance in driving:
steering and collision detection, vision in vehicles 11. Dublin,
Ireland
76. Barlow HB (1953) Summation and inhibition in the frog’s
retina. J Physiol 119:69–88
77. Brown HA, Allison DD, Samonds JM, Bonds AB (2003) Nonlocal
origin of response suppression from stimulation outside the
classical receptive field in area 17 of the cat. Vis Neurosci
20:85–96
78. Das A, Gilbert CD (1999) Topography of contectural modula-
tions mediated by short-range interactions in primary visual
cortex. Nature 399:655–661
79. Stettler DD, Das A, Bennett J, Gilbert CD (2000) Lateral con-
nectivity and contextual interactions in macaque primary
visual cortex. Neuron 36:739–750
80. Bullier J, Hupe JM, James AC, Girard P (2001) The role of
feedback connections in shaping the responses of visual cor-
tical neurons. Prog Brain Res 134:193–204
81. Moore T, Armstrong KM (2003) Selective gating of visual sig-
nals by microstimulation of frontal cortex. Nature 421:370–373
82. Roberts MJ, Zinke W, Guio K, Robertson R, McDonald JS, Thiele
A (2005) Acetylcholine kinetically controls spatial integration
in Marmoset primary visual cortex. J Neurophysiol
93:2062–2072
83. Ledgeway T, Hess RF (2002) Rules for combining the outputs
of local motion detectors to define simple contours. Vis Res
42:653–659
84. Field DJ, Hayes A, Hess RF (1993) Contour integration by the
human visual system: evidence for a local ‘association field’.
Vis Res 33:173–193
Driving Under Reduced Visibility Conditions for Older Adults 437
85. Kovacs I (1996) Gestalten of today: early processing of visual
contours and surfaces. Behav Brain Res 82:1–11
86. Aspell JE, Wattam-Bell J, Braddick O (2006) Interaction of
spatial and temporal integration in global form processing.
Vis Res 46:2834–2841
87. Bair W, Movshon JA (2004) Adaptive temporal integration in
direction-selection neurons in Macaque visual cortex.
J Neurosci 24:7305–7323
88. Shadlen MN, Movshon JA (1999) Synchrony unbound: a critical
evaluation of the temporal binding hypothesis. Neuron 24:67–77
89. Baker CL, Braddick OJ (1985) Temporal properties of the short-
range process in apparent motion. Perception 14:181–192
90. Andersen GJ, Cortese JM (1989) 2-D contour perception
resulting from kinetic occlusion. Percept Psychophysic 46:49–55
91. Kellman PJ, Shipley TF (1992) Perceiving objects across gaps in
space and time. Curr Dir Cogn Sci 1:193–199
92. Lee S-H, Blake R (1999) Visual form created solely from tem-
poral structure. Science 284:1165–1168
93. Blake R, Lee S-H (2005) The role of temporal structure in
human vision. Behav Cogn Neurosci Rev 4:21–42
94. Thorton WJL, Raz N (2006) Aging and the role of working
memory resources in visuospatial attention. Aging
Neuropsychol Cogn 13:36–61
95. Greenwood PM, Parasuraman R (2004) The scaling in visual
search and its modification in healthy aging. Percept
Psychophys 66:3–22
96. Andersen GJ, Ni R, Bower JD, Watanabe T (2010) Perceptual
learning, aging, and improved visual performance in early
stages of visual processing. J Vis 10(13):4, 1–13
97. Richards E, Bennett PJ, Sekuler AB (2006) Age related
differences in learning with the useful field of view. Vis Res
46:4217–4231
438 Dynamic Environment Sensing Using an Intelligent Vehicle
Dynamic Environment Sensing Using
an Intelligent Vehicle
HUIJING ZHAO, LONG XIONG, YIMING LIU, XIAOLONG ZHU,
YIPU ZHAO, HONGBIN ZHA
Key Lab of Machine Perception (MOE), Peking
University, Beijing, China
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Sensor and System Architecture
Sensor Calibration and Data Alignment
Contextual Map Generation and Representation
Future Directions
Bibliography
Glossary
Classification Annotating a data segment or an object
by a class label.
Contextual map A map containing high-level knowl-
edge beyond geometry, such as object, class, and
motion.
Data alignment Integrating the instantaneous mea-
surements from sensors’ coordinate system(s) to
a global coordinate system.
Dynamic mapping A mapping technology that uses
a dynamic procedure or generates a map of
a dynamic environment.
Multi-laser sensing system A sensor system that
makes collaborative use of a number of laser range
scanners.
Range image A 2D image where the value of each pixel
is a range distance, with the pixel index
corresponding to range angle and scanning
sequence, so that a 3D coordinate can be retrieved
for each pixel of the range image.
Scene understanding Converting from low-level
knowledge to high-level knowledge of an
environment.
Segmentation Making partitions on a data set, where
in each partition cell (i.e., segment), data has the
property of certain homogeneity.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Sensor calibration Finding a set of parameters that
describes internal or external sensor geometry.
Definition of the Subject and Its Importance
There have been numerous research efforts toward
generating 2D/3D urban models using mobile robots.
In addition, research has focused on robot-centric map-
ping and moving object detection/tracking for online
perception and navigation. However, to date, there has
been little work on generating a realistic 3D copy of
a dynamic environment that describes the state of both
static and dynamic objects at the moment. Toward the
goal of developing omni-directional range sensing in
a dynamic urban scene using an intelligent vehicle,
research has mainly focused on the fundamental issues
of multi-laser sensor system calibration and scene under-
standing in contextual map generation. Here, both sys-
tem and algorithmic development are presented as well as
experimental research demonstrating that a geometric
and contextual representation of static objects such as
buildings, trees, and roads, as well as the motion of
dynamic entities such as people, bicycles, and cars, can
be achieved using a vehicle robot.
Introduction
Traditionally, urban modeling has been thought of as
generating maps of cities, in either 2D or 3D, for
applications such as city planning, facility manage-
ment, tourism, navigation, and cultural heritage.
Numerous research efforts in photogrammetry and
remote sensing [1] have been devoted to the study of
aerial or satellite-based mapping technologies for the
reconstruction of urban objects [2–4], yielding a vast
amount of geographic databases. A summary of world-
wide urban modeling projects can be found in [5].
Normally, aerial sensing can cover relatively wide
area, but it fails to capture urban details due to the
limitation of spatial resolution and a bird’s eye view-
point. Mobile mapping systems (MMS), which
emerged at the end of the last decade [6–8], were
initially considered as a complimentary mapping
approach to aerial ones. This technological trend is
described in [9, 10]. In such systems, ground vehicles
are used as moving platforms carrying multi-modal
# 2012, DOI 10.1007/978-1-4419-0851-3

sensors, so that city objects can be measured from
nearby viewpoints on street. Early results [11–13] dem-
onstrate thatrich data sets can be acquired, and highly
realistic city models can be reconstructed using such
a mapping technique. With the success of Google Earth
[14] and Microsoft Virtual Earth [15], a broader com-
munity has found interest in visualizing cities in multi-
levels of detail by combining data from both aerial and
ground-based mapping techniques. Applications
linking geographical maps with embedded 3D/2D visu-
alization from street viewpoints are increasing. One of
the most successful applications is Google StreetView
[16], and a number of commercial systems using
mobile mapping technologies [17–19] provide 3D/2D
urban data. A review of the major approaches to
large-scale modeling is given in [20]. However, these
systems represent only the static entities of the environ-
ments, while the dynamic objects, such as cars and
people, are either removed from the models or left on
2D images that do not reflect the current situation of the
environment.
Mapping has been an active area of research in
robotics and artificial intelligence for decades. Differ-
ing from photogrammetry and remote sensing, robotic
mapping generates and maintains a spatial model of
the robot’s working environment, which is used to help
the robot’s online perception and decision making.
Maps vary in scale, dimension, content, and
representation, according to the robot’s function.
A comprehensive survey of robotic mapping tech-
niques, i.e. the key issues, major approaches and future
challenges, is given in [21]. In recent years, interest in
ground vehicle robots has grown rapidly. Especially
with the DARPA Grand Challenges 2004, 2005 [22]
and Urban Challenge 2007 [23], considerable enthusi-
asm and research interest have been inspired in the
robotics and intelligent vehicle (IV) communities
toward developing autonomous driving techniques
and advanced driver assistance systems. The ground
vehicle robot or intelligent vehicle must function in
a complex outdoor environment [24, 25], where
a map that eprovides knowledge of the highly dynamic
environment is essential for the robot to achieve
advanced online perception and decision. Generating
maps of dynamic outdoor environments that contain
the representations of both static and dynamic entities
using mobile robots is still a fundamental issue.
Dynamic Environment Sensing Using an Intelligent Vehicle 439
There have been numerous research efforts to gen-
erate 2D/3D urban models using mobile robots [26–29].
There has also been work on robot-centric mapping
and moving object detection/tracking for online per-
ception and navigation [30–34]. However, to date,
there has been little research on generating realistic
3D copies of a dynamic environment that describes
the current situation of both static and dynamic objects
at the moment. This research focuses on developing an
intelligent vehicle system that conducts omni-
directional range sensing to a dynamic urban scene,
where a geometric and contextual representation of
static objects such as buildings, trees, and roads,as
well as the motion at the moment of dynamic entities
such as people, bicycles, and cars, can be achieved after
a run. In the following, sensor and system architecture
is first addressed, followed by the description to the two
major modules, sensor and data alignment, and map
generation. Algorithms and experimental results are
presented, and finally future directions are discussed.
Sensor and System Architecture
Sensor System
Over the past decade, laser range scanners, either 2D or
3D, have emerged as the dominant exteroceptive sensor
for mobile robotics applications. Instead of using the
3D laser scanners, such as Velogyne HDL-64E [35],
which has 64 laser heads and can achieve omni-
directional 3D range sensing with a scan rate of
5–15 Hz, this research developed a multi-laser sensor
system by integrating a number of single-row laser
scanners (called LIDAR, for Light Detection and Rang-
ing) in order to achieve low sensor cost and flexibility
for a variety of applications. It is desirable that the
software be adaptable to different numbers and layouts
of laser scanners according to application
requirements.
Figure 1 shows an example of sensor layout with
five SICK LMS2∗∗s [36] and a GPS (Global Positioning
System)/IMU (Inertial Measurement Unit) based nav-
igation unit. L1 is a horizontal scanning unit mounted
in the center of the front bumper, which is used in
combination with GPS/IMU so as to estimate vehicle
position while generating a horizontal reference, i.e.,
a 2D grid map of the environment, and extracting
motion trajectories of dynamic objects at a horizontal
440 Dynamic Environment Sensing Using an Intelligent Vehicle
GPS IMU
L4
Laser scanner
L3,L1,L2
Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 1
A test-bed vehicle system designed for omni-directional range sensing [41]
level (SLAMMODT [31]: Simultaneous Localization
and Mapping [37, 38] with Moving Object Detection
and Tracking [39, 40]). Other laser scanners are used to
acquire 3D data about the environment along streets:
L2 scans downwardly, measuring the lower part of
objects on roadside, as well as road surface; L3 scans
the upper part of objects on roadside, as well as those
above the road surface, such as traffic signs and signal
lights; L4 and L5 scan vertically to the left and right
sides of the car. The data from each laser scanner can be
represented in the form of 2D range images, as shown
in Fig. 2, where the horizontal axis is time, and vertical
strips correspond to laser scan lines, with each pixel
valued by converting range distance into an intensity in
[0, 255]. A piece of range segment, i.e., area “A” in
Fig. 2, is enlarged in Fig. 3. It can be seen that a dynamic
environment containing not only the stationary objects
such as trees and poles, but also the moving objects
such as pedestrians and cyclists, is vividly captured
through such a range measurement.
Software Modules
Figure 4 describes software modules and major data
flows (input and output). Module 1 is a sensor data
logger, which records both GPS/IMU navigation inputs
and all laser scan data, and then forwards the raw
sensor logs to other modules through a wired local
network, e.g., the data of L1 and GPS/IMU are
L5
Laser # Scan mode Major purpose
L1 Horizontal SLAM with MODT
L2 Downward Road geometry
L3 Upward Objects above the road
L4 Vertical Right Objects to the right of the car
L5 Vertical Left Objects to the left of the car
forwarded to module 2, while those of L2–L5 are
forwarded to processors 3 and 4.
In and urban environment, bridges, trees, tunnels,
and buildings might obstruct GPS signals, and
multi-path degrades the positioning accuracy of
GPS/IMU based navigation units dramatically [43].
An algorithm of SLAMMODT was developed in the
authors’ previous work [44], coupling a horizontal
laser scanner with GPS/IMU, so that estimations of
the robot’s motion parameters can be refined by
matching inter-frame sensing data to static objects
(localization), meanwhile generating a consistent map
in 2D grids, and detecting and tracking the data of
dynamic entities at the horizontal level. The
SLAMMODT is conducted in a horizontal level with
an assumption that the ground surface is flat, and three
kinds of outputs can be issued online-vehicle pose,
a 2D grid map of static objects, and 2D trajectories of
moving objects. A result after a cyclic run around
a large building complex in a dynamic campus is
shown in Fig. 5, where the course length is about 1 km.
In order to make collaborative usage of the data
from different laser scanners, a calibration of sensor
geometries is conducted in module 3 by referencing the
sensor’s coordinate system of L1. With the data of both
sensor geometry and vehicle position, laser scans cap-
tured by different sensors at different vehicle positions
can be integrated into a global coordinate system in
module 4, so as to generate a map of the outside
 Dynamic Environment Sensing Using an Intelligent Vehicle 441

To the left of the vehicle

L2

Lower part of objects Road surface To the road

on road sides

To the right of the vehicle

L3 To the left of the vehicle

Upper part of the Cars

A sign board A road light To the air
Trees objects on road sides

To the right of the vehicle

L4

Building

To the right of the vehicle

L5 Trees A
People Cars

To the left of the vehicle

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 2

Range images of laser scan data L2–L5 [42]. Horizontal axis is scan line number (corresponding to time). Each vertical strip
represents a line of range values of a certain laser scan

Tree

Road light

Cyclists
Pedestrian

Bush

Road Surface

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 3

An enlarged range image by laser scanner L5 [42]
442 Dynamic Environment Sensing Using an Intelligent Vehicle

GPS/IMU
Module 1 L2....L5
L1
.... Sensor data logger
L5
Module 3
GPS/IMU Sensor calibration
L1
Module 2 2D Grid map
Sensor geometry
SLAM with MODT Vehicle pose
Module 4
Data alignment

Geo-referenced data
2D Trajectory
(moving objects) Module 5
Map generation

3D Contextual map
static & dynamic objects

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 4

Software modules and major data flows

Static objects (Black)

Trajectory of the robot (Green dots)
The data of moving objects (Color)

Start and end points

of the run

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 5

A result of SLAMMODT after a cyclic run in a dynamic campus

environment using the low-level primitive of 3D laser
points. With such a representation, an operator can
obtain visual knowledge of the environment, however,
it is not informative for a robot, as such a map reflects
only spatial existence through point sampling of object
surface. In order for a robot to have semantic knowl-
edge of the environment, such as objects, types, and
their spatial relationships, a segmentation and classifi-
cation of 3D laser points is conducted in module 5.
A contextual map is finally reconstructed with each 3D
object annotated, and the motion data of dynamic
entities from processor 2 are integrated, which can be
used to infer the relationships between different
objects.
Currently, modules 1, 2, and 4 conduct online
processing, module 3 is pre-processing directly after
each sensor system setting, and module 5 is an off-
line procedure after a vehicle run. The algorithm details
as well as experiments with modules 3, 4 and 5 are
described below; module 2 is described in [44].
Sensor Calibration and Data Alignment
Literature Review
There have been dozens of intelligent vehicle systems
using laser scanners for perception. For example, all 11
finalists in the DARPA Urban Challenge relied upon
laser scanners (LIDARs) as their primary exteroceptive
sensing modality [24, 25]. Laser scanners can provide
angle and range measurements, which are the sam-
plings on object surface, and can be easily converted
into 2D or 3D coordinates. However, a single laser
scanner might have limited viewing angles and
occluded sensing space, e.g., the widely accepted SICK
LMS 2∗∗ conducts 2D scanning on a plane within
180
. Therefore, it is important for a robot to possess
a number of laser scanners to achieve a better under-
standing of the whole 3D environment. In order to
integrate the data from different sensors, calibration is
required to find sensor geometries, i.e., transformation
parameters between different sensor’s coordinate
systems.
Normally, calibration methods are decomposed
into intrinsic and extrinsic calibrations. Intrinsic cali-
bration finds the local geometry of each sensor, such as
the camera’s focal length, optical distortion, principal
point, and so on. These parameters can be estimated
Dynamic Environment Sensing Using an Intelligent Vehicle 443
independently with other sensors. They experience only
slight changes in temperature and humidity. Extrinsic
calibration finds the relative geometry, e.g. rotation and
transformation parameters, between different sensors
that compose a rigid body. These parameters change
during sensors’ removal and set-up, and need to be
calibrated after the sensors are mounted on the vehicle.
Calibration of the sensor system using laser scanner(s) is
poorly documented, except for a few papers on intrinsic
calibration of laser scanners [45, 46], extrinsic calibra-
tion of laser scanners with camera [47–49], and multi-
laser systems [50].
The procedure for extrinsic calibration, especially
for a large sensor system, is notoriously labor intensive.
Relative geometry between different sensors’
coordinate systems can be estimated by registering
their data sets, and a relationship between the data
spaces can be established. Registration of range data
has been studied for decades [51–53], where sensor
geometry is supposed to be fixed during the measure-
ment of each set of range data. However, registration of
mobile laser scan data is different from that of
a stationary platform.
This research focuses on extrinsic calibration of
a multi-laser sensor system that uses a number of
single-row laser scanners (LIDAR). A calibration
method has been developed that is not restricted to
the sensor system shown in Fig. 1, but is adaptable to
different sensor numbers and layouts. Here, it is neces-
sary to stress that (1) the laser beam of the laser scan-
ners (e.g., SICK LMS2∗∗) is invisible (near infrared
band); (2) a traditional checkerboard is not useful here,
as a single-row laser scan shoot only one line on it; and
(3) locating a reflective target to catch a laser beam is
labor intensive, especially with laser beams of upward
scanning. A calibration method is desirable if it
(1) does not require special facility and (2) uses the
objects in the scene as landmarks.
Problem Statement
Without loss of generality, a laser scanner’s coordinate
system is defined as follows (see Fig. 6). The origin is at
the principle point of the laser scanner. Laser beam
scans start from the x-axis, moving toward the y-axis.
The z-axis is vertical to the laser beam’s scanning plane,
and xyz-axes compose a right-hand coordinate system.
444 Dynamic Environment Sensing Using an Intelligent Vehicle

A range point pir ¼ ðki ; ri ; si Þ is measured by the laser
scanner p at scan line number ki (corresponding to and
denoted as “time” below), with range distance ri , and
data number si. ðki ; si Þ locates the pixel index in range
image (see Fig. 2), the value of which is assigned by
ri 256
rmax . Let ares be the angular resolution for range
sampling, e.g., ares ¼ 0:5 in this research, si is converted
to a scanning angle by ai ¼ si  ares, pir is converted
into a 2D coordinate at laser scanner p’s coordinate
system, and is represented in the following form.

pi ¼ ðri cos ai ; ri sin ai ; 0; 1Þt

Let Tpv be the geometric transformation from laser
scanner p to the vehicle body frame, which is defined
on that of L1, pi can be converted to the vehicle body
frame as follows.

pv ¼ ðxv ; yv ; zv ; 1Þt ¼ Tpv  pi

Given an initial set of sensor geometry (Tpv s), the

laser points that are measured by L1–L5 at a certain
Dynamic Environment Sensing Using an Intelligent moment can be integrated into the vehicle body frame
Vehicle. Figure 6 as shown in Fig. 7, where colors denote for different
Instantaneous laser measurements at vehicle body sensor data: light blue for L1, dark blue for L2, orange
frame for L3, green for L4, and red for L5. It can be found that, at

z
qi measured at time ki

x
y
Tqv
Laser scanner q
No o
verla
p
Vehicle’s body
frame at time ki Tvw(ki)
z
pi measured at time kj

x Tpv Q ={qw}
y
Laser scanner p overla
p

P ={pw}
Vehicle’s body Tvw(kj)
frame at time kj

Global coordinate system

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 7

Sensor geometry and data geo-referencing [42]

any moment, an instantaneous measurement obtains
data only on five 2D planes, which do not compose a
3D measurement of the whole environment. Also, even
Tpv s are erroneous, it cannot be discriminated from the
instantaneous laser scan lines, as there are no or few
intersections between the data spaces of different sensors.
On the other hand, with module 2 – SLAMMODT,
the vehicle robot is able to build a 2D grid map of the
environment, meanwhile it determines its location and
orientation within the map as shown in Fig. 5. The map
can be a relative one, if referencing to the vehicle body
frame at the starting point, or a global one if incorpo-
rating GPS/IMU inputs, where a fiducial vehicle body
frame is defined on that of L1. Position estimation finds
the location and orientation of the vehicle body frame
with respect to the reference one (denoted as global
coordinate system), as a result, a transformation matrix
ki
Tvw is obtained at each time instance ki, associating the
vehicle’s body frame to the global one. With both
ki
sensor geometry Tpv and vehicle position Tvw at time
ki , pi can be converted to the global coordinate system
as follows (see Fig. 7). Pw ¼ fpw g denotes the geo-
referenced data set of laser scanner p.
pw ¼ ðxw ; yw ; zw ; 1Þt ¼ Tvw
ki
 Tpv  pi
In the same way, let q be another laser scanner, qir ,
qi , qw , and Qw denote a range point, 2D coordinates at
L2
L5 L4
Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 8
Geo-referenced 3D points of each individual laser scanner and their integration [42]
Dynamic Environment Sensing Using an Intelligent Vehicle 445
the sensor’s coordinate system, a 3D point in the global
coordinate system, and the geo-referenced data set of
the laser scan measurements, respectively.
Figure 8 shows the geo-referenced 3D laser points,
where the top-left, top-right, and bottom-left drawings
are the data of each individual laser scanner, while the
bottom-right is an integrated version. For the purposes
of discrimination, colors in the drawings in the top row
denote the local surface normal of the laser points,
while colors in the drawings in the bottom row denote
data from different sensors. It can be found that
although the data generated by each individual sensor
contains only partial knowledge of the whole environ-
ment, e.g., the data of L2 represents road surface only,
the laser points have good consistency that describes
object geometries in its vision field. However, when
integrating the data from different sensors into
a single one, displacement can be found in the dupli-
cated measurements of common objects. It reflects
that, although laser scanners have no intersection in
their instantaneous measurement, a single object could
be measured by different sensors at different time
ki
instances. When using the same vehicle position Tvw s,
the displacements come from the erroneous sensor
geometries (Tpv s). This suggests that sensor geometries
can be refined by minimizing the displacements
between the overlapped measurements of different sen-
sors after data geo-referencing, i.e., data registration.
L3
Integration
446 Dynamic Environment Sensing Using an Intelligent Vehicle
1
Let Tpq ¼ Tpv  Tqv denote the relative geometry
between laser scanner p and q. Calibrating the sensor
geometry between laser scanner p and q allows conver-
sion into the following data registration problem:
Tpq ¼ arg min DðzðPw Þ; zðQw ÞÞ
Tpq
where zðÞ is a function projecting Pw and Qw to
a matchable data space, and Dð ; Þ is a distance mea-
sure evaluating the displacement (i.e., distance) between
two data sets. In solving the problem, a coarse-to-fine
procedure is taken. Note that if q is L1, as the vehicle
body frame is defined on L1, Tpq ¼ Tpv .
Coarse Registration
A geometric transformation Tpq is first initialized
according to hardware design, then refined through
matching their data sets, i.e., Pw and Qw . If a direct
correspondence between a number of landmark points
in Pw and Qw could be established, Tpq can be solved
using a Least Squares Method. However, when laser
scanners face different directions, partial observations
and occlusions make direct correspondence an
extremely difficult issue.
Recall the problem statement in previous section.
A fiducial vehicle body coordinate system is defined on
that of L1. Sensor calibration can be stated more spe-
cifically to achieve an alignment of all other laser scan-
ners (L2–L5) with reference to L1. With the output of
module 2 – SLAMMODT, a 2D grid map generates
a horizontal reference of the environment, while Tvw k Dh ðzh ðpÞ; zðqÞÞ ¼ C2 ðp; qÞ  C1 ðqÞ ! min
locates a vehicle at any time k on the map as shown
in Fig. 5. In a calibration procedure, vehicle positions
k
Tvw s are considered adequately accurate, leaving Tpq s
for refinement. In this research, a two-step procedure is
designed in coarse registration: horizontal and vertical
matching, both of which generate projections to a 2D
grid map on a horizontal level.
Horizontal Matching Horizontal matching gener-
ates a projection zh ðÞ, so that the laser points of vertical
features, such as buildings, poles, and tree branches, are
projected to a horizontal grid map. The grid value vðÞij
can be 1 or 0, representing that the number of laser
points projected to the grid cell is greater than or less
than a predefined threshold. The reason for
thresholding is to filter out irregular laser points and
enhance vertical features. Horizontal matching is
conducted by taking a laser scanner p as reference,
adjusting Tpq to minimize the distance between
zh ðPw Þ and zh ðQw Þ. A distance measure is designed
based on the following two criteria.
XX
C1 ¼ vð  Þij ! min
i j
C1 means that the sum of all grid values should be
minimized. As each grid cell has a value of either 1 or 0,
C1 means also that the number of grid cells that has
a value equal to 1 should be minimized. Ideally,
a vertical wall is converted to a line by zh ðÞ, a pole
becomes a point and so forth. Laser points belong to
the vertical features, such as walls or poles, tend to be
projected to the limited number of grid cells
corresponding to the lines and points. A concentrated
projection is highly evaluated comparing to a distrib-
uted projection.
XX
C2 ¼ vðpÞij  vðqÞij ! min
i j
C2 means that the sum of a production between the
grid values vðpÞij and vðqÞij should be maximized. As
each grid cell has a value of either 1 or 0, it also means
that the number of matched grid cells that have a value
equal to 1 should be maximized. A distance measure is
designed by summarizing the above criteria.
Vertical Matching Vertical matching generates
a projection zv ðÞ, so that the laser points on ground
surface are projected to a horizontal grid map. The grid
value uðÞij represents the ground elevation at the grid
cell, which is assigned by taking a mean of the z values
of all laser points that are projected to the grid cell.
A distance measure is designed as follows to evaluate
the displacement between the elevation values:
XX
Dv ðzv ðpÞ; zv ðqÞÞ ¼ juðpÞij  uðqÞij j ! min
i j
Figure 9 shows an example of extracting the hori-
zontal and vertical elements from the data of L4. A local
surface normal is estimated for each laser point using
its neighborhood data on scanning order. Either the
 Dynamic Environment Sensing Using an Intelligent Vehicle 447

Vertical elements (Blue) Horizontal matching

Ground data (Grey) Vertical matching

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 9

Example of extracting horizontal and vertical elements

2D grid map by the SLAMMODT of L1 (Black)

Vertical elements of L4 (Blue)

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 10

Example of horizontal matching of vertical elements from L4 and L1

horizontal and vertical elements are extracted by
thresholding on their local surface normal. The vertical
elements are used for horizontal matching, where an
example of registering L4 on the 2D grid map of L1 is
shown in Fig. 10. The horizontal elements, i.e., ground
limited by such a representation. Find registration
approach is to refine the coarse registration results.
Given the sensor geometries of Tpv and Tqv , for any
pw measured by laser scanner p, a mirror range point
pj ¼ ðkj ; rj 0 ; sj Þ is estimated at the range image frame
q r

data, are used for vertical matching, where an example
of registering L3 with L4 and L5 is shown in Fig. 11. In
this research, both horizontal and vertical matching is
conducted by adjusting the parameters of Tpq , so as to
minimize Dh and Dv , respectively, in an exhaustive way.
Fine Registration
In coarse registration, data are represented in 2D grid
maps at a horizontal plane, where resolution in describ-
ing object details as well as registration accuracy are
of laser scanner q. Let Qr ¼ fqjr g denote the range
image measured by laser scanner q, while Qpr ¼ fq pjr g
for the mirror range image generated from the data
measured by laser scanner p. The objective of fine
registration is to find a Tpv that minimizes the differ-
ence between Qr and Qpr . It is defined as follows:
Tpv ¼ arg min DðQ r ; Qpr Þ
Tpv
Given pw , a mirror range point q pjr ¼ ðkj ; rj 0 ; sj Þ is
estimated, where kj is the number of a scan line at laser
448 Dynamic Environment Sensing Using an Intelligent Vehicle

Ground elevation map by L3

Elevation difference map of L3 with L4&L5

−0.1 0 +0.1

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 11

Example of vertical matching of ground data from L3 with L4 and L5

scanner q’s data space, sj is the data number at kj , rj 0 is Scanning plane of laser
the range distance that pw be measured on the scanning scanner q at time kj rx

plane of kj at range beam angle aj ¼ sj  ares . A mirror p⬘w

pw

range point is estimated in an order of kj ! sj ! rj 0 ,

ry dj
where estimation of each parameter, as well as
a definition to Dð ; Þ is described below.

Vehicle path
Estimation of kj For any laser scan kj at the data
space of laser scanner q, an orthogonal projection is
done from pw to the scanning plane as shown in Fig. 12,
Dynamic Environment Sensing Using an Intelligent
where rx is the orthogonal distance, ry is the Euclidean
Vehicle. Figure 12
distance from sensor q’s viewpoint at kj to the orthog-
r Locating a scan line kj at the data space of q for a laser
onal point pw 0 , and dj ¼ arctanðrxy Þ. kj is located as the
point pw
laser scan of minimal dj and dj < t, where t is
a threshold representing the diffusion of laser beam.

Estimation of sj and rj 0 With the orthogonal point

pw 0 , a mirror point in sensor q’s coordinate system is
estimated as follows, Definition of Dð ; Þ Let

pj ¼ ðq xi ; q yi ; q zi  0; 1Þt ¼ ðTqv Þ1  ðTvw
kj 1 q p q p
q
Þ  pw0 1 if dij  q dij  ðjjrij  q rij jj < B
signij ¼
0 otherwise
So that sj and rj 0 are estimated as
q p
where dij and q dij are binary values, denoting whether
q
yi . q p
the range value rij at Qr and q rij at Qpr , respectively, are
sj ¼ arctanðq Þ ares
xi valid (1 for valid). B is a threshold, defining the bound
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
for matched pair of range values. signij ¼ 1 denotes that
rj 0 ¼ ðq xi Þ2 þ ðq yi Þ2
the range points of Q r and Qpr at pixel index (i,j) are
 Dynamic Environment Sensing Using an Intelligent Vehicle 449

matched ones. Dð ; Þ is designed based on the follow- the original range measurement by L5. A match is made
ing two criteria: between L30 and L5, as shown in (d), where residuals
between corresponding range points are represented in
W X
X H a color band of [blue-black-red], while the range points
C3 ¼ signij ! max of L5 that do not have matched pair in L30 are shown in
i¼0 j¼0 grey. Some other results are also shown in Fig. 14,
W X
X H which are represented in both the residual maps
q p
C4 ¼ signij  jjrij  q rij jj ! min between mirror and original range images, and their
i¼0 j¼0 edge maps. In the edge map of L30 and L5, it can be
DðQ ; Qpr Þ ¼ C4 =C3  C3
r clearly found that a cyclist is captured by L5, however
not in L30 . This might happen in the measurements to
where C3 counts for the number of matched range moving objects, where due to the difference in sensors’
points, and C4 is the sum of residuals between the viewpoints, a single object (place) might be measured
matched range values. Dð ; Þ is designed to maximize by sensors at a different time period. Although the
the number of matched range points, while minimizing cyclist is captured by L5, when L3 measured the place,
the average of residuals between matched range values. it was not there, and L3 failed to capture the cyclist all
Figure 13 explains the procedure of fine registration the time due to its high dynamic motion. On the other
using experimental results, where the data of L3 (a) is hand, this phenomenon suggests that by matching the
projected to the range image frame of L5, and a new data from different sensors, the dynamic factors in the
range image L30 is generated as shown in (b). (c) shows environment can be detected. Due to occlusion,

L3

a
The same road light

L3’ by projecting L3 to the image plane of L5

b

L5
c

Residual between L3’ with L5 −1.0m 0m +1.0m

d

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 13

Explanation of the fine registration procedure using experimental results
450 Dynamic Environment Sensing Using an Intelligent Vehicle
Residual map
–1.0m 0m
L3’ & L5
L3’ & L4
L2’ & L5
a L2’ & L4
Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 14
Some results of fine registration. Left column: residual map between mirror and original range images; Right column: edge
points extracted from mirror (blue) and original (red) range images
viewpoint difference, and the dynamic entities of the
environment, matching between the mirror and origi-
nal range image is conducted manually in this research,
where an automated method is to be developed
through future research.
After sensor alignment, the laser points from dif-
ferent laser scanners can be integrated into a global
coordinate system with more consistency, as shown in
Fig. 15, where the ground data of L2 is shown in gray,
non-ground data of L2 containing those of road edges
are shown in green; L4 is in blue and L5 in red. It can be
found that good consistency between the data of dif-
ferent sensors is achieved through a calibration proce-
dure, and an omni-directional measure of the whole
environment is achieved through the collaborative
exploration of multiple single-row laser range scanners.
Contextual Map Generation and Representation
An environmental map can be generated directly using
the low-level geometric primitive of 3D laser points.
However, such a map has limited capacity in
Edge map
A cyclist captured by L5 only
1.0m
L3’ (blue) & L5 (red)
L3’(blue) & L4 (red)
L2’ (blue) & L5 (red)
b L2’ (blue) & L4 (red)
representation, as it describes only spatial existence.
An operator can easily understand from the data
where objects are, and what kinds of objects they are,
while a robot cannot. In order for a robot to have
semantic knowledge of the environment, such as
objects, types, and their spatial relationships, an auto-
matic technique of converting a low-level map repre-
sentation into a high-level one is important.
Literature Review
Through a bottom-up procedure, laser points can first
be processed to find data clusters, i.e., laser points that
are most likely to be the measurements of the same
objects. They can then be recognized as certain kinds of
objects, e.g., planar surfaces [54], line feature objects
[55], cars [56], or natural objects [57]. In order to
tackle the large number of laser points, [58, 59] tessel-
lated the 3D space and projected laser points to voxels.
The sequence of laser points can also be used as a whole
to represent the geometric appearance of local

Building
Road
Trees
A cyclist
Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 15
3D map of the environment created by integrating multi-laser sensing data [42]
surroundings, and a higher level knowledge of the
robot’s concurrent location, such as a doorway, corri-
dor, or room, can then be inferred through machine
learning techniques [60]. In addition, each individual
laser measurement is considered to be dependent on its
neighborhood in [61]. Their relationships are modeled
using Markov networks, where labeling of each laser
point is influenced by the labeling of others in its local
vicinity. [62] also associated image cues with each laser
point and varied the probabilistic framework using
Conditional Random Field. Reference [63] further
extended the method of [61], so that each node in the
Markov network corresponds to a data patch, i.e.,
a superpixel (image patch) with corresponding laser
points in successive scans, rather than a laser point. In
addition, a recent report can be found in [63], where an
aerial laser scan data is processed to label the small
objects, such as posts, lights, and cars, in an urban
environment. The problem is solved by localization,
segmentation, representation, and classification proce-
dures in a sequential way.
On the other hand, laser scan data can be
represented in the form of a range image, where each
pixel represents a depth value, and its index corre-
sponds to the sequential order of measurements.
Thus, beam origin and angle of each depth value can
Dynamic Environment Sensing Using an Intelligent Vehicle 451
Building
A Pedestrian
Road
A car
be retrieved, and depth value can then be easily
extended to a 3D coordinate. There is a large body of
work addressing range image segmentation. A famous
report comparing the major segmentation methods
can be found in [64]. Many of the methods are moti-
vated by the needs for recognizing industry parts [65]
or registering the data taken at different locations [66].
These works always assume simple or well-defined
object geometry. There are still a few research works
processing range images of real-world scenes. [67] con-
siders a real-world indoor and outdoor scene by
modeling the man-made objects using planes and
conics, modeling free-form objects using splines, and
modeling trees using 3D histogram; segmentation and
model fitting for each segment is formulated in a data-
driven Markov Chain Monte Carlo procedure.
Image segmentation and semantic interpretation
have been studied extensively in the field of computer
vision. As the data form of a range image is consistent
with that of a visual image, many methods developed in
the field of computer vision are of great reference for
the processing of range image [68–71].
As shown in Figs. 2, 8, and 15, laser scan data can be
represented in the forms of 2D range images or 3D laser
points, where for a 2D range image, each pixel can be
converted into a 3D point in a global coordinate system
452 Dynamic Environment Sensing Using an Intelligent Vehicle
as addressed previously. In this research, range image is
chosen as an interface for data representation. More
specifically, this research concerns the segmentation
and classification of range images in the real world,
however estimations are conducted by retrieving the
3D coordinates of each range pixel. In dealing with the
problem, a sequential flow as shown in Fig. 16a is
generally used [72], where classification is conducted
subsequently to segmentation. The framework is
straightforward and easy to implement. However,
when facing a complex environment with different
kinds of objects, segmentation failures might happen
due to the heterogeneous geometric properties between
different kinds of objects. Unified frameworks are also
proposed [68–71] that cope with the segmentation and
classification problems simultaneously (see Fig. 16b).
This research examines the performances of both
sequential and simultaneous processing flows in solv-
ing the problem of segmentation and classification of
the range images in complex outdoor environments.
The algorithmic and implementation details as well as
experimental results are described below.
Segmentation
First, the data of the ground surface are discriminated
and removed from the range image, so making the data
Comparison of off-
Range image
the-shelf algorithms
Contour-based
segmentation
Graph-base
Comparison of the
classifications on
Classification
different features
Scan line
Objects with Point cloud
semantic label
Both
a b
Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 16
Sequential (a) and simultaneous (b) processing flows for range image segmentation and classification
of vertical objects spatially disconnected. Ground ele-
vation at the point of host vehicle can be assumed, as
the vehicle stands directly on the ground, and the
sensor-vehicle is composed of a rigid body. For each
laser scan line, segmentation is conducted, and the
nearly horizontal line segment that has at least one
end point within the bound of assumed ground eleva-
tion is extracted as the ground data, as the grey points
in Fig. 17.
Contour-Based Segmentation In segmenting the
rest of range image, contour points are first extracted,
where the Euclidean distance between the 3D coordi-
nates of each contour point with its four neighbors
in range image is larger than a given threshold;
a point-based region growing is then conducted to
extract the data segments with spatial continuity (dif-
ferent color in Fig. 17 denotes a different segment). The
method is efficient in extracting the data segments of
most of the objects, and they are spatially disconnected.
As shown in Fig. 17, although over-segmentation
(i.e., a single object can be separated into a number of
segments) occurs, especially for trees and pedestrians,
most of the segments contain the data of a single object;
with under-segmentation, different objects are merged
into a single segment. Failure A merged the data of
Scan line
segmentation
Range image
Extracting segments
of planarity
Over-segmentation
Contour point
extraction
Joint merge Extracting segments
with classification of continuity
Calculating merge
Objects with probability
semantic label simultaneously
with classification

Pedestrian
a
Failure A
b Parked cars
Failure B
c Parked cars
Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 17
Results of contour-based segmentation
a piece of wall and a nearby car. Failure B merged the
data of building surface and a clump of bush, where in
some places, they are spatially close.
Graph-Based Segmentation Data acquired by laser
scanner usually spatially connect with its neighbors in
a range image. This property suggests a graph-based
segmentation. The set of pixels can be represented as
a weighted undirected graph G ¼ ðV ; EÞ, where V is
the set of nodes standing for pixels and E is the set of
edges between pixels and their neighbors in 3D space.
Thus, segmentation can be formulated as the problem
of cutting an undirected graph G, where many mature
algorithms exist [73, 74]. This research chose the FH
algorithm [74] for its good performance in catching the
non-local properties and adaptability to online proce-
dure. A range segment s is composed of a set of range
points as well as weighed edges that represent the
spatial relationships between neighbor range point
pairs. The weight wof an edge eis assigned by the
Euclidean distance between the 3D coordinates of two
Dynamic Environment Sensing Using an Intelligent Vehicle 453
Building
Parked cars
Trees
neighbor range points, which represents their spatial
continuity. For any segment s that has ns range points,
an internal difference (IntðsÞ ¼ max wðeÞ) as well as
e2s
a regulation term (tðsÞ ¼ nsk ) is defined, where k
is a predefined experimental factor. For any pair of
segments (sa and sb ), a minimum internal difference
MIntðsa ; sb Þ ¼ minðIntðsa Þ þ tðsa Þ; Intðsb Þ þ tðsb ÞÞ is
defined. The pair of segments sa and sb can be merged
if and only if it has MIntðsa ; sb Þ < wðea;b Þ, where ea;b is
an edge that has its two end points in sa and sb , respec-
tively. Experimental results reflect that the parameter
setting of k in the regulation term (tðsÞ ¼ nsk ) is crucial
to the segmentation result as described in Fig. 18.
A small k causes objects to be over-segmented, while
enlarging kmight cause the data of different objects to
merge. Figure 19 shows some of the results with k ¼ 1.
It can be found that the problem of over-segmentation
is reduced compared with the results of a contour-
based one, while under-segmentation is even heavier.
Failure A merged the data of a number of trees and
a clump of bush. Failure B merged the data of a banner
and two neighboring trees.
454 Dynamic Environment Sensing Using an Intelligent Vehicle

K=0.1 K=0.25 K=0.55 K=1 K=5

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 18

Parameter setting in graph-based segmentation

Parked cars Pedestrians

Failure A Failure B

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 19

Results of contour-based segmentation

to extract data cues from the set of 3D laser points, and

Classification devise a classifier by evaluating their classification
properties. On the other hand, a single-row laser scan-
A segment contains a set of 3D laser points, which
ner measures the environment in a mode of scan-line
describes the geometry of an object surface through
by scan-line. A scan-line measurement on a planar
the representation of point clouds. It is straightforward

object, such as the surface of a building or road, can be
modeled using a line segment. The measurement of
a free-form object, such as a tree, yields many small
line segments and isolated points. An isolated point can
be considered as a specific type of line segment.
A segment can also be considered as a set of scan line
segments [75, 76], where data cues are extracted and
evaluated to compose a classifier. Of course, a classifier
can also be devised by a combinatory evaluation on the
properties of 3D laser points and line segments. In this
research, three classifiers are developed on line seg-
ments, 3D laser points, and their combinations,
which are described below as well as an experimental
comparison.
Classification on Scan Line Segments Let L denotes
the set of object classes, i.e., L ¼ fljbuilding ; road;
tree; person; car   g, y be the label of a segment s,
and yk be the label of the kth line segment. Given
data cue I, classification of segment s on scan line
segments is formulated as follows.
Y
y ¼ l ¼ arg max Pðyk ¼ ljIÞ
l k
Let I ¼ fd1 ; d2 :::g, the probability of a line
segmentlabeled l, i.e., Pðyk ¼ ljd1 ; d2 ; :::Þ can be fur-
ther extended according to Bays’ rule (i.e., a naı̈ve
Baysian classifier).
Y
Pðyk ¼ ljd1 ; d2 ; :::Þ ¼ Pðyk ¼ lÞ Pðdi jyk ¼ lÞ
i off-the-shelf method, LIBSVM [77], is used.
where Pðyk ¼ lÞ is a prior knowledge of a scan line
segment labeled l. It is initialized as an equal distribu-
tion and updated after each iteration according to the
percentages of data labeling.
Pðdi jyk ¼ lÞ is the likelihood measure (denoted by
yk
ll ðdi Þ), when given a label yk ¼ l, the probability di of
the scan line segment can be observed. The data cues
extracted from each scan line segment are listed in
Table 1. For any pair of data cues di and object label
yk
yk ¼ l, a likelihood measure ll ðdi Þ is trained using
a set of manually labeled scan line segments.
A histogram is first generated on data samples, then
Gaussian fittings are conducted on each distinctive
picks, followed by a normalization so that integration
of the graph is 1. The likelihood measures trained in
this research are shown in Fig. 20.
Dynamic Environment Sensing Using an Intelligent Vehicle 455
Dynamic Environment Sensing Using an Intelligent
Vehicle. Table 1 Data cues of a scan line segment
Feature Definition
d1 Length of the scan line segment
d2 Maximal height value
d3 Minimal height value
d4 Z-coordinate of the directional vector
d5 Mean of line regression
d6 Variance of line regression
Classification on 3D Laser Points Let Y denote the
label on the cloud of 3D laser points. Classification is
formulated as follows:
y ¼ l ¼ arg max PðY ¼ ljIÞ
l
Data cues extracted from a cloud of 3D laser points
are listed in Table 2. Differing from that of scan line
segments, classification of 3D laser points does not
evaluate the properties of each individual laser point
but treats them as a whole. The likelihood measures
yk
ll ðdi Þ of a certain pair of data cues and object labels
are generated as demonstrated in Fig. 21. However, due
to the limited and unbalanced number of sample data
(see Table 3), a naı̈ve Baysian classifier did not yield
satisfying results. In classification of 3D laser points, an
Classification of Both 3D Laser Points and Scan Line
Segments A classifier of both 3D laser points and
scan line segments is designed through the multiplica-
tion of probabilistic estimations on each individual
primitive, as follows:
Y
y ¼ l ¼ arg max PðY ¼ ljIÞ  Pðyk ¼ ljIÞ
l k
Comparative Experimental Results In training the
classifiers, an interactive tool is developed to generate
training data samples. A range image is first processed
to find local surface normal for each laser point, edge
point that has discontinuous change in range value,
scan line segments, super segments, etc. These results
are compared by an operator to discriminate the
456 Dynamic Environment Sensing Using an Intelligent Vehicle

0.03 0.06
yk yk
0.025 λl (d1) 0.05 λl (d2)

0.02 0.04

0.015 0.03

0.01 0.02
0.005 0.01
0 0
–1 0 1 2 3 4 5
–1 –0.5 0 0.5 1 1.5 2 2.5 3

0.03
Build
yk yk
0.025 λl (d3) λl (d4)
Road
0.02 Tree
0.015 Car

0.01 Human
Bush
0.005
Bus
0
–1 –0.5 0 0.5 1 1.5 2 –1 –0.5 0 0.5 1 1.5

0.02 0.012
0.018 yk
0.016
yk
λl (d5) 0.01 λl (d6)
0.014
0.008
0.012
0.01 0.006
0.008
0.006 0.004
0.004
0.002
0.002
0 0
–5 0 5 10 15 –0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
x 10–1

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 20

Likelihood measures for the classification of a scan line segment [41]

boundaries of individual objects. For each object, the
operator will manually draw a boundary and assign
a label. The software will output the scan line segments
that are inside the boundary, as well as the label, for
training classifier PðykjIÞ. The software will also output
all laser points within the boundary, as well as the label,
for training classifier PðY jIÞ.
In experiments, two simultaneously measured range
images from sensors L4 and L5 are manually labeled.
The manually labeled data of L4 is used in training
classifiers. The number of training samples with respect
to each object class and classification method are listed
in Table 3. Three classifiers are trained on the features of
scan lines (classifier 1), 3D laser points (classifier 2), and
their combination (classifier 3). On the other hand, the
range image of L5 is first segmented then classified, and
the result is compared with those of manually labeled
ones. Here, contour-based segmentation is exploited,
and the segments are classified using each of the classi-
fiers for comparison. A set of results is shown in Fig. 22,
where a number of objects are highlighted. Object 1
includes cars that are correctly classified by all classifiers.
Object 2 is a bus that is correctly classified by classifier 1
and 3, but failed in classifier 2. A reason might be the
limited number of training samples. Objects 3 and 4 are
pedestrians, which are failed in all three classifiers,

Dynamic Environment Sensing Using an Intelligent
Vehicle. Table 2 Data cues extracted from a cloud of laser
points
Feature Definition
d1 Minimal height value
d2 Maximal height value
d3 Ratio of boundary point number versus total
point number
d4 Mean of height values
d5 Variance of a histogram distribution on normal
vectors
d6 Major picks of a histogram distribution on
normal vectors
d7 Ratio of width versus length
pointing out that classification performance related to
pedestrians needs to be improved through future work.
Object 5 is a small house, the shape of which is similar to
a bus. Classifier 1 misclassified it into a bus, while two
others answered correctly.
Simultaneous Segmentation and Classification
As discussed previously, no matter whether it is
a contour-based segmentation or a graph-based one,
the same problem occurred. The balance between over-
segmentation and under-segmentation can be guided
through a parameter setting that is universal for all
kinds of objects.
Segmentation is making a partition in data, where
in each partition cell (i.e., segment), data has the prop-
erty of certain homogeneity. However, in facing
a complex environment that contains different kinds
of objects, different kinds of objects might have differ-
ent homogeneities. A segmentation method that applys
varied criteria for different object classes is required,
where the segmentation and classification issues are
considered in a unified form.
The processing flow is shown in Fig. 17b, where
a super-segmentation is conducted using the bottom-
up heuristics, such as scan-line segments and contour
points at different scales, and then the super-segments
are merged considering the factor of object class, which
is modeled as a problem of joint merger with
Dynamic Environment Sensing Using an Intelligent Vehicle 457
classification. In super-segmentation, scan line seg-
ments are first extracted [69, 70] and merged according
to their planarity; a region growing is then conducted
to merge the left points and isolated scan line segments
according to their spatial connectivity, stopping when
a contour point is met. Super-segmentation requires
that each segment be a partial observation of an object;
it should not be a mixture of different objects. There-
fore, strict criteria are used in the above region growing
procedures. In a merge procedure, a key problem is
defining a model that evaluates the homogeneities for
a variety of data segments. The probabilistic formula-
tion is described below.
Probabilistic Model Let si and sj be a pair of neigh-
boring segments with the label of yi and yj , respectively.
Let siþj denote the merged segment of si and sj , and yiþj
for its label. The probability for si merged with sj ,
i.e., Pðsiþj jIÞ, can be estimated as follows;
X
Pðsiþj jIÞ ¼ Pðyi ¼ ljIÞPðyj ¼ ljIÞPðsiþj jyiþj ¼ l; IÞ
l2L
Where, given data cues I, Pðyi ¼ ljIÞ and Pðyj ¼ ljIÞ
evaluate the probabilities of si and sj labeled to l,
respectively. Here, the classifier of both 3D laser
points and scan line segments is exploited.
Pðsiþj jyiþj ¼ l; IÞ is a likelihood measure, where, given
the knowledge of object class yiþj ¼ l and data cues I,
the probability of si and sj are the measurements to
a single object, i.e., si and sj are merged to siþj . The
likelihood measures with respect to different kinds of
objects are defined below.
Definition to Likelihood Measures As the object
class is given, an evaluation based on a prior knowledge
of the object class is required. They are defined exper-
imentally below.
1. y = building or road :
Normally only a partial surface of building or
road is measured, which could be modeled using
a planar surface, with a certain volume (e.g., e =
20 cm) representing the errors in laser range
measurements and modeling generalization. Let S
denote the total number of laser points in the range
segment, N the number of laser points within the
volume, and a the angle between the surface normal
458 Dynamic Environment Sensing Using an Intelligent Vehicle

102
8
λλl lU(d
U
7 (d11) ) Build
6
Road
5
Tree
4
Car
3
2 Human

1 Bush
0
−1.03 −1.04 −1.02 −1 2.53 2.55 2.34 4.92 .08 4.53 4.88

0.05 4 U
0.045 λλllU(d
U
(d22
) λl (d3)
3.5
0.04
3
0.035
0.03 2.5
0.25 2
0.02 1.5
0.015
1
0.01
0.000 0.5
0 0
−1.00 −1 0.11 0.0 0.15 .25 0.72 .07 −1.25 −1.2 −1.11 −1.1 −1.00 −1 −0.00 −0.5 −0.55 −0.0 −0.75

0.03 0.018
U
λl (d4)
λλl lU(d
U
0.016 (d55) )
0.25
0.014
0.02 0.012
0.01
0.015
0.008
0.01 0.006
0.004
0.000
0.002
0 0
−0.5 −1 −1.15 −0.2 −0.25 −0.3 −0.35 −0.4 −0.45 0.5 0.6 0.7 0.8 0.9 1 1.1

0.012 0.03
U U
λl (d6) λl (d7)
0.01 0.025

0.008 0.02

0.006 0.015

0.004 0.01

0.002 0.005

0 0
−1.1 −1 −0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 −0.74 −0.735 −0.73 −0.725 −0.72 −0.715 −0.70 −0.705

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 21

The likelihood measures for classifying a cloud of laser points [41]
 Dynamic Environment Sensing Using an Intelligent Vehicle 459

and a vertical normal vector ð0; 0; 1Þt . The likeli- N  cos a

Pðsjy ¼ road; IÞ /
hood measure is define S

N  sin a 2. y = car:
Pðsjy ¼ building ; IÞ / Normally data of a car can be restricted within
S
a cube, so that tW , tL , and tH are defined according
to the largest width, length, and height of a normal
car. A cubic model is used to fit on the laser points
Dynamic Environment Sensing Using an Intelligent of the segment and obtain a width (Wf ), a length
Vehicle. Table 3 The number of sample data in training (Lf ), and a height (Hf ). A step function is defined
for classifiers (see Fig. 23)
Scan line Segments of 3D Laser y ¼ f ðx; a1 ; a2 Þ
Class segments points
and
Building 9,394 96
Z a2 þe
Road 10,714 23
y dx
Tree 4,122 148 maxða1 e;0Þ

Car 6,080 41 The likelihood measure is defined

People 394 120
Pðsjy ¼ car; IÞ / f ðWf ; 0; tW Þ
Bush 1,176 39
 f ðLf ; 0; tL Þ  f ðHf ; 0; tH Þ
Bus 253 1
3. y = bus:
Total 32,133 468
Similar with that of a car.

Segmentation

Classifier 1

Classifier 2 Object 2

Object 5 Object 4
Classifier 3 Object 3 Object 1

Manu labeling

Building Road Tree Car Bush Bus Others Unlabeled

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 22

Comparative results of different classifiers
460 Dynamic Environment Sensing Using an Intelligent Vehicle

4. y = person: to fit on the laser points of the segment and obtain

Normally data of a person can be restricted a radius (rf ) and a height (hf ). The likelihood
within a cylinder, so that a radius (tr ) and measure is defined
a height threshold (th ) is defined according to the
tr th
size of a normal person. A cylindrical model is used Pðsjy ¼ person; IÞ / 
maxðtr ; rf Þ maxðth ; hf Þ

y ε
5. y = tree:
Normally the range segment of a tree consists
of many small line segments and edge points. Let S
denote the total number of laser points of the
range segment, and E the number of laser points
on scan line segments. The likelihood measure is
defined
α1 α2 x
E
Pðsjy ¼ tree; IÞ /
Dynamic Environment Sensing Using an Intelligent
S
Vehicle. Figure 23 6. y = bush:
A step function for likelihood evaluation [41] Similar with that of a car or bus.

Range image

Super segmentation

Segmentation A

C
Classification

B D

Manu labeling

Building Road Tree Car Bush Bus Others Unlabeled

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 24

Results of simultaneous segmentation and classification
 Dynamic Environment Sensing Using an Intelligent Vehicle 461

Range image

Super segmentation

Segmentation

Classification

Manu labeling

Building Road Tree Car Bush Bus Others Unlabeled

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 25

Results of simultaneous segmentation and classification [41]

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 26

A map of the environment represented by unlabeled 3D laser points of L4 and L5 [41]
462 Dynamic Environment Sensing Using an Intelligent Vehicle

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 27

A map of the environment represented by annotated 3D laser points of L4 and L5 through simultaneous segmentation
and classification [41]

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 28

A map of the environment represented by annotated 3D laser points of L4 and L5 through manual labeling [41]

Experimental Results Figures 24 and 25 show exper-
imental results of simultaneous segmentation and clas-
sification, where in each figure, the first row is an
original range image, and the second row shows the
segments after super-segmentation. The third and
fourth rows are simultaneously obtained results,
which are the segments after merge and their class of
the highest probability, respectively. For comparison,
manually labeled results are also demonstrated at the
last row. Generally speaking, the method has good

Dynamic Environment Sensing Using an Intelligent Vehicle. Figure 29
A map of the environment with the representation of both static and dynamic entities
Building
Building
A pedestrian and
its trajectory
Dynamic Environment Sensing Using an Intelligent
Vehicle. Figure 30
Rendered from a viewpoint on street
performance with a majority of objects, especially
large-scale ones, such as buildings and trees. Compared
with the results of contour-based segmentation, over-
segmentation in Fig. 24 is reduced greatly, and the
problem of under-segmentation (as Failure B in
Fig. 17) is solved in this result (highlighted by A in
Fig. 24). However, classification errors still exist, where
B indicates two cars that are misclassified as buildings,
C is trees that are also misclassified as buildings, and
Dynamic Environment Sensing Using an Intelligent Vehicle 463
D is clumps of bushes, however misclassified as cars.
Classification accuracy will be improved through
future work.
Contextual Map of Dynamic Environment
As discussed previously, a 3D coordinate can be
retrieved for each range pixel, and a range image can
be converted into a 3D representation using laser
points. Figure 27 shows a map representation of an
environment using the geo-reference 3D laser points
of L4 and L5, where colors denote different sensor data.
In such a map, laser points represent object geometry
but do not suggest any contextual knowledge of the
objects. Through segmentation and classification,
range segments, i.e., laser point clouds, are extracted
that represent the data of the same objects, and they are
annotated by certain object classes. After segmentation
and classification, the laser points of Fig. 26 are anno-
tated and shown in Fig. 27. Here, the color legend is the
same as that of Figs. 24 and 25. The result is compared
with a manually labeled one, as shown in Fig. 28, where
those laser points (i.e., range pixels) that are difficult in
manual discrimination (as the black points in Figs. 24
and 25) are not shown.
In geo-referencing laser measurements of Figs. 26
and 28, the vehicle position of Fig. 5 is used, which is an
464 Dynamic Environment Sensing Using an Intelligent Vehicle
output of module 2 –SLAMMODT by combinatory
exploration of GPS/IMU inputs and a horizontal laser
scanner L1. Except for vehicle position, a set of motion
trajectories is also obtained as the output of module 2,
which represents the motion of moving objects at
a horizontal level that are captured by L1. With the
result of Figs. 27 and 28, a 3D map representation of the
potentially dynamic objects, such as cars and pedes-
trians, as well as those of static ones in the environment
are generated. The motion trajectories from module 2
are the only direct measurements of the system to the
dynamic entities at the environment, however, they are
restricted within a horizontal plane. By fusing both of
the data as shown in Fig. 29, where the motion trajec-
tories are overlapped onto the 3D map as shown in
light blue points, those true dynamic entities at the
moment can be inferred. For example, there are almost
no light blue points associated with the parked cars in
Fig. 29, which reflect that the potentially moving
objects are static at the moment. A map is rendered in
Fig. 30 from a street viewpoint, where a pedestrian as
well as its motion is highlighted. The 3D shape of the
pedestrian is vividly represented in the 3D map. A long,
light blue tail can be associated with the pedestrian,
which reflects its walking path. In addition, there is
a building entrance nearby where the pedestrian is
heading, suggesting a possible reason for such
a motion.
Future Directions
In order to generate an omni-directional range sensing
of a dynamic urban scene using an intelligent vehicle,
a system and algorithmic developments for the funda-
mental issues of multi-laser sensor system calibration,
and scene understanding in contextual map generation
are presented, as well as experimental examinations.
Although there are still open problems left in each of
the key issues, this research intends to demonstrate that
a geometric and contextual representation of static
objects such as buildings, trees, and roads, as well as
the motion at the moment of dynamic entities such as
people, bicycles, and cars can be obtained using
a vehicle robot system.
It is not only that such a representation is helpful in
a robot’s online perception and decision, but it is also
useful in a broad range of potential applications. A city
has not only static objects, such as buildings, trees, and
roads, but also moving objects, such as people and cars.
A study to analyze the interactions between static and
dynamic objects is only available when a sensing
technique is able to generate a copy of their simulta-
neous states. Knowledge of the statistical properties of
the relationship between static and dynamic objects is
important for an advanced online perception, such as
surveillance, or offline systems, such as behavior
analysis, architectural design, security planning, and
so on. This research focused on a sensing system
using an intelligent vehicle, where a 3D map represen-
tation of both the static and dynamic entities as well as
motions in a dynamic environment can be generated.
However, how to best make use of such a sensing tech-
nology for potential applications needs further
research. The data sets that are generated in this
research are in [78].
Bibliography
1. International Society of Photogrammetry and Remote Sens-
ing. http://www.isprs.org/
2. Collins R, Hanson A, Riseman E (1994) Site model acquisition
under the UMass RADIUS project. In: Proceedings of arpa
image understanding workshop, Monterey, CA, pp 351–358
3. Gruen A (1998) TOBAGO – a semi-automated approach for the
generation of 3-D building models. ISPRS J Photogramm
Remote Sens 53(2):108–118
4. Forstner W (1999) 3D-city models: automatic and semiauto-
matic acquisition methods. In: Proceedings photogrammetric
week, University of Stuttgart, Institute for photogrammetry,
pp 291–303
5. Shiode N (2001) 3D urban model: recent developments in the
digital modeling of urban environments in three-dimensions.
GeoJournal 52(3):263–269
6. Ellum C, El-Sheimy N (2000) The development of a backpack
mobile mapping system. Int Arch Photogramm Remote
Sensing XXXII(B2):184–191, Amsterdam
7. He G, Orvets G (2000) Capturing road network data using
mobile mapping technology. Int Arch Photogramm Remote
Sensing XXXIII(B2):272–277, Amsterdam
8. Silva JFC, Camargo PO, Oliveira RA (2000) A street map built by
a mobile mapping system. Int Arch Photogramm Remote
Sensing XXXIII(B2):510–517, Amsterdam
9. El-Sheimy N (2000) Mobile multi-sensor systems: the new
trend in mapping and GIS applications. IAG J Geodesy, vol.
121, Geodesy beyond 2000: the challenges of the first decade.
Springer, Berlin/Heidelberg, pp 319–324
10. Li R (1997) Mobile mapping: an emerging technology for
spatial data acquisition. Photogramm Eng Remote Sens
63(9):1085–1092

11. Früh C, Zakhor A (2004) An automated method for large-scale,
ground-based city model acquisition. Int J Comput Vis
60(1):5–24
12. Ikeuchi K, Sakauchi M, Kawasaki H, Sato I (2004) Constructing
virtual cities by using panoramic images. Int J Comput Vis
53(3):237–247
13. Zhao H, Shibasaki R (2003) Special issue on computer vision
system: reconstructing textured CAD model of urban environ-
ment using vehicle-borne laser range scanners and line cam-
eras. Mach Vis Appl 14(1):35–41
14. Google Earth (2004) http://earth.google.com
15. Microsoft Virtual Earth (2006) http://www.microsoft.com/
virtualearth
16. Google StreetView (2007) http://maps.google.com/help/
maps/streetview
17. StreetMapper (2007) http://www.streetmapper.net
18. City Grid (2006) http://www.cybercity.tv
19. Cyber City (2007) http://www.cybercity.tv
20. Hu J, You S, Neumann U (2003) Approaches to large-scale urban
modeling. IEEE Comput Graph Appl 23(6):62–69
21. Thrun S (2002) Robotic mapping: a survey. CMU-CS-02-11
22. DARPA (2004) DARPA grand challenge rulebook. http://www.
darpa.mil/grandchallenge05/Rules_8oct04.pdf
23. DARPA (2006) DARPA grand challenge rulebook. http://
www.darpa.mil/grandchallenge/docs/Urban_Challenge_Rules_
121106.pdf
24. Journal of Field Robotics: Special issue on the 2007 DARPA
urban challenge, Part I 25(8)
25. Journal of Field Robotics: Special issue on the 2007 DARPA
urban challenge, Part II 25(9)
26. Nuchter A, Lingemann K, Hertzberg J, Surmann H (2007) 6D
SLAM – 3D mapping outdoor environments. J Field Robot
24(8/9):699–722
27. Zhao H, Shibasaki R (2003) A vehicle-borne urban 3D acquisi-
tion system using single-row laser range scanners. IEEE Trans
SMC Part B: Cybern 33–4:658–666
28. Allen P, Atamos I, Gueorguiev A, Gold E, Blaer P (2001) AVE-
NUE: automated site modeling in urban environments. In:
Proceedings of the 3rd international conference on 3D digital
imaging and modeling, Quebec City, pp 357–364
29. Georgiev A, Allen PK (2004) Localization methods for a mobile
robot in urban environments. IEEE Trans Robot Automat (TRO)
20(5):851–864
30. Hahnel D, Burgard W, Fox D, Thrun S (2003) An efficient
FastSLAM algorithm for generating maps of large-scale cyclic
environments from raw laser range measurements. In: Pro-
ceedings of IEEE/RSJ international conference on intelligent
robots and systems, Las Vegas, pp 206–211
31. Wang CC (2004) Simultaneous localization, mapping and
moving object tracking. PhD dissertation, Carnegie Mellon
University, CMU-RI-TR-04-23
32. Vu T, Aycard O, Appenrodt N (2007) Online localization and
mapping with moving object tracking in dynamic outdoor
environment. In: Proceedings IEEE intelligent vehicle sympo-
sium, Istanbul, pp 190–195
Dynamic Environment Sensing Using an Intelligent Vehicle 465
33. Weiss T, Schiele B, Dietmayer K (2007) Robust driving path
detection in urban and highway scenarios using a laser scan-
ner and online occupancy grids. In: Proceedings of the IEEE
intelligent vehicle symposium, Istanbul, pp 184–189
34. Zhao H, Chiba M, Shibasaki R, Shao X, Cui J, Zha H (2008) SLAM
in a dynamic large outdoor environment using a laser scanner.
In: Proceedings of the IEEE international conference on robot-
ics and automation (ICRA), Pasadena, CA, pp 1455–1462
35. Velodyne HDL-64E (2007) http://www.velodyne.com/lidar/
products/overview.aspx
36. SICK LMS2** (2001) http://www.sick.com
37. Durrant-Whyte H, Bailey T (2006) Simultaneous localization
and mapping: Part I. IEEE Rob Autom Mag 13(2):99–108
38. Bailey T, Durrant-Whyte H (2006) Simultaneous localization
and mapping: Part II. IEEE Rob Autom Mag 13(3):108–117
39. Streller D, Dietmayer K (2004) Object tracking and classifica-
tion using a multiple hypothesis approach. In: Proceedings of
the IEEE intelligent vehicles symposium, Parma, pp 808–812
40. Fayad F, Cherfaoui V (2007) Tracking objects using a laser
scanner in driving situation based on modeling target shape.
In: Proceedings of the IEEE intelligent vehicle symposium,
Istanbul, pp 44–49
41. Zhao H, Liu Y, Zhu X, Zhao Y, Zha H (2010) Scene understand-
ing in a large dynamic environment through a laser-based
sensing.In: IEEE international conference on robotics and
automation, Anchorage, pp 127–133
42. Zhao H, Xiong L, Jiao Z, Cui J, Zha H (2009) Sensor alignment
towards an omni-directional measurement using an intelli-
gent vehicle. In: Proceedings of the IEEE intelligent vehicle
symposium, Kobe, pp 292–298
43. Zhao H, Shibasaki R (2001) High accurate positioning and
mapping in urban area using laser range scanner. In: Proceed-
ings of IEEE intelligent vehicles symposium, Tokyo,
pp 125–132
44. Zhao H, Chiba M, Shibasaki R, Shao X, Cui J, Zha H (2009)
A laser scanner based approach towards driving safety and
traffic data collection. IEEE Trans Intell Transport Syst
10(3):534–546
45. Cheok GS, Leigh S, Rukhin A (2002) Calibration experiments of
a laser scanners. NISTIR 6922:121
46. Santala J, Joala V (2003) On the calibration of a ground-based
laser scanner. TS12.4, FIG working week
47. Mahlisch M, Hering R, Ritter W, Dietmayer K (2006) Heteroge-
neous fusion of video, Lidar, ESP data for automotive ACC
vehicle tracking. In: Proceedings of IEEE international confer-
ence on multisensor fusion and integration for intelligent
systems, Heidelberg, pp 139–144
48. Rodriguez F, Sergio A, Fremont V, Bonnifait P (2008) Extrinsic
calibration between a multi-layer LIDAR and a camera. In: Pro-
ceedings of IEEE international conference on multisensor fusion
and integration for intelligent systems, Seoul, pp 214–219
49. Zhao H, Chen Y, Shibasaki R (2007) An efficient extrinsic cali-
bration of a multiple laser scanners and cameras’ sensor sys-
tem on a mobile platform. In: IEEE intelligent vehicles
symposium, Istanbul, pp 422–427
466 Dynamic Environment Sensing Using an Intelligent Vehicle
50. Gao C, Spletzer JR (2010) On-line calibration of multiple
LIDARs on a mobile vehicle platform. In: Proceedings of IEEE
international conference on robotics and automation,
Anchorage, pp 279–284
51. Besl PJ, McKay ND (1992) A method for registration of 3-D
shape. IEEE Trans Pattern Anal Mach Intell 14:239–256
52. Chen Y, Medion G (1992) Object modeling by registration of
multiple range images. Image Vis Comput 10(3):145–155
53. Zhang Z (1994) Iterative point matching for registration
of a free-from curves and surfaces. Int J Comput Vis
13:119–152
54. Hahnel D, Burgard W, Thrun S (2003) Learning compact 3d
models of indoor and outdoor environments with a mobile
robot. Rob Autom Syst 44:15–27
55. Althaus P, Christensen H (2003) Behavior coordination in struc-
tured environments. Adv Robot 17(7):657–674
56. Mendes A, Nunes U (2004) Situation-based multi-target detec-
tion and tracking with laserscanner in outdoor semi-
structured environment. In: Proceedings IEEE/RSJ interna-
tional conference on intelligent robots and systems, Sendai
57. Posner I, Cummins M, Newman P (2008) Online generation of
scene descriptions in urban environments. Rob Autom Syst
56(11):901–914
58. Douillard B (2009) Vision and laser based classification in
urban environments. PhD thesis, University of Sydney
59. Nuchter A, Lingemann K, Hertzberg J, Surmann H (2009) 6D
SLAM – 3D mapping outdoor environments. J Field Robot
24(8–9):699–722
60. Martinez-Mozos O, Stachniss C, Burgard W (2005) Supervised
learning of places from range data using AdaBoost. In: IEEE
international conference on robotics and automation (ICRA),
Barcelona, pp 1742–1747
61. Triebel R, Kersting K, Burgard W (2006) Robust 3D scan point
classification using associative Markov networks. In: IEEE inter-
national conference on robotics and automation (ICRA),
Orlando, FL
62. Douillard B, Fox D, Ramos FT (2008) Laser and vision based
outdoor object mapping. In: Proceedings of robotics: science
and systems, Zurich
63. Posner I, Cummins M, Newman P (2009) A generative frame-
work for fast urban labeling using spatial and temporal con-
text. Auton Robot 26(2–3):153–170
64. Hoover A, Jean-Baptiste G, Jiang X, Flynn PJ, Bunke H, Goldgof
D, Bowyer K (1996) A comparison of range image
segmentation algorithms. IEEE Trans Pattern Anal Mach Intell
18(7):673–689
65. Katsoulas D, Bastidas CC, Kosmopoulos D (2008) Superquadic
segmentation in range image via fusion of region and
boundary information. IEEE Trans Pattern Anal Mach Intell
20(5):781–795, 2008
66. Weingarten J, Siegwart R (2005) EKF-based 3D SLAM for
structured environment. In: IEEE/RSJ international conference
on intelligent robots and systems, Edmonton, Alberta,
pp 2089–2094
67. Han F, Tu Z, Zhu SC (2004) Range image segmentation by an
effective jump-diffusion method. IEEE Trans Pattern Anal
Mach Intell 26(9):1138–1153
68. Borenstein E, Ullman S (2008) Combined top-down/bottom-up
segmentation. IEEE Trans Pattern Anal Mach Intell
30(12):2109–2125
69. Malisiewicz T, Efros AA (2008) Recognition by association via
learning per-exemplar distances. In: Proceedings of the IEEE
conference on computer vision and pattern recognition
(CVPR), Anchorage
70. Porway J, Wange K, Zhu SC (2008) A hierarchical and contex-
tual model for aerial image understanding. In: Proceedings of
the IEEE international conference on computer vision and
pattern recognition (CVPR), Anchorage, pp 1–8
71. Tu Z, Chen X, Yuille A, Zhu SC (2005) Image parsing: unifying
segmentation, detection and recognition. Int J Comput Vis
63(2):113–140
72. Golovinskiy A, Kim VG, Funkhouser T (2009) Shape-based
recognition of 3D point clouds in urban environments. In:
IEEE international conference on computer vision, Kyoto, pp
2154–2161
73. Shi J, Malik J (2000) Normalized cuts and image segmentation.
IEEE Trans Pattern Anal Mach Intell 22:888–905
74. Felzenszwalb PF, Huttenlocher DP (2004) Efficient graph-
based image segmentation. Int J Comput Vis 59:167–181
75. Jiang X, Bunke H (1994) Fast segmentation of range images into
planar regions by scan line grouping. Mach Vis Appl 7:115–122
76. Rosin PL, West GAW (1995) Nonparametric segmentation of
curves into various representations. IEEE Trans Pattern Anal
Mach Intell 17:1140–1153. http://users.cs.cf.ac.uk/Paul.Rosin
77. Chang CC, Lin CJ (2001) LIBSVM: a library for support vector
machines. http://www.csie.ntu.edu.tw/ cjlin/libsvm
78. PKU Omni Smart Sensing – POSS (2010) http://www.poss.pku.
edu.cn

Electric, Hybrid Electric and Fuel Cell
Vehicles, Architectures of
YIMIN GAO
Department of Electrical and Computer Engineering,
Texas A&M University, College Station, TX, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Architectures of Electric Vehicles
Architectures of Hybrid Vehicles
Architectures of Fuel Cell Vehicles
Future Directions
Bibliography
Glossary
Architecture The mechanical and/or electrical
structure that defines the energy flow routes in a
vehicle.
Electric coupling The method by which two electric
powers are added together.
Electric vehicle An electric vehicle is referred to as a
vehicle that uses electric energy as its energy
source, generally chemical batteries and/or
ultracapacitors and electric motor as its power
source.
Fuel cell vehicle A fuel cell vehicle is referred to as
a vehicle that uses fuel cells and electric motor as its
power source.
Hybrid electric vehicle A hybrid electric vehicle is
referred to as a vehicle in which two energy sources
and correspondingly two power sources are
installed, which are selectively used to power the
vehicle. The primary energy source and power
source may be chemical fuels and internal combus-
tion engine. The secondary energy source and
power source may be chemical batteries and electric
motor.
Mechanical coupling The method by which two
mechanical powers are added together.
Operation mode Drive train operation status in which
power sources are selectively used as the active
power mover to propel the vehicle.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 467
Definition of the Subject
The architecture design of an electric vehicle is to
design its energy flow route from the energy source to
the energy destination. Usually, chemical batteries are
used as its energy source and electric motors are used to
produce mechanical power to propel the vehicle. For
satisfying vehicle performance, the batteries, electric
motor, and transmission should be properly designed.
A hybrid electric vehicle uses two power sources to
power the vehicle. There are many methods to connect
the two power sources together, each of which has its
special operation characteristics. For different vehicles
that have different mission requirements and operation
environments, a special architecture should be used to
fully use its advantages and avoid its shortcomings.
Fuel cell vehicle is the vehicle which is powered by
a fuel cell system. Compared with internal combustion
engine, fuel cell is more efficient and cleaner. Architec-
tural design of a fuel cell vehicle consists in designing
the drive train structure and components to meet the
vehicle performance requirements.
Introduction
Electric vehicle (EV) is referred to as a vehicle that
employs electric energy storage as its energy source
and electric machines as its power source. Electric
energy is a multisource energy type which can be
obtained from many primary energy sources, such as
traditional fossil energy sources (coal, petroleum, and
natural gas), nuclear energy, hydraulic power, bioenergy,
solar energy, wind energy, etc. With continuous deple-
tion of petroleum resources and more and more con-
cerns about environmental issues, it has been well
recognized that electric vehicle is one the most viable
substitutes to the current petroleum-fueled vehicle. Elec-
tric vehicle also possesses other advantages of absence
emissions, high efficiency, quiet and smooth operation,
etc. With continuous progress on the technologies of
chemical batteries, electric propulsion systems, and elec-
tronic control, electric vehicles are much closer to meet-
ing a user’s requirements than ever before.
Compared with petroleum fuels, the energy density
of chemical batteries is much lower, which results in
shorter driving range per battery charge. Long time
battery charging also causes incontinence to users.
# 2012, DOI 10.1007/978-1-4419-0851-3
468 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
At present, the battery-powered electric vehicles cannot
challenge the petroleum-powered vehicles in terms of
performance and use convenience. Hybrid electric
vehicles are the combination of conventional vehicles
and battery-powered electric vehicles. They can adopt
the advantages and avoid the shortcomings of both.
It has been recognized that hybrid electric vehicles,
especially plug-in hybrid vehicles, are the major sub-
stitutes of conventional vehicles before technologies of
chemical batteries have substantive progress in energy
density, safety, and cost.
Fuel cell vehicles have been considered to be the
final solution as the future high-efficiency and clean
vehicles. Compared with conventional vehicles, fuel cell
vehicles are independent from the petroleum resource
and potential has high efficiency and zero emission.
Compared with battery-powered electric vehicles,
they potentially have the advantages of long driving
range and quick fuel replenishing. However, at present,
the hydrogen sources and onboard storage are the
major challenges.
Architectures of Electric Vehicles
In principle, electric vehicles share the common char-
acteristics of internal combustion engine–powered
vehicles. A chemical battery pack replaces the fuel
tank as its energy storage. Electric machine replaces
the internal combustion engine as the power source.
Both share the similar performance requirements, such
as, speed, acceleration, gradeability, and driving range.
The design principles as discussed in the chapter of
Mechanical
transmission
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 1
Primary electric vehicle drive train
▶ Vehicle Dynamics and Performance are still valid
for electric vehicles. However, unique operating char-
acteristics of the electric propulsion system have to be
involved into the design process.
Previously, electric vehicles were designed to have
a similar drive train structure to conventional internal
combustion engine–powered vehicles as shown in
Fig. 1. It simply replaces the engine with an electric
motor and fuel tank with a chemical battery pack. With
development of the technologies of light weight, com-
pact electric drive system, and advanced control, an
electric vehicle drive train can be designed to various
configurations as shown in Fig. 2 [1].
Figure 2a shows the first alternative configuration
that is most alike a conventional one. An electric
motor (EM) drive replaces the internal combustion
engine. A clutch is used to connect the motor drive to
or disconnect from the final drive and differential.
A transmission gear box (GB) is used to modify the
output speed–torque profile for meeting the perfor-
mance requirement and supply a reverse gear. This
drive train may be used for the motor drives which
are only capable of operating in one or two quadrants,
forward motoring and/or forward regeneration. Since
a transmission gearbox is used, the torque-speed pro-
file may not be required to close the ideal profile,
which has constant torque in low speed range and
constant power in high speed range. A motor drive
with simple control may meet the requirement, such
as series-excited DC motor.
Similar to the configuration shown in Fig. 2a, b
shows the configuration in which clutch is not further
Electric
motor
drive
Electric
energy
storage

Batt. MC EM GB Diff.
a b
Diff.
FGB
Batt. MC EM
c d
FGB
MC M
Batt.
MC M
FGB
e f
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 2
Possible electric vehicle configurations [1]
needed. The motor starts producing its torque
from zero speed. The transmission gear box may be
either multi-gear or single gear, depending on the
motor torque-speed profile. With a multi-gear trans-
mission, the motor torque needs to be stopped during
shifting. This can be carried out by cutting the current
supply to the motor drive by the motor controller.
Figure 2c illustrates a configuration with a more
compact design. A fixed gear ratio single gear transmis-
sion (FGB) is installed between the motor drive and the
final drive (differential). This design requires the motor
drive to be capable of producing a torque-speed profile
which is closer to the ideal one. It also requires the
motor be capable of operating in four quadrants, that
is, forward and backward motoring and generating.
Figure 2d illustrates a configuration in which two
motor drives are used. Each motor drives an individual
wheel through a fixed gear ratio single gear box (FGB)
which is used to match the motor speed to the wheel
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 469
Batt. MC EM GB Diff.
FGB
Batt. MC M
M
MC
FGB
M
MC
Batt.
MC
M
speed. A differential function is carried out by control-
ling the two motors. One advantage of this configura-
tion is that it can fully utilize the road adhesive
capability to improve the drivability of the vehicle.
For example, when one wheel is on a slip road and
loses its drive force, another wheel can produce a big
driving force to help the vehicle get out of trouble.
The drive train shown in Fig. 2e is very similar to
the one in Fig. 2d. The difference is that the two fixed
gear ratio single gear boxes (FGB) are attached to
wheels rather than to the vehicle frame. This design
can save much space and may make packaging easier.
Figure 2f shows a simplest drive train structure, in
which the wheel is designed as the rotor of the motor. This
configuration is the most compact one and saves a lot
space. This structure is generally termed as wheel motor.
In the design of the drive trains shown in Fig. 2e, f, the
impact of additional wheel weight to vehicle handing and
vibration performance should be considered.
470 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of

In all the architectures, the electric motors are
powered by a battery pack and corresponding motor
controller(s).
Except electric motor(s) and corresponding trans-
missions, a modern electric vehicle also needs other
components for fulfilling its functions as shown in
Fig. 3. The drive train consists of three major subsys-
tems: electric propulsion subsystem, energy storage
subsystem, and auxiliary subsystem. These three sub-
systems are interfaced through power routes and con-
trol signal routes. The electric energy from power grid
is charged to the energy storage (batteries) through an
energy charge system (battery charger). The energy
storage delivers its energy to or absorbs energy
from the electric motor through an electronic power
converter which functions as a motor controller.
The electric motor operation is controlled by a vehicle
controller which commands the torque requirement to
the motor controller based on the accelerator or brake
pedal position and energy source operation parameters
and energy management algorithm. The energy source
also supplies energy to the vehicle auxiliaries. A battery
management system (BMS) is used to monitor the
operation of the battery pack, generate operation infor-
mation, such as state-of-charge (SOC), and protect
battery pack from abusing.
Architectures of Hybrid Vehicles
Hybrid vehicles (HV) are referred to as the vehicles that
are powered by at least two energy sources and two
power sources, in which at least one energy source and
one power source have bidirectional energy flow capa-
bility, delivering energy to drive wheels and accepting
energy from braking. When electric energy source and
electric power source are used, they are referred to as
hybrid electric vehicles (HEV). It is possible to replace
the electric system with a hydraulic system. This kind of
hybrid vehicle is referred to as hydraulic hybrid vehicles
(HHV). Fuel cell vehicles (FCV) are usually referred to

Electric propulsion subsystem wheel

Brake

Electronic
Vehicle Electric Mechanical
power
controller motor transmission
converter
Accelerator

wheel

Energy Auxiliary Power

management Batteries power steering
unit and BMS supply unit
Steering
wheel
Energy Auxiliary
charge power units
unit
Energy source
subsystem Auxiliary subsystem

Mechanical link
Electric link
Control link

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 3
Conceptual illustration of an electric vehicle configuration [1]
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 471

the vehicles in which fuel cells are used as the primacy
power source. A fuel cell vehicle may use two energy
sources, fuel cells and chemical batteries, which is
referred to as fuel cell hybrid vehicles (FCHV).
Since there are two power sources and only one
energy sink (wheels), there are many ways to combine
the two powers together. The design of the energy flow
routes are referred to as drive train architecture design.
Control of the energy flows in hybrid drive train is
referred to as vehicle control strategy.
Concept of Hybrid Vehicle
In a hybrid vehicle, two power sources supply powers to
drive wheels. The total power should meet the vehicle
performance requirements, in terms of maximum speed,
gradeability, acceleration time requirement, and driving
range. The vehicle performance requires the vehicle
possessing sufficient power, and sufficient energy for
continuous driving range. In operation, any one of the
two power sources or both can be used for traction.
Thus, there are many possible operation modes as
shown in Fig. 4 [2].
The primary energy source and energy converter
allows unidirectional energy flow, whereas, the
secondary energy source and energy converter are
required to have the capability of bidirectional energy
flow. With today’s technologies, the common primary
energy sources are chemical fuels and primary energy
converters are internal combustion engines (ICE)
and fuel cells. The secondary energy sources are
chemical batteries and/or ultracapacitors and the
secondary energy converters are electric motors. The
final destination of the energy flow (energy sink)
is the vehicle wheels with the form of mechanical energy.
As shown in Fig. 4, the wheels can be powered by
selecting power sources, which are referred to as oper-
ation modes, which include:
1. Primary power source alone propelling
2. Secondary power source alone propelling
3. Both power sources propelling (hybrid drive)
4. Regenerative braking (secondary power source
absorbs energy form load)
5. Primary power source splitting for propelling and
secondary power source charging
So many available operation modes create higher
flexibility over single power source vehicles. It provides
the drive train with opportunities for maximizing its
operation efficiency and/or minimizing its emissions at
any driving conditions by selectively using a best oper-
ation mode.
Vehicle load in normal driving varies randomly as
shown in Fig. 5. Conceptually, the random load is com-
posed of two components, one is the steady power
(average) and the other is the dynamic power with zero
average [2]. Ideally, hybrid vehicle can be designed and
controlled in such a way that one power source supplies
a constant power for the average load, which is referred to
as energy source, since all the energy comes from it, and
the other power source supplies dynamic power, which is
named peaking power source, since it consumes zero
energy in a whole driving cycle.
In normal driving, average power is relative small
(few 10 kW). Thus, the primary power source can be

Unidirectional
Primary energy
Primary energy
converter Load
source
(power source)

+
Unidirectional
Secondary energy
Secondary
converter
energy source
(power source)

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 4
Concept of hybrid electric vehicles
472 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
Load power
Average power
0 0
= +
Time
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 5
Steady power and dynamic power [2]
designed with a small power capacity. When the inter-
nal combustion engine is used as a primary power
source, it can be downsized and operated near its
most efficient region. In addition, other types of
engines may be used, such as Sterling engine, gas tur-
bine engine, and others. The secondary power source
may be batteries, ultracapacitors, flywheels and their
combination, and electric motors.
Architectures of Hybrid Vehicles
Architecture of a hybrid vehicle is loosely defined as the
ways by which the components of a hybrid drive train
are connected and, therefore, the energy flow routes are
schemed. Traditionally, hybrid vehicles are classified
into series, parallel and series/parallel. However, the
definitions of these configurations have not been clear.
As discussed in a prior section, a hybrid vehicle has two
power sources and one energy sink. Thus, there has to
exist at least point at which two powers are merged
together. At this point, the power forms have to be
identical, mechanical or electrical. In this chapter, the
terms of electric coupling and mechanical coupling are
used rather than the traditional names. Figure 6 dem-
onstrates the concepts of four configurations.
Figure 6a shows the configuration of an electric
coupling, traditionally, a series configuration. Two
powers are merged together at the power converter in
the form of electrical. An internal combustion engine
drives an electrical generator to supply electric power
to the electronic converter, and at the same time, the
battery pack also contributes its power to the power
converter. The output power from the power converter
flows to the electric motor to produce mechanical
power and then drives vehicle wheels. The power shares
are controlled by the power converter.
Figure 6b shows the configuration of mechanical
coupling, traditionally, a parallel configuration. The
internal combustion engine directly contributes its
Dynamic power
0
Time Time
mechanical power to a mechanical coupler. At the same
time, the battery pack supplies its power to the electric
motor through a power converter which functions as the
motor controller for controlling the motor torque. Two
mechanical powers are merged together at the mechan-
ical coupling and then delivered to drive wheels.
Figure 6c shows the configuration in which three
power couplers are used, two mechanical and one elec-
trical. There are plentiful available operation modes in
the configuration that will be discussed in detail in
following sections.
Figure 6d shows a configuration which is similar to
Fig. 6c. The only difference is that the electric coupling
function is moved from the power converter to the
batteries and one more power converter is added
between the motor/generator and the batteries [2].
Electric Coupling HEV (Series)
The electric coupling configuration (series) is schemat-
ically shown in Fig. 7, on which, the possible power
flow routes are also shown. This drive train may have
been developed originally from pure electric vehicles by
adding a “range extender” engine/generator set. The
engine/generator unit is used to contribute part of the
electric energy for compensation of battery energy
shortage. An electric power converter is used for energy
management purpose based on the states of charge
(SOC) of the battery pack. As shown in Fig. 7, there
are several operation modes available:
1. Pure electric propelling mode, in which the engine/
generator is shut down and only batteries are used
to propel the vehicle.
2. Pure engine/generator traction, in which batteries
are isolated from the power supply route and
engine/generator power follows the load power.
This operation mode is also used by diesel-electric
locomotives
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 473

Electrically coupling (Series)

Fuel IC
tank engine Mechanical coupling (Parallel)

Fuel IC
Gene- tank engine
rator Mech.
coupler
Power Electric
Power Battery
converter motor
converter Electric
Battery
(elec. motor
coupler) b
a

Mechanical and electric coupling

(Series-parallel)
Mechanical and electric coupling (Complex)
Fuel IC Mech.
Fuel IC Mech
tank engine coupler .
tank engine coupler

Motor /
Mech. Power Motor /
Generator Mech.
coupler Converter generator coupler

Power
converter Electric
Battery Battery Power Electric
(elec. motor
(elec.coupler) converter motor
c coupler) d
Eletrical link Hydraulic link Mechanical link

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 6
Various configurations of hybrid vehicles [1]

Batteries

Electric
Engine Motor Trans-
Generator coupling motor
controller mission
device

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 7
Electric coupling (series) hybrid drive train configuration and power flow scenario

3. Hybrid propelling mode, in which both batteries
and engine/generator supply power to the traction
motor. The power shares between batteries and the
engine/generator are controlled by coordinate con-
trols of the power electronic converter and engine/
generator.
4. Engine/generator power split mode, in which the
power of engine/generator is separated by the
power converter, driving wheels and charging
batteries.
5. Regenerative braking mode, in which the electric
motor is operated to function as electric braking to
474 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
slow down the vehicle and get the braking energy
back to the batteries. The engine/generator may be
shut down meanwhile.
6. Battery charging mode, in which the engine/gener-
ator is operated only for charging batteries while
vehicle stop.
7. Hybrid battery charging mode, in which both the
engine/generator and the regenerative braking
charge the batteries.
Electric Coupling Devices
The electric coupling device is the sole device at which
two electrical powers are combined together. The oper-
ation modes discussed above can be carried out by
coordinately controlling the electric coupling device
and engine/generator. The simplest structure may be
to directly connect the three terminals together as
shown in Fig. 8.
The configuration in Fig. 8 has the features of
identical voltage at the coupling point. The DC bus
voltage is equal to the battery voltage. The power flows
are only controlled by the voltage of the generator/
rectifier, which depends on the operation speed of the
generator and magnetic field. For battery charging, the
engine/generator voltage has to be higher than the open
circuit voltage of the batteries. In regenerative braking,
the bus voltage has to be higher than the battery volt-
age. However, the voltage generated by the motor in
generating mode is proportional to the motor speed.
command
Vehicle
controller
+
Engine Generator/
rectifier
–
+ –
Batteries
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 8
Directly connected power sources and motor drive
Thus, the regenerative braking capability at low speeds
may be rather limited.
A DC/DC converter can be added between the
batteries and the DC bus to release the voltage con-
straint as shown Fig. 9. This design may significantly
improve the performance of the drive train. It has
the following advantages: (1) variations in the battery
voltage do not affect the bus voltage, thus the
engine/generator can be operated at a constant speed;
(2) the battery energy can be fully used, which is very
important for a plug-in hybrid drive train; (3) low
battery voltage and less battery cells can be used,
which may lead to a small and light battery pack, low
cost and high reliability; and (4) charging current of the
batteries can be regulated during regenerative braking
and charging from the engine/generator.
It is obvious that the DC/DC converter has to be
bidirectional. It is considered to be the appropriate
design with battery voltage lower than DC bus voltage.
It boosts the battery voltage to match the DC bus
voltage for propelling. On the other hand, it bucks
the DC bus voltage to the level of the battery voltage
for battery charging. In the situation of regenerative
braking with the generated voltage by the motor higher
than the battery voltage, the DC/DC operates in
buck mode. However, at low speeds, the DC bus may
be lower than the battery voltage. In this case, the
DC/DC converter has to be operated in boost
mode. The DC/DC operation mode is illustrated in
Table 1 [2].
Driving
Battery SOC
Motor controller Motor
(Power electronics)
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 475

Driving
command

Battery SOC
Vehicle
controller

Batteries

DC/DC

+ +
Engine Generator Motor controller Motor
/rectifier – (Power electronics)
−

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 9
Configuration with a DC/DC converter

Electric, Hybrid Electric and Fuel Cell Vehicles, Architec- Fig. 11. Switch S1 is operated on and off periodically.
tures of. Table 1 Basic function a DC/DC converter [2] During the on period of S1, the batteries supply current
Energy flow to the DC bus through S1, Ld, and D3 as shown in
Fig. 11a, and during the off period, the batteries are
Battery Battery
discharging charging isolated from the DC bus, and the current continuously
flows through D2, Ld, and D3. The bus voltage depends
Battery alone traction Boost –
mode
on the duty cycle d, of S1, which is defined as the
ratio of the on period, ton, to whole period
Battery charging from – Buck
T ðT ¼ ton þ toff Þ. When d = 0, DC bus voltage is
engine/generator
zero, and when d = 1, DC bus voltage equals to the
Regenerative braking – Buck/boost battery voltage:
Vbus ¼ d Vb : ð1Þ
The pulse width modulation (PWM) method is
used to control the on and off operation of S1 to obtain
Figure 10 shows the topology of the four-quadrant the desired DC bus voltage. This case may not be used
DC/DC converter, operating in forward buck/boost with a design that the battery voltage is lower than the
and backward buck/boost by turning on or off the DC bus voltage.
switches S1, S2, S3, and S4. The operation of the DC/
DC converter is explained below.
Forward Boosting When battery voltage is lower
Forward Bucking Forward operation is defined as than the DC bus voltage, the DC/DC converter is oper-
the energy flow direction from batteries to DC bus. ated for boosting the battery voltage for supplying
Forward bucking operation is that the DC/DC con- current to the DC bus. In this case, switch S1 is on, S2
verter supplies current to the DC bus which has and S3 are off, and S4 is operated on and off periodically
a voltage lower than battery voltage. In this operation, as shown in Fig. 12. During the on period of S4, batte-
switches S2, S3, and S4 are turned off as shown in ries charge the inductor Ld through S1 and S4 as shown
476 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of

Rectifier
+

Engine/
generator

Inverter
DC/DC –
converter Bus
voltage

+ S1 D1 S3 D3
Ld +
Battery

A Traction
VDC B
C motor
C –
– S2 D2 S4 D4

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 10
Bidirectional DC/DC converter [2]

S1 D1 S3 D3 S1 D1 S3 D3
+ +

Ld Ld
Vb Vb

– S2 D2 S4 D4 – S2 D2 S4 D4

a b

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 11
Forward bucking operation

S1 D1 S3 D3 S1 D1 S3 D3
+ +

Vb Ld Vb Ld

– S2 S4 –
D2 D4 S2 D2 S4 D4

a b

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 12
Forward boosting operation
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 477

S1 S3 S1 S3
+ D1 D3 + D1 D3

Vb Ld Vb Ld

– S2 S4 –
D2 D4 S2 D2 S4 D4

a b

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 13
Backward bucking operation

S1 D1 S3 D3 S1 S3
+ + D1 D3

Vb Ld Vb Ld

– S2 S4 –
D2 D4 S2 D2 S4 D4

a b

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 14
Backward boosting operation

in Fig. 12a. During the off period of S4, both batteries
and the inductor Ld supply current to the DC bus,
which has a higher voltage than batteries. The DC bus
voltage depends on the duty cycle of S4:
Vb
Vbus ¼ ð2Þ
1d
Backward Bucking Backward operation means the
DC bus is supplying current to batteries from regener-
ative braking or engine/generator. Bucking operation is
to reduce the DC bus voltage to battery voltage level. In
this operation, S1, S2, and S4 are off and S3 is operated
on and off periodically. During the on period of S3, the
DC bus supplies current to batteries through S3 and D1
as shown in Fig. 13a. During the off period of S3, the
DC bus is isolated from the batteries. The batteries and
DC bus have a voltage relation of
Vb ¼ d Vbus : ð3Þ
This mode may be used in the situation of charging
the batteries from engine/generator and regenerative
braking at high vehicle speeds.
Backward Boosting This operation is to boost a low
DC bus voltage to a high battery voltage. It may be used
in regenerative braking at low vehicle speeds at which,
the voltage generated by the motor is low. In this
operation mode, S1 and S4 are off and S3 is on and S2
is operated on and off periodically. During the on
period of S2, the DC bus charges the inductor Ld
through S3 and S2 as shown in Fig. 14a. During the
off period of S2, both the DC bus and inductor Ld
supply energy to the batteries. The DC bus and batte-
ries have the voltage relation:
Vbus
Vb ¼ : ð4Þ
1d
478 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
Advantages and Disadvantages of Electric Coupling
(Series) Drive Train
Electric coupling (series) hybrid vehicles have the
advantages of:
1. Internal combustion engine is completely mechan-
ically decoupled from wheels. Thus the engine oper-
ation can be operated in its most efficient speeds
range. On the other hand, since batteries supply
supplementary power, engine power is also
decoupled from load power. Therefore, the engine
can be operated in its most efficiency power range
as shown in Fig. 15.
2. Since the engine can be potentially operated in
a very narrow speed and power region, it becomes
much easier for further improvement of fuel econ-
omy and emission characteristics by using some
engine special design and control technologies in
this narrow region than in the whole operation
domain.
3. Completely mechanical decoupling of engine from
driving wheels allows using engines that may not be
Maximum power curve
Optimal operation region
Power
100% 107%
Speed (rpm)
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 15
Illustration of a typical engine fuel consumption map
suitable for direct vehicle propelling, such as gas
turbine engine, sterling engine, etc.
4. A multi-gear transmission and torque converter
may not be required, since the vehicle is solely
driven by an electric motor, which has proper
torque-speed profile for vehicle traction.
5. In-wheel motors may be used similar to a pure
electric vehicle.
6. The control of the drive train may be simple, com-
pared with other configurations.
However, electric coupling (series) drive train is
also suffered from some disadvantages such as:
1. Two times energy form transformations from
engine to wheels (mechanical to electrical and
then to mechanical again) may cause significant
energy losses.
2. Additional electric machine (generator) adds
weight and cost.
3. Traction motor(s) have to be designed big enough
to meet vehicle performance, such as maximum
speed, acceleration, and gradeability.
115%
135%
155%
195%
300%
Relative
specific fuel
consumption
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 479

Mechanical Coupling Hybrid Drive Trains obtains:

In mechanical coupling hybrid vehicles, mechanical
coupling devices are used to couple two mechanical
powers together. Mechanical power composes of force
and speed in linear motion, and torque and angular
speed in rotary motion. Accordingly, a mechanical cou-
pling device may add to torques together or add two
angular speeds together, depending on its mechanical
structure.
Torque Couplers A torque coupler is a three-port
mechanical device which adds two input torques
together and delivers them to output port as concep-
tually illustrated in Fig. 16. Port 1 is a unidirectional
input port (e.g., engine port), port 2 is bidirectional
input port (e.g., electric motor port), and port 3 is
a bidirectional output port (e.g., wheel port).
Ignoring losses and inertia effect, the total input
power is always equal to the output power. Supposing
port 1 and port 2 are inputs and port 3 is output, one
Port 1 Mechanical
torque
T1, w1
coupling
T3 o3 ¼ T1 o1 þ T2 o2 : ð5Þ
Inside the device, two torques, T1 and T2, are added
together in the following way:
T3 ¼ k1 T1 þ k2 T2 ; ð6Þ
where k1 and k2 are structural parameters which are
constants or variables, depending on the structure design
of the device. T1 and T2 are independent of each other
and then can be independently controlled. Solving
Eqs. 5 and 6 gives the relationship among the speeds:
o1 o2
o3 ¼ ¼ ð7Þ
k1 k2
Obviously, three speeds are linked together by k1
and k2, and have a fixed relationship with each other.
When a torque coupler is used in a hybrid vehicle, the
engine torque can be decoupled from the wheel torque
and motor torque is used for compensation. However,
the engine speed varies with the vehicle speed. This
feature provides the engine with an opportunity of
always operating at its operation line at any vehicle
T3 , w3 load as shown in Fig. 17. Figure 18 shows the com-
Port 3 monly used torque couplers.

Port 2 T2 , w 2
Speed Couplers Similar to the torque coupler (also
refer to Fig. 16), a mechanical speed coupler can be
Electric, Hybrid Electric and Fuel Cell Vehicles, Architec- described as:
tures of. Figure 16
Concept of a mechanical coupling device o3 ¼ k1 o1 þ k2 o2 ð8Þ

1.0
Optimum
Generalized engine power

0.8 operation
5
5
26
28

line
0

0
32

5 35
25
0
Fu

0.6
40

0
el

50
co eas
inc
ns ed

0
60
r
um

0.4
700 0
pti

80
on

0.2
Engin specific fuel
consumption, g/kWh
0
1000 2000 3000 4000 5000
Engine speed, rpm

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 17
Engine fuel consumption map and its optimal operation line
480 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of

Gear Box

w 1 T1
w1 T1
Z1
Z1
w3 T
3
Z3
w 2 T2 w3 T
3
w 2 T2

Z2
Z2
z3 z3 z2
k1 = , k2 = k1 = , k2 = 1
z1 z2 z1

z1, z2, z3 ---Tooth number of the gears z1, z2---Tooth number of the gears

Pullery or chain assembly

w1 T r1 r3 r1
1
w1 T1
w3 T3

r2
r2 w2 T2 w3 T3
w2 T2 r4

r2 r3 r2
k1 = , k2 = k1 = , k2 = 1
r1 r4 r1

r1, r2, r3 and r4 ---Radius of the pulleys r1, and r2 ---Radius of the pulleys

shaft

w 2 T2
w1 T1 w3 T3 k1 = 1
k2 = 1

Rotor
Stator

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 18
Commonly used torque coupling devices [2]

where k1 and k2 are the constants or variables T1 T2

associated with the device structure design.
Among these three speeds, two of them are
independent and then can be controlled indepen-
dently. Combining with Eq. 5, the torques are linked
together by:
T3 ¼ ¼ ð9Þ
k1 k2
A typical speed coupler is the planetary gear unit
as shown in Fig. 19. It consists of a sun gear 1, a ring
gear 2, and a yoke 3 which is hinged to the centers
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 481

2 w 2,T2 Another interesting speed coupling device is an

3 electric motor with a floating stator [2], in which the
stator, generally fixed to a stationary frame in conven-
R1 tional motor, is released and forms a double rotor
R3 motor as shown in Fig. 20. In this device, port 2 is an
electric port, at which electric power is converted into
R2 w 1,T1 mechanical power at the air gap between the inner and
w3 outer rotor to generate a relative speed o2. Obviously,
1 T3
o3 ¼ o1 þ o2 ; ð11Þ

and
Electric, Hybrid Electric and Fuel Cell Vehicles, Architec- T3 ¼ T1 ¼ T2 : ð12Þ
tures of. Figure 19
Here, k1 = 1 and k2 = 1 in Eqs. 8 and 9.
Planetary gear unit used as a speed coupler [2]
With a speed coupler, the engine speed is decoupled
from the vehicle speed, thus the engine can always be
operated within its optimal speed range.
Port 2
Selective uses of electric coupling, mechanical
Port 1 torque, and speed couplings can create many hybrid
vehicle configurations as discussed in following
Port 3
w2 T2 sections.

w1 T1 w3
Inner rotor
Outer rotor
Electric, Hybrid Electric and Fuel Cell Vehicles, Architec-
tures of. Figure 20
Electric motor with double rotors as a speed coupler [2]
of several planetary gears (usually three or four gears
for force balance). The relationship among o1, o2, and
o3 is
1 gr
o3 ¼ o1 þ o2
1 þ gr 1 þ gr
where gr is the gear ratio of ring gear to sun gear
(R2/R1 = z2/z1, where z2 and z1 are the tooth number
of ring gear 2 and sun gear 1). Obviously, constants
k1 and k2 in Eqs. 8 and 9 are k1 = 1/(1+gr) and
k2 = gr/(1+gr). The directions of the torques as shown
in Fig. 19 (counterclockwise for sun and ring gears and
clockwise for yoke) are defined as positive. The torque
relationship can be expressed by Eq. 9. It can be seen
from Fig. 19 that the torques on sun gear and ring gear
always have the same direction and the torque on the
yoke always has an opposite direction.
T3
Hybrid Drive Train with One Torque Coupler
Hybrid drive train with one torque coupler is the sim-
plest configuration which is traditionally referred to as
parallel configuration. The torque coupler can be
designed with different k1 and k2 values as indicated
in Eqs. 6 and 7 and Fig. 16. A two-shaft configuration
with gear box as its coupler and two transmissions is
shown in Fig. 21.
In this configuration, the torque coupler can be
designed with any parameter values of k1 and k2. In
front of the torque coupler, two transmissions can be
ð10Þ used, which may be multi-gear, single gear, or contin-
uous varying transmission (CVT), depending on the
engine and motor torque-speed characteristics and
vehicle performance requirements. Actually, two trans-
missions and the torque coupler together can be taken
as one assembly, which has the inputs of engine and
motor and the output to final drive. The torque cou-
pling assembly still has the same characteristic of the
single torque coupler explained by Eqs. 6 and 7, except
gear ratios of two transmissions being added, which
can be expressed as:
To ¼ k1 ie Te þ k2 im Tm : ð13Þ
482 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
Clutch
Engine Engine
transmission
Motor
Motor
transmission
Batteries Motor
controller
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 21
General torque coupling hybrid drive train [2]
and
oe om
oo ¼ ¼ ; ð14Þ
k1 ie k2 im
where oe , Te , om , and Tm are the angular speeds and
torques of the engine and motor, ie and im are the
transmission gear ratios of the engine and motor trans-
missions, respectively.
Obviously, two multi-gear transmissions produce
many tractive effort profiles [2]. For example, two
three-gear transmissions produce nine tractive effort
profiles (3  3). High vehicle acceleration and
gradeability performance can be obtained. Further,
high overall efficiency of the drive train can also be
obtained by selecting proper gears for operating the
engine and motor within their best region. This design
also provides opportunity for using the engine and
motor which may have less favorable characteristics.
However, this design makes it complicated for the
control system for selecting proper gears.
Instead, a single gear transmission for the motor
may be used, since electric motor generally has good
torque-speed characteristics. The three-gear transmis-
sion for engine is used to modify the engine torque-
speed profile and provides opportunity for engine
operating closer to its high-efficiency region.
Torque
coupler Final drive and
differential
Another possible design is to use single gear trans-
mission for engine and multi-gear transmission for
engine. This design is considered to be an unfavorable
design because it does not take advantage of the two
power plants.
The simplest design is to use two single gear trans-
missions. When the engine and motor are designed to
meet gradeability and acceleration performance, it is
considered to be a good design.
Another two-shaft torque coupling hybrid configu-
ration is to place a transmission behind the torque cou-
pler as shown in Fig. 22 [1], in which, the engine and
motor torque-speed profiles and the tractive effort are
also illustrated. Obviously, the transmission modifies the
engine and motor torques with the same scale. The gear
ratio of the torque coupler (k1 and k2) is designed to
matches the speeds of the engine and motor. This
design may be a proper selection for the one with larger
engine and small motor, such as mild hybrid drive
train.
A simple shaft may be used as the torque coupler,
which constitutes the configuration, traditionally
called one-shaft parallel drive train. A transmission
may be placed in front of the torque coupler as shown
in Fig. 23. In this configuration, the rotor of the electric
motor functions as the torque coupler, in which one
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 483

Torque
Te Clutch coupling Ft

we Engine
Tm V
Trans.

Motor

wm

Motor
Batteries controller

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 22
Two-shaft configuration with a transmission behind the torque coupler [2]

Batteries

Motor Torque
Controller Ft
coupler

Engine Trans.
V

Te Tm

we wm

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 23
One-shaft configuration with a transmission in front of the torque coupler [1]

end is connected to the engine and the other end to the
transmission. Obviously, here, k1 = 1 and k2 = 1. This
configuration has the similar operation characteristics.
Since the motor directly drives the final drive, the
motor speed is proportional to the vehicle speed, and
the final drive gear ratio should be designed to match
the motor speed and vehicle speed.
An alternative configuration is to place the trans-
mission behind the motor as shown in Fig. 24.
In this configuration, torques of both engine and
motor are enhanced by the transmission. However,
the engine and motor are directly coupled together;
hence, they have to have the same operation
speed range. The motor can serve as the engine start
when a clutch is placed between the motor and trans-
mission. When this configuration is used in a mild
hybrid drive train, where motor alone operation
mode is used, the clutch can be placed between the
484 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of

Batteries

Motor
Controller Torque Ft
coupler

Engine Trans.

V
Te
Tm

we
wm

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 24
One-shaft configuration with a transmission behind the torque coupler [2]

Engine
Trans. Motor Trans.

Motor
controller

Batteries

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 25
Front- and rear-wheel-driven hybrid vehicle

motor and transmission. The motor functions as an
engine torque booster, engine start, generator, and
regenerative braking.
Actually, the vehicle chassis and road can also used
as a torque coupler. Figure 25 shows a hybrid drive
train in which the engine and motor drive the front
and rear wheels, respectively. The tractive efforts pro-
duced by the engine and motor are added together by
the vehicle chassis and road. This configuration offers
some advantages of keeping the original engine and
transmission driveline and providing the vehicle with
a four-wheel-drive capability.
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 485

Hybrid Drive Trains with Torque and Speed A detailed drive train structure is shown in Fig. 27
Couplings [1]. The drive train uses a planetary gear unit as its speed
coupler. Engine is connected to its yoke and a motor/
As discussed in prior sections, the torque coupler can
generator to its sun gear. The ring gear as the output port
decouple the engine torque from the wheel’s torque
is connected to a torque coupler which is a fixed shaft
demand, and speed coupler can decouple the engine
gear set. Another two ports of the torque coupler are
speed from the wheel speed. Thus, it is possible to
connected to a traction motor and wheels.
decouple the engine torque and speed from the
As indicated in Eqs. 8 and 9, Fig. 19, the torque of
wheels with two couplers. This concept is illustrated
the motor/generator always has an opposite direction as
in Fig. 26.
that of the engine and can rotate in two directions. When
the rotating speed is in the same direction as its torque, it
is a motor, otherwise, a generator. When it operates in
Speed Motor/ the generating mode, it absorbs part of the engine power.
Engine
coupler generator Due to this reason, this hybrid drive train gets another
name of engine power split hybrid drive train.
The torque of the motor/generator is always propor-
tional to the engine torque. Thus, an ideal operation
Motor/ Torque
coupler
characteristic is a constant torque at the whole positive
generator
and negative speed domain. Also, the three controllable
components (the engine, motor/generator, and traction
Electric, Hybrid Electric and Fuel Cell Vehicles, Architec- motor) provide very plentiful operation modes, and
tures of. Figure 26 hence, greater opportunity for high operating efficiency.
Concept of torque and speed coupling drive train However, very complex control is required [2].

Planetary
gear unit
Clutch
Motor/Generator

Za Lock

Engine Zs

M/G
Yoke controller

Zr
Zb

Traction
Zc
motor

Motor
Batteries
controller

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 27
Hybrid drive train with speed and torque couplers
486 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
Architectures of Fuel Cell Vehicles
Fuel Cell System
A fuel cell is a chemical device that converts the chemical
energy into electric energy by means of redox reaction
with the operation principle exactly the same as in
a chemical battery. In a battery cell, the reactive chemicals
are pro-installed in the cell. When the active chemicals are
used up, the cell is dead and has to be recovered by
charging it from an outside electrical source. Instead,
a fuel cell is fed with its reactive chemicals continuously
from outside of the cell. It is more like an internal
combustion engine which is fed with fuel and produces
mechanical energy. However, the energy conversion effi-
ciency in a fuel cell is higher than internal combustion
engine due to its free from the Carnot cycle limitation.
Compared to chemical batteries, it has the advantages
of quick refueling and much more energy in an outside
energy storage, similar to a fuel tank.
A fuel cell has a very similar structure of chemical
battery cell, as shown in Fig. 28. Fuel and the oxidizing
agency are continuously and separately fed into the
anode and cathode electrodes, where they are ionized.
Electrolyte is used to conduct positive ions from the
anode to the cathode, and at the same time, electrons
are conducted from the anode to the cathode through
an electrical load.
e–
load
− I
Fuel
Positive ions
+
+
Anode Electrolyte
electrode
Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 28
Illustration of fuel cell operation
A single fuel cell has very low voltage. In practice,
a fuel cell stack may include hundreds of single
cells serially connected together. A fuel stack needs
auxiliaries to support its operation. The auxiliaries
mainly include air circulating pump, coolant circu-
lating pump, ventilation fan, fuel supply pump,
and electric control devices. Part of the power devel-
oped from the fuel cell stack is used to support
the operation of the auxiliaries as shown in Figs. 29
and 30.
Fuel Cell Technologies
There are several types of fuel cells, depending on their
electrolytes. Table 2 lists the operation temperature and
electrolytes.
Proton Exchange Membrane Fuel Cells (PEMFCs)
PEMFCs work with hydrogen as fuel and oxygen as
oxidant. Hydrogen is supplied to the anode where it is
oxidized, losing electrodes in the process. Positive
hydrogen ions (protons) migrate across the electrolyte
through the membrane to the cathode, at the same
time, the electrons travel round the external circuit to
the cathode. The oxygen is supplied to the cathode
where it is reduced, picking up the electrons and the
ions from the hydrogen to form water.
+
Oxidant
(O2 or air)
Exhaust
Cathode
electrode
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 487

Auxiliaries
1. Air circulating pump
2. Coolant circulating pump
3. Ventilation fan
4. Controller
Hydrogen
storage
Power
converter
H2
Iaux
+ +

Ifc ILoad

Load
Fuel cell stack

– –
Air
Water Waster heat

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 29
A hydrogen–air fuel cell system

Electric, Hybrid Electric and Fuel Cell Vehicles, Architec-

1.0 tures of. Table 2 Operating data of various fuel cell
systems [3]
0.8 Cell voltage, V Operating
Cell system temperature,  C Electrolyte
0.6
System efficiency Proton exchange fuel 60–100 Solid
cells (PEMFCs)
0.4
Alkaline fuel cells (AFCs) 80–230 Liquid
Net power density W/cm2 Phosphoric acid fuel 60–200 Liquid
0.2
cells (PAFCs)

0 Molten carbonate fuel 600–1000 Liquid

0 0.2 0.4 0.6 0.8 1.0 1.2 cells (MCFCs)
Net current density (A/cm2)
Solid oxide fuel cells 1000–1200 Solid
Electric, Hybrid Electric and Fuel Cell Vehicles, Architec- (SOFCs)
tures of. Figure 30 Direct methanol fuel 100 Solid
Cell voltage, system efficiency, and net power density cells (DMFCs)
varying along net current density of a hydrogen–oxygen
fuel cell [2]
the cost of expensive catalysts, some fuel cells are
The electrodes of PEMFCs are made of porous designed to work at elevated temperatures.
carbon which allows the active gases to pass through. The platinum catalyst used in PEMFCs is very
The electrode surfaces are coated with platinum as expensive and extremely sensitive to poisoning by
a catalyst which increases the rate of oxidation at even small amounts of carbon monoxide. Thus a high
the anode and the rate of reduction at the cathode. purity of hydrogen gas is necessary.
The platinum catalysts allow the chemical reaction to Electrolyte is a thin, fragile sheet of acidic, solid
take place at a lower temperature. Alternatively to avoid organic polymer about 50 mm thick which permits
488 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of
the passage of hydrogen ions but is impermeable to
electrons. Acidic compounds are fluids with free
hydrogen ions and these are the fuel cell’s charge
carriers [4].
The first PEMFCs were developed in the 1960s for the
need of a US-manned space program. It is nowadays the
most investigated fuel cell technology for automotive
applications. It operates at 60–100 C and can offer
a power density of 0.35–0.6 W/cm2. PEMFCs offer
some definite advantages for electric and hybrid vehicle
applications that include low temperature operation,
high power density among all the available types of fuel
cells, solid electrolyte that does not change, movement
and vaporization from the cell, and most importantly,
potentially high efficiency and being free of toxins and
greenhouse gas emissions. The major problems are
their requirements of expensive noble metal catalyst
and membrane, and a need of high purity of hydrogen
as its fuel.
Alkaline Fuel Cells (AFCs) Alkaline fuel cells use
aqueous electrolytes of potassium hydroxide (KOH)
to conduct ions between electrodes. Hydrogen and
oxygen are used as the fuel and oxidant. The ion carried
by the alkaline electrolyte is a hydroxide ion (OH).
The half reactions are:
Anode : 2H2 þ 4OH ! 4H2 O þ 4e 
Cathode : O2 þ 2H2 O þ 4e  ! 4HO
AFCs are capable of operating over a wide range of
temperature (80–230 C) and pressure (2.2–45 atm).
AFCs are capable of achieving high operating efficiency
because of the fast kinetics permitted by the hydroxide
electrolyte. The fast kinetics in AFCs allows using silver
or nickel as catalysts instead of expensive platinum.
The performance of AFCs can be further improved
by circulating the electrolyte, which offers some advan-
tages that include easy thermal management, more
homogenous electrolyte concentration, easy replace-
ment of electrolyte, and possible removal of the elec-
trolyte when it is turned off. Of course, circulation of
electrolyte needs additional equipments, such as circu-
lation pump, heat exchanger, and evaporator. Further,
it increases the risk of leakage (potassium hydroxide is
highly corrosive and has a natural tendency to leak even
through the tightest seals).
AFCs were some of the earliest practical cells and
were used in the Apollo space program, generating
drinking water as well as electric power. Although
they are inexpensive compared with PEM cells, operat-
ing efficiencies of 60% are possible. Unfortunately they
have a low power output and the catalyst is prone to
poisoning from carbon dioxide in the atmosphere.
Phosphoric Acid Fuel Cells (PAFCs) PAFCs use
acidic electrolyte to conduct hydrogen ions and the
anode and cathode reactions are the same as in
PEMFCs. PAFCs run at high temperature of around
200 C, delivering high power of a MegaWatt or more
but with relatively low efficiency of around 35%. The
poor conversion efficiency is the high heat generation
in the fuel cell stack. Because of the high working
temperature the efficiency losses can be mitigated by
using the waste heat in combined heat and power
(CHP) applications.
The advantage of PAFCs is the cheap electrolyte and
disadvantages are the need of expensive catalyst (plat-
inum), corrosion by acidic electrolyte, CO poising, and
low efficiency.
Molten Carbonate Fuel Cells (MCFCs) MCFCs run
at high temperatures of 650–1,000 C. They use molten
carbonate salt to conduct ions, usually, lithium-
potassium carbonate or lithium-sodium carbonate.
Their unique chemistry needs carbon dioxide from
the air as part of the process. Efficiencies achieved are
45% or more and power outputs of over 1 MW are
typical in grid supply applications. Because of their
high working temperature, they can operate directly
with hydrocarbon gases which are reformed within
the cell and do not need a separate hydrogen supply.
The high temperature also means that less expensive
catalysts are needed, but the molten electrolyte
imposes special requirements on containment and
anticorrosion measures.
Solid Oxide Fuel Cells (SOFCs) Solid oxide fuel cells
also operate in the same or higher temperatures as the
molten carbonate cells with the same fuel and catalyst
advantages. The ceramic electrolyte which can run as
hot as 800 C has the advantage that the electrolyte stays
solid. They can deliver powers of several Megawatts but
at a lower efficiency of around 35%.
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 489

Direct Methanol Fuel Cells (DMFCs) Direct meth- Nevertheless the ability to use liquid fuel coupled with
anol fuel cells are similar to PEM cells. Instead of the elimination of the reformer make these fuel cells
pure hydrogen, they directly use liquid Methanol as very attractive.
fuel. It is sure that there are definite motivations for
using DMFCs to vehicles. Methanol is a liquid fuel at Fuel Cell Vehicles
room temperature that can be easily stored, distrib-
Fuel cell–powered vehicles potentially have the advan-
uted, and marketed through current infrastructures.
tages of no toxic and greenhouse emissions and high
Further, methanol is the simplest organic fuel that
operation efficiency. Compared to internal combustion
can be economically and efficiently produced on
engine, fuel cells have a prominent advantage of high
a large scale from fossil fuel, such as coal and natural
operation efficiency at low power output which is coin-
gas. Further it can be also produced from biomass
cident to a vehicle’s normal driving situation. This
products.
means that fuel cells in a fuel cell–powered vehicle
Anode and cathode electrode of DMFC also need
will naturally operate in their high-efficiency region
platinum or platinum alloys as catalyst. The electrolyte
in normal driving conditions.
can be a trifluoromethane sulfonia acid or proton
Fuel cell–powered vehicles can be either fuel cell
exchange membrane (PEM). The chemical reactions
alone or hybridized with batteries and/or ultra-
in a DMFC are:
capacitors. Fuel-cell-alone-powered vehicles have
Anode : CH3 OH þ H2 O ! CO2 þ 6Hþ þ 6e  a very simple configuration, similar to pure electric
vehicles by replacing the battery pack with a fuel cell
3
Cathode : O2 þ 6 Hþ þ 6e  ! 3H2 O system as shown in Fig. 31.
2
A fuel-cell-alone-powered vehicle is a single power
3 source vehicle. The fuel cell power follows the load pro-
Overall : CH3 OH þ O2 ! CO2 þ 2H2 O
2 file. The power rating of the fuel cell system is designed to
Like PEM fuel cells, DMFCs work at low tempera- meet the vehicle performance requirement. A fuel cell
tures in the range of about 50–120 C. They have system management is required to manage the fuel cell
a relatively low efficiency and power density. Output operation, such as air supply, fuel supply, and cooling of
power using current technology is limited to about the fuel cells for operating in their safe range.
1.5 kW which is enough to power most consumer Fuel cell is a unidirectional energy converter. It
goods but not sufficient for automotive applications. cannot operate in opposite directions to absorb energy

Driving
command

Vehicle Feedback
...

controller signals

Fuel cell Motor Trans-

system motor mission
controller

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 31
Configuration of fuel-cell-alone-powered vehicle drive train
490 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of

from an electric energy source. Thus, a fuel-cell-alone- powers of the fuel cells and secondary power
powered vehicle is not capable of recovering braking source are coupled together at an electronic inter-
energy. face which is used to control the power flows.
A vehicle controller is used to control the drive
Fuel Cell–Powered Hybrid Vehicles train based on the driver’s command and system’s
feedback signals.
Fuel cell system, at today’s technology, has low
There are several operation modes that can be
power density. For satisfying vehicle performance,
selectively used which are:
a bulky and heavy fuel cell system is needed, which
in turn, harms the vehicle performance and causes 1. Fuel cell alone propelling: in which the electronic
high energy losses. In addition, lack of regenerative interface isolates the secondary power source from
braking capability is a major shortcoming of this the power line and fuel cell pack alone supplies
drive train. power to the electric motor.
Hybridization of fuel cell system with a secondary 2. Secondary power source alone propelling: in which
power source, such as a battery pack and/or the fuel cell pack is isolated from the power line
ultracapacitor pack is an effective technology to over- and the secondary power source alone supplies
come these shortcomings. A fuel cell hybrid vehicle power to the traction motor.
drive train is shown in Fig. 32. 3. Hybrid propelling: in which both the fuel cell sys-
Fuel cell hybrid vehicle drive train consists tem and the secondary power source provide pow-
mainly of a fuel cell system as primary power ers to traction motor simultaneously.
source, a secondary power source that may be 4. Fuel cell power split mode: in which fuel cell power
a battery pack and/or ultracapacitor pack. The is split into two parts by the electronic interface.

Accelerator Brake
pedal pedal

Traction Brake
command command
signal signal
Speed signal
Vehicle
Motor control signal
controller
control signal

Electric
control
Fuel cell

signal
Secondary power signal

power signal

wheel
Fuel cell

FC Electronic Motor Elec.

system interface controller Motor

Secondary Transmission
power
source

wheel

Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of. Figure 32
Configuration of a fuel cell–powered hybrid drive train
 Electric, Hybrid Electric and Fuel Cell Vehicles, Architectures of 491

motor drive with high power density, high efficiency,

1.0 and long extended speed range is the key technology.
The success of an electric vehicle is almost determined
Cell voltage, V
0.8 by the performance of the battery system. The basic
requirements for electric vehicle batteries are the high
0.6 System efficiency energy capacity for extended range, high power for
vehicle performance, high safety, and low cost. At pre-
0.4 sent, lithium-ion batteries are considered to be the
most promising candidates. For hybrid vehicles,
Net power density W/cm2
0.2 advanced primary power sources are very important
that may include advantage engine technologies and
0 alternative fuel engines. Advanced control strategies
0 0.2 0.4 0.6 0.8 1.0 1.2
Optimal
and real-time control algorithms are also very impor-
Net current density, A/cm2
operating region tant. The key technologies for fuel cell vehicles include
enhancement of fuel cell power density, reduction of
Electric, Hybrid Electric and Fuel Cell Vehicles, Architec- cost, and, more importantly, hydrogen production and
tures of. Figure 33 onboard storage.
Optimal operation region of a fuel cell system

One part is used to power the traction motor and
the other charges the secondary power source.
5. Regenerative braking: in which the traction motor
functions as a generator to recover the braking
energy and stores it in the secondary power source.
By electively using the operation modes, the fuel
cells can be operated in their most efficient region as
shown in Fig. 33.
The electronic interface plays a critical role in the
drive train control. It should be designed with three
ports and the ports to the secondary power source and
electric motor are bidirectional. The configuration and
operation principle are very similar as that used in
engine/generator-based electric coupling (series)
hybrid drive train.
Future Directions
The electric vehicle has a relatively simple power train
structure. The most important components are the
motor drive and the battery system. An advanced
Bibliography
Primary Literature
1. Chan CC, Chau KT (2001) Modern electric and vehicle technol-
ogy. Oxford University Press, New York
2. Ehsani M, Gao Y, Emadi A (2010) Modern electric, hybrid
electric and fuel cell vehicles–Fundamentals, theory and
design, CRC, FL, USA
3. Berlowwitz PJ, Darnell CP (2001) Fuel choice for fuel cell
powered vehicles. Society of automotive engineers (SAE),
Journal, Paper No. 2001-02-0003
4. http://www.mpoweruk.com/hydrogen_fuel.htm. Accessed
Oct 2010
Books and Reviews
Miller JM (2004) Propulsion systems for hybrid vehicles. The Insti-
tution of Electrical Engineers, London
Messerle HK (1969) Energy conversion statics. Academic, New York
Larminie J, Dicks A (2000) Fuel cell system explained. Wiley,
New York
Thomas CE, James BD, Lomax FD Jr, Kuhn IK Jr (2002) Societal
impact of fuel options for fuel cell vehicles. SAE Journal, Paper
No. 982496
492 Electric, Hybrid, and Fuel Cell Vehicles, Introduction
Electric, Hybrid, and Fuel Cell
Vehicles, Introduction
MEHRDAD EHSANI
Department of Electrical & Computer Engineering,
Texas A&M University, College Station, TX, USA
The success of internal combustion engine vehicles
over the past century is one of the greatest technolog-
ical achievements of mankind with a profound impact
on virtually every aspect of human life on the plant.
However, the very widespread success of this technol-
ogy has produced associated problems which can be the
seeds of its own long-term demise. Solution of these
problems will guarantee the long-term survival of per-
sonal and mass transit vehicles which are so essential to
the economy and welfare of human beings. The proper
combination of these solutions can bring about what is
termed sustainable vehicles.
This section of the encyclopedia describes the
important sustainable vehicle technical topics and the
present state of the art on each topic. The following is
a brief review of the entries in this section.
▶ Sustainable Transportation describes the techni-
cal and environmental problems that have to be
addressed to achieve sustainable vehicles for the future.
Issues of fuel, environment, and vehicle technologies
are introduced.
▶ Electric, Hybrid Electric and Fuel cell Vehicles,
Architectures of describes the basic configuration of the
drivetrain of electric vehicles and hybrid electric vehi-
cles. The various types of hybrid electric vehicles’ drive-
trains are presented. The basic operation of each of
these systems and their components are described.
▶ Vehicle Dynamics and Performance describes
vehicle drivability, fuel economy, braking performance,
handling characteristics, noise, vibration, and harsh-
ness. Vehicle performance discussed in this entry is
restricted to propelling and braking performance in
terms of vehicle speed, acceleration, road grade
climbing, braking deceleration, and braking force dis-
tribution on the front and rear wheels.
▶ Vehicle Energy Storage: Batteries gives an over-
view of batteries for vehicle applications and discusses
the research and development roadmap of next-
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
generation batteries for vehicle applications. It
describes how the battery is used in EV and HEV and
what the vehicle requirements on it are. It also presents
the basic state of the art on the available batteries today.
▶ Battery Technologies describes the internal work-
ing of the electrochemical battery, the relationship
between the internal resistance of the battery cell, max-
imum power output, energy efficiency in charging and
discharging, and the issue of cell heating. A review of
the characteristics of lead acid, nickel metal hydride,
lithium-ion batteries, and super capacitors are then
presented.
▶ Energy storage: Ultracapacitor describes the new
technology of ultracapacitors for high-power energy
storage and delivery. This is another approach to
providing energy storage for hybrid vehicles.
Ultracapacitors can be used either alone or in combi-
nation with batteries. In this approach, the batteries
can be optimized for energy density and cycle life, and
ultracapacitors can provide the power both for accel-
eration and regenerative braking.
▶ Hybrid Energy Storage Systems for Vehicle
Applications describes the possibility of composing
a high-energy, high-power hybrid energy storage sys-
tem for vehicle applications, based on the analysis of
inherent characteristics of different energy storage
methods. The basic components in this system are
chemical batteries, ultracapacitors, and flywheels
(mechanical batteries). The objective is to fully utilize
the advantages and suppress the disadvantages of each
component to compose a compact energy storage sys-
tem with high energy capacity and high power
capability.
▶ Batteries, Battery Management, and Battery
Charging Technology describes issues related battery
charging and battery management. Battery manage-
ment is one of the important engineering challenges
of using batteries in vehicles. It starts with an overview
of different types of rechargeable battery chemistries.
Then, it discusses circuits for charging and managing
vehicle batteries.
▶ Vehicle Traction Motors describes the role of the
electric motor in EV and HEV drivetrains. The require-
ments that the vehicle put on the electric motor are
explained. This is used as the basis to present the
various traction motor drive technologies.
# 2012, DOI 10.1007/978-1-4419-0851-3

▶ AC Machines: Permanent Magnet Synchronous
and Induction Machines describes the most common
electric machines, motors, and generators that are used
in EV and HEV drivetrains. It presents AC machines
including induction machine (IM) and permanent
magnet synchronous machine (PMSM). It is shown
that both the IM and PMSM take advantage of the
rotating magnetic field that is produced by their
three-phase stator windings.
▶ Switched Reluctance Motor Drives for Propul-
sion and Regenerative Braking in EV and HEV
describes a new basis for the selection of vehicle trac-
tion motor technology. The concept of extended speed
range at constant power and its benefit in vehicle trac-
tion application is introduced. On this basis, the mod-
ern switched reluctance motor drive technology is
shown to be very attractive for future EV and HEV
applications. In addition to having a wide speed
range, these motors have a rugged and low-cost struc-
ture, reliable drive power converter topology, and high
efficiency over a wide range of speed.
▶ Regenerative Braking describes the important
technology of regenerative braking in EV and HEV
drivetrains. Regenerative braking is responsible for
most of the battery energy and fuel savings in these
vehicles. This entry presents the principles of regener-
ative braking operation, design, and its hardware.
▶ Hybrid Electric and Hydraulic Technology Appli-
cations in Off-Road Vehicles describes a special call of
HEV that are more suitable for off-road and heavy-
duty applications. These include military vehicles,
construction and mining vehicles, and similar indus-
trial and special-purpose heavy uses. The special use
of hydraulics in the drive train as part of the hybrid-
ization of the vehicle is of particular interest in this
entry.
▶ Plug-in Hybrid Electric Vehicles describes
a special class HEV that is called plug-in HEV
(PHEV). It is shown that the PHEV has certain advan-
tages over pure EV and HEV, such as (1) displacement
of significant amount petroleum fuel by electric energy
Electric, Hybrid, and Fuel Cell Vehicles, Introduction 493
that comes from utility grid, (2) longer range than
a pure EV, (3) no fuel consumption and emission
during pure EV mode, and (4) lower overall fuel con-
sumption and emission.
▶ PHEVs and BEVs in Coupled Power and Trans-
portation Systems describes the role of utility electric
power systems in the EV and HEV transportation sys-
tem. The impact of plug-in HEV and battery-operated
EV (PHEV/BEV) transportation systems on the utility
power systems is discussed. The mutual benefits of
charging the EV/HEV from the utility grid and some
of their draw backs are addressed.
▶ Internal Combustion Engines, Alternative Fuels
for describes the technical implications of alternative
fuels on vehicle engines, their performance and emis-
sions characteristics. The characteristics of the most
dominant alternative fuels are defined and compared
to conventional gasoline and diesel fuel. Of particular
interests are the renewable alternative fuels.
▶ Vehicle Biofuels describes the nature and origins
of liquid fuels for vehicle engines that are derived from
biomass. Liquids biofuels are the most valuable because
they are easily stored, have a high energy density, and
are readily metered into engines. It is shown that
biofuels are infinitely renewable provided appropriate
agricultural practices are followed.
▶ Sustainable Vehicle Fuels, Well-to-Wheel Analy-
sis describes the important stages, results, and key
issues associated with the well-to-wheel stages of
biofuels, hydrogen, and electricity as alternative trans-
portation fuels. The importance of comparing
alternative vehicle engine fuels with the conventional
fossil fuels is presented from the fundamental point
of view.
▶ Fuel Cell Powered HEV Design and Control
describes different types of fuel cells and their electro-
lytes, sources of fuels for fuel cells, those applicable to
the transportation sector, and the hybrid vehicle archi-
tecture. The ICE hybrid is also presented as a base
model, since its operation is well understood by the
general public.
494 Energy Storage: Ultracapacitor
Energy Storage: Ultracapacitor
ANDREW F. BURKE
Institute of Transportation Studies, University of
California-Davis, Davis, CA, USA
Article Outline
Glossary
Definition
Introduction
Energy Storage Characteristics of Ultracapacitors and
Batteries
Systems Considerations and Advantages of Combining
Ultracapacitors and Batteries
Laboratory and Vehicle Field Experiences Using
Ultracapacitors
Hybrid Vehicle Simulations Using Ultracapacitors
Alone
Plug-in Hybrid Vehicle Simulations Using Advanced
Batteries And Ultracapacitors
Fuel Cell Vehicle Simulations Using Ultracapacitors
Economic Considerations
Future Directions of Ultracapacitor Development and
Vehicle Applications
Summary
Bibliography
Glossary
Battery A cell in which electrical energy is stored and
discharged via Faradaic electrochemical reactions
occurring at a predetermined voltage dependent
on the active materials in the two electrodes.
Control (power split) strategy The strategy used in
a hybrid-electric vehicle to determine the fraction
of the power demanded that is provided by the
electric motor and engine, respectively.
Hybrid-electric vehicle (HEV) A vehicle having both
an electric motor and internal combustion engine
in which the battery is maintained near a fixed
state-of-charge via electrical energy generated by
the electric motor/generator driven by the engine.
Plug-in hybrid electric vehicle (PHEV) A vehicle
having both an electric motor and internal combus-
tion engine in which the battery is depleted by all-
electric driving and recharged from the wall plug.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Pulse power efficiency The fraction of the discharge
or charge energy from the battery or ultracapacitor
during a short, high-power pulse that is available as
electrical energy for use by the vehicle.
Ultracapacitor (electrochemical capacitor) A device
in which electrical energy is stored and discharged
via the electrostatic separation of charge in the
micropores of activated carbon in two electrodes.
Vehicle simulations Computer calculations of the
operation of a vehicle based on models of the var-
ious components in the driveline of the vehicle and
a specified control strategy for its operation.
Definition
Most hybrid vehicles use batteries for energy storage.
The batteries must provide both the energy and power
required by the vehicle. In all cases, this requires com-
promises in designing the battery in that it cannot be
optimized for either energy density or power capability.
In addition, in most cases the battery unit is oversized
(stores more energy than is used by the vehicle) in
order to achieve long cycle life even for the shallow
cycles experienced in charge-sustaining hybrid vehicles
like the Toyota Prius. As discussed in this entry, another
approach to providing energy storage for hybrid vehi-
cles, both charge-sustaining and plug-in designs, is to
utilize ultracapacitors either alone or in combination
with batteries. In this approach, the batteries can be
optimized for energy density and cycle life and
ultracapacitors can provide the power both for accel-
eration and regenerative braking. The ultracapacitors,
which are inherently high-power devices, would be
deep discharged to one-half rated cell voltage and still
provide cycle life of 500,000 to one million cycles.
The development of ultracapacitors suitable for use
in vehicle powertrains started in about 1990 and steady
progress in improving the performance of the devices
has occurred since that time. This entry is concerned
with the present status of commercially available cells
and modules and projections of future developments
based on the performance of prototype cells of both the
carbon/carbon and hybrid carbon/pseudo-capacitive
designs. The cost of ultracapacitors is currently the
primary obstacle to the use of ultracapacitors in vehicle
applications. In this entry, the future costs are projected
# 2012, DOI 10.1007/978-1-4419-0851-3

and compared with high-power batteries based on the
ultracapacitor material costs. Finally, applications of
ultracapacitors in hybrid-electric vehicle powertrains
are discussed. Key considerations are the relationships
between the Wh stored in the ultracapacitor unit, sys-
tem control strategy, and the fuel economy improve-
ments attained for various driving cycles. These
relationships are studied using detailed vehicle simula-
tion results from Advisor.
Introduction
In the first section of the entry, the energy storage
characteristics of lithium batteries and ultracapacitors
are presented and compared. Of particular interest is
the comparative power capability of lithium batteries
and carbon/carbon ultracapacitors for charge/dis-
charge conditions to be encountered in hybrid-electric
vehicles. In the next section, system considerations and
advantages of combining ultracapacitors and batteries
for energy storage are discussed. Next laboratory and
field experiences with ultracapacitors in hybrid-electric
vehicles are reviewed for transit bus, light-duty, and
fuel cell applications.
The final sections of the entry discuss in detail
simulations of vehicles on various driving cycles using
ultracapacitors alone and in combination with
advanced batteries. The vehicle simulations treat
charge-sustaining micro- and mild-hybrid designs,
plug-in hybrids, and fuel cell–powered vehicles. The
charge-sustaining hybrid vehicles using ultracapacitors
alone in place of batteries are discussed in terms of the
sizing of the ultracapacitor unit and control strategies
that optimize the fuel economy improvements achiev-
able. For the plug-in hybrids, the emphasis is on the use
of ultracapacitors with advanced batteries having
energy densities greater than 200 Wh/kg. It is likely
that these batteries will not have proportionally high
power density and will require the assist of
ultracapacitors to achieve all-electric operation even
on the FUDS driving cycle. The next section of the
entry deals with fuel cell–powered vehicles and con-
siders the use of ultracapacitors to load level the fuel
cell to increase system efficiency and thus improve
hydrogen consumption. In the final section of the
entry, the cost of batteries and ultracapacitors are com-
pared for charge-sustaining hybrid applications.
Energy Storage: Ultracapacitor 495
Energy Storage Characteristics of Ultracapacitors
and Batteries
Lithium batteries and ultracapacitors have been tested
in the laboratory at the University of California, Davis
[1–3]. A summary of the test results for the batteries is
given in Table 1 and for ultracapacitors in Table 2.
For both energy storage technologies, the devices
with the highest energy density typically have the low-
est power capability. The pulse power capabilities
shown in the tables were calculated using the following
relationships:
Batteries: P = EF(1
EF) Voc2/R
Ultracapacitors: P = 9/16(1
EF) Vrated2/R
where EF is the efficiency of the pulse (EF = Vpulse/Voc).
The matched impedance power which is often cited
for both battery and ultracapacitor devices is calculated
as follows:
Pmatchimped: ¼ V2 =4R
These power relationships are derived in [4].
For charge-sustaining hybrids, it seems reasonable
to cite the power capability of devices for pulse power
efficiencies of 90–95%. For PHEVs and EVs operating
in charge-depleting modes for the battery, it is reason-
able to cite the power for efficiencies of 75–80%. In all
instances, the power capability is proportional to V2/R
indicating that high power capability requires a low
resistance R.
Comparisons of the Power Capabilities of Lithium
Batteries and Ultracapacitors
There has been considerable discussion in the literature
[4–6] comparing the power capability of lithium-ion
batteries and ultracapacitors. The conclusions vary
from statements that lithium batteries have power
capability equal to that of ultracapacitors to statements
that ultracapacitors have an order of magnitude higher
power capability than lithium batteries. Detailed com-
parisons of the power capability of ultracapacitors and
batteries taken from [4] are shown in Table 3. As
indicated in the table, neither of the extreme statements
is valid in general and that comparisons should be
made between specific devices for specific applications.
Comparisons are often made based on the matched
496 Energy Storage: Ultracapacitor

Energy Storage: Ultracapacitor. Table 1 Summary of the performance characteristics of lithium-ion batteries of various
chemistries

Battery developer/ Electrode Voltage Resist. W/kg W/kg Density

cell type chemistry range Ah (mOhm) Wh/kg 90% effic.a match. imped. Wt. (kg) (gm/cm3)
Enerdel HEV Graphite/ 4.1–2.5 15 1.4 115 2,010 6,420 0.445 –
Ni MnO2
Enerdel EV/PHEV Graphite/ 4.1–2.5 15 2.7 127 1,076 3,494 0.424 –
Ni MnO2
Kokam prismatic Graphite/ 4.1–3.2 30 1.5 140 1,220 3,388 0.787 2.4
NiCoMnO2
Saft cylind. Graphite/ 4.0–2.5 6.5 3.2 63 1,225 3,571 0.35 2.1
NiCoAl
GAIA cylind. Graphite/ 4.1–2.5 40 0.48 96 2,063 5,446 1.53 3.22
NiCoMnO2
7 3.6 78 1,310 3,472 0.32 –
A123 cylind. Graphite/ 3.6–2.0 2.2 12 90 1,393 3,857 0.07 2.2
iron
phosph.
Altairnano LiTiO/ 2.8–1.5 11 2.2 70 990 2,620 0.34 1.83
prismatic NiMnO2
Altairnano LiTiO/ 2.8–1.5 3.8 1.15 35 2,460 6,555 0.26 1.91
prismatic NiMnO2
Quallion cylind. Graphite/ 4.2–2.7 1.8 60 144 577 1,550 0.043 2.6
NiCo
Quallion cylind. Graphite/ 4.2–2.7 2.3 72 170 445 1,182 0.047 2.8
NiCo
EIG prismatic Graphite/ 4.2–3.0 20 3.1 165 1,278 3,147 0.41 –
NiCoMnO2
EIG prismatic Graphite/ 3.65–2.0 15 2.5 113 1,100 3,085 0.42 –
iron
phosph.
Panasonic EV Ni metal 7.2–5.4 6.5 11.4 46 395 1,093 1.04 1.8
prismatic hydride
a
power density P=Eff.*(1
Eff.) Voc2/R, Pmatch. imped. = V2/4R

impedance power of the two types of devices. These
comparisons indicate that most ultracapacitors have
a power capability (W/kg) of three to six times that of
lithium batteries. However, for vehicle applications the
matched impedance power is not appropriate and
should not be the basis of comparison.
For charge-sustaining HEV applications, a good
basis of comparison is the W/kg at 95% efficiency
at the SOC at which the devices will be used in the
vehicle. On this basis, there are lithium batteries with
the same power capability as some carbon/carbon
ultracapacitors, but there are some ultracapacitors
with power capability twice that of the highest power
lithium batteries presently available for vehicle appli-
cations. In other words, it is not possible to make
general statements that are applicable to all devices of
either type. The issue is further complicated when one
notes that the density of the lithium batteries is about
twice that of carbon/carbon ultracapacitors (2.2 g/cm3
for the batteries and 1.2 g/cm3 for the ultracapacitors).
 Energy Storage: Ultracapacitor 497

Energy Storage: Ultracapacitor. Table 2 Summary of ultracapacitor device characteristics

Wh/kg W/kg W/kg
Device V rated C (F) R (mOhm) RC (s) (1) (95%) (2) match. imped. Wt. (kg) Vol. (l)
a
Maxwell 2.7 2,885 0.375 1.08 4.2 994 8,836 0.55 0.414
Maxwell 2.7 605 0.90 0.55 2.35 1,139 9,597 0.20 0.211
Skeleton technologies 2.8 1,600 1.3 2.1 5.8 800 7,140 0.22 0.13
b
Yunasko 2.7 55 4 0.22 5.5 5,695 50,625 0.009 –
b
Yunasko 2.7 450 1.3 0.58 5.89 2,766 24,595 0.057 0.045
b
Yunasko 2.7 510 0.9 0.46 5.0 2,919 25,962 0.078 0.055
Ness 2.7 1,800 0.55 1.00 3.6 975 8,674 0.38 0.277
Ness 2.7 3,640 0.30 1.10 4.2 928 8,010 0.65 0.514
Ness (cyl.) 2.7 3,160 0.4 1.26 4.4 982 8,728 0.522 0.379
Asahi glass (propylene 2.7 1,375 2.5 3.4 4.9 390 3,471 0.210 0.151
carbonate) (estimated)
Panasonic (propylene 2.5 1,200 1.0 1.2 2.3 514 4,596 0.34 0.245
carbonate)
EPCOS 2.7 3,400 0.45 1.5 4.3 760 6,750 0.60 0.48
LS Cable 2.8 3,200 0.25 0.80 3.7 1,400 12,400 0.63 0.47
BatScap 2.7 2,680 0.20 0.54 4.2 2,050 18,225 0.50 0.572
Power sys. (activated 2.7 1,350 1.5 2.0 4.9 650 5,785 0.21 0.151
carbon, propylene
carbonate)b
Power sys. (graphitic 3.3 1,800 3.0 5.4 8.0 486 4,320 0.21 0.15
carbon, propylene
3.3 1,500 1.7 2.5 6.0 776 6,903 0.23 0.15
carbonate)b
JSR micro (AC/graphitic 3.8 1,000 4 4 11.2 900 7,987 0.113 0.073
carbon)
2,000 1.9 3.8 12.1 1,038 9,223 0.206 0.132
(1) Energy density at 400 W/kg constant power, Vrated – 1/2 Vrated
(2) Power based on P=9/16*(1
EF)*V2/R, EF=efficiency of discharge
a
Except where noted, all the devices use acetonitrile as the electrolyte
b
Except for these, all devices are packaged in metal containers

Hence on a volume basis W/l at 95% efficiency, the
differences between the batteries and ultracapacitors
are often quite small. Hence for HEVs, batteries alone
and ultracapacitors alone can be an option with the
decision being based on cycle life and cost in addition
to relative power capability [7].
For plug-in hybrid and battery electric vehicle
applications, the maximum usable power density
from the lithium-ion battery can be higher than in an
HEV because the peak power of the driveline is used
less frequently and consequently charge/discharge effi-
ciently is less important. For example, a pulse power
efficiency of 80% is probably sufficient and most of the
lithium batteries have a power capability of greater
than 1,000 W/kg, 2,200 W/l for that efficiency.
In addition, the battery is larger (heavier) in these
vehicles and as a result, the power density requirement
is less demanding. For PHEVs and EVs, the best appli-
cation of ultracapacitors is likely to be in combination
with batteries designed for high energy density, long
498 Energy Storage: Ultracapacitor

Energy Storage: Ultracapacitor. Table 3 Comparisons of the power capabilities of various devices for HEVs and PHEVs
using the different methods of calculation

Device
Lithium batteries Matched USABC Efficient pulse Efficient pulse
60% SOC impedance Min/max EF =95% EF =80%
Kokam NCM 30 Ah 2,893 2,502 550 1,848
Enerdel HEV 5,491 4,750 1,044 3,507
NCM 15 Ah
Enerdel EV 2,988 2,584 568 1,908
NCM 15 Ah
EIG NCM 2,688 2,325 511 1,721
20 Ah
EIG FePhosph. 2,141 2,035 458 1,540
15 Ah
Altairnano LiTiO 1,841 1,750 350 1,180
11 Ah
Altairnano LiTiO 4,613 4,385 992 3,341
3.8 Ah
Ultracapacitors
V0 = 3/4VRated
Maxwell 2,890 F 8,836 4,413 994
Nesscap 3,100 F 8,730 4,360 982
BatScap 2,700 F 18,224 9,102 2,050
Yunasko 450 F 22,838 11,406 2,569
LS Cable 3,200 F 12,446 4,609 1,038
JSR 2,000 F 9,228 6,216 1,400

cycle life, and low cost. In those cases, as discussed in
a later section of the entry, the ultracapacitors greatly
reduce the peak currents and dynamic stress on the
batteries and thus extend their cycle life.
Systems Considerations and Advantages of
Combining Ultracapacitors and Batteries
Vehicle Design Considerations
The energy storage requirements vary a great deal
depending on the type and size of the vehicle being
designed and the characteristics of the electric
powertrain to be used. Energy storage requirements
for various vehicle designs and operating modes are
shown in Table 4 for a midsize passenger car. Require-
ments are given for electric vehicles and both charge-
sustaining and plug-in hybrids. These requirements
can be utilized to size the energy storage unit in the
vehicles when the characteristics of the energy storage
cells are known. In some of the vehicle designs consid-
ered in Table 4, ultracapacitors are used to provide the
peak power rather than batteries.
In the cases using only ultracapacitors, the key issue
is the minimum energy (Wh) required to operate the
vehicle in real-world driving because the energy density
characteristics of ultracapacitors are such that the
 Energy Storage: Ultracapacitor 499

Energy Storage: Ultracapacitor. Table 4 Energy storage unit requirements for various types of electric drive midsize
passenger cars

Type of electric System Usable Maximum pulse power at Cycle life (number Usable depth-
driveline voltage V energy storage 90–95% efficiency (kW) of cycles) of-discharge
Electric 300–400 15–30 kWh 70–150 2,000–3,000 deep
70–80%
Plug-in hybrid 300–400 6–12 kWh battery 50–70 2,500–3,500 deep
100–150 Wh 60–80%
ultracapacitors
Charge- 150–200 100–150 Wh 25–35 300 K–500 K Shallow
sustaining hybrid
ultracapacitors 5–10%
Micro-hybrid 45 30–50 Wh 5–10 300 K–500 K Shallow
ultracapacitors 5–10%

power and cycle life requirements will be met in
most cases if the unit is large enough to met the
energy storage requirement. As shown in Table 4, for
passenger car applications, the energy storage in the
ultracapacitor can be 150 Wh or less even if the
ultracapacitor is used alone for the energy storage.
Reasons for Combining Ultracapacitors and
Batteries
Most hybrid vehicles use batteries for energy storage.
The batteries must provide both the energy and power
required by the vehicle. In all cases, this requires com-
promises in designing the battery in that it cannot be
optimized for either energy density or power capability.
In addition, in most cases the battery is oversized
(stores more energy than is used by the vehicle) in
order to achieve long cycle life even for the shallow
cycles experienced in charge-sustaining hybrid vehicles
like the Toyota Prius. Combining ultracapacitors with
batteries permits the design of the batteries to be opti-
mized for energy density and cycle life and limits the
magnitude and frequency of maximum power pulses
experienced by the batteries during vehicle operation.
The result will be a lighter, smaller, and lower cost
battery which will have much longer cycle life (by as
much as a factor of 2–3) and will operate over signifi-
cantly narrower ranges of voltage and temperature than
would be the case using only a battery. In addition, the
cooling requirements and complexity of the battery
management unit of the battery will be reduced and
be less costly.
The ultracapacitors would provide the high-power
pulses in discharge and accept all the charge pulses
during regenerative braking of the vehicle. If needed,
the ultracapacitors can be recharged at high rates
during regenerative braking and/or using power from
the engine or fuel cell. The ultracapacitors would be
deep discharged (up to 75% of their rated energy)
when that energy is needed to power the vehicle. Even
for deep discharge operation, the ultracapacitors can be
expected to have a cycle life of at least 500,000 cycles.
As noted in Table 4, the energy stored (Wh) in the
ultracapacitors can be quite limited being less than
150 Wh for midsize passenger cars. Larger
ultracapacitor units would be needed for buses and
trucks, but their advantages would be similar to those
used in passenger cars. The weight and cost of
the ultracapacitor unit would be proportional to the
energy stored so it is advantageous to limit the
energy stored to the minimum required for
a particular application. As shown in Fig. 1, for most
applications, it will be necessary to utilize interface
electronics to control the power flow from the
ultracapacitors because their voltage changes signifi-
cantly during discharge. This permits a large fraction
of the usable energy of the ultracapacitors to be utilized
and minimizes their cost.
500 Energy Storage: Ultracapacitor

Motor Trans-
Ultracap Motor
controller mission

Battery DC/DC Engine

Clutch

Motor Trans-
Battery Motor
controller mission

Ultracap DC/DC Engine

Clutch

Energy Storage: Ultracapacitor. Figure 1

Schematics of powertrains using ultracapacitors and batteries

Sizing Components: Weight and Volume Energy storage units for charge-sustaining hybrid
powertrains for midsize passenger cars are shown
Whether or not a vehicle designer will consider using
in Table 5. For each application, the usable power
ultracapacitors alone or in combination with batteries
and energy, total energy stored, and incremental
instead of a battery pack for energy storage will depend
state-of-charge for the unit and its weight and volume
to a large extent on the relative weight, volume, and
are shown in the table. The results shown in Table 5
cost of the two systems.
indicate that for charge-sustaining hybrids, either bat-
Consider first the case of charge-sustaining hybrid
teries or ultracapacitors can be used alone. All the units
vehicles (HEVs). In this case, the energy storage unit is
listed should function satisfactorily for the applications
sized by both usable power (kW) and energy storage
indicated with the decision concerning their relative
(Wh) requirements. For batteries, the key issues are the
attractiveness based to a large extent on cost, which at
power requirement and the minimum usable energy
the present time is uncertain for all the systems. The
consistent with high cycle life for shallow cycles.
results in Table 5 also indicate that a combination of
The total energy stored in the battery unit is of second-
lead-acid batteries and ultracapacitors would result in
ary importance as far as operation of the vehicle is
a relatively low weight unit for a micro-hybrid having
concerned, but it has a strong effect on the weight,
a maximum electric motor power of 5–10 kW. The
volume, and cost of the unit. For ultracapacitors, the
economics of such a combination should be attractive
key issue is the minimum energy (Wh) required to
because of the low cost of the lead-acid batteries.
operate the vehicle in real-world driving because the
Next consider energy storage units for plug-in
energy density characteristics of ultracapacitors are
hybrid vehicles (PHEVs). A key design parameter for
such that the power and cycle life requirements will
PHEVs is the all-electric range. Energy storage units
be met if the unit is large enough to meet the energy
will be considered for all-electric ranges of 10, 15, 20,
requirement.
 Energy Storage: Ultracapacitor 501

Energy Storage: Ultracapacitor. Table 5 Characteristics of various technologies/types of batteries for use in charge-
sustaining hybrid vehicle applications

Charge-sustaining hybrids (HEVs)

Technology C or Ah System Δ SOC% Total Max. Pulse Weight of Volume of
voltage energy Wha Power kWb efficiency% cells kg cells l
Ultracapacitors
Carbon/carbon 2,400 160 75 100 >40 90 22 15
microporous C/ 2,400 160 75 150 27 90 18 13
graphitic C
Lithium-ion batteries
Graphite/Ni Co Al 9 160 6.6 1,500 31 91 24 11
Iron phosphate 11 160 5.5 2,000 27 91 23 11
2.2 Ah cells
Lithium titanate 11 160 5.5 1,800 27 90 26 14
11 Ah
Lithium titanate 4 160 16.6 600 27 90 16 9
2.5 Ah
Nickel metal 14 90
hydride (Prius)
10 160 6.25 1,600 25 80 35 19
Micro-hybrids
Lead-acid 50 45 5 1,800 5 90 53 19
Lead-acid/ 28 10 1,000 29 10
ultracapacitors
4,000 45 75 50 15 95 10 8
39 total 18 total
Nickel metal 24 45 10 1,100 10 90 24 13
hydride (Prius)
a
Total energy stored
b
Power at the pulse efficiency indicated

30, and 40 miles. The acceleration performance of all
the vehicles will be the same (0–60 mph in 8–9 s). For
the batteries, the usable depth of discharge will be taken
to be 70%. For the ultracapacitors, it is assumed that
they can be cycled from rated to one-half rated voltage
using 75% of the total stored energy. In the PHEVs, the
battery will be sized by the energy needed to sustain
a specified all-electric range. Hence the weight and
volume of the battery pack follow directly from its
energy density (Wh/kg, Wh/l). This means that battery
technologies with high energy density will be strongly
favored for PHEV applications. However, the batteries
must also be able to meet the power requirements of
the PHEV powertrain which can approach those of the
battery-powered EV. Unfortunately, battery designs to
attain the maximum energy density in most cases
require a sacrifice in power capability as shown in
Table 1. Calculated power density requirements for
PHEVs of various ranges and maximum power [8]
are given in Table 6 for batteries having energy densities
of 70, 100, and 200 Wh/kg. The values shown are for
a midsize passenger car having an energy usage of
250 Wh/mile in the all-electric mode and maximum
electric motor power of 50 and 70 kW.
502 Energy Storage: Ultracapacitor

Energy Storage: Ultracapacitor. Table 6 Battery sizing and power density for various ranges and motor power
Battery 200 Wh/kg Battery 100 Wh/kg Battery 70 Wh/kg
Range kWh kWhb 50 kW 70 kW 50 kW 70 kW 50 kW 70 kW
a b
miles needed stored kg kW/kg kW/kg kg kW/kg kW/kg kg kW/kg kW/kg
10 2.52 3.6 18 2.78 3.89 36 1.39 1.94 51 0.98 1.37
15 3.78 5.4 27 1.85 2.59 54 0.92 1.30 77 0.65 0.91
20 5.04 7.2 36 1.39 1.94 72 0.69 0.97 103 0.49 0.68
30 7.56 10.8 54 0.93 1.30 108 0.46 0.65 154 0.32 0.46
40 10.1 14.4 72 0.69 0.97 144 0.35 0.49 206 0.24 0.34
a
Vehicle energy usage from the battery: 250 Wh/mi
b
Usable state-of-charge for batteries: 70%, weights shown are for cells only

Energy Storage: Ultracapacitor. Table 7 Battery/ultracapacitor combinations for various all-electric ranges

Wh/kg 100 200 100 90 70 70

Range miles GAIA Quallion 200 Quallion 100 A123 Altairnano Panasonic Nickel mt.hyd.
a
10 N/N Y/Y N/N N/N N/N Y/Y
15 N/N Y/Y N/N N/N N/N Y/Y
20 N/N Y/Y N/N N/N N/N Y/Y
30 N/N Y/Y N/N N/N N/N Y/Y
40 N/N N/N N/N N/N N/N Y/Y
a
X/X means need at 50 kW/need at 70 kW yes Y or no N

The required power densities for all-electric ranges
of 20 miles and less are high especially for high energy
density batteries. When the power density required
exceeds that of the battery, it is of interest to consider
using an ultracapacitor to augment the power of the
battery. The usable pulse power from a battery depends
on the efficiency that is acceptable for a particular
application. The power densities for efficiencies of
95% and 80% for the various battery technologies are
shown in Table 3. If it is assumed that 80% efficiency is
satisfactory for use of the battery in the charge-
depleting mode of the PHEV, it can be concluded
from Tables 3 that ultracapacitors would be needed by
the 200 Wh/kg lithium-ion and the 70 Wh/kg nickel
metal hydride batteries. This result is shown in Table 7.
The energy storage unit weights for the PHEVs of
various all-electric ranges and battery energy density
are given in Table 8. The combination of
ultracapacitors and 200 Wh/kg lithium-ion batteries
have the lowest weight for all the ranges. For higher
energy density batteries (>250 Wh/kg), it seems even
more likely that ultracapacitors will be needed to meet
the power demands of PHEVs. The combination of
ultracapacitors and batteries will also yield the highest
efficiency during the charge-sustaining mode of vehicle
operation after the energy of the batteries has been
depleted.
As noted above, ultracapacitors can be used with
70 Wh/kg nickel metal hydride batteries for PHEV
applications due to their limited power capability.
Even using ultracapacitors, the all-electric range with
nickel metal hydride batteries is limited to about
 Energy Storage: Ultracapacitor 503

Energy Storage: Ultracapacitor. Table 8 Storage unit weights using a combination of batteries and ultracapacitors for
various all-electric ranges

Wh/kg 5 200 100 70

Range
miles Ultracap kga Battery Kgb Combination kg Battery kg Combination kg Battery kg Combination kg
10 20 18 38 36 56 51 71
15 20 27 47 54 74 77 97
20 20 36 56 72 92 103 123
30 20 54 74 108 128 154 174
40 20 72 92 144 164 206 226
a
The ultracapacitor unit stores 100 Wh usable energy
b
Weights shown are for cells only, packaging into modules not Included

15 miles as otherwise the weight of the energy storage
unit would exceed 100 kg.
Power Split Strategies
As noted previously, interface electronics is needed
(Fig. 1) in hybrid powertains using a combination of
batteries and ultracapacitors to control the power split
between the two energy storage units. Selection of
a control strategy that will both limit the maximum
power demanded of the batteries and improve the effi-
ciency of the powertrain is critical to the design of
attractive HEVs and PHEVs. One approach for the
PHEV applications is to have the battery meet the aver-
age vehicle power requirement and the ultracapacitor
meet the variable, pulse power requirement. The average
vehicle power demand can be calculated over some
specified time period (30–120 s) depending on the driv-
ing conditions. As discussed later, vehicle simulations
indicate that this approach results in relatively slow
variations in battery power and should lead to signifi-
cantly increased battery cycle life and reduced need for
high-power cells. Using this approach, the interface elec-
tronics could be directly attached to either the batteries
or ultracapacitors (see Fig. 1).
As in the case of the HEVs, the key element of the
system control strategy is how the ultracapacitors are
recharged utilizing engine power via the motor/gen-
erator. The vehicle simulations have shown that the
ultracaps should be recharged during periods in which
the engine is needed to meet the vehicle power
requirements. In this way, the engine operates nearly
all the time at high efficiency and not at low output
torque conditions resulting in large improvements
(>35%) in fuel economy compared to the reference
engine alone case.
Laboratory and Vehicle Field Experiences Using
Ultracapacitors
Laboratory Testing of Ultracap/Battery
Combinations
Most of the research [9–11] involving the testing of
ultracap/battery combinations has been performed on
bench-top arrangements. In most cases, these tests have
included state-of-the-art interface electronics that con-
trolled the currents from the ultracapacitors. The intent
of these studies has been to show that the stress on the
batteries due to repeated high current pulses was
greatly reduced using ultracapacitors resulting in
extended battery cycle life. The test data show clearly
that the mean currents from the batteries are greatly
reduced, but the effect on battery cycle life is not yet
clear from the data. Much further testing of this type,
especially using lithium batteries, is needed. The
test results [12] from a simpler test utilizing small
spiral wound lithium batteries (1.4 Ah) and 310
F ultracapacitors without interface electronics are
shown in Fig. 2. The data show the effect of load
504 Energy Storage: Ultracapacitor
3.0
End of discharge voltage (V)
2.5
2.0
Valence IFR18650E Li-ion battery
1.5
0 500
Energy Storage: Ultracapacitor. Figure 2
Pulsed discharge at average rate of 1 C comprising 5 C 5 s pulses at a 20% duty factor. Each cycle 80% (1.12 Ah)
leveling on the cycle life of the cells which were pulsed
at the 5C rate. The magnitude of the effect (a factor of
7–8) is likely much greater than can be expected with
larger batteries, but nevertheless, the data do indicate
that load-leveling batteries can be expected to signifi-
cantly improve cycle life.
Vehicle Field Experience Using Ultracapacitors
There has been continuing experience in demonstrat-
ing the use of ultracapacitors in hybrid-electric
powertrains in heavy-duty [13–15] and light-duty
[16–18] vehicles. In these applications, the capacitors
provide the peak power for acceleration and recover
energy during periods of braking. An engine generator
or fuel cell is used to generate onboard electricity to
recharge the capacitors as needed. The control strategy
for the hybrid vehicle should be designed to operate the
engine or fuel cell as close to the maximum efficiency
AFS trinity cycle testing
pulse testing Δ80% SOC
Hybrid pack UUT3 vs. valence cell UUT6
Hybrid pack: valence IFR18650E Li-ion batteries
+ 2 Maxwell BCAP0310 P250 ultracapacitors
1000 1500 2000 2500 3000 3500 4000
Cycle number
BoostCap
500Ω / 10mA
+ 0310 P250
–
BoostCap 500Ω / 10mA
0310 P250
condition as the driving cycle permits. Capacitor units
for use in hybrid vehicles must be high voltage as the
peak power for these applications is high (up to
200 kW). For passenger cars and SUVs, the voltage is
likely to be about 300 V and for transit buses about
700 V. This would require 125 cells in series for the
light-duty vehicles and 250 cells in series for the buses.
Transit Bus Applications In the USA, the company
most active in utilizing ultracapacitors in hybrid-
electric powertrains for buses and large trucks has
been the ISE Corporation in San Diego, California
[13]. ISE has developed a 360 V capacitor unit
consisting of 144 2,600 F Maxwell cells connected in
series (see Fig. 3). The weight and volume are 114 kg
and 189 l, respectively. The unit stores 0.325 kWh of
energy (0.245 kWh usable). In a transit bus, two of the
units are used in series resulting in a voltage of 720 V
and energy storage of 0.650 kWh. The peak power

Nominal voltage
Peak voltage
Rated current
Capacitance
Total energy stored nominal / max.
Leakage current
Operating temperature
Weight
Dimensions wxlxh
Standard pack
Fire suppression system
Energy Storage: Ultracapacitor. Figure 3
The ISE ultracapacitor unit (360 V, 0.325 kWh)
capability of the combined unit is over 300 kW. ISE
utilizes this capacitor unit with a 225 kW electric motor
in series hybrids using gasoline and diesel engines and
hydrogen fuel cells. Since the capacitor unit stores only
about 0.5 kWh, it can provide power only during vehi-
cle acceleration and recovers energy during braking.
The engine or fuel cell must provide all the power
during cruise and hill climbing. ISE has built over 100
buses using the ultracapacitor energy storage units for
transit companies in Southern California. The buses
are in daily revenue service. Field experience has shown
that the capacitor unit functions as planned with a high
reliability. Field operation of the buses has shown a fuel
economy improvement of 25% compared to the stan-
dard diesel engine buses [13].
Hybrid-Electric Passenger Car Applications Unlike
the transit bus applications described in the previous
section, there has not been a commercialized use of
ultracapacitors in passenger cars as of 2010. In the last
couple of years, there has been considerable discussion
[19, 20] of marketing of micro-hybrids with stop-start
Energy Storage: Ultracapacitor 505
360V*
403V**
400A
18.05F
0.325k Wh/0.407k Wh
5mA nominal
–35 to 65°C
240lbs
24″×40″×12″
144 capacitors
Heat activated halotron system
capability to improve fuel economy particularly in
Europe. It is likely that when these cars are finally
marketed some of them will use ultracapacitors.
There have been several noteworthy demonstrations
of ultracapacitors in hybrid-electric passenger cars. As
discussed in the following paragraphs, these projects
were very successful and showed that ultracapacitors
can be used in place of batteries in mild hybrid vehicles
with fuel economy improvements at least as high as
with batteries.
The first project was termed “SUPERCAR” and was
funded by the European Community (EC). It was
a joint project between EPCOS [16], the ultracapacitor
developer, and Siemens VDO [21], the vehicle integra-
tor. The parallel hybrid passenger car (VW Golf) com-
bined the ultracap module and lead-acid battery into
a 42 V, 10 kW (peak) electric driveline with a 66 kW
engine. The vehicle was tested on both the chassis
dynamometer and the road. The tests showed a
16–18% improvement in fuel economy compared to
the standard ICE car. The capacitor unit consisted of 24
3,600 F cells connected in series (see Table 2 for the cell
506 Energy Storage: Ultracapacitor
Energy Storage: Ultracapacitor. Figure 4
The EPCOS 60 V ECC module (150 F, 56 Wh)
characteristics). The rated voltage of the unit is 60 V
and its capacitance and resistance are 150 F and
8 mOhm, respectively. The unit is shown in Fig. 4.
A second, more recent project (2009), is described
in detail in [17]. This project was performed by the
National Renewable Energy Laboratory (NREL) and
General Motors (GM). It involved installing
ultracapacitors in a 2007 Saturn Vue hybrid along
with the nickel metal hydride battery. The vehicle
could be operated on either of the energy storage
units alone in order to make a direct comparison of
its operation on the ultracapacitors and battery. The
ultracapacitor energy storage unit consisted of one or
two 48 V, 165 F modules from Maxwell. Each module,
which consisted of 18 3,000 F cells connected in series
(see Table 2 for the characteristics of the cells), stored
about 35 Wh. A special UCAP state estimator was
utilized to maintain the ultracapacitors in the required
range of state-of-charge.
The 2007 Saturn Vue hybrid is a mild hybrid using
a belt alternator/starter in parallel with a 127 kW, var-
iable valve timing gasoline engine. The alternator/
starter electric machine has a continuous rating of
3 kW and a peak power of 6–8 kW. The nickel metal
hydride battery has a nominal rating of 36 V, but
vehicle testing indicated it operates in the 38–42 V
range most of the time at 75% state-of-charge. The
characteristics of the energy storage units are compared
in Table 9.
Extensive dynamometer testing of the vehicle was
done operating on the ultracapacitors (both with one
module and two modules) and the nickel metal hydride
battery. The vehicle was tested on several driving cycles
(FUDS, FHWY, and US06). The three energy storage
Energy Storage: Ultracapacitor. Table 9 Characteristics
of the energy storage units used in the Saturn Vue test
vehicle
Weight Volume Energy
Storage unit (kg) (l) stored
Nickel metal hydride 24.7 15.4 1.1 kWh
battery (estimated)
Ultracapacitors
One 48 V module 14.8 11.2 35 Wh
Two 48 V modules 29.6 22.4 70 Wh
options could be utilized interchangeably using the
same vehicle operating strategy. The vehicle operated
essentially the same on each of the energy storage
options. The testing indicated that only 25–30 Wh
was used from the ultracapacitor units even on
the US06 driving cycle. When using only one
ultracap module on the FUDS cycle, the state-of-charge
(V/Vrated) of the ultracap unit varied from 75–90%.
There was no difference in the measured fuel economy
when the vehicle was operated on the ultracapacitors or
the nickel metal hydride battery for any of the three
driving cycles. Test track acceleration tests of the
vehicle indicated no differences in the acceleration
times for 0–60 or 40–60 mph using the one module
ultracapacitor and nickel metal hydride battery units.
Fuel Cell Vehicle Applications Ultracapacitors can
also be used to meet the peak power transients in vehi-
cles powered by fuel cells [22, 23]. The most publicized
use of ultracapacitors in fuel cell vehicles has been that
by Honda [22] in the “Honda FCV.” In the Honda FCV,
the capacitors were used to load level the fuel cell
during accelerations and to recovery energy during
regenerative braking. In the Honda fuel cell system,
the ultracapacitors are used without interface electron-
ics. The capacitors functioned very well and resulted in
significant improvements in system energy efficiency.
A detailed study of the use of ultracapacitors in fuel
cell–powered vehicles is given in [24]. The study
showed that ultracapacitors are well suited to be used
with fuel cells. This is especially true when the fuel cell
and ultracapacitors are connected in parallel because
the effective rest voltage of the capacitor is variable and
their high-power capability permits them to meet the

Energy Storage: Ultracapacitor. Figure 5
Photograph of the ultracapacitor/fuel cell test setup
maximum power requirement of the system at all volt-
ages between the rest voltage and the minimum voltage
at which the fuel cell is to be operated. In this mode of
operation, the system is self-controlling with the
ultracapacitor and fuel cell sharing the power demand
based on the system voltage of operation. Tests [24]
were performed on a system (Fig. 5) consisting of the
Ballard 1.2 kW Nexa fuel cell and an ultracapacitor
bank made up of 17 Ness 2,500 F capacitors. The
system was tested for 10 s pulses up to 5 kW. An
analytical model of the fuel cell/capacitor system was
derived and programmed using an EXCEL spreadsheet.
Comparisons of the measured and modeled response
of the system at various power levels (Fig. 6) indicated
that the system operated as modeled and that the self-
control of the fuel cell was as expected. SIMPLEV
simulations [24] on the FUDS driving cycle were
also performed for a midsize SUV powered by
ultracapacitors connected in parallel with a fuel cell. It
was found that the ultracapacitors significantly load
leveled the fuel cell and the behavior of the system
was much like that measured with the Ballard 1.2 kW
Nexa fuel cell.
Hybrid Vehicle Simulations Using Ultracapacitors
Alone
Powertrain Control Strategies with Ultracapacitors
Alone in Place of Batteries
Schematics of hybrid powertrains using ultra-
capacitors and batteries are shown in Fig. 1. If an
Energy Storage: Ultracapacitor 507
ultracapacitor unit is used alone, there is no need
for special electronics if the motor electronics can
handle the increased voltage swing of the ultra-
capacitors. If that is not the case, then interface elec-
tronics is needed between the ultracapacitors and the
motor electronics.
When ultracapacitors are used alone as the
energy storage unit in a charge-sustaining hybrid
(HEV), the objective of the control strategy is to permit
the engine to operate near its maximum efficiency.
As shown in [25–27], this can be done by operating
the hybrid vehicle on the electric drive only when the
power demand is less than the power capability of the
electric motor; when the vehicle power demand exceeds
that of the electric motor, the engine is operated to
meet the vehicle power demand plus the power to
recharge the ultracapacitor unit. In this mode, the
electric machine is used as a generator and the engine
operating point is selected along its maximum effi-
ciency line (torque vs. RPM). The recharging power is
limited by the power of the electric machine because
ultracapacitors can have pulse power efficiency greater
than 95% for W/kg values of over 2,000 W/kg (see
Table 3). This control strategy is referred to as the
“sawtooth” strategy because a plot of the ultracapacitor
state-of-charge (SOC) has the form of a saw blade. It
can also be referred to as the “high engine efficiency”
strategy.
Vehicle Simulation Results
Simulations of midsize passenger cars using the
ultracapacitors in micro-hybrid and charge-sustaining
hybrid powertrain designs have been performed using
the Advisor vehicle simulation program modified with
special routines at UC Davis [25–27]. All the
powertrains were in the same vehicle having the fol-
lowing characteristics: test weight 1,660 kg, Cd =0.3,
AF = 2.25 m2, RRCF = 0.009. The engine map used in
the simulations was for a Ford Focus 2 l, 4-cylinder
engine. The rated engine power was 120 kW for the
conventional ICE vehicle and the micro-hybrid
and 110 kW for the charge-sustaining hybrids. All the
hybrids use the single-shaft arrangement similar to
the Honda Civic hybrid. The same induction electric
motor map was used for all the vehicle designs.
The simulation results are summarized in Table 10
for a conventional ICE vehicle and each of the
508 Energy Storage: Ultracapacitor

5000

4000

3000 Total power

Capacitor power
Fuel cell power
Total power
Power - W

Calculated total power

2000
Capacitor power calc
Fuel cell power calc

1000
Fuel cell power

Modeled

0
3.08
20.4
43.2
81
77.8
14.6
1
178
145
162
173
185
212
225
246
283
279
296
313
330
317
333
309
292
414
437
447
454
Measured Capacitor power

–1000
Time

Energy Storage: Ultracapacitor. Figure 6

Comparison of the measured and modeled ultracapacitor and fuel cell powers for the 5 kW pulse power test

hybrid designs. Results are given for fuel usage in terms of hybrid designs, the rated engine power used was the
both l/100 km and mpg for various driving cycles. It same as that in the conventional ICE vehicle in order
is clear from Table 10 that large improvements in that the performance of the hybrid vehicle when the
fuel usage are predicted for all the hybrid power- energy storage in the ultracapacitors is depleted would
trains using ultracapacitors for energy storage. The sim- be the same as the conventional vehicle. The
ulation results will be discussed separately for each ultracapacitors were used to improve fuel economy
hybrid design. with only a minimal change in vehicle acceleration
performance. The control strategy used was the “saw-
Micro-hybrids The results for the micro-hybrids are tooth” strategy discussed in the previous paragraph. As
particularly interesting and somewhat surprising, shown in Fig. 7, this resulted in a large improvement in
because of the large fuel economy improvements average engine efficiency from 19% in the ICE vehicle
predicted. These improvements were about 40% on to 30% in the micro-hybrid even though the electric
the FUDS and ECE-EUD cycles and 20% on the Federal motor had a peak power of only 6 kW.
Highway and US06 cycles using the carbon/carbon Additional computer simulations were made for
ultracapacitors. The improvements are significantly higher motor power (up to 12 kW) and larger
less using the advanced hybrid carbon units because ultracapacitor energy storage (up to 50 Wh). It was
of their lower round-trip efficiencies. In the micro- found that the improvements in fuel economy were only
 Energy Storage: Ultracapacitor 509

Energy Storage: Ultracapacitor. Table 10 Summary of the vehicle fuel economy simulation results using ultracapacitors
for various driving cycles

Driveline Voltage and L/100 km/mpg

type Energy storage type weight cells (kg) EM Peak kW FUDS HWFET US06 ECE-EUDC
ICE baseline 10/23.8 6.9/34.4 9.6/24.7 9.7/24.6
Micro-HEV Lead-acid/ultracaps 48
Carbon/carbon 6 kg 6 5.7/41.7 5.3/44.7 7.8/30.6 5.9/40.2
Hybrid carbon 3 kg 6 7.3/32.8 6.3/38.0 8.9/26.7 7.1/33.4
Charge- Ultracaps 200
sustaining
Carbon/carbon 30 kg 35 5.4/43.8 5.0/47.9 7.1/33.6 5.5/43.2
hybrid
Hybrid carbon 13 kg 35 5.8/40.9 5.2/45.8 8.0/29.9 5.8/41.3
Plug-in 12 kWh Li battery 300 70 kW with
hybrid. (200 Wh/kg) and 45 kW from
ultracaps caps
Carbon/carbon 40 kg 45 5.5/43.2 5.0/47.7 7.0/33.9 5.5/42.9
Hybrid carbon 18 kg 45 5.8/41.2 5.2/46.2 7.9/30.2 5.8/41.2

Fuel converter efficiency Fuel converter efficiency

0.35 0.35

0.3 0.3

0.25 0.25
Efficiency

Efficiency

0.2 0.2

0.15 0.15

0.1 0.1

0.05 0.05

0 0
0 200 400 600 800 1000 1200 1400 0 500 1000 1500 2000 2500 3000 3500 4000 4500
Time (s) Time (s)
Engine operating efficiency for the ICE vehicle- Engine operating efficiency for the micro-hybrid-
average engine efficiency .19 average engine efficiency .30

Energy Storage: Ultracapacitor. Figure 7

A comparison of engine efficiencies for a conventional ICE vehicle and a micro-hybrid on the FUDS cycle using carbon/
carbon ultracapacitor

marginally greater. However, using a motor power of 3 kW
reduced the fuel economy improvement on the FUDS by
more than 50%. Note from Table 10 that the fuel economy
improvements using the carbon/carbon ultracapacitors
were for all the cycles greater than those using the
advanced hybrid carbon devices. This was the case because
the round-trip efficiencies for the carbon/carbon units
were 95–98% and those of the hybrid carbon units were
75–90% for the various driving cycles. As shown in
Table 2, the advanced hybrid carbon devices had higher
energy density, but even though their power density for
95% efficiency was relatively high (1,050 W/kg), it was not
510 Energy Storage: Ultracapacitor

proportionally higher – that is twice as high – as the Plug-in Hybrid Vehicle Simulations Using
carbon/carbon devices with lower energy density. These Advanced Batteries and Ultracapacitors
results show that it is essentially to develop high energy
Control Strategy for Plug-in Hybrid Vehicles
density ultracapacitors with proportionally higher power
(PHEVs)
density; otherwise their use in vehicle applications will be
compromised. For plug-in hybrid vehicles, batteries and
ultracapacitors would be used in combination with
the batteries providing the energy and the
Charge-Sustaining Mild/Strong Hybrids The fuel
ultracapacitors the power. The powertrain schematics
economy simulation results for mild/strong charge-
are shown in Fig. 1. The control strategy in the charge-
sustaining hybrids are also shown in Table 10 for a mid-
depleting mode is to limit the power from the battery
size passenger car using both carbon/carbon and
to the average power needed by the vehicle with the
advanced hybrid carbon ultracapacitors. Using the car-
ultracapacitor providing the additional power during
bon/carbon ultracapacitor unit, the fuel savings are
vehicle accelerations. The ultracapacitors also accept all
about 45% for the FUDS and ECE-EUD cycles and
the energy recovered during regenerative braking. If
about 27% for the Federal Highway and US06 cycles.
engine operation is needed, the “sawtooth” strategy is
These improvement values are higher than for the
used with the ultracapacitors being recharged using
micro-hybrid, but not by as large a factor as might be
engine power. In the charge-sustaining mode of oper-
expected. The prime advantage of the high-power elec-
ation of the PHEV, the electric drive is operated using
tric driveline in the mild/strong charge-sustaining
only the ultracapacitors for energy storage like that
hybrid is that it yields large fuel economy improve-
previously described for mild/strong charge-sustaining
ments even for high-power-requirement driving cycles
hybrid vehicles. The PHEV operating strategies are
like the US06. The fuel economy improvements using
summarized in Fig. 8.
the advanced hybrid carbon ultracapacitor unit are not
much less (5–10%) than those with the carbon/carbon
Simulations of Plug-in Hybrid Vehicles Using
unit even though the round-trip efficiency of the
Advanced Batteries and Ultracapacitors
hybrid carbon unit is only 85–90% compared to 98%
for the carbon/carbon unit. Since the weight/volume of A detailed study of plug-in hybrids using advanced
the hybrid carbon unit is relatively small – 43% of that batteries is presented in [8]. The characteristics of the
of the carbon/carbon unit, it appears that the mild/ advanced batteries used in the simulations are given in
strong charge-sustaining hybrid application is a better Table 11.
one for the advanced hybrid carbon technology than The battery and ultracapacitor units used in the
the micro-hybrid application. simulations are given in Table 12. Simulations were

Low battery SOC & High power request

Low ultracap SOC (> opt. engine power)
OR &
High power request Low battery SOC &
(> max. motor power) Low ultracap SOC
OR
Low ultracap SOC &
High power demand HEV mode HEV mode
(> max. battery power)
EV mode Engine optimum Engine normal
operation operation

High ultracap SOC Low power request

(> ultracap mean SOC) (< opt. engine power)
& &
Low power request High ultracap SOC
(< max. motor power) (> ultracap mean SOC)

Energy Storage: Ultracapacitor. Figure 8

Summary of the operating strategy used in the simulation of plug-in hybrid vehicles
 Energy Storage: Ultracapacitor 511

Energy Storage: Ultracapacitor. Table 11 Characteristics of present and future battery cell technologies for EVs and
PHEVs

Chemistry Anode/ Cell voltage EV Wh/ HEV W/ EV W/ Cycle life Thermal

cathode Max/nom. Ah Wgt. kg R mOhm kg kg 95% kg 75% (deep) stability
Present technology batteries
Graphite/NiCoMnO2 4.2/3.6 30 0.787 1.5 140 521 2,060 2,000– fairly stable
3,000
Graphite/Mn spinel 4.0/3.6 15 0.424 2.7 127 540 2,120 1,500 fairly stable
Future technology batteries
Graphite/composite 4/3.6 5 0.09 20 200 250 1,350 – fairly stable
MnO2
Silicon carbon 4/3.6 20 0.24 4.5 295 621 2,250 – fairly stable
composites/
composite MnO2
Rechargeable zinc–air 1.3/1.15 20 60 6.6 385 156 616 – very stable
Present technology power devices
supercapacitor 2.7/1.35 500 F 0.068 1.3 5.5 2,320 11,600 500 K Very stable
activated carbon/
activated carbon
Power battery lithium 2.8/2.5 4 0.23 1.15 40 1,310 5,170 20–50 K Very stable
titanate oxide

Energy Storage: Ultracapacitor. Table 12 The advanced currents limited to about 300 A even in the cases of the
battery and ultracapacitor units used in the simulations batteries alone. In all cases, the batteries were depleted
Lithium-ion batteries to 30% SOC from 100% SOC in the charge-depleting
EIG 30 kg 20 Ah 60 in 3.5 kWh
mode of vehicle operation.
NiCoMnO2 cells series usable
Composite 32 kg 40 Ah 60 in 4.6 kWh Simulation Results for the PHEVs All the PHEV
MnO2 cells series usable simulations were performed using the following vehicle
Silicone 22 kg 30 Ah 60 in 4.6 kWh inputs:
composite cells series usable CD = 0.27, AF = 2.2 m2, fr = 0.008, test weight =1,650 kg
Zinc–air batteries (approx.)
Zn–air 32 kg 60 Ah 180 in 9.5 kWh Engine: Honda 1.3 l, iVTEC engine map, scaled to
cells series usable 90 kW
Electric motor: Honda hybrid Civic AC PM 2006 effi-
Carbon/carbon ultracapacitors
ciency map, scaled to 70 kW
Symmetric 20 kg 1,350 F 110 in 100 Wh
DC/DC inverter: constant efficiency 0.96
C/C cells series usable
Transmission: 5-speed manual (3.11, 2.11, 1.55, 1.0,
0.71, FD=3.95), automatically shifted in the
model, but future models would incorporate the
performed with the batteries alone and in combination DCT (dual clutch transmission) as a convenient
with the ultracapacitors. The nominal energy storage means to have smooth, fast shifting and to decouple
unit voltage was 240 V in all cases with the maximum the engine when it is not needed to provide power
512 Energy Storage: Ultracapacitor

Fuel and Electricity Use Characteristics The simu-
lation results are summarized in Tables 13 and 14
showing results for vehicle operation in the charge-
depleting and charge-sustaining modes of the PHEV.
With the batteries in combination with the
ultracapacitors, the PHEVs were able to operate in
the all-electric mode down to the battery SOC=30%
on the FUDS and FDHW driving cycles. In all cases for
the US06 driving cycle, the vehicle had blended oper-
ation (engine and electric drive both needed) in the
charge-depleting mode. The use of the ultracapacitors
with the batteries permitted all-electric operation over
a wide range of driving conditions with higher Wh/mi
for all the driving cycles. Hence in the charge-
depleting mode, the fuel economy (mpg) is higher by
50–100% using the ultracapacitors for all the batteries.
The fuel economy in the charge-sustaining mode is
also higher for all the driving cycles using the
ultracapacitors by 15–40%. The acceleration times of
the vehicle were lower using the ultracapacitors than
for the batteries alone. With the ultracapacitors, the
acceleration times were 2.7 s for 0–30 mph and 6.9 s
for 0–60 mph. For the batteries alone, the acceleration
times varied somewhat with the battery used ranging
from 2.9 to 3.2 s for 0–30 mph and from 8.6 to 9.8 s for
0–60 mph. Hence in all respects, vehicle performance
was improved using the ultracapacitors for all the
batteries studied.
Reductions in Battery Stress Using Ultracapacitors
The current and voltage responses of the batteries with
and without the ultracapacitors are shown in Figs. 9–12
for the silicone carbon lithium-ion and the zinc–air
batteries for the FUDS and US06 driving cycles. The
effects of the load leveling of the power demand from
the batteries using the ultracapacitors are evident in the
figures. Both the average currents and the peak currents
from the batteries are lower by a factor of 2–3 using the
ultracapacitors. The minimum voltages of the batteries
are significantly higher using the capacitors and the
voltage dynamics (fluctuations) are dramatically
reduced. Hence, the stress on the battery and resultant
heating are much reduced. The simulation results in
Figs. 9–12 also show that the ultracapacitors are uti-
lized over a wide voltage range indicating that a large
fraction of their usable energy storage (100 Wh) is
being used to load level the batteries. This is only

Energy Storage: Ultracapacitor. Table 13 Simulation results for the advanced batteries with ultracapacitors
Battery Range kW max. kW Eff. kW Eff. Wh/ Operat. mpg mpg Ch. Sus.
Type (1) Cycle mi. control max bat. Bat. max. Cap. Cap. mi Bat. mode 20mi 40mi HEV mpg
Compos. FUD 22 40 18 0.94 40 0.97 215 AE none 97 52.8
MnO2
32 kgbat HW 20 45 18 0.91 45 0.96 227 AE none 109 56.3
20 kgcap US06 30 68 21 0.91 68 0.94 180 blended 71.9 56 38.3
Si Carb/ FUD 20 40 18 0.94 40 0.97 220 AE none 99 52.8
Compos.
MnO2
22 kgbat HW 20 45 19 0.91 45 0.97 225 AE none 110 56.8
20 kgcap US06 30 68 21 0.91 68 0.94 190 blended 71.1 52 38.4
Rech. FUD 40 45 19 0.87 45 0.97 228 AE none none 54.5
Zn–air
32 kgbat HW 38 45 19 0.81 45 0.97 242 AE none none 57.7
20 kgcap US06 66 68 21 0.82 68 94 149 blended 62.4 60 38.8
(1) Weight of cells only
 Energy Storage: Ultracapacitor 513

Energy Storage: Ultracapacitor. Table 14 Simulation results for the batteries alone
Range kW max. kW max Eff. Wh/mi Operat. mpg mpg Mpg Ch.
Battery Type (1) Cycle mi. control bat. Bat. Bat. mode 20mi. 40mi sus. HEV
EIG FUD 27 30 30 0.94 125 blended 134 85 47
NiCoMn
30 kg HW 24 20 20 0.93 137 blended 110 87 47
US06 57 58 58 0.88 blended 48 45 37
Compos. MnO2 FUD 36 30 30 0.92 135 blended 134 104 46.9
32 kgbat HW 31 20 20 0.91 147 blended 167 113 46.6
US06 64 58 58 0.87 92 blended 48 48 34.1
Si Carb/Compos. FUD 35 30 30 0.93 138 blended 138 106 46.9
MnO2
22 kgbat HW 32 20 20 0.92 148 blended 169 114 46.9
US06 64 58 58 0.88 87 blended 48 48 35.7
Rech. Zn–air FUD 66 30 30 0.84 139 blended 139 137 39.4
32 kgbat HW 63 20 20 0.83 156 blended 169 169 41.1
US06 93 36 36 0.72 101 blended 48.5 48.5 30.1
(1) Weight of cells only

possible using a DC/DC converter between the battery
and the DC bus.
The simulation results also indicate that using
ultracapacitors, batteries with a wide range of power
characteristics can be used in PHEVs and also EVs
without sacrificing vehicle performance and subjecting
the batteries to high stress and resultant shorter cycle
life. This could be especially important in the future as
high energy density batteries such as zinc–air and pos-
sibly lithium–air are developed. It is likely that those
battery types will not have commensurate increases in
usable power density and without ultracapacitors,
the battery unit in PHEVs and EVs would be sized
by the maximum power requirement (kW) rather
than the range (mi)/energy requirement (kWh). This
would significantly increase weight, volume, and
the cost of the battery unit. It is also unlikely that the
air electrode will have high charge acceptance capabil-
ity and thus regenerative braking performance
approaching that of ultracapacitors or even lithium-
ion batteries. This is another advantage of the use of
ultracapacitors with the air-electrode batteries.
Fuel Cell Vehicle Simulations Using
Ultracapacitors
Fuel Cell Powertrain Configurations and Power Split
Strategies
A fuel cell vehicle powertrain consists of three elements:
(1) a fuel cell unit that consists of a fuel cell stack,
air and hydrogen supply, and water and thermal man-
agement systems; (2) an energy storage unit
(ultracapacitors or batteries) that can store the electric-
ity generated by the fuel cell as needed; (3) an interface
electronics unit that controls the power split between
the fuel cell and the energy storage unit. Fuel cells in
combination with energy storage can create high power
for vehicle traction with fast dynamic response, effi-
cient capture of regenerative braking energy, and
reduced stress on the fuel cell due to the elimination
of high current pulses. Potentially the use of energy
storage can also permit downsizing the fuel cell unit
in some applications. As indicated in Fig. 13, there are
several approaches to hybridizing the fuel cell system.
These various powertrain configurations for fuel cell
 514
Energy Storage: Ultracapacitor

Energy Storage: Ultracapacitor. Figure 9

The Si carbon lithium battery on the FUDS with and without ultracapacitors
Energy Storage: Ultracapacitor. Figure 10
The Si carbon lithium battery on the US06 with and without ultracapacitors
Energy Storage: Ultracapacitor
515
 516
Energy Storage: Ultracapacitor

Energy Storage: Ultracapacitor. Figure 11

The zinc–air battery on the FUDS with and without ultracapacitors
Energy Storage: Ultracapacitor. Figure 12
The zinc–air battery on the US06 with and without ultracapacitors
Energy Storage: Ultracapacitor
517
518 Energy Storage: Ultracapacitor
Fuel Motor & Trans-
cells controller mission
a b
Fuel Motor &
cells controller
Energy
DC/DC
storage
c d
Energy Storage: Ultracapacitor. Figure 13
Powertrain configurations for hybridized fuel cells vehicles
vehicles have been simulated [28–30] to determine the
effect of the configuration used on vehicle hydrogen
consumption and the dynamic stress on the fuel cell
stack. Of particular interest in this entry are
powertrains using ultracapacitors in place of batteries.
The key considerations in configuring the
powertrain are whether interface electronics is used
and where the electronics is positioned relative to the
fuel cell and the energy storage unit. A powertrain
without interface electronics is possible using
ultracapacitors, but not with batteries. For fuel cell
vehicles with ultracapacitors coupled directly in paral-
lel with the fuel cell, a power diode is employed to
prevent the flow of reverse current during braking
periods.
In systems without interface electronics, the power
sharing between the fuel cell and ultracapacitor is
determined by the current–voltage characteristics of
fuel cell and ultracapacitor because the voltages must
be equal at all times.
Vcap ¼ Vfc
Vdiode
The voltage response of the fuel cell on various
driving cycles is modeled in detail in [28–30].
Fuel cell vehicles with DC/DC interface elec-
tronics placed between the fuel cell and energy
storage units have more flexibility for balancing
Fuel Motor & Trans-
cells controller mission
Ultra-
capacitor
Trans- Energy Motor & Trans-
mission storage controller mission
Fuel
DC/DC
cells
the power flow between the fuel cell and the
energy storage. The general control objective is to
operate the fuel cell unit only in its high efficiency
region and as close to constant or slowly varying
power as possible. Two different control strategies,
power assist and load leveling, can be employed
[30]. Both control strategies maintain the SOC of the
energy storage unit within a specified range. The power
assist control strategy sets the current demand for the
fuel cell based on the fuel cell voltage and the energy
storage SOC. If the fuel cell voltage remains relatively
high, it provides most of the current to the electric
drive. When the fuel cell voltage becomes low and the
energy storage SOC is high, the energy storage unit
provides a large fraction of the current. In the load-
leveled control strategy, the fuel cell provides relatively
steady power and the energy storage unit provides
transient power. The fuel cell power command is the
vehicle demand averaged over a specified time period
such as 60 s.
Simulation Results for Fuel Cell Vehicles
The performance and fuel consumption of midsize fuel
cell vehicles were simulated for the different powertrain
configurations and control strategies using the vehicle
models developed at UC Davis [28–30]. Simulations
 Energy Storage: Ultracapacitor 519

Energy Storage: Ultracapacitor. Table 15 Vehicle simu- UC coupled to FC DC-Link via a converter
1.30
lation parameters
1.25
Vehicle and system parameters

Fuel economy improvement

Drag coefficient 0.3 1.20
2
Frontal area (m ) 2.2 1.15
Rolling resistance 0.01
1.10
Vehicle hotel load (kW) 0.3
Vehicle mass without energy storage (kg)a 1,500 1.05
Power assist on FUDS
Electric motor (kW) 75 1.00 Power assist on US06
Fuel cell stack and auxiliaries Load leveling on FUDS
0.95
Load leveling on US06
Max. net power (kW) 87.6
0.90
Gross power (kW) 106 60 80 100 120 140 160 180 200 220
Ultracapacitor capacity (Wh)
Number of cells 440
2
Cell area (cm ) 510 Energy Storage: Ultracapacitor. Figure 14
Compressor (kW) 17.2 Fuel economy improvements using ultracapacitors
a
Weight includes all powertrain components except energy
storage
Li-ion battery coupled to FC DC-link via a converter
1.30
Power assist on FUDS
were performed for the hybrid fuel cell vehicles with 1.25 Power assist on US06
a dynamic fuel cell system operating in the optimal Load leveling on FUDS
Fuel economy improvement

1.20 Load leveling on US06

varying back pressure mode. The simulation parame-
ters are listed in Table 15. 1.15
The simulations were performed over the FUDS
and US06 driving cycles. The simulation results are 1.10
discussed in detail in [30]. It was found that the use
1.05
of energy storage significantly reduced the stress on the
fuel cell stack by reducing the peak currents as well as 1.00
the current fluctuations. There were no significant dif-
0.95
ferences in the stack currents using batteries or
ultracapacitors if interface electronics was positioned 0.90
between the energy storage unit and the fuel cell bus; 5 6 7 8 9 10
Battery capacity (Ah)
the results using ultracapacitors without interface elec-
tronics were nearly as advantageous as those with Energy Storage: Ultracapacitor. Figure 15
electronics. Fuel economy improvements using lithium batteries
The use of energy storage in the fuel cell vehicles
resulted in improved fuel economy (reduced hydrogen
consumption). As shown in Figs. 14 and 15, the of the improvement in fuel economy for fuel cell vehi-
improvements increase for larger energy storage units cles using energy storage is due to energy recovery
for both ultracapacitors and batteries. In general, the during regenerative braking. Note from Figs. 14
magnitudes of the improvements are much smaller and 15 that the fuel economy improvements with
than in hybrid-electric vehicles. This is due to the ultracapacitors are slightly larger than with batteries
higher efficiency of the fuel cell and its smaller variation primarily due to the higher efficiency (lower losses) of
with power fraction than is the case for engines. Most the ultracapacitors.
520 Energy Storage: Ultracapacitor
Energy Storage: Ultracapacitor. Table 16 Material costs for a 2.7 V, 3,500 F capacitora
Carbon Electrolyte ACN
F/gm gmC/dev. $/kg $/l $/kg salt
75 187 50 10
120 117 100 10
75 187 5 2
120 117 10 2 50
a
4.5 Wh/kg, 1,000 W/kg-95% eff.
Economic Considerations
Reducing the present high cost/price of ultracapacitors
is a key issue in achieving high market penetration in
the future especially of midsize and large devices. There
are many applications for which ultracapacitors are
presently precluded or even seriously considered
because they remain too expensive even though their
selling price has decreased significantly in recent years.
The cost of manufacture of any product is closely tied
to production volume with the cost decreasing rapidly
with increased volume up to relatively high production
rates. Sales of capacitors in the many millions of units
per year are necessary to reduce the unit costs to levels
at which large markets can develop. Semi-automated
production facilities presently exist at a number of
companies for ultracapacitors of all sizes. In fact, pro-
duction capabilities exceed sales volumes for most
devices and that is the reason the price of devices has
decreased markedly in recent years. It is common to
speak of the price of devices in terms of cents per Farad
(cents/F) or $/Wh stored. It is easier to interpret the
price information on the cents/F basis as it does not
depend on the cell voltage or what fraction of energy
stored can be used in a particular application. For
example, for a 10 F device, if the price is quoted as
10 cent/F, the device cost would be $1. Similarly,
a 2,500 F device would cost $25 at 1 cent/F.
The cost to manufacture a carbon/carbon device
depends on the material and production costs. At the
present time, material costs are high. The cost of car-
bon suitable for use in ultracapacitors can be as high as
$100/kg with the average price being in the $30–50/kg
range. The cost of the electrolyte solvent is also high in
the range of $5–10/l for both propylene carbonate and
Device Unit Costs
Total mat. $ $/kg $/Wh $/kW Ct/F
125 17.0 29 6.4 29 0.48
125 15.5 26 6 26 0.44
50 3.6 6.0 1.3 6 0.10
2.5 4.2 0.93 4.2 0.070
acetonitrile. The ionic salts that dissociate in the sol-
vent into the positive and negative ions that move into
and out of the double layer in the microporous carbon
to store the energy are also expensive being $50–100/
kg. Since the analysis of ultracapacitors is straightfor-
ward, material costs can be calculated [31, 32] with
good accuracy. The result of a typical costing exercise
is shown in Table 16. Note the strong dependency of the
cents/F and $/Wh unit costs for the device on the unit
material costs. Presently the price of ultracapacitors is
high because both the material and manufacturing
costs are high. With more automated production and
reduced material costs, it is anticipated that the price of
ECCs in high volume can be in the range of 1–2 cents/F
for small devices and 0.25–0.5 cents/F for large devices
like those needed for vehicle applications.
Ultracapacitors cannot compete with batteries in
terms of $/Wh, but they can compete in terms of $/kW
and $/unit to satisfy a particular vehicle application.
Both energy storage technologies must provide the
same power and cycle life and sufficient energy (Wh)
for the application. The weight of the battery is usually
set by the system power requirement and cycle life and
not the minimum energy storage requirement. Satisfying
only the minimum energy storage requirement would
result in a much smaller, lighter battery than is needed to
meet the other requirements. On the other hand, the
weight of the ultracapacitor is determined by the mini-
mum energy storage requirement. The power and cycle
life requirements are usually easily satisfied. Hence, the
unit can be a more optimum solution for many applica-
tions and its weight can be less than that of the battery
even though its energy density is less than one tenth that
of the battery.
 Energy Storage: Ultracapacitor 521

Energy Storage: Ultracapacitor. Table 17 Relationships between capacitor and battery costs
Battery cost Battery Ultracap cost Ultracap cost Ultracap costb Ultracap cost $/kW
$/kWh costa $/kW cents/F Vcap =2.6 cents/F Vcap =3.0 $/kWh Vcap =3.0 Vcap =3.0
300 30 0.25 0.34 3,626 7.3
400 40 0.34 0.45 4,800 9.6
500 50 0.42 0.56 5,973 11.9
700 70 0.59 0.78 8,320 16.6
900 90 0.76 1.0 10,667 21.3
1,000 100 0.84 1.12 11,947 23.9
a
battery 100 Wh/kg, 1,000 W/kg
b
capacitor 5 Wh/kg, 2,500 W/kg

Consider the example of a charge-sustaining hybrid
like the Prius. If the energy stored in the capacitor unit
is 125 Wh and that in the battery unit is 1,500 Wh, the
unit costs of the capacitors and battery for equal unit
cost are related by
ð$=WhÞcap ¼ :012ð$=kWhÞbat
The corresponding capacitor unit costs in terms of
cents/Farad and $/kWh are given by
ðcents=FÞcap ¼ :125  10
3  ð$=kWhÞbat Vcap 2
 capacitance or Faradaic materials in at least one
ð$=kWhÞcap ¼ 9:6  104 ðcents=FÞcap Vr 2
and shown in Table 17 for a range of battery costs.
The results shown in Table 17 indicate that for the
charge-sustaining hybrid application, ultracapacitor
costs of 0.5–1.0 cents/F are competitive with lithium
battery costs in the range of $500–700/kWh. Note also
that the $/kW cost of the ultracapacitor are about one
fourth those of the batteries.
Future Directions of Ultracapacitor Development
and Vehicle Applications
Ultracapacitor Developments
Ultracapacitors for vehicle applications – that is devices
having a capacitance greater than several hundred
Farads – have been under development for over
20 years. This development is expected to continue
with improvements in energy density as special
carbons developed especially for the ultracapacitor
application become available. Energy densities of
about 10–15 Wh/kg can be expected using the
advanced carbons and electrolytes permitting voltages
of 3–3.5 V. The cost of ultracapacitors using activated
carbon are presently 1–2 cents/F. It is expected that
the cost will continue to decrease as the volume of
production of ultracapacitors continues to increase.
If the cost can be reduced to 0.25–0.50 cents/F, the
market for ultracapacitors in vehicle applications will
become large (millions of vehicles per year using the
devices).
Development of ultracapacitors using pseudo-
electrode will continue. These devices can use either
activated carbon or an advanced carbon in one of the
electrode. The energy density of these advanced devices
could be greater than 20 Wh/kg. The key issues with the
advanced devices are their power capability (W/kg for
95% pulse power efficiency) and cycle life. Research to
date has indicated it is difficult to achieve large
improvements in energy density without significant
sacrifices in cycle life and power capability.
Vehicle Applications of Ultracapacitors
There have been numerous demonstrations of the use
of ultracapacitors in vehicle applications, but except for
a limited number of transit buses and heavy-duty
trucks produced for sale, there has been no significant
market for ultracapacitors for hybrid vehicles. This may
be about to change (2011) as several auto companies
are discussing the inclusion of ultracapacitors in
micro-hybrids combined with lead-acid batteries. The
522 Energy Storage: Ultracapacitor
success of this application is highly dependent on the
cost of the ultracapacitors and will require a continuing
decrease in their cost. If the energy density of the
carbon/carbon devices can be increased to 8–10 Wh/kg
with continuing cost reductions to 0.5–1.0 cents/F, it
seems likely ultracapacitors will be used in charge-
sustaining HEVs to further improve the fuel economy
of those vehicles. This could be a very large market for
ultracapacitors.
Applications of ultracapacitors in vehicles further
in the future could involve their use in fuel cell–
powered vehicles and electric and hybrid vehicles
using advanced, high energy density (>250 Wh/kg)
lithium batteries, which may not have corresponding
higher power capability and need ultracapacitors to
meet vehicle power requirements. These applications
can be the focus of laboratory studies and field dem-
onstrations in the next 5–10 years.
Summary
In this entry, various aspects of the use of ultracapacitors
alone and in combination with batteries are discussed in
relation to general system considerations and the perfor-
mance of the energy storage systems in charge-sustaining
and plug-in hybrid vehicles and fuel cell–powered vehi-
cles. The performance of ultracapacitors and lithium
batteries is reviewed based on test data taken
at the University of California-Davis. It was found
that ultracapacitors are commercially available with
energy densities of 4.5–5 Wh/kg and power density of
1–2 kW/kg with a pulse efficiency of 95%. Lithium
batteries of various chemistries were also tested. It
was found that batteries consisting of graphite and
manganese oxide/spinel had energy densities of
140–160 Wh/kg and power densities of 0.5–1.0 kW/kg
with pulse efficiencies of 95%. High-power lithium
batteries using lithium titanate oxide in the negative
electrode were found to have an energy density of about
40 Wh/kg and a power density of 1.3 kW/kg with pulse
efficiencies of 95%. Comparisons of the power densi-
ties of ultracapacitors and lithium batteries indicated
that the best ultracapacitors had power densities
a factor of 2–3 higher than the high-power batteries,
but there are some batteries with higher power capa-
bility than the lower power capability capacitors. For
use in hybrid vehicles, high power density is very
important for ultracapacitors.
The application of ultracapacitors in hybrid-electric
vehicles was considered in detail utilizing vehicle sim-
ulation results. The use of ultracapacitors alone is ana-
lyzed for micro- and mild/full charge-sustaining hybrids
and in combination with high-energy density batteries
for plug-in hybrids. It was found that ultracapacitors
could be used in place of batteries in the charge-
sustaining hybrids resulting in fuel economy improve-
ments of 25–50% even for relatively small electric drive
systems of 6–10 kW. The ultracapacitor units only stored
50–100 Wh of usable energy and their efficiency was high
being 95–98%. In all cases, the ultracapacitors were used
with interface electronics that permitted the use of a large
fraction of their stored energy.
Vehicle simulations were also performed for plug-in
hybrid vehicles that used ultracapacitors combined
with advanced, high energy density batteries. The bat-
teries had energy densities of 200–400 Wh/kg, but only
moderate power capability of 300–1,000 W/kg for 90%
efficient pulses. The simulation results indicated that
the advanced batteries combined with ultracapacitors
could provide all-electric operation on the FUDS
and Federal Highway cycles in the charge-depleting
mode and excellent fuel economy in the charge-
sustaining mode. Vehicle operation with the advanced
batteries without the ultracapacitors required blended
operation on all driving cycles. Comparisons of the
current/voltage/power profiles of the batteries with
and without the ultracapacitors indicated the peak
currents and thus the stress on the batteries were
reduced by about a factor of 3 using the ultracapacitors.
This reduction is expected to lead to a large increase in
battery cycle life.
Vehicle simulation results are also discussed for fuel
cell vehicles using ultracapacitors or batteries for
energy storage. It was found that the vehicle and fuel
cell system functioned essentially the same with
ultracapacitor or batteries. In the case of fuel cell vehi-
cles, the ultracapacitor unit could be used without
interface electronics with the resultant power split
between the ultracapacitors and the fuel cell being
self-controlled by the voltage–current characteristics
of the respective devices. The fuel economy (hydrogen
consumption) improvements using energy storage

were 10–15% for the FUDS cycle and 5–10% for the
US06 cycle with ultracapacitors and a few percent less
using lithium batteries.
The long-term costs of both lithium batteries and
ultracapacitors are uncertain. An analysis of
ultracapacitor unit costs indicated that a battery cost
of $500/kWh and ultracapacitor cost of 0.5 cents/F
would result in a comparable energy storage unit cost
for charge-sustaining hybrid vehicles.
Bibliography
1. Burke AF, Miller M (2009) Performance characteristics of
lithium-ion batteries of various chemistries for plug-in
hybrid vehicles, EVS-24, Stavanger (paper on the CD of the
meeting)
2. Burke AF, Miller M (2009) Electrochemical capacitors as energy
storage in hybrid-electric vehicles: present status and future
prospects, EVS-24, Stavanger (paper on the CD of the
meeting)
3. Burke AF, Miller M (2009) The UC Davis emerging lithium
battery test project, Report UCD-ITS-RR-09-18
4. Burke AF, Miller M (2011) The power capability of
ultracapacitors and lithium batteries for electric and hybrid
vehicle applications. J Power Sources 196(1):514–522
5. US Department of Energy (2008) Battery test manual for plug-
in hybrid electric vehicles, INL/EXT-07-12536
6. Yoshio M, Brodd RJ, Kozawa A (eds) (2009) Lithium-ion Batte-
ries, chapter 12. In: HEV Applications. Springer, New York
7. Anderman M (2005) Could ultracapacitors become the
preferred energy storage device for future vehicles?
In: Proceedings of the 5th international advanced automotive
battery conference, Honolulu, 15–17 June 2005
8. Burke A, Miller M (2010) Lithium batteries and ultracapacitors
alone and in combination in hybrid vehicles: fuel economy
and battery stress reduction advantages. Paper presented at
the electric vehicle symposium 25, Shenzhen
9. Bohn T (2007) Plug-in hybrid vehicles: decoupling battery load
transients with ultracapacitor storage. Paper presented at the
advanced capacitor world summit 2007, San Diego
10. Miller JM, Bohn T, Deshpande U (2008) DC-DC converter buff-
ered ultracapacitor in active parallel combination with lithium
battery for plug-in hybrid electric vehicle energy storage. SAE
paper 2008-01-1003, SAE World Congress, Detroit
11. Dixon J, Nakashima I, Ortuzar M (2010) Electric vehicle using
a combination of ultracapacitors and ZEBRA battery. IEEE
Trans Ind Electron 57(3):943–949
12. Bender J (2009) Life cycle testing of a ultracapacitor/lithium
battery combination. Private communication from AFS Trinity
13. Bartley T (2005) Ultracapacitor energy storage in heavy-duty
hybrid drive update. In: Proceedings of the advanced capaci-
tor world summit 2005, San Diego
Energy Storage: Ultracapacitor 523
14. King RD et al (2003) Ultracapacitor enhanced zero emissions
zinc air electric transit bus – performance test results.
In: Proceedings of the 20th international electric vehicle sym-
posium, Long Beach
15. Okamura M, Hayashi K, Ohta H (2006) Status report on ECaSS
and Nanogate-capacitors. In: Proceedings of the 16th interna-
tional seminar on double-layer capacitors & hybrid energy
storage devices, Deerfield Beach
16. Schwake A (2006) EC-funded project “SUPERCAR”:
ultracapacitor modules for mild hybrid applications. In: Pro-
ceedings of the second international symposium on large
ultracapacitor (EDLC) technology and application, Baltimore,
16–17 May 2006
17. Gonder J, Pesaran A, Lustbader J, Tataria H (2009) Fuel econ-
omy and performance of mild hybrids with ultracapacitor
simulations and vehicle test results. Presentation at the 5th
international symposium on large EC capacitor technology
and applications, Long Beach
18. Armiroli P et al (2008) Valeo stars+system, ultracapacitors: the
way for regen for micro-mild hybrids. In: Proceedings of the
fourth international symposium on large ultracapacitor tech-
nology and application, Tampa
19. Peasaran A (2007) Factors and conditions for widespread
use of ultracapacitors in automotive applications. In: Pro-
ceedings of the advanced capacitor world summit 2007,
San Diego
20. Burke AF (2002) Cost-effective combinations of ultracapacitors
and batteries for vehicle applications. In: Proceedings of the
second international advanced battery conference, Las Vegas,
4–7 Feb 2002
21. Knorr R (2006) SUPERCAR – results of a European mild hybrid
project. In: Proceedings of the 6th international advanced
automotive battery and ultracapacitor conference, Baltimore,
17–19 May 2006
22. Knight B (2003) Information on the ultracapacitors used in the
Honda fuel cell vehicle
23. Thounthong P, Rael S, Davat B (2009) Energy management of
fuel cell/supercapacitor hybrid power source for vehicle appli-
cations. J Power Sources 193:376–385
24. Burke AF, Miller M (2003) Ultracapacitor and fuel cell applica-
tions. In: Proceedings of the 20th international electric vehicle
symposium, Long Beach
25. Burke AF, Zhao H, Van Gelder E (2009) Simulated performance
of alternative hybrid-electric powertrains in vehicles on vari-
ous driving cycles, EVS-24, Stavanger, (paper on the CD of the
meeting)
26. Burke AF, Van Gelder E (2008) Plug-in hybrid-electric vehicle
powertrain design and control strategy options and simula-
tion results with lithium-ion batteries. Paper presented at EET-
2008 European Ele-drive conference, Geneva, 12 Mar 2008
(paper on CD of proceedings)
27. Burke AF (2007) Batteries and ultracapacitors for electric,
hybrid, and fuel cell vehicles, IEEE Journal, Special issue on
Electric Powertrains
524 Energy Storage: Ultracapacitor
28. Zhao H, Burke AF (2009) Optimum performance of direct
hydrogen hybrid fuel cell vehicles, EVS-24, Stavanger (paper
on the CD of the meeting)
29. Zhao H, Burke AF (2009) Optimization of fuel cell system
operating conditions for fuel cell vehicles. J Power Sources
186(2):408–416
30. Zhao H, Burke AF (2010) Effects of powertrain configurations
and control strategies on fuel economy of fuel cell vehicles.
Paper presented at the electric vehicle symposium 25,
Shenzhen
31. Burke AF (2010) Electrochemical capacitors. In: Handbook of
batteries, 4th edn. McGraw-Hill, New York
32. Burke AF (2004) Key issues for the implementation and
marketing of ultracapacitors in vehicle applications. In:
Proceedings of the 14th international seminar on double-layer
capacitors and similar energy storage devices, Deerfield Beach
 Fuel-Cell-Powered HEV Design and Control 525

Fuel-Cell-Powered HEV Design and Proton exchange membrane (PEM) A fuel cell that
uses either a sulphonated (low temperature) mem-
Control brane or polybenzimidazole (high temperature)
ETIM U. UBONG1, JIM GOVER2 membrane to generate direct electricity.
1
Center for Fuel Cell Research and Powertrain Reformate A fuel derived from fossil fuel (natural gas)
Integration, Electrical & Computer Eng, Kettering after reformation that is a mixture of hydrogen,
University, Flint, MI, USA carbon monoxide, and carbon monoxide.
2
Department of Electrical Engineering, Kettering Solid polymer fuel cell (SPFC) Same as PEM above.
University, Flint, MI, USA
Definition of the Subject
Article Outline
A hybrid vehicle implies a vehicle that is powered by
Glossary
more than one source of power. It can be an internal
Definition of the Subject
combustion engine and a battery, an ICE and fuel cell,
Introduction of Fuel Cells
and ICE and solar cells or fuel cell and batteries. The
Candidate Fuel Cells for Hybridization
current trend of hybridization of internal combustion
Hybrid Technology Using Gasoline or Diesel, Fuel
engines (ICE) is a result of the inefficiency of ICE. Spark
Cells, Ultra-Capacitors, and Batteries
ignition internal combustion engine, SI-ICE, is about
Ultra-Capacitors
25–27% efficient, while compression ignition internal
Conclusion
combustion engine, CI-ICE, (diesel) is 34–38% efficient
Bibliography
[1, 2]. The remaining percentage of the input energy
into ICE is expended in the form of heat loss to the
Glossary
coolant and exhaust, by convection, conduction, and
Balance of plant (BOP) Auxiliary components that radiation, and by combustion inefficiency. The electric
complement a power unit as a functional or oper- drive has a higher efficiency around 90%, as such, it is
ational unit. justifiable to incorporate electric drive into the hybrid-
Electric vehicle (EV) A vehicle that uses battery or any ized powertrain. With the increased use of electronics
other onboard electrical power source to operate in cars for various applications, such as visual display of
the vehicle. GPS, fuel economy, maintenance schedule and diag-
Fuel cell A device that uses fuel and oxidant in nostics, voice activation of devices and DVD players,
an electrochemical reaction to generate direct etc., there is a need to use additional power sources to
electricity. maintain this level of power demand without
High-temperature proton exchange membrane fuel compromising the power required for ignition, and
cell (HT-PEMFC) A PEM fuel cell that operates operate the electronic control unit (ECU), anti-brake
between 120
C and 200
C. systems (ABS), and the fuel injection units.
Hybrid electric vehicle (HEV) A vehicle that uses an In the hybridized ICE-powered vehicle, in the
auxiliary electrical power source, such as a fuel cell series mode of connection, the mechanical drive
or a battery, along with an internal combustion shaft that transmits power from the crankshaft
engine (ICE) for propulsion. through the gearbox to the wheels is replaced with
Internal combustion engine (ICE) A device that con- a generator and electric wires to the electric motors
verts the chemical energy of fuel in a cylinder to spinning the wheels. The design configuration of the
mechanical work through the piston, connecting ICE-battery hybrid presents the base design of other
rod and crank shaft mechanisms. complex hybrid vehicles architecture. The design
Low-temperature PEM fuel cell (LT-PEMFC) A structure of the fuel cell–powered hybrid electric vehi-
PEM fuel cell that operates between 50
C and cle includes fuel cell as the primary power source and
80
C. a battery as a secondary power source for start-up,

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
526 Fuel-Cell-Powered HEV Design and Control

lighting, and power electronics support. The power
transmission shaft reminiscent of the internal com-
bustion engines (ICE)–powered vehicle is absent,
rather electric motors spin the four wheels. Fuel cells
generate direct electricity that is used to drive the
electric motors in hybrid vehicles. As such, reliability
is enhanced as there are fewer rotating components.
The complexity and maintenance are reduced in elec-
tric machinery compared with mechanically driven
machinery. The following report will present different
types of fuel cells and their electrolytes, sources
of fuels for fuel cells, those applicable to the transpor-
tation sector and the hybrid vehicle architecture.
The ICE hybrid will also be presented as a base
model, since its operation is well understood by the
general public.
Introduction of Fuel Cells
Fuel cells are electrochemical electricity–generating
devices that oxidize the fuel at the anode to electrons
and protons and reduce oxygen at the cathode to produce
heat and water. The splitting of hydrogen into protons
and electrons is done at the catalyst layer (anode).
The electrons move across the external circuit generat-
ing electricity, while the protons permeate through the
PEM to the cathode. The proton exchange membrane
fuel cells are of interest in the mobility sector as it is
modular (scalable) and, as such, can be built to desirable
sizes without compromising the efficiency or
performance. Various petroleum and nonpetroleum
fuels, chemical compounds, and hydrogen carriers can
be used as fuels in fuel cells after preprocessing. However,
hydrogen is the most preferred fuel with the end prod-
ucts, in case of the proton exchange membrane (PEM)
membrane being water, with the release of heat. Hydro-
gen is widely used in most fuel cells ranging from the
low to the high-temperature PEM to solid oxide fuel
cells (SOFCs).
Commonly Used Fuels in Fuel Cells
Fuels commonly used in fuel cells are: hydrogen,
methanol, potassium hydroxide; phosphoric acid;
methane (natural gas); carbon monoxide; carbonate
salts of sodium, lithium, and potassium; and yttrium-
stabilized zinc oxide as the electrolyte. Table 1 sum-
marizes various fuels and applicable fuel cells that
utilize them.
Low-Temperature and High-Temperature PEMs
The low temperature PEM fuel cell shown in Fig. 2a
relies on membrane humidification for ion transport
from the anode to the cathode through Grotthuss mech-
anism or ion-hopping mechanism. If the membrane is
not well humidified or is dry, vehicular mechanism
is involved in the ion transport. For the high-
temperature PEM fuel cell based on polybenzimidazole
(PBI) membrane, which is presented in Fig. 2b,

Fuel-Cell-Powered HEV Design and Control. Table 1 Different types of fuel cells

PEM
Parameters LT-PEM HT-PEM DMFC AFC PAFC MCFC SOFC
Fuels H2 H2, Reformates Methanol KOH (H2) PO4 (H2) H2, CO, CH4 H2, CO, CH4

Charge carrier H +
H+
H +
OH H +
CO32 O
a b
Electrolyte PEM PBI PEM 30–50% 100% conc.
conc. KOH in water in PO4
Temperature 50–80
C 130–200
C 50–80
C 200
C 150–220
C 650
C 650–1,000
C
range
LT-PEM low-temperature PEM that are made of poly-sulphonated membranes, such as NAFION marketed by Du Pont, GORE membrane,
ASAHI glass, 3M, etc.
HT-PEM High temperature membrane, such as polybenzimidazole (PBI) membranes made by BASF (USA), etc.
a
Liquid molten carbonate salts, such as Na2CO3 and Li2CO3, in a ceramic matrix of LiAl02
b
Solid oxide ceramic (Yttrium-stabilized zirconium (ZrO2/Y2O3)

the membrane is treated in sulfuric or phosphoric acid.
Protons are transported in an acidic medium when the
temperature is raised up to 130
C and above. The
membrane performs much better at an elevated tem-
perature of 170–200
C. Precautionary measures
require that the membrane be raised to above 130
C
before the load is applied. Better results are obtained if
the cell is allowed to run and equilibrate at a reduced
load for about 100 h before readings are taken. This
allows the membrane to equilibrate. During the shut-
down period, the membrane should be purged with
nitrogen gas till it cools to ambient conditions to avoid
leaching of the acid and degrading of the membrane
performance [3].
Source of Fuels for Fuel Cell
The current earthquake in Japan and Virginia, USA, is
a reminder that the nuclear power–generating plants
are very vulnerable to natural disasters. Nuclear plant
(prior to the earthquake in Japan) has been projected
to be the main source of hydrogen production, as it
is cleaner and more efficient. The Chernobyl
(1986) nuclear disaster and the Japanese (2011) disaster
triggered by an earthquake of 9.0 magnitude on the
Richter Scale, clearly drew a line between science, engi-
neering, and nature. Despite the precautions and tech-
nological advancement of Japan, it could not contain
such magnitude of earthquake. Couple of months after
the incident, the levels of radiation at the Japanese
nuclear meltdown have been upgraded to level 7,
same as the Chernobyl radiation after the disaster.
The Virginia quake of summer 2011 rattled Washing-
ton D.C., metropolis with a magnitude of 5.8 on the
Richter Scale. This spontaneously shutdown the
nuclear reactor in Virginia, raising serious concerns of
possible nuclear radiation to the atmosphere. These
recent incidents pre-suggest that another sustainable
source of hydrogen production (and not nuclear
power plant) is needed for the hydrogen economy.
Following the above Japanese event, Japan, Germany,
and Switzerland have frozen proliferation of building
further new nuclear reactor plants.
The question now is which is the most environmen-
tally and sustainable way of producing hydrogen for the
hydrogen economy? Current pathways of hydrogen pro-
duction include: steam methane reformation (SMR),
Fuel-Cell-Powered HEV Design and Control 527
partial oxidation (POX) process, and autothermal refor-
mation (ATR) process of fossil fuels. However, the pro-
duction and availability of fossil fuels are projected to dip
after 2020, and as such, a renewable source of hydrogen is
required. Electrolysis of water is a viable source; however,
energy is required to electrolyze water and compress it to
350 bar (5,000 psi) or 700 bar (10, 000 psi) in order to
store enough hydrogen to meet the same driving dis-
tance as its gasoline equivalent and package the fuel tank
into the available fuel storage space. First, an analysis of
the five types of fuel cells will be presented highlighting
their potentials for hybrid vehicle propulsion.
In the final analysis, water electrolysis method to
obtain hydrogen is a clean technology, but requires
energy for the process and storage of the gas under
pressure or possibly liquefaction of the fuel. The cur-
rent technology of steam reformation relies on fossil
fuels as feedstocks. Future energy carrier should be
sustainable, clean, and renewable. Fuel cells for trans-
portation in hybrid vehicles are feasible with the scal-
able fuel cell categories, such as PEM, direct methanol
fuel cell (DMFC), alkaline fuel cell (AFC), and to
a limited extent, with the solid oxide fuel cell (SOFC)
as an auxiliary power unit (APU).
Fuel Cell as a Power System
Generally, fuel cells have the following advantages over
other power-producing devices:
1. Low to zero emissions, if a reformate or hydrogen is
used as fuel, which makes it environmentally very
friendly.
2. Performance and efficiency are very high as there
are no moving parts (excluding the system’s balance
of plant, BOP).
3. It is a bridge to hydrogen economy. The bridge may
be in the form of electric vehicles (EV) or hybrid
electric vehicles (HEV) – in which both fuel cell and
a smaller internal combustion engine (ICE) are
used for propulsion in the medium and low-
power applications.
4. Absence of transmission lines which decreases the
complexity of the BOP.
5. Flexible, scalable, and adaptable where remote
power systems are required, for example, remote
stations, booster stations, traffic and bacon lights,
road signs, street lights, etc.
528 Fuel-Cell-Powered HEV Design and Control

6. For space missions. Fuel cells have long been used
for manned space missions, for example, the Apollo
Mission of the 1960s and the Shuttle Orbiter.
In the following analysis, much attention is given to
PEM because of its potential as the viable candidate
fuel cell for use in transportation. It is noteworthy that
not all types of fuel cells are suitable for use in the
transportation industry because of the power density,
type of electrolyte, size, and thermal cycling.
Analysis of Various Fuel Cells
The Proton Exchange Membrane (PEM) Fuel Cell The
PEM fuel cell derives its name from the electrolyte
used, which is a conducting membrane sandwiched
between two catalyst layers, gas diffusion layers and
plates. The schematic is shown in Fig. 1. A PEM fuel
operates by inducting hydrogen into the anode, where
it is oxidized into electrons and protons as per the
equation:
H2 ! 2Hþ þ 2e ð1Þ
The two electrons are not membrane permeable; hence,
they travel through the anode electrical interconnect
to the load, generating electricity. From there, the
electrons travel to the cathode catalyst layer, where
they react with the supplied oxygen and the protons
that permeate through the membrane to the cathode
catalyst layer to form water in a reduction reaction. The
process is exothermic.
At the cathode, the reaction is as follows:
1=2O
2 þ 2e þ 2Hþ ! H2 O þ Heat ð2Þ
Overall reaction:
H2 þ 1=2O2 ! H2 O þ heat ð3Þ
The schematics of a PEM fuel cell, a 50 cm2 LT-PEM
and HT-PEM fuel cell hardware are shown in Figs. 1, 2a
and 2b respectively, while and multi-test station is
shown in Fig. 3.
Figure 3 features a SERC (Shartz Energy Research
Center) four test station (TS) platform for testing and
validation of a single cell to 1 kW stack. The two TS to

e e
e Load e

H2 H2 H2
e– O2 O2
Air (O2) In
e– H +

H2 H2 H +
O2

e– PEM
e–
Membrane

H2 H2
O2

H +

H +
H2
Unused Fuel H2 H2 O2 Air + H2O Out

Catalyst Layer
Plate Plate
Gas Diffusion Gas Diffusion
Layer (GDL) Layer (GDL)

Fuel-Cell-Powered HEV Design and Control. Figure 1

PEM fuel cell components

a It can contribute to slipperiness of the road and can
b tages: Both are credited with good startability – they start
Fuel-Cell-Powered HEV Design and Control. Figure 2
(a) LT-PEM single cell of area 50 cm2 assembled at the
Center for Fuel Cell Research, Kettering University,
Michigan (Courtesy Kettering PEM Fuel Cell Lab). (b) HT-PBI –
PEM Single cell of area (Courtesy: Kettering PEM fuel cell
research and powertrain integration, laboratory)
the left of the computer monitor operate in a dead-end
mode and are capable of testing single cells and low-
end stacks rated up to 250 W. While TS 3 and 4 are
rated up to 1 kWe and operate in a flow-through mode.
Low-temperature PEM relies on humidification of the
reactants and employs various types of humidifiers. For
example, the 4-cell PEM stack on TS 2 has an-inbuilt
membrane humidifier while TS 4 has an externally
mounted humidifier shown in Fig. 4. That aside, PEM
fuel cell stacks require a cooling system which also
serves as the medium for maintaining the cell/stack
Fuel-Cell-Powered HEV Design and Control 529
temperature at a constant temperature. The test equip-
ments are designed and sized for testing various output
power ranges of cells/stacks. As the power output of the
stack increases, the measuring devices ranges and pipe
sizes are progressively increased to deliver the desired
amount of reactants. Figures 5 and 6 show test equip-
ment for low and high-temperature PEM fuel cell test
equipments capable of testing and validating a 5 kW
PEM stack.
For the mobility sector, low-temperature PEM
requires a complex balance of plant for its smooth
operation. That aside, water from the exhaust is an
issue of consideration during city driving in winter.
be a hazard to other road users. On the other hand, the
high-temperature fuel cells offer an advantage in that
vapor exits from the exhaust and can be used for
heating the vehicle in winter. However, both low- and
high-temperature PEM require heated pads to keep the
stacks above 10
C during winter season. Freezing of
liquid within the stack’s flow field, in the gas diffusion
layer and condensates in the exit flow manifold are all
unacceptable.
PEM membranes are categorized as low-temperature
and high-temperature membranes. A review of both
types highlights the following advantages and disadvan-
to generate electricity shortly after start-up (LT-PEM
from 60
C and HT-PEM at 130
C). Performances of
PEM membranes are generally degraded by the presence
of carbon monoxide. For LT-PEM, CO concentration
above 10 ppm is not recommended, while HT-PEM
based on a PBI can tolerate up to 3% CO in the fuel.
Higher concentrations have been tested up to 8% at
a much reduced voltage [3–6]. Operation around
190
C favors higher CO concentration and the loss in
performance is reversed if the cell is operated later with
pure hydrogen [3].
Both LT and HT-PEM require thermal management
systems – LT-PEM uses water as the coolant while
HT-PEM uses mineral oil. However, LT-PEM in addi-
tion relies on humidification of the membrane for
proton transport. The functional humidification
range is between 80% and 100% relative humidity
(RH). Above 100% RH, the cell floods and impairs
the flow of the reactants to the catalyst sites. As
a consequence, the cell reaches its mass concentration
530 Fuel-Cell-Powered HEV Design and Control

Fuel-Cell-Powered HEV Design and Control. Figure 3

Multi–test station (TS) for LT- and HT-PEM research

Fuel-Cell-Powered HEV Design and Control. Figure 4

Humidifier for a LT-PEM fuel cell at Kettering University Center for Fuel Cell Research

limit, a point where the flow of reactants to the catalyst This inadequacy of the LT-PEM led to the development
sites diminishes and the fuel cell voltage falls off. of a more tolerant membrane, capable of operating
At a lower RH, the membrane dries up and impairs without humidification and its complexities. Its
proton transfer from the anode to the cathode. electrochemical kinetics are enhanced because of its

Fuel-Cell-Powered HEV Design and Control. Figure 5
A 5 kW Greenlight low-temperature PEM test equipment at Kettering University Center for Fuel Cell Research
high-temperature operation; also, the activation over-
voltage is reduced. It is more tolerant to a higher CO in
the fuel composition.
Future fuels require flexibilities to fuel contami-
nants. Experience has shown that conventional fuels’
dispensing, storage, and transportation systems
(trucks, pipelines, etc.) harbor huge deposits of foreign
materials that degrade the quality of fuel with time.
High-temperature membranes are more resilient and
less sensitive to increased CO constituents in the
fuel than sulphonated membranes and can tolerate
reformates. This enables reformers to be coupled
directly to the fuel cell which eliminates the complexity
of post-reformation processing of the fuel to obtain
hydrogen of 99.997% purity.
Direct Methanol Fuel Cell (DMFC) The direct
methanol fuel cell has the same architecture as
hydrogen-air PEM fuel cell in that it uses sulphonated
membranes. It is a low-temperature fuel cell that
operates between 50
C and 80
C. Methanol is fed to
the anode while air is fed to the cathode. The anode
reaction proceeds as follows:
CH3 OH þ H2 O ! þCO2 þ 6Hþ þ 6e ð4Þ
Fuel-Cell-Powered HEV Design and Control 531
At the cathode, the reactions proceed as:
11=2O2 þ 6Hþ þ 6e ! 3H2 O ð5Þ
The overall reaction is:
2CH3 OH þ 3O2 ! 4H2 O þ 2CO2 ð6Þ
The reaction in DMFC is very slow in both electrodes
despite its apparent advantages of releasing six
electrons. To remedy this, a bi-metal catalyst Pt/Ru is
used at the anode with a tenfold increase in catalyst
loading compared with hydrogen-air PEM fuel cell
[7–10]. The activation overvoltages in both electrodes
of DMFC is considered, unlike hydrogen–air PEM
where the anode activation overvoltage is ignored
because it is by many orders of magnitude smaller
than the cathodic activation overvoltage. DMFC is
a candidate for portable devices such as cell phones,
PDAs, soldier power, laptops, DVD players, iphones,
ipads, ipods, kindles, and other portable devices. It
works well in remote areas where power is required
for a prolonged period such as remote sensing
stations, railway crossings, traffic lights, bacon lights,
professional cameras, etc. It is one candidate for light
HEV application but the technology has not been fully
532 Fuel-Cell-Powered HEV Design and Control
Fuel-Cell-Powered HEV Design and Control. Figure 6
A 5 kW high-temperature PEM fuel cell test stand at
Kettering University Center for Fuel Cell Research
developed for such applications yet. DMFC has a low
power density (kW/L); hence, its role as a power source
for transportation is diminished.
Alkaline Fuel Cell The AFC uses liquid KOH
electrolyte for its operation and it is popular for its
past space missions in Apollo Spacecraft in the 1960s
and later in the Space Shuttle Orbiter. Due to the
renewed interest in the development of solid polymer
membranes in the early 1990s, less attention has been
given to AFC. The advantage of AFC is overwhelming.
The electrolyte is cheap and has a proven technology
for the production of KOH. However, the electrolyte is
sensitive to the presence of atmospheric CO2 in the air
which combines with the potassium to form an
insoluble K2CO3 that renders the electrolyte insoluble,
hence degrading its performance. AFCs are installed in
utility vehicles which operate within a restricted area
where pollution by automobiles is restricted. It can be
operated as a mobile and stationary power source.
The basic anode reaction
2H2 þ 4OH ! 4H2 O þ 4e ð7Þ
At the cathode
2H2 O þ O2 þ 4e ! 4OH ð8Þ
Phosphoric Acid Fuel Cell (PAFC) Phosphoric acid
fuel cell PAFC derives its name from the type of
electrolyte used. It is a medium temperature fuel cell
used for stationary power generation. The operational
temperature is in the 200–220
C range. It is one of the
oldest and matured fuel cells with vast production in
the ranges of 200 kW. However, more powerful units
ranging in MW have been introduced into service. It
finds wide applications as back-up power units in
credit card companies’ headquarters, police precincts,
hospitals, banks, airports, etc. The USA and Japan are
the two leading countries in the production and
application of PAFCs. Phosphoric acid fuel cell is
equally susceptible to CO poisoning as hydrogen-air
fuel cell, but can tolerate up to 0.5% or 5,000 ppm of
CO [4–7].
Molten Carbonate Fuel Cell (MCFC) Molten carbonate
fuel cell is a stationary power generation system that
operates around 650
C. It has the advantage of not using
noble metals at catalyst as a result of high operational
temperature. It can use multiple fuels, such as methane
(CH4), CO, and hydrogen, as fuel. One advantage of
MCFC as well as SOFC is the use of exit gases for
cogeneration. That implies using waste heat to operate
either steam or gas turbine, thereby bottoming the cycle
and increasing the efficiency of the utilized fuel in the
entire system. At the exit temperature of 600–650
C,
the waste heat is rerouted to the boiler to generate
steam to operate a steam turbine. The turbine in turn
is connected to a generator or it can be geared to operate
several devices of the balance of plant. The final waste
heat at a lower enthalpy can be used for heating a facility.
This is achieved through routing the city water and the
waste heat though a heat exchanger, thereby reducing or
eliminating the extra cost of maintaining the facility
warm water system.
The common electrolytes used in MCFC are the
carbonate salts of sodium, lithium, and potassium.
The mobile ion is the CO32 that moves from the
cathode to the anode [7–10].

The anode reaction using hydrogen as fuel is:
2H2 þ 2CO3 2 ! 2H2 O þ 2CO2 þ 4e
The cathode reaction is:
O2 þ 2CO2 þ 4e ! 2CO3 2
The reaction using CO as fuel is as follows:
At the anode, it is:
2CO þ 2CO3 2 ! 4CO2 þ 4e
The cathode reaction is:
O2 þ 2CO2 þ 4e ! 2CO3 2
Molten carbonate fuel cell finds a wide application as
a stationary power utility plant and is not suitable for
hybrid electric vehicles purposes.
The Solid Oxide Fuel Cell (SOFC) The solid oxide fuel
cell can be used both as a stationary and auxiliary (for
mobile) power unit (APU) depending on its application.
Current research favors using it as an auxiliary power
unit for inter-state trucks, recreational vehicles, auxiliary
power unit, and recharging unit for soldier power,
power tools, etc. SOFC uses yttrium-stabilized zinc
oxide as the electrolyte [7–10]. The operating
temperature is between 650
C and 1,000
C. Due to its
high operating temperature, SOFC as of now is not
considered a candidate for hybrid electric vehicle
applications, but probably for future locomotive or
metro-rail transportation systems.
The reaction at the anode using hydrogen as fuel is:
2H2 þ 2O¼ ! 2H2 O þ 4e
The cathode reaction is:
O2 þ 4e ! 2O¼
The reaction using CO as fuel.
At the anode, the reaction is:
2CO þ 2O¼ ! 2CO2 þ 4e
The cathode reaction is:
O2 þ 4e ! 2O¼
Candidate Fuel Cells for Hybridization
The above overview shows three fuel cells as candidates
for HEV: PEM, DMFC, and AFC in a descending order.
Out of these three, AFC is least considered for
Fuel-Cell-Powered HEV Design and Control 533
automotive applications for the following obvious rea-
sons: (1) The electrolyte is easily degraded in the pres-
ð9Þ
ence of atmospheric CO2. When potassium carbonate
is formed, less OH ions are available for ionization,
ð10Þ and as such, the performance of the cell is degraded.
(2) The fuel cell lacks flexibility in all possible orienta-
tions. The advantages are overwhelming: (1) The elec-
trolyte is cheap and easily replaceable. (2) The
ð11Þ technology for KOH production is available which
makes the fuel source to be sustainable. (3) The absence
of bipolar plates reduces the cost and technological
ð12Þ process of AFC production [4].
Solid oxide fuel cell SOFC on the other hand can be
considered in an auxiliary capacity to support recrea-
tion vehicles (RV), large trucks as an APU, but not as
the main power unit. The disadvantages are: It requires
some hours to bring it up to the operating temperature
and shutdown – thermal cycling, and it operates at very
high temperatures of 650–1,000
C and, as such, is not
suitable for HEV application.
Hybrid Technology Using Gasoline or Diesel, Fuel
Cells, Ultra-Capacitors, and Batteries
Hybrid Using Gasoline- or Diesel-Powered Engines
The technology of gasoline and diesel (internal com-
bustion engine) operation is very well understood;
hence, little time will be devoted here for its explana-
tion. Internal combustion engines (ICE) are devices
that convert the chemical energy of the fuel (gasoline
or diesel) to mechanical work through the piston,
ð13Þ
connecting rod, and crankshaft mechanisms. The
architecture of the gasoline or diesel hybrid consists of
an ICE/generator, electric motor, battery, drivetrain,
ð14Þ
and wheels. The schematic is shown below in Fig. 7
in which ICE/generator for oversimplification is
represented as ICE. The mechanical power from ICE
is converted into electrical power through the electric
ð15Þ
generator connected on the same shaft with the engine.
Electric power is transmitted to the motor. Turbo-
charged ICE has an immense advantage over its natu-
ð16Þ rally aspirated versions. The exhaust gas energy from
the turbine is harnessed by coupling an electric motor
generator to the same shaft as the turbo-compressor
taking advantage of the exhaust gas energy to generate
extra electricity. During acceleration when the vehicle
requires more torque/air, the electrical motor is
534 Fuel-Cell-Powered HEV Design and Control

IDLE MODE ACCELERATION/TRANSIENT

EM B EM B

DIFF
DIFF
ICE TRANS ICE TRANS

CLUTCH CLUTCH

a d

CITY DRIVING
CONVENTIONAL ICE
(low load traffic)

EM B

DIFF
DIFF

TRANS ICE TRANS

ICE

CLUTCH CLUTCH

b e

NORMAL GENERATION REGENERATIVE BRAKING

EM B EM B
DIFF

DIFF

ICE TRANS ICE TRANS

CLUTCH CLUTCH
ENERGY PATH

c f
ENERGY PATH

Fuel-Cell-Powered HEV Design and Control. Figure 7

Hybrid ICE

activated with the extra energy from the turbo-gener- bottoming the cycles of the exhaust gas use in hybrid
ator. When there is excess exhaust energy, that portion vehicle [11]. The architecture in Fig. 7 shows various
would be harnessed to recharge the batteries, thereby modes of ICE hybrid propulsion. Figure 7a presents
 Fuel-Cell-Powered HEV Design and Control 535

idle operation where the generated power is used Fuel Cell Hybrid Architecture
to charge the battery while (b) depicts the normal
The fuel cell hybrid vehicle architecture incorporates
operation of a conventional ICE. The energy path
PEM fuel cell, onboard compressed hydrogen tank, and
is indicated with the arrows. Under normal opera-
power conditioner, batteries, ultra-capacitors, motor
tion, the power from the engine is used both for pro-
and control electronics. A schematic of the PEMFC-
pulsion and charging the battery as shown in (c) while
powered hybrid vehicle using heavy-duty batteries and
additional energy is needed during acceleration and
ultra-capacitors is presented in Fig. 8. The PEMFC
transients from the second power source (battery)
balance of plant includes compressed hydrogen in
shown in (d). As an extender, the ICE is turned off
a storage tank at a pressure of 350 bar (5,000 psi) or
during traffic in city driving and the energy from the
700 bar (10,000 psi); cooling system; humidification
battery is used for propulsion (e). The concept of
system; compressed air system (fan, etc.); and an elec-
regenerative braking involves capturing the kinetic
tronic display panel for controlling the flows into the
energy during braking and deceleration to charge the
fuel cell. The steps involved in integrating the entire
batteries (f).
system include sizing the fuel cell to provide average
power over the entire operating range [12, 13]. The fuel
cell plant has input and output data logging inlets and
Hydrogen
Tank System outlets to monitor the reactants’ flow rates, the voltage,
Controller Ultracaps
Battery current, power, power density, and stoichiometries of
reactants. A DC–DC converter is required between the
Fuel DC–DC Motor FC and the load. Sets of batteries and ultra-capacitors
Cell Converter
are also required [12]. Charging of the batteries is done
when the vehicle idles, or prior to using the vehicle. The
Speed
Controller
DC–DC converter provides supplementary power for
the ultra-capacitor’s recharging functions while the
Fuel-Cell-Powered HEV Design and Control. Figure 8 ultra-capacitors provide power needs during tran-
Fuel cell hybrid architecture sients, such as peak loads and transients during

Fuel-Cell-Powered HEV Design and Control. Figure 9

Maxwell ultra-capacitors
536 Fuel-Cell-Powered HEV Design and Control
acceleration in fuel cell–powered vehicles. Some of the
roadblocks in quick implementation of FC-HBVs are
hydrogen infrastructure, sub-zero temperature opera-
tion, cold start constraints in the colder regions of the
country and issues of dealing with fuel cell exhaust
water during winter months on the road during city
driving.
Ultra-Capacitors
The factor limiting the range and effectiveness of
a rechargeable-battery-powered electric vehicle is the
battery (or the energy storage system). Ultra-capacitors
represent another generation of energy storage devices
and are electrochemical double layer capacitors that
supply peak energy requirement during transients,
such as acceleration and cold start in fuel cell–powered
vehicles and other hybrid electric vehicles (HEV).
Figure 9 shows Maxwell’s 2,600 F ultra-capacitors.
Ultra-capacitors are widely accepted in fuel cell vehicles
and other HEVs for their quick charge/discharge func-
tions. Some of the advantages are their flexibilities in
charging/recharging. They cannot be overcharged and
can be maintained at any voltage above/below the rated
value. While rechargeable batteries generally exhibit
a very slow charge/response functions with signifi-
cantly reduced life cycles, ultra-capacitors work best
in this condition by recuperating within a few seconds
as a result of their low equivalent series resistance
(ESR). They possess a high cycle-life with less deterio-
ration when used frequently and infrequently. They
respond within a wide temperature range of 65–40
C.
They exhibit high columbic efficiency which implies
total charge removed over the replenished charge
[13–15].
Conclusion
In conclusion, both low- and high-temperature
proton exchange membrane fuel cells will be used in
fuel cell hybrid vehicles for transportation as they
provide overwhelming advantages over the internal
combustion engines, such as zero emissions when
hydrogen is used as fuel, reduced noise, and higher
efficiency.
The use of integrated technologies such as fuel cell,
batteries, ultra-capacitors, and other novel techniques
will occupy the center stage in fuel cell vehicle devel-
opment within this decade.
Bibliography
1. Heywood JB (1985) Fundamentals of internal combustion
engines. McGraw Hill, New York
2. Ferguson C, Kirkpatrick A (2001) Internal combustion engines
and applied thermosciences. Wiley, New York
3. Ubong EU, Shi Z, Wang X (2009) Three dimensioning modeling
and experimental study of a high temperature PBI based PEM
fuel cell. J Electrochem Soc 156(10):B1276–B1282
4. Ubong EU, Dimitrov B (2010) Regression of the response
variable of a high temperature PEMFC-PBI based membrane.
J Electrochem Soc 157(7):B1059–B1067
5. Ubong EU (2008) A single cell testing and evaluation
of a high temperature PBI based membrane. In: Fuel cell
seminar & exposition. Convention Center, Phoenix, 27–30
Oct 2008
6. Ubong E, Das SK, Berry KJ, Antonio R (2008) Experimental
performance evaluation of a high temperature PEM fuel cell
at different temperatures. In: ASME 6th international fuel cell
conference. Denver, 16–18 June 2008
7. Larminie J, Dicks A (2003) Fuel cell systems explained, 2nd edn.
Wiley, New York. ISBN 0-470- 84857-X
8. Babir F (2005) PEM fuel cells: theory and practice. Elsevier,
Amsterdam/London
9. Mench MM (2008) Fuel cell engines. Wiley, Hoboken
10. O’Hare R, Cha SW, Colella W, Prinz FB (2006) Fuel cell
fundamentals. Wiley, New York
11. Sahed SM (2005) The power of turbocharging. Automotive
Engineering, International, Sep 2005, SAE
12. Anzicek J, Joel Berry K, Lerner S, Slota G, Thompson M, Ubong E
et al (2004) Fuel cell hybrid vehicle at Kettering University.
In: Electrical manufacturing and coil winding conference
(EMCWA conference), Indianapolis, 23–25 Sep 2004
13. Ubong EU, Mizell C, Slota G (2005) Ultracapacitors with Ballard
Nexa in a GEM vehicle in hybrid mode. In: Electrical
manufacturing and coil winding conference) – EMCWA
conference, Indianapolis, 20–22 Sep 2005
14. Schneuwly A, Maher B, Auer J. Maxwell Technologies Inc.,
www.maxwell.com
15. Namisnyk AM, Zhu JG. A survey of electrochemical
supercapacitors technology. Faculty of Engineering, University
of Technology, Sydney
 High Speed Rail, Technology Development of 537

High Speed Rail, Technology Trans-European Transport (TEN-T) Network is an

important part of this program aiming to establish
Development of and develop the key links and interconnections
EDA BEYAZIT, MOSHE GIVONI needed to eliminate existing bottlenecks to mobility
Transport Studies Unit, School of Geography and the fill in missing sections and complete the main
Environment, University of Oxford, Oxford, UK routes – especially their cross-border sections
cross natural barriers and improve interoperability
on major routes [33]. TEN-T includes projects such
Article Outline as road, rail, seaport, inland waterway, airport, ship-
ping management, air traffic management and nav-
Glossary
igation networks. Number of total rail projects
Definition of the Subject
makes almost 50% of all transport projects and
Introduction
high-speed rail network is one of the most important
Technological Development of HSTs
projects of TEN-T. With projects such as the PBKAL
The Spatial Development of HST
(Paris-Brussels-Köln-Amsterdam-London) network
Main Planning Principles of HSTs
and Southwest Europe network, high-speed rail lines
Economic Costs of HSTs and Financing Strategies
are expected to create substantial reduction in travel
Impacts of HSTs
times, increase the capacity and promote alternative
Future Directions: Future of the High-Speed Train
transport to air and road transport.
Bibliography
TGV TGV (Train à Grande Vitesse) is the name given to
the high-speed trains in France, which later on
Glossary adopted by other countries using the same technol-
ogy. French TGV is the first high-speed train in
Dedicated lines In transport and telecommunication,
Europe, started to serve between Paris and Lille in
dedicated lines are special rail, road infrastructure,
1981. Unlike Shinkansen, TGV operates on conven-
cable or other facilities dedicated to a specific appli-
tional tracks as well, thus, it can serve in city centers as
cation. In transport technology examples can be
well as regions by reducing its speed and this leads to
found in bus, train, and subway systems running
a better connectivity between regional and urban
on dedicated lines.
network and significant cost savings. French recorded
Induced traffic/demand Induced traffic (or demand)
the highest speed ever reached on wheel on steel trains
is the volume of traffic that is drawn to a new road
as high as 574 kph in 2007 during test drives [34].
by additional capacity [8]. Induced demand occurs
Tunneling Effects A notion brought up by Graham
from a number of different sources such as new
and Marvin [18] referring to the possible effects of
infrastructure, capacity increase, and infrastructure
transport systems connecting specific areas in a
improvement or land use and location changes.
region and bypassing others. This leads to some
Shinkansen Shinkansen (new trunk line in Japanese)
highly mobile socio-economic groups traveling
refers to the high-speed trains in Japan but also to the
through these places without even noticing the
world’s first HSR corridor, Tokaido Shinkansen,
existence of them. Connecting places via high-
which was opened between Tokyo and Osaka in
speed systems may create adverse impacts as it
1964. The main reason for the development of the
may push disconnected places to the periphery
Shinkansen network was to increase the network capac-
(see also space-time compression by Harvey [21]).
ity between major cities of Japan. Shinkansen runs on
dedicated railways designed only for high speed.
Definition of the Subject
TEN Trans-European Network (TEN) Program
comprises of a large number of projects in High-Speed Rail is an advanced railway technology for
telecommunications, transport, and energy [13]. carrying passengers at high speeds. It can be defined as

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
538 High Speed Rail, Technology Development of
the infrastructure capability to support high-speed
rolling stocks. In the transport literature, rail (or rail-
way) and train are often used synonymously [15]. In
this section of the Encyclopedia of Sustainability the
term will be used as High-Speed Train with reference to
the technological development of both the vehicle (the
train) and the track (the railways) but also with
a reference to the train as a carrier and high-speed
train not as a single element but as a system. Moreover,
the main focus is on the passenger transport, although
some emphasis will be made on the freight transport.
Late twentieth century witnessed the return of the train
as an important passenger transport mode and the
reason for this is the rapid development of the high-
speed rail technology. Twenty-first century is
witnessing even a larger impact of train as a transport
mode. World’s biggest economies are racing with each
other to implement high-speed train systems in order
to enhance their economies with fast transfer between
major centers of business, finance, residential develop-
ment, and tourism. However, these ambitions come
with a price: high-speed train can enhance economies
and societies but at the same time it can create envi-
ronmental, spatial, and socio-economic deprivation.
This review evaluates the challenges that the future
HST development faces within the context of sustain-
able development.
Introduction
" Transport technologies seldom make a comeback, save
in nostalgia trips for well-heeled tourists. Stagecoaches
have not made a reappearance on the Bath Road, nor
sedan chairs on the streets of London. But there is
a spectacular exception: railways, written off thirty
years ago as a Victorian anachronism destined to atro-
phy before the steady growth of motorway traffic, have
suddenly become one of the basic technologies of the
twenty-first century.
The reason of course is the high-speed train. . .
(Hall and Banister 1993 [19, p. 157])
Speed has been one of the biggest ambitions of the
mankind, not only in transport but also in different
aspects of life: fast broadband, fast food, fast vehicles,
and fast monetary transactions are only a few examples
for the desire of a fast life. The reason behind this
passion can be explained sociologically and can be
attributed to the way in which modernism was shaped
around the Fordist ideas of production and daily life in
the twentieth century, however, one can discuss that it
dates back to pre-Fordism, to the industrial revolution
when the train emerged as one of the most important
modes of transport of the time. Although rail lost its
power over the years, especially after the introduction
of the motorcar, in the last 40 years the number of
passengers using rail has increased substantially within
the European Union (EU15 see [14]) and in Japan.
Nevertheless, the situation differs from one country
to another. According to the figures of European Com-
mission’s 2010 Statistical Pocketbook, the Eastern
European countries witnessed a significant decline in
the rail use over the years while in the Western Euro-
pean countries the rail use increased significantly
within the same period [12]. This might be related to
the differences between countries in terms of “invest-
ment in rail services, the quality and frequency of
services offered, the growth in car ownership, levels of
expendable income, and the location of rail services
with respect to the population” [14, p. 4].
High-Speed Train development can also be seen as
one of the reasons of the differences between rail uses
in countries since majority of the Western European
countries have HST services, whereas HST develop-
ment has recently started in Eastern Europe. However,
HSTs still do not accommodate the majority of the
trips in many countries. In countries where the HST
growth took place more effectively, such as France,
Germany, Sweden, Spain, and Italy, in 2008 share of
HST in total rail transport was 61.9%, 28.5%, 27.2%,
22.9%, and 17.8%, respectively [12]. However, within
the European Union, HST network accounts only
23.9% of the EU27 passenger kilometer in 2006 [12].
Although these figures show that the HSTuse in Europe
is only a small fraction of the total rail usage and is
mainly applied by the Western European countries,
more and more countries are becoming interested in
HST systems. By 2025 the HST network around the
world will reach more than 40,000 km and one third of
the whole network will be in China, a new emerging
force in HST development ([36], See Fig. 4, [4]). More-
over, the EU White Paper for 2010 emphasizes the
importance of an effective passenger high-speed rail
network integrated with the air travel [10]. Recently,
in the USA, the government committed $8 billion to
 High Speed Rail, Technology Development of 539

HST projects on 13 corridors across 31 states. All these Technological Development of HSTs
investments around the globe shows that the govern-
The idea of increasing the speed of the railways dates
ments started to become more aware of the benefits of
back to as early as 1903 when in Germany test drives
a HST network as it not only increases the passenger
achieved 200 kph [9]. Since then and until the inaugu-
capacity between places but it also can be a driving
ration of the modern HST, tests continued in many
force of development in many scales from local to
countries. Minor technological attempts in Germany
regional and national.
during 1930s were followed by the achievement of
However, this fast development of HST network
Italian ETR in 1938, reaching up to 203 kph during
creates new challenges, especially in terms of the goals
test drives. However, these tests were interrupted by the
of sustainability. As discussed earlier, speed is one of
Second World War. The attempts took off again in 1955
the main elements of defining the movement of peo-
in France where the highest speed by that time was
ple and goods and in transport research travel time
recorded as 331 kph. Although the first attempts were
savings has been one of the main concerns: “[t]
made in Europe, it was Japan who first succeeded in the
ransport appraisal is almost totally (80%) dependent
commercial operation of HSTs after its success in
on the user benefits resulting from travel time savings
recording 256 kph in test drives in 1963.
and hence the overwhelming desire to speed up traf-
Although much of the reference is to the increased
fic” [2, p. 955]. Since traveling in high speed is very
speed while defining High-Speed Trains, the main rea-
attractive not only for businesses but also for tourism
son for constructing such a system is to increase the
and daily commuting, HST often increases the demand
capacity of routes. HST lines can increase the capacity
for traveling which is one of the main obstacles of
first, by supplementing the existing lines with more
sustainability and one of the main arguments of the
capacity and second, by shortening the headway
sustainable mobility paradigm [1]. In transport litera-
between trains. The latter generates as the frequency
ture, the volume of traffic created on a route with new
between services increases due to high speeds, longer
infrastructure or expansion in the current road struc-
trains and the most up-to-date signaling systems [15].
ture is called the induced demand. With the help of the
Higher frequency increases the capacity, therefore
new improvements, cars, passengers, and trucks are
makes the HST an attractive investment for countries.
drawn from other routes or even new demand is gen-
For instance, Shinkansen carried 400,000 passengers
erated with the relief in traffic on some routes. As
between Tokyo and Osaka per day in 2010 [34].
Cervero [5] states “[. . .] road improvements, critics
It is difficult to talk about a single definition of
charge, provide only ephemeral relief—within a few
High-Speed Train (HST), however, “the reference
years, most facilities are back to square 1, just as
is always to passenger services and not to freight”
congested as they were prior to the investment” [5, p.
[15, p. 594]. In order to define HST, one should look
3]. This phenomenon points out one of the main
at the meaning of high speed: “[h]igh speed can relate
challenges for HST development in terms of sustain-
to the infrastructure capability to support high speed,
ability: “how should the railway take a larger share of
the rolling stock capability to achieve high speed and/
the current demand for transport without inducing
or the actual operation speed achieved” (ibid. 594).
new demand” [14, p. 1].
According to the European Union (EU), high speed is
Under the light of the discussions above, this
250 kph for dedicated new lines and 200 kph for
review will first give explanation of the meaning of
upgraded lines in respect of the infrastructure capabil-
the High-Speed Train and evaluate the technological
ities [38]. Therefore, EU describes High-Speed Train as;
development of HSTs by demonstrating different
HST types. It will then discuss the principles of ● Those built specially for High-Speed travel.
HSTs briefly and mention the latest debates on the ● Those specially upgraded for High-Speed travel.
transport, socio-economic, and environmental They may include connecting lines, in particular
impacts of HSTs. The review will be finalized with junctions of new lines upgraded for High Speed
a discussion on the future of High-Speed Train with town center stations located on them, on
system. which speeds must take account of local conditions.
540 High Speed Rail, Technology Development of

Thus, high-speed lines would comprise: be implemented widely in the future or will be
completely abandoned due to extremely high construc-
● Specially built High-Speed lines equipped for
tion costs compared to other HST models.
speeds generally equal to or greater than 250 km/h.
● Specially upgraded High-Speed lines equipped for
Shinkansen Model: The Bullet Train
speeds of the order of 200 km/h.
● Specially upgraded High-Speed lines which have Japanese Shinkansen is regarded as the first modern
special features as a result of topographical, relief HST in operation [15]. The Tokaido Shinkansen
or town-planning constraints, on which the speed opened in 1964 between Tokyo and Osaka as a celebra-
must be adapted to each case. tion of the Olympics hosted by Japan [20]. However,
the main reason was to increase the passenger capacity
Currently, with some HSTs operating at speeds of
between these cities. Shinkansen travels 552.6 km,
350 kph, 200 kph might not seem high speed anymore.
connecting Tokyo, Nagoya, and Osaka at a maximum
“In practice, however, speed has not always been the
speed of 270 kph in 2 h and 25 min. As of March 2010,
best indicator, since commercial speed in many services
approximately 138 million passengers are carried per
is often limited due to, for example, proximity to
annum [23] and there are 400,000 passengers on board
densely urbanized areas or the existence of viaducts or
per day [34]. Remarkably, since its opening no fatal
tunnels” ([40], p. 20). Thus, speed cannot be regarded
accidents or injuries have been recorded [23].
as the only determining factor of the HSTs and
Due to the unique geographic features of Japan and
according to the International Union of Railways
technical differences between conventional tracks and
there are two important principles composed of differ-
new gauges, it was not possible to achieve high speeds
ent measurements [35]:
on the existing narrow gauge lines. Thus, the
● High-Speed Railway is not element but systems. Shinkansen was built on a new line, dedicated only to
They are in fact very complex system, comprised high-speed trains. With this feature, Shinkansen differs
by state of the art of the following components: from other HSTmodels in most of Europe. The need for
Infrastructure, station emplacement, rolling stock, construction of new lines, high usage of tunnels and
operation rules, signaling systems, marketing, high land values within city centers are some of the
maintenance systems, financing, management, factors that increase the building costs of Shinkansen
legal aspects [15] (Fig. 1).
● High speed is not unique and must be adapted to all
countries and circumstances: As well as other rail-
way systems, high speed can be considered under
different conceptions, regarding commercial
aspects, operation aspects, etc.
The differences between countries in terms of geo-
graphic and demographic characteristics, needs, and
resources have led to the evolution of different types
of HSTs [15]. These models which developed in the
conditions of different countries then adopted by
others, can be grouped into four: The Shinkansen,
TGV, The tilting HST and The MAGLEV. However,
in the last 10 years these models have been evolved
to include each other’s features. Examples such as
tilting technology adopted by the TGV and
Shinkansen trains can be found. Furthermore, MAG-
LEV technology is still in the first stages of its develop- High Speed Rail, Technology Development of. Figure 1
ment and it is not certain whether this technology will Shinkansen (Source: BakaOnigiri, Wikipedia)

The TGV
French TGV (Train à Grande Vitesse) In Europe,
where the high-speed system emerged in its modern
form in 1970–1980s, the rationality behind building
high-speed lines and developing it in such a way that
the lines in different countries would connect with each
other and other transport networks, was to build up
a Trans-European High-Speed Rail Network in the
future. TEN-T network comprises of road, rail, air,
and water transport networks designed to provide
seamless transport within the European Union and
High-Speed Rail network has a crucial role in achieving
the cohesion goals of the EU. According to Vickerman,
the reasons for creating such a network were [37, p. 22]:
● To overcome limited capacity on critical links of the
rail network, where some new investment was
needed and a completely new alignment appeared
the most cost effective solution.
● To increase speeds on particularly slow sections of
the trunk network where the developments in rail
technology could not be fully exploited.
● To improve the accessibility of more remote
regions.
However, it was initially France, within the EU
members, where the high-speed lines were developed
in a network and later on expanded to connect with
other national lines.
Although French began test drives as early as 1950s,
it was not until 1980s that they put the first modern
HST into service in Europe. The first TGV line opened
between Paris and Lyon in 1981. It serves the same
purpose as Shinkansen but differs in design due to
different features of France. “The most significant dif-
ference between the TGV and the Shinkansen is prob-
ably the ability of the former to operate on
conventional tracks as well, which allows the TGV to
use the conventional lines as it enters and leaves the city
center, leading to significant cost savings” [15]. This
feature of the TGV creates a network extending to
regions without HST and allows connections between
major cities and small towns where the demand is not
high enough to justify the construction of a dedicated
line (ibid.). This operational model is an example to the
mixed high-speed model [40]. The TGV can reach up
to 320 kph for commercial speed. French recorded the
High Speed Rail, Technology Development of 541
High Speed Rail, Technology Development of. Figure 2
TGV at Gare de Lyon (Source: Sese Ingolstadt, Wikipedia)
highest speed ever reached on wheel on steel trains as
high as 574 kph in 2007 during test drives [34]. Never-
theless, operating at this speed is not possible at the
moment (Fig. 2).
Since 1981, the opening of the first TGV line 1,400
million passengers have been carried [34]. Currently
the French TGV consists of eight lines [35]: TGV Sud-
Est (Paris – Lyon); TGV Rhône-Alpes (Lyon – Valence);
TGV Méditerranée (Valence – Marseille); TGVAtlantiqué
(Paris – Le Mans & Tours); TGV Interconnexion-Est (east
of Paris); TGV-Nord (Paris – Lille –London via Channel
Tunnel, Brussels – Amsterdam – Cologne); TGV Est
(Paris – Strasbourg); TGV Peripignan Figures (France –
Spain). The current network under operation is 1,896 km
with a further 210 km under construction and 2,616 km
planned (ibid.).
The Tilting Trains
Tilting train is a type of train with a sophisticated com-
puterized mechanism running on conventional lines.
The mechanism helps the train to tilt while it is going
through the curves so that the train can travel on con-
ventional lines at high speeds and passengers do not feel
the discomfort from the centrifugal force. Tilting tech-
nology was initially invented in the UK in early 1980s,
however, it was later abandoned. Nowadays, Italy is the
leading country in tilting train technology with its
Pendolino trains. Opposite to the Japanese, French,
Spanish, and German HST trains which all use a newly
built track on the sections where high speed is achieved
(except the German ICE-T), in Italy the tilting
542 High Speed Rail, Technology Development of
technology is developed on routes where the demand is
not high enough to justify the cost of constructing new
tracks that allow high-speed operation [15].
The German ICE-T, the Swedish X-2000 and the
Virgin Trains in the UK connecting London, Birming-
ham, Manchester, Liverpool, and Glasgow are the exam-
ples of tilting trains adopting the technology from Italy’s
Pendolino (ETR-450) trains. The Italian Pendolino
reaches up to the maximum speed of 250 kph, whereas
the Swedish X-2000 and the UK Virgin Trains
Pendolino are slightly slower reaching up to 210 kph
and 225 kph, respectively. Tilting mechanism is adopted
by other HST models such as Shinkansen trains which
use this technology to increase their speed.
The MAGLEV
The idea of running the train on magnetic plates dates
back to as early as 1970s, to the first tests conducted in
Germany. “The technology relies on electromagnetic
forces to cause the vehicle to hover above the track
and move forward at theoretically unlimited speeds”
[15, p. 597]. In practice, the aim is for an operation
speed of 500 kph (Taniguchi [39] cited in [15, p. 597].
Japan is one of the leading countries in testing the
MAGLEV technology. In 1997, tests started on the
Yamanashi Maglev Test Line [24]. According to Japa-
nese Railway Technical Research Institute that is
responsible of conducting the MAGLEV tests in coop-
eration with the JR Central, in 2003, a MAGLEV three-
car train set attained a maximum speed of 581 kph in
a manned vehicle run [29]. In November 2004, two
train sets succeeded in passing each other at relative
speeds of up to 1,026 kph [29]. Even though the MAG-
LEV technology has been proved to be successful in the
test drives, it has not been put into practice widely due to
high construction costs and no compatibility with the
existing lines. In Japan, Yamanashi Maglev Test Line has
recently been agreed to be extended for another 23 km
and is planned to become a part of the Tokaido
Shinkansen Bypass, making it the first long-distance
MAGLEV line in commercial operation [25]. In China,
a short MAGLEV line was opened in December 2003
connecting Shanghai Airport and the city’s Pudong
financial district with trains running at maximum
speed of 430 kph. However, as mentioned earlier, the
investment on MAGLEV trains is very small in number
MAGLEV
None
Construction cost
Shinkansen
Compatibilty
TGV
Full Tilting
200 350 500
Operating speed (kph)
High Speed Rail, Technology Development of. Figure 3
Operating speed, construction cost, and compatibility
(with the conventional network) characteristics of the four
high-speed train models ([15, p. 509], updated by the
authors)
compared to the rest of the HST development all around
the world.
As explained above, there are differences in princi-
ple between the models of HSTs. Givoni [15] compared
these models in terms of compatibility, construction
costs, and speed (Fig. 3).
The Spatial Development of HST
According to the reports of the International Union of
Railways [34], the HST lines in operation and con-
struction cover many of the developed and developing
countries. Nowadays there are high-speed train services
in 14 countries and planned lines in other ten countries
[36]. According to the UIC 2011 report, there is
14,700 km of high-speed line in operation, 9,703 km
under construction and 17,594 km planned HST sys-
tem in the world, that will make 41,997 km of high-
speed line in 2025, 17,769 km of which in Europe and
21,460 km in Asia [36] (Fig. 4). However, considering
more than 150 countries in the world have railway
systems [28]; the number of countries investing in
HST system is very low. Thus, it is questionable how
much the HSTs can be a part of the solution for the
increasing travel demand and whether it is the high
costs or lack of planning or else that is making some
countries reluctant of investing in this technology.
Europe
Besides France and Italy, HST development takes places
in other European countries as well. The Spanish AVE
 High Speed Rail, Technology Development of 543

SITUATION In operation Under construction Planned Total

COUNTRIES
Belgium 209 209
France 1896 210 2616 4722
Germany 1285 378 670 2333
Italy 923 395 1318

E The Netherlands 120 120

U Poland 712 712
R Portugal 1006 1006
O
P Russia 650 650
E Spain 2056 1767 1702 5525
Sweden 750 750
Switzerland 35 72 107
United Kingdom 113 204 317
Europe Total 6637 2427 8705 17769
China 4175 6058 2901 13134
Taiwan-China 345 345
India 495 495
A Iran 475 475
S Japan 2534 508 583 3625
I
A Saudi Arabia 550 550
South Korea 412 412
Turkey 235 510 1679 2424
Asia Total 7701 7076 6683 21460
Morocco 200 480 680
Other HST Systems

Argentina 315 315

Brazil 511 511

USA 362 900 1262

Total other HSTs 362 200 2206 2768

WORLD TOTAL 14700 9703 17594 41997

High Speed Rail, Technology Development of. Figure 4

The HST lines in operation, under construction and in the planning stage. Based on [36]

(Alta Velocidad Espanola, or Spanish High Speed),
opened in 1992, can be seen as another type of TGV as
it uses TGV-type rolling stocks in order to create inte-
gration with rest of Europe, mainly with France. How-
ever, like Shinkansen, “it runs on a dedicated line
throughout, because the Spanish conventional network
is wider than the standard International Union of
Railways (UIC) gauge used across most of Europe”
[15, p. 596]. Since it runs on dedicated lines and not
compatible with the rest of the domestic network, it is
possible to classify AVE in the Shinkansen category. In
order to solve the problem of integration within the
country, tracks have been developed to have three rails,
thus, both types of trains (conventional and AVE) can
run on them. The Spanish AVE achieved 350 kph in test
drives and runs on 300 kph max speed commercially,
544 High Speed Rail, Technology Development of
connecting Madrid-Barcelona (689 km) in 2 h 38 m
and Madrid – Sevilla (471 km) in 2 h 30 m.
Germany’s HST, the ICE (Inter-City Express) that
started to operate in 1991, today reaches the speeds of
up to 320 kph [7]. Unlike other systems, the initial
network was planned for both passenger and freight
transport, using a technology deviating from the
TGV and Shinkansen models. However, this mixed-
use turned out to be a disadvantage due to high
construction costs and low utilization of the lines,
thus, no other country attempted to replicate it.
Today the ICE network consists of different technol-
ogies with trains running on both dedicated lines and
existing lines using the tilting technology where the
demand is high enough to run high-speed train ser-
vices but not high enough to construct dedicated
lines [7]. Five different lines operate between several
major European cities including international destina-
tions in Switzerland, Austria, Belgium, and the
Netherlands.
Portugal is also one of the leading countries in
terms of the extent of the network to be developed.
According to UIC [36], 1,006 km of new HST line has
been planned to connect major cities with each other
and with Spain. Portugal is followed by Sweden,
Poland, and Russia that will have, respectively,
750 km, 712 km, and 650 km new high-speed lines in
the near future. By the extension of 72 km HST lines in
Switzerland and 204 in the UK, the HST lines in Europe
(currently 6,637 km) will reach 17,769 km in 2025 [36].
In the UK, currently the only operating dedicated high-
speed line is the Channel Tunnel Rail Link (CTRL,
nowadays referred to HS1) which connects London to
the Channel Tunnel.
Asia (Including Middle-East)
In Asia, following Shinkansen’s success, China is aiming
to be the new leader in HST development. Currently,
there are 4,175 km of high-speed lines in operation
which are planned to be extended to 13,161 km by
2025 [36]. When finished, China will have one third
of the world’s total HST network. Although South
Korea (KTX – Korea Train eXpress – 412 km) and
Taiwan (345 km) have put the HST lines into operation
before China, they have no future plans to develop the
network within the next 15 years. With the new lines in
India, Iran, Saudi Arabia, and Turkey, the existing
7,701 km HST network in Asia will reach to
21,460 km in total in 2025 [36].
Other High-Speed Systems
In the USA, the only HST line in operation is the Acela
Express tilting train running on the North East Corri-
dor line between Boston and Washington, DC.
Recently, the USA government announced the eight
billion US dollars of state funds to be spent for the
development of future HST lines. Currently, only the
State of California is planning to develop a HST
network between Los Angeles, San Francisco, and
Sacramento which will add 900 km of HST line to the
existing network [36]. In South America, Brazil, and
Argentina planned 511 km and 315 km HST lines,
respectively, which are aimed to be completed by
2020. Moreover, Morocco has 200 km of HST lines
under construction and a further 480 km planned
network [36].
Main Planning Principles of HSTs
As mentioned earlier, the main reason behind develop-
ing the high-speed rail technology is to increase the
capacity of routes and reduce the travel time. However,
increased speed is not solely sufficient to reach these
goals; other measures need to be met as well. These are
the integration of the HST network with the existing
transport system, creating seamless transfer between
the different modes of transport system and choosing
the location and number of the stations.
First of all, there are three main models of station
location: at the city center and in the outskirts (a third
model, station at the edge of the city, can be included in
either of the two models mentioned. For a detailed
analysis and explanation of the subject, see Menéndez
et al. [27]). Each option has its own pros and cons, and
the balance is always context specific. From an accessi-
bility perspective, the city center station location model
might be better than the outskirts model as demand is
likely to be higher due to higher population density
and/or level of commercial activity. It is also likely to be
much better served by public transport including
regional and national train services. On the other
hand, this model is significantly more costly and
more complicated than the outskirts model due to

restricted land for development and high costs of land
intake. The outskirts model is likely to be cheaper in
terms of land acquisition, simpler to construct and
might offer more land for redevelopment and likely
better connection opportunities for regional accessibil-
ity from a road transport perspective. Therefore, pros
of one model become generally the cons of the other.
Moreover, as Givoni and Banister mention the design
of the stations is as important as the location in terms
of allowing “seamless transfer between HST services
and other rail (or air) services” [14, p. 9]. The transfer
centers need to be designed in order to ease the pas-
senger flows between different modes and also to
accommodate different needs of passengers especially
in international stations.
The number of stations is another important ele-
ment in the planning of HSTs. In simple economic
terms, the number of stations is a function of a trade-
off between reducing the cost of accessing the station by
opening an additional station, thus shortening the
access journey, and the cost of adding a station which
is the cost of providing it (construction) and the cost
penalty of stopping the train (longer travel time) [17].
The characteristics of HST suggests it should have very
low number of stops between the start and end stations.
HSTs are feasible mainly on high-demand routes with
stations located in major conurbations with 12–15
million passengers a year [37]. Most cities, except
large metropolises like Tokyo and Paris, have one
HST station. Increased number of stations means lon-
ger travel time, as the train will stop frequently and that
would in turn reduce the desirability of the high-speed
train. For instance, in France, an important number of
trains between Paris and Lyon do not stop for 447 km
which is different than the UK Channel Tunnel rail link
that has three stops between London and the Channel
tunnel which is 109 km apart ([20, 22]). However, there
are also negative consequences of limiting the number
of stations and planning them only for the major cities.
“Since the HST often replaces or supplements an
existing conventional railway service, these bypassed
cities along the HST line will see substantial reduction
in rail services because they will lose the through traffic
to the HST and local demand would not support the
same level of service. In turn, this reduction in rail
accessibility can contribute to economic decline” [14,
p. 6]. Thus, in routes where the demand is high enough
High Speed Rail, Technology Development of 545
to run extra services, cautions are taken due to possible
negative impacts of HSTs. The different service levels
can be found in Japan (see the “Socio-economic
Impacts and Spatial Imbalance” section). However, as
stated earlier, these measures can be applied in high-
demand routes, therefore, for other places with low
demand the integration of the HST network with the
existing road, rail, and air system is essential.
There are three levels of integration: integration with
the urban transport network, the national intercity
transport network and the international, long-distance
air transport network. The attractiveness of using HST
compared to aircraft depends on the ease to get to the
station vs. the airport and the time of doing so. More-
over, integration with the regional network is one of the
essentials of HST development. Since the spatial spread
of HST network is constrained by the station location
choice and number of stations, its integration with the
regional rail network is crucial for sustaining an
increased accessibility from different regions. As Givoni
and Banister suggest “the location of the HST stations
must consider the entire rail network and not only local
demand from cities” in order to spread the accessibility
benefits from HST services across the entire network
[14, p. 9]. Also, the integration with the rail network
is more important than the road network as the latter
would not create opportunities for a decrease in the car
use due to the efficiency of the car in many ways such as
door-to-door travel. Lastly, integration with air travel is
another important strategy to spread the benefits of
HST network. More emphasis will be made on the
integration of air and rail in the next section.
It is those guiding principles for the development of
HST that will determine the scope and scale of its
impacts. Therefore, they are crucial in the planning of
HST systems.
Economic Costs of HSTs and Financing Strategies
Construction costs and financing strategies of HSTs
vary between countries and pretty much depend on
the type of technology that is used. Different technol-
ogies are adopted by countries in order to overcome the
restrictions that occur from geographic features or the
existing technology of the railways.
The construction costs have been briefly discussed
earlier when describing different HST models and
546 High Speed Rail, Technology Development of
different systems were compared in Fig. 3. However,
a more detailed comparison of construction costs
requires more information as the costs depend on so
many criteria that are mentioned in the previous par-
agraph. Nevertheless, “the comparison of construction
costs between different HSR projects is difficult since
the technical solutions adopted in each case to imple-
ment these features do not only differ widely
(depending on topography and geography), but also
evolve over time” ([40], p. 22).
The UIC [41] points out three major types of costs
for building a new HST structure. These are planning
and land costs, infrastructure building costs, and super-
structure costs [41]. According to a study comparing
the construction costs of HSTs from different countries
by Campos and de Rus [40], the range of costs varies
between €9 and €39 million per kilometer (2005 fig-
ures). In Europe “France and Spain have slightly lower
building costs than Germany, Italy, and Belgium. The
HSR construction in Asia, except China, seems to be
more expansive than in Europe” (ibid. 23). In addition
to the building costs, operation costs are important in
planning HST systems. According to Campos and de
Rus [40], these costs can be grouped in two: infrastruc-
ture operating costs and rolling stock and train oper-
ating costs (for a detailed explanation of the operating
costs see [40]).
Some of the well-known financing strategies can be
discussed here. The TGV was financed by SNCF
(Société Nationale des Chemins de fer Français –
French National Railway Corporation) with expecta-
tions of minimum 12% financial rate of return.
Unexpectedly, SNCF’s investment was fully amortized
by the end of 1993 after 12 years of service, proving that
it was hugely profitable [37]. In 2010, SNCF’s revenue
from high-speed train services including TGV was
€7.2 billion [30]. The SNCF is considered a state-
owned industrial and commercial enterprise under
French law [31]. Thus, the investment and operation
of the TGV is under the control of the state. In Japan
where the Shinkansen was initially constructed using
government and World Bank funds the situation is
slightly different. The Shinkansen lines were privatized
in 1987 making the financing strategy a public-private
partnership. The JNR (Japan National Railways) was
divided to six passenger and one freight rail companies
collectively called Japanese Railways Group (JR Group)
that include East Japan Railway Company, West Japan
Railway Company, and Central Japan Railway Com-
pany, all operating and maintaining different parts of
the HST network. The initial investment for the
Shinkansen was recovered by 1971, in 7 years of oper-
ation. Revenue from the Shinkansen has been an
important source for financing local lines. However,
these two examples (French TGV ad Shinkansen) are
exceptional and the outcomes cannot be attributed to
other HST systems.
Recently, in the US, the President Obama
announced a spending of nearly $8billion for a new
high-speed railway project funded by the Recovery Act
in order to move the US in the direction of developing
a high-speed rail service in 13 travel corridors covering
31 states all across this country with the goal of making
high-speed rail accessible to 80% of Americans within
25 years [6]. In this sense, the financing strategy in the
USA is similar to that of in France where the state
controls the investment and operations through state-
owned companies.
In the UK, the first high-speed line, CTRL, was
initially planned as a state investment. However, due
to the financial difficulties the governments signed
several consortiums until 2010 when a 30-year conces-
sion to a consortium of Canadian investors was
announced. For the HS2 the government set up
an initiative under the name of HS2 Limited Company
in 2009 for the design and development of the new
HST line.
Impacts of HSTs
Transport Impacts
The main benefit of high-speed train which also fosters
other benefits at economic and social levels is the
increased route capacity. Here, the attribute is given
not to the increased speed but to the reduced headways
due to the high speeds. Higher speeds and the up-to-
date signaling systems which are prerequisites of high-
speed systems create short headway between trains and
thus, trains run at higher frequencies [15] and provide
more capacity on the routes where the existing network
cannot meet the demand. Thus, changes in modal share
occur due to the demand shift between transport
modes and generation of additional demand. In TGV
line between Paris and Lyon the share of the train

EU-27 HST
YEARS Thousand million pkm CHANGE
1990 15,23
1995 32,94
2000 58,8
2005 80,11
2008 97,6
High Speed Rail, Technology Development of. Figure 5
Number of passengers transported by HST and conventional railways per km in EU27. Based on [12]
journeys increased by 32% within the 3 years of its
opening opposite to the decreased aircraft, car, and
bus traffic (ibid.) “Paris-Lyon air traffic halved between
1980 and 1984, Paris-Geneve fell by around 20% and
Paris-Sud Est in general grew at less than half the rate of
other radial routes from Paris” [37, p. 24]. A similar
result was seen in Spain after the inauguration of the
AVE between Madrid and Seville where the train jour-
neys increased by 35% between 1991 and 1994 [15].
Likewise, in the first 5 years of operation ICE, passen-
gers more than doubled from just over 10 million to
nearly 23 million [37]. “The AVE has been remarkably
successful in terms of traffic growth and diversion. [. . .]
[A]ir traffic has suffered with an estimated 60% reduc-
tion. Some 32% of travelers have diverted from air,
25% from car, and 26% were newly generated” (ibid.
30). However, it is important to mention that develop-
ing HSTs come at the expense of other railway net-
works. For instance, in EU 27 the number of
passengers traveled by conventional railways showed
no significant increase, whereas this increase is notable
in HST trips (Fig. 5). This shows the fast increase in
HST use while the rail use stays constant.
Apart from modal shift, high-speed lines also create
substantial reduction in travel times which makes HSTs
an attractive mode of transport. Before the inaugura-
tion of the HST in Japan, it took 7 h to travel between
Tokyo and Osaka on the conventional line; it was then
reduced to 4 h following the inauguration of the
Shinkansen (Matsuda, 1993 cited by [15]). The current
travel time between Tokyo and Osaka is 2 h 25 min with
a maximum speed of 270 kph. Traveling with TGV
between Paris and Lyon takes about 2 h at the moment.
Likewise, in central and southern China almost
1,000 km is taken in 3 h and in northern China around
High Speed Rail, Technology Development of 547
RAIL
Thousand million pkm CHANGE
400,7
116% 350,5 −13%
79% 370,7 6%
36% 377 2%
22% 409,2 9%
120 km is traveled in 30 min by Chinese HST, bringing
countries’ largest cities closer to each other.
With reductions in travel time and changes in
modal share, HSTs become competitors to airline com-
panies. However, rather than competition, cooperation
is more beneficial for airlines. “Cooperation between
air and HST can take place if the airport is situated on
the rail and HST networks allowing airlines to retain
connecting traffic from short-haul routes to their more
profitable long-haul services, thus giving up flights
without compromising their Hub and Spoke opera-
tion” [14, pp. 6–7]. In Europe, Amsterdam Schiphol,
Paris Charles de Gaulle, and Frankfurt International are
examples where the seamless connection between the
airport and high-speed train already exists.
Figure 6 demonstrates the possibility of airline-HST
cooperation and integration of the HST network with
the existing regional railway network. According to
this, rather than providing services between minor
cities, airlines can reduce the number of short-haul
flights and operate on long-distance routes which are
more profitable. The connection of airports with HST
network which is also connected to the conventional
rail network would not reduce the profits of the airline
companies; on the contrary this type of connection
would decrease their costs occurring from short-haul
flights by keeping the number of long-haul flight pas-
sengers steady.
Sustainability Impacts
Socio-economic Impacts and Spatial Imbalance
Transport investments that supply fast, affordable, reli-
able, and secure systems are seen as a condition for
economic development either creating new economic
548 High Speed Rail, Technology Development of
AIRCRAFT
HST
HST
High Speed Rail, Technology Development of. Figure 6
Integration of HST network with airport can reduce short-haul flights
areas with the help of agglomeration economies or
increasing the size of actual markets (for a detailed
understanding of the subject see [3]). “Development
opportunities from new networks will occur at
locations with good access to the new network and
which provide nodal connections to other net-
works”[37, p. 35].
It is possible to observe this argument in the case of
high-speed rail investments as well. Evaluating the pre-
vious studies on Shinkansen HST, Banister and
Berechman [3] states that there are three conditions
needed for growth: high incidence of information
exchange industries; sufficient opportunities for higher
education; good accessibility to a Shinkansen station
[3, p. 279]. At the local level, Shinkansen’s impacts
occurred differently. The stations located in existing
cities barely had any impacts, conversely, new stations
at the edge of existing cities boosted the population and
employment growth rates. First TGV line between Paris
and Lyon created similar impacts in some parts on the
route. A new station in Lyon Part-Dieu, close to central
Lyon 2 h from Paris by TGV Sud-Est, has been largely
affected from the opening of the TGV line especially in
terms of the significant increase in the demand for
office space (ibid.). Moreover, further local impacts
have been observed as the TGV has been extended to
the west and north. “Substantial growth has taken place
at Le Mans, Nantes, and Vendôme in the TGV
Atlantique” (ibid. 281). Also on the TGV-Nord line
opened later in 1993 in order to connect Paris, London,
and Brussels (northern Europe), a technopole concept,
Euralille, was planned (ibid.). Initially, the network was
planned strategically that Lille would become “a busi-
ness bridge-head on the continent for London; city of
services for Brussels and an attraction of relocations
HST
from Paris” ([42], p. 130). Recently, a major office
development has been extended southward in Lille
[20]. Likewise in Spain, AVE is known to have created
new markets with the help of the shuttle services
Ciudad Real and Puertollano around 200 km from
Madrid [37].
However, new HST lines do not generate the same
economic effects everywhere: “only those centers which
are prepared to support it with complementary invest-
ment will stand to gain” [37, p. 35]. As seen in the
example of The Channel Tunnel, where the stations
are not located in major conurbations (but in the
Kent countryside), the development potential accrues
in other centers (ibid.), having maybe less impact that it
would have had they been located in major centers.
Nevertheless, a major office development around sta-
tion, King’s Cross and St Pancras after the completion
of the CTRL (Channel Tunnel Rail Link) in 2007, can
be seen [20].
The accumulation of investments only in major
centers might not always have positive influences, espe-
cially when the subcenters or minor cities are
concerned. As mentioned earlier, HSTs are feasible
mainly on high-demand routes. This economic neces-
sity makes it difficult for other areas – minor cities – to
connect to the high-speed system which in turn creates
economic and social imbalances. According to Graham
and Marvin [18], “advanced infrastructure networks
targeted on valued parts of the metropolis and drawing
them into intense interaction with each other” create
“tunnel effects” [18, p. 201]; also mentioned by
Plassard, 1991 cited in [37]). In this case, HSTs
connecting major cities within and beyond countries
can create so-called tunneling effects, changing the
proximity of places: making some closer to each other

while pushing others to the periphery. Hall [20] calls
this effect “peripheralization of the periphery” empha-
sizing that the regions which are already in the periph-
ery, due to the lack of economic viability, are even more
peripheralized as a result of making investments in
other parts of the region. Maps of Europe as
a “Shrinking Continent” have been created by
Spiekermann and Wegener [32]. According to the
time-space maps of rail travel times which are based
on a deeper geographical understanding of space and
time (see for example, [21] for more detailed under-
standing of the space-time compression), the core cities
are drawn closer in terms of time-distance, while the
more distant places which lack high-speed networks are
bypassed and thus, become relatively more remote
(ibid.). For instance, in a recent study by Martı́nez
Sánchez-Mateos and Givoni [26] where the authors
explore possible impacts of London – Birmingham
HS2 link, they demonstrate that the towns in Birming-
ham catchment area will be affected from the construc-
tion of the HST stations in different ways. According to
the study that explores the possible accessibility
changes in the area after the introduction of the HS2
link, the towns that are relatively close to Birmingham
and have good access to London will not benefit from
the HS2, on the contrary, they might be adversely
affected resulting in a decrease in their accessibility to
London. Whereas, the settlements on the North of
Birmingham that have less accessibility to London
will benefit from the HS2 service more than the others.
Thus, this would result in changing the geography of
the county [26]. This example not only demonstrates
the power of transport systems on geography but also
shows the importance of station locations and integra-
tion between the transport modes.
In order to solve this problem, ”in Japan the
Shinkansen was deliberately structured on the basis of
two service levels, fast (Hikari) serving only main cities
and semi-fast (Kodama) serving intermediate cities,
and this has been maintained; a super-fast (Nozomi)
service was added in 1992” [20, p. 65]. In France and
Germany similar measures were taken and additional
stations were planned in small cities (ibid.). Likewise,
in the UK, the Channel Tunnel Rail Link (HS1) was
seen as a tool to regenerate the economy and Ebbsfleet
and Stratford stations formed part of the Thames Gate-
way regeneration project east of London (ibid.).
High Speed Rail, Technology Development of 549
However, these solutions are only possible where the
demand is high enough to compensate this investment.
Furthermore, since high-speed trains are relatively
more expensive compared to conventional trains, net-
work solutions which aim to spread the impacts of
HSTs would not be sufficient to make the system
socially inclusive as not all socio-economic groups
would be able to afford them. Here the crucial question
is who benefits from such an expensive system.
Apart from economic and social impacts of HSTs
that have been argued above, the safety benefits of HSTs
are also important for the society. According to
Campos and de Rus [40], “any comparison of accident
statistics for the different transport modes immediately
confirms that HSR is – together with air transport – the
safest mode in terms of passenger fatalities per billion
passenger-kilometers” (ibid. 25). As mentioned earlier,
no fatal accidents or injuries on Shinkansen lines have
been recorded since its opening [23].
Despite the social and economic inequalities that
the HSTs might create in some regions, they are still
important tools for the social and economic sustain-
ability of the regions as they play an important role in
the regeneration of the regions in the longer term and
creating safe travel for societies. However, the govern-
ments need to be careful in planning these networks as
like other transport investments they can be “the maker
of some cities, but the breaker of the others” (Clark,
1957 cited in [20]).
Environmental Sustainability The environmental
impacts of HSTs are not negligible. Although high-
speed train systems are widely perceived to be climate
friendly, there are serious environmental costs that they
generate both during the construction and the opera-
tion. The impacts range from land take, barrier effects,
visual intrusion, noise, and air pollution to contribu-
tion to global warming ([40]; [15]). The amount of
energy used during construction and operation of
HSTs, the source of energy and the level emissions
created by the HSTs are important determinants of
how green the HSTs are. “HSTs are predominantly
electric powered and therefore emissions from HST
operations are considered to be linearly related to
energy consumption and the sources used to generate
the electricity” [15, p. 606]. Thus, the cleaner the source
of energy, the more climate friendly HSTs are.
550 High Speed Rail, Technology Development of
In France, the main energy source of the TGV trains is
nuclear energy. Nuclear power is seen as an important
source of energy as it generates electricity in higher
amounts than any other clean energy source. However,
nuclear power carries high risks in terms of radioactive
elements produced during the energy production pro-
cess and the problems of radioactive wastes; thus, great
attention needs to be paid in planning these systems
especially in places where geographical hazards are
more likely to happen.
As discussed earlier, the modal shift from cars and
aircrafts to HSTs as a result of increased connectivity
between major cities is known to be beneficial for the
environment. The CO2 emissions per passenger gener-
ated by HSTs are significantly lower than the emissions
from aircraft [16]. While the cleanest aircraft produces
56.7 kg of CO2 per passenger between London and
Paris, the HST produces 14.4 kg per passenger on the
same route (ibid. 80). Thus, it is quite common to
speak about CO2 benefits as a result of HST operations.
However, it is not possible to achieve reductions in CO2
emissions unless HSTs fully replace road and air
traffic. According to Givoni and Banister [14] London
and Paris route has not experienced such a decline in
air traffic after the introduction of HST services. On the
contrary, carriers like British Airways and Air-France
continue to offer frequent services on this route even
though the HST operator captures over 70% of the
market [14, p. 6].
Givoni and Banister [14] mention two conditions
that are necessary for the HST to reduce air transport
impact on the environment. “First, the HST service
must be integrated with other aircraft services provided
by the airline and thus it must complement it rather
than compete with it. Second, the airline (or airport)
should not use the freed runway slots for other flights.
With the development of the HST network and suc-
cessful integration of air and rail services (for example,
at CDG) the first condition is now being met at
a number of airports” [14, p. 7]. However, unless new
regulations are introduced to the airway companies,
the second necessary condition is unlikely to happen.
Future Directions: Future of the High-Speed Train
The fast development of the modern HST in the last
50 years and the construction of high-speed lines by
a growing number of countries all around the world
every year show that the HSTs are regarded as crucial
elements for the future of the transport systems.
First of all, the HST development needs to be seen
as a network development integrated both within and
in-between countries rather than single lines
connecting major cities, which is often the case. Thus,
integration of HSTs with other transport modes is
crucial. As discussed earlier, the development of HST
lines in Europe under the Trans-European Network is
an important step toward achieving a more inclusive
network in Europe. However, further development is
needed especially in the Eastern European countries.
This can be achieved not only in Europe but also in
other parts of the world by “adopting a HST technol-
ogy where the trains can also run on the conventional
network (the TGV model) and by designing the HST
stations to allow fast and seamless transfer between
HST services and other (national and regional) ser-
vices” [14, p. 17]. Moreover, a global integration of
the whole system can be achieved by possible connec-
tivity of HSTs all around the world via airports.
Second, the uneven distribution of HST network is
an important obstacle for the regional development
and equality. Although it is not possible to build HST
lines connecting every single city due to the high con-
struction and operation costs, it is an important task
for planners, engineers, and authorities to make the
HST service available to the wider community. Thus
the question, as the UIC [34] suggests will be “how to
reduce the door-to-door travel time when serving large
built-up areas?” [34, p. 1]). In this sense, the use of
existing rail systems and their integration with the HST
lines is crucial.
Third, as discussed earlier, HSTs can help shifting
the demand from other modes of transport to rail as
they offer reductions in travel time. This might result in
reducing the impacts of transport on the environment.
However, there is evidence that the HSTs can create
new demand [15]. In this case, “the introduction of
HST services might be counterproductive to sustain-
able transport, even when it promotes rail use” [14,
p. 6]. In order to overcome this problem there is a need
to ensure the development of HST result in modal shift
(e.g., from aircraft) and not generation of new travel.
Moreover, as integration is a crucial element in creating
sustainable transport, technical incompatibilities need

to be solved to integrate the systems fully. Differences in
electrical power, safety requirements, track belts, load-
ing gauge sizes are the main technical problems to
overcome [20]. Technical standardization has been
a major issue for the EU. In order to improve the
interlinking and interoperability of the national rail
networks as well as access to them, rail interoperability
directives have been published [11].
Finally, HSTs can be seen as locomotives of the
regional development and can regenerate the economy
in regions where the socio-economic deprivation has
started as a result of lack of public services and private
investment. However, they might also create spatial
imbalance and social inequalities. Therefore, introduc-
tion of HSTs need to be evaluated within the regional
and national socio-economic plans.
In conclusion, High-Speed Train can be consid-
ered an important element of sustainable develop-
ment as it can increase the social integration, act as
a catalyst for the economic development and provide
greener transport with less energy consumption and
high passenger volumes only if the conditions that
are discussed in this text are met. Only then it would
be possible to discuss the benefits of HSTs not only
for a sustainable transport system but also for sus-
tainable local, regional, and national development by
all means. It would then be possible to state that the
high-speed train marks the second age of the
railways.
Bibliography
Primary Literature
1. Banister D (2008) The sustainable mobility paradigm. Transp
Policy 15:73–80
2. Banister D (2011) The triology of distance, space and time.
J Transp Geogr 19:950–959
3. Banister D, Berechman J (2000) Transport investment and
economic development. UCL Press, London
4. Bradsher K (2010) China is eager to bring high-speed rail
expertise to the US. NY Times, 7 April 2010
5. Cervero R (2002) Induced travel demand: research design,
empirical evidence, and normative policies. J Plan Lit 17(1):3–20
6. Cooper M (2011) Administration pitches big rail projects. NY
Times, 8 Feb 2011. www.nytimes.com/2011/02/09/us/09rail.
html?ref=highspeedrailprojects
7. Deutche Bahn (2011) The ICE. Deutsche Bahn’s fastest train.
http://www.bahn.com/i/view/GBR/en/trains/overview/ice.shtml.
Accessed 10 Apr 2011
High Speed Rail, Technology Development of 551
8. Ewing R, Lichtenstein A (2002) Induced traffic and induced
development: a literature review, Prepared for the Chester
County Planning Commission
9. Ellwanger G, Wilckens M (1994) High speed for Europe. JPN
Railway Transp Rev 3:17–25
10. European Commission (2001) White paper, European trans-
port policy for 2010: time to decide. www.ec.europa.eu.
Accessed 28 May 2011
11. European Commission (2004) M1 directive of the European
parliament and of the council of 19 March 2001 on the inter-
operability of the conventional rail. www.eur-lex.europa.eu.
Accessed 28 May 2011
12. European Commission (2010) EU energy and transport in
figures statistical pocketbook 2010. www.ec.europa.eu.
Accessed 14 May 2011
13. European Commission (2011) Trans-European networks.
www.ec.europa.eu/ten/index. Accessed 12 May 2011
14. Givoni M, Banister D (2008) Reinventing the wheel – planning
the rail network to meet mobility needs of the 21st century.
TSU working paper, Ref. 1036
15. Givoni M (2006) The development and impact of the modern
high speed train. Transp Rev 26(5):593–611
16. Givoni M, Brand C, Watkiss P (2009) Are railways ‘climate
friendly’? Built Environ 35(1):70–86
17. Givoni M, Rietveld P (2010) Access to rail in urban areas –
examination of the number of stations. TSU working paper,
Ref. 1044
18. Graham S, Marvin S (2001) Splintering urbanism. Routledge,
London
19. Hall P, Banister D (1993) The second railway age. Built Environ
19(3/4):157–162
20. Hall P (2009) Magic carpets and seamless webs: opportunities
and constraints for high-speed trains in Europe. Built Environ
35(1):59–69
21. Harvey D (1990) The condition of postmodernity: an enquiry
into the origins of cultural change. Blackwell Publishing, Oxford
22. HS1 (2011) Key facts. http://www.highspeed1.com/about/
facts/. Accessed 12 May 2011
23. JR-Central (2011) Central Japan Railway Company, about the
Shinkansen, safety. http://english.jr-central.co.jp/about/
safety.html. Accessed 11 Apr 2011
24. JR-Central (2011) Central Japan Railway Company, company
profile. http://english.jr-central.co.jp/company/company/
about/history.html. Accessed 11 Apr 2011
25. JR-Central (2011) Central Japan Railway Company, FY2011 key
measures and capital investment. Report can be accessed on
http://english.jr-central.co.jp/index.html. Accessed 11 Apr 2011
26. Martı́nez Sánchez-Mateos HS, Givoni M (2009). The accessibil-
ity impact of a new high-speed rail line in the UK: a preliminary
analysis of winners and losers. TSU working paper series,
Ref. 1041
27. Menendez JM, Guirao B, Coronado JM, Rivas A, Rodriguezi FJ,
Ribalaygua C, Urefia JM (2002) New high-speed rail lines and
small cities: locating the station. In: Brebbia CA, Martin-Duque
JF, Wadhwa LC (eds) The sustainable city II. WIT Press, UK
552 High Speed Rail, Technology Development of
28. Railisa (2011) The UIC statistics database. www.railisa.tsf.it/
railisa/ password required and can be obtained from the UIC
website
29. RTRI (2011) Japan Railway Technical Research Institute, Over-
view of MAGLEV R&D. http://www.rtri.or.jp/rd/division/rd77/
yamanashi/english/maglev_frame_E.html. Accessed 18 Nov
2011
30. SNCF (2010) Profile and key figures. http://medias.sncf.com/
resources/en_EN/medias/MD0006_20100210/file_pdf.pdf.
Accessed 11 Apr 2011
31. SNCF (2011) Legal framework. http://www.sncf.com/en_EN/
html/media/CH0012-Suppliers/BR1051-Important-laws/MD0005
_20100804-Read-article.html
32. Spiekermann K, Wegener M (2008) The Shrinking continent:
accessibility, competiveness, and cohesion. In: Faludi A (ed)
European spatial research and planning. Lincoln Institute,
Cambridge, MA, pp 115–139
33. Tentea (2011) TEN-T projects, Trans-European Transport Net-
work Executive Agency. www.tentea.ec.europa.eu/en/ten-t_
projects/. Accessed 11 May 2011
34. UIC (2010) High speed rail: fast track to sustainability.
www.uic.org/IMG/pdf/20101124_uic_brochure_high_speed.
pdf. Accessed 11 Apr 2011
35. UIC (2011) High speed principles and advantages. http://www.
uic.org/spip.php?article443#outil_sommaire_0. 11 Apr 2011
36. UIC (2011b) High speed lines in the world – UIC High
Speed Department. http://www.uic.org/spip.php?article573.
Accessed 11 Apr 2011
37. Vickerman R (1997) High-speed rail in Europe: experience and
issues for future development. Ann Regional Sci 31:21–38
38. European Commission (1996) Council Directive 96/48/EC of
23 July 1996 on the interoperability of the trans-European
high-speed rail system. Official Journal of the European Com-
munities, 17 Sep 1996. http://eur-lex.europa.eu/LexUriServ/
LexUriServ.do?uri=OJ:L:1996:235:0006:0024:EN:PDF. Accessed
22 Aug 2011
39. Taniguchi M (1993) The Japanese magnetic levitation trains.
Built Environ19(3/4):235–243
40. Campos J, de Rus G (2009) Some stylized facts about high-
speed rail: A review of HSR experiences around the world.
Transp Policy 16:19–28
41. UIC (2005) High Speed Rail’s leading asset for customers and
society. UIC Publications, Paris. Accessible at: /www.uic.asso.fr
cited in Campos J, de Rus G (2009)
42. Ampe F (1995) Technopole development in Euralille. In:
Banister D (ed) Transport and urban development. E & F.N.
Spon, London, pp 128–135
Books and Reviews
Ampe F (1996) Technopole development in Euralille. In: Banister D
(ed) Transport and urban development. E & F.N. Spon, London
Banister D, Goodwin MT (2011) Quantification of the non-transport
benefits resulting from rail investment. J Transp Geogr
19(2):212–223
Brons M, Givoni M, Rietveld P (2009) Access to railway stations and
its potential in increasing rail use. Transport Res A 43:136–149
Campos J, Gagnepain P (2009) Measuring the intermodal effects of
high-speed rail. In: Rus G (ed) Economic analysis of high speed
rail in Europe. BBVA Foundation, Madrid
Coto-Milla´n P, Inglada V (2004) Social benefit of investment
projects: the case for high-speed rail. In: Coto-Millan P (ed)
Essays on microeconomics and industrial organisation.
Springer, Berlin
De Rus G, Nombela G (2007) Is the investment in high speed rail
socially profitable? J Transp Econ Policy 41(1):3–23
Department for Transport (2010) High speed rail London to the
West Midlands and beyond – A report to the government by
high speed two limited. www.dft.gov.uk
Garmendia M, Ureña JM, Ribalaygua C, Leal J, Coronado JM
(2008) Urban residential development in isolated small cities
that are partially integrated in metropolitan areas by high
speed train. Eur Urban Reg Stud 15:249–264
Givoni M, Banister D (2006) Airline and railway integration. Transp
Policy 13:386–397
Givoni M (2007) Environmental benefits from mode substitution -
comparison of the environmental impact from aircraft and
high-speed train operation. Int J Sust Trans 1(4):209–230
Givoni M, Holvad T (2009) The prospects for European railways: is
the second railway age still here or yet to begin? Built Environ
35(1):5–10
Givoni M, Rietveld P (2007) The access journey to the railway
station and its role in passengers’ satisfaction with rail travel.
Transp Policy 14:357–365
González-Savignat M (2004) Will the high-speed train compete
against the private vehicle? Transp Rev 24(3):293–316
Hall P (1999) High-speed train and air: competitive or complemen-
tary? Paper presented at the Lake Arrowhead transportation
symposium, 25 Oct 1999
Holvad T (2009) Review of railway policy reforms in Europe. Built
Environ 35(1):24–42
Mokhtarian PL, Salomon I (2001) How derived is the demand for
travel? Some conceptual and measurement considerations.
Transport Res A 35:695–719
Murikami J, Cervero R (2010) California high-speed rail and
economic development: postindustrial market profiles and
public policy implications, symposium, Berkeley, 2–3 Dec
2010
Norman C, Vickerman R (1999) Local and regional implications of
trans-European transport networks: the channel tunnel rail
link. EnvironPlann A 31:705–718
Preston J (2009) Trends in European railways over the last two
decades. Built Environ 35(1):11–23
Rietveld P (2002) Why railway passengers are more polluting in the
peak than in the off-peak; environmental effects of capacity
management by railway companies under conditions of fluc-
tuating demand. Transport Res D 7(5):347–356
Sánchez-Borràs M, Nash C, Abrantes P, López-Pita A (2010) Rail
access charges and the competitiveness of high speed trains.
Transp Policy 17:102–109

Sasaki K, Ohashi T, Ando A (1997) High-speed rail transit impact on
regional systems: does the Shinkansen contribute to disper-
sion? Ann Regional Sci 31:77–98
Steer Davies Gleave (2004) High speed rail: international compar-
isons. Final report for the commission for integrated transport
(February). http://cfit.independent.gov.uk/pubs/2004/hsr/
research/pdf/hsr.pdf
High Speed Rail, Technology Development of 553
Takagi R (2005) High speed railways: the last 10 years. JPN Railway
Transp Rev 40:4–7
Ureña JM, Menerault P, Garmendia M (2009) The high-speed rail
challenge for big intermediate cities: a national regional and
local perspective. Cities 26:266–279
Whitelegg J, Hultén S, Flink T (1993) High speed trains: fast tracks
to the future? Leading Edge Press, Hawes
554 High-Occupancy Vehicle and Toll Lanes

High-Occupancy Vehicle and Toll Definition of the Subject

Lanes Traffic congestion continues to be a significant issue in
metropolitan areas throughout the country. Transpor-
KATHERINE F. TURNBULL
tation agencies at the federal, state, metropolitan, and
Department of Landscape Architecture & Urban
local levels are using a variety of techniques and
Planning, Texas Transportation Institute, College of
approaches to improve traffic flow, enhance mobility,
Architecture – The Texas A&M University System,
and provide travel options.
College Station, TX, USA
High-occupancy vehicle (HOV) and high-
Article Outline occupancy toll (HOT) lanes are being used in some
metropolitan areas to ease traffic congestion and
Glossary
enhance mobility. HOV lanes provide travel-time sav-
Definition of the Subject
ings and improved trip-time reliability by encouraging
Introduction
travelers to change from driving alone to carpooling,
Types of HOV Lanes
vanpooling, or riding the bus. HOV/HOT lanes expand
Access Treatments
the allowed user groups to include solo drivers or
Allowed Vehicles and Vehicle-Occupancy
lower-occupant vehicles, who can access the lanes by
Requirements
paying a fee.
Operating Hours
The use of HOV/HOT lanes has evolved since the
Supporting Facilities and Services
late 1960s. Today, HOV lanes and HOV/HOT lanes
Planning
are in operation in 32 metropolitan areas in North
Design
America. These facilities represent an important ele-
Enforcement
ment of the transportation system in these areas,
Public Information and Marketing
enhancing mobility, managing congestion, and provid-
Case Studies
ing travel options.
Future Directions
Bibliography
Glossary Introduction
Buses Vehicles designed to carry multiple passengers. The development and operation of HOV/HOT
Depending on the size, buses may carry between facilities in the USA have evolved over the past
12 and 100 passengers. 42 years. The opening of the bus-only lane on the
Carpools Vehicles carrying two or more persons. Shirley Highway (I-395) in northern Virginia/Wash-
High-occupancy toll (HOT) lanes HOV lanes that ington, D.C., in 1969, the contraflow bus lane on the
also allow single-occupant vehicle use through pay- approach to New York-New Jersey’s Lincoln Tunnel in
ment of a toll. 1970, and the El Monte Busway in Los Angeles in 1973
High-occupancy vehicle (HOV) lanes Roadway lanes represents the first freeway HOV applications in the
designed and operated for buses, vanpools, and country.
carpools at all times of the day or during specific Many of the initial HOV lanes were bus-only appli-
time periods. cations or allowed buses and vanpools. In an effort to
Hybrid vehicles Vehicles which operate under maximize use, carpools became the dominant use
a combination of gasoline and electric power. group on most projects during the 1970s and 1980s.
Low-emission and energy-efficient vehicles Vehicles The vehicle-occupancy requirements for carpools have
which typically use non-gasoline power sources and evolved over time. A three-person per vehicle (3+)-
do not emit high levels of pollutants. occupancy level was initially used on many projects,
Vanpools Vans organized to carry commuters to and but most current facilities use a two-person per vehicle
from work. (2+) carpool designation.

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3

In the 1990s, a few areas began experimenting with
value pricing projects, often called HOT lanes, which
were allowed under the Value Pricing program in pre-
vious federal legislation. HOV/HOT lanes expand the
allowable user groups to include single-occupant or
lower-occupant vehicles for a fee, while maintaining
free travel to qualifying HOVs. In addition, the intro-
duction of electronic toll collection (ETC) and other
advanced technologies provided greater opportunities
for pricing applications on HOV lanes. HOV/HOT
lanes enhance mobility and travel options in congested
corridors.
The Safe, Accountable, Flexible, Efficient Transpor-
tation Equity Act: A Legacy for Users (SAFETEA-LU)
provides operating agencies with additional flexibility
by allowing exempt user groups to use HOV lanes
when available capacity exists on an HOV lane. Tolling
single-occupant vehicles (SOVs) represents one of
the exempt user groups. Low-emission and energy-
efficient vehicles, including some types of hybrid vehi-
cles, are allowed to use HOV/HOT lanes in some states.
Today there are some 130 HOV freeway projects in
the 32 metropolitan areas in North America. Further,
there are12 HOV/HOT lanes in operation in eight
metropolitan areas, and more projects in the planning
and development stage. Further, 10 states allow low-
emission and energy-efficient vehicles to use HOV/
HOT lanes.
HOV/HOT lanes involved different designs and
operating strategies. Some facilities, such as the I-25
Express Lanes in Denver, are barrier-separated facilities
located in the center median of the freeway, with lim-
ited access points. Other HOV/HOT lanes are sepa-
rated from the adjacent general-purpose lanes only by
normal paint striping. Access and egress to these facil-
ities may be limited to specific sections or unlimited.
The I-394 HOV/HOT lanes in Minneapolis include
sections of both types of treatments.
Some facilities are reserved for HOV/HOT vehicles
at all times (24/7). Other facilities are reserved for
HOV/HOT vehicles only during the morning and
afternoon peak periods and are open to all traffic at
other times of the day.
Many HOV/HOT projects throughout the country
carry more people than the adjacent freeway lane. In
addition, they provide trip-time savings and trip-time
reliability while enhancing mobility. Further, the
High-Occupancy Vehicle and Toll Lanes 555
revenues from the HOV/HOT lanes in some areas are
used to support transit services, improve overall oper-
ations, and maintain the facilities.
This article provides an overview of key aspects
related to HOV/HOT facilities. The types of HOV
facilities and supporting elements are described next.
Factors to consider in planning, designing, marketing,
and enforcing HOV/HOT lanes are then described.
Case studies of the HOV/HOT lanes in Denver,
Houston, and Minneapolis are presented. The article
concludes with a discussion of future directions for
HOV/HOT lanes.
Types of HOV Lanes
There are four general types of HOV lanes or busways
in separate rights-of-way – exclusive freeway HOV
lanes, concurrent flow freeway HOV lanes, contraflow
freeway HOV lanes, and arterial HOV lanes [1–4].
Busways in separate rights-of-way are special lanes
in separate rights-of-way reserved for buses only dur-
ing all operating hours. Busways are in operation in
Pittsburgh, Miami, Minneapolis-St. Paul, Boston, and
Ottawa, Canada.
Exclusive HOV lanes are separated from general-
purpose lanes by barriers or wide painted buffers.
Exclusive lanes may be reversible. This approach is
more expensive due to separation, but easier to enforce
and safety enhanced.
Concurrent flow HOV lanes are lanes in the same
direction of travel as the general-purpose lanes sepa-
rated by paint striping and in some cases a 2–4 ft buffer.
Concurrent flow HOV lanes are the most common type
of HOV facility found in the USA.
Contraflow HOV lanes use a lane in the off-peak
direction of travel for HOVs in the peak direction of
travel. They are typically separated from the general-
purpose lanes by pylons or moveable barriers. These
types of lanes are appropriate when there is available
capacity in the off-peak direction of travel. They are
typically peak-only operation, and in some cases, only
the morning peak period. There are only five
contraflow lanes in operation: three in New York/
New Jersey, one in Boston, and one in Dallas.
HOV lanes are also in operation on some arterial
streets. Most of these are lanes reserved for buses only.
Most are also found in downtown areas or on streets
556 High-Occupancy Vehicle and Toll Lanes
approaching downtown areas. There are a few arterial
street HOV lanes open to carpools and vanpools.
Access Treatments
A number of different access alternatives may be used
with HOV/HOT lanes. These alternatives include con-
tinuous access, slip ramp and limited access, direct
access, and freeway-to-freeway HOV connections [1].
Continuous access is common with many concur-
rent flow HOV lanes. Vehicles may move into and out
of the HOV lanes at any point. This approach offers
more flexibility, but is more difficult to enforce and
may cause safety concerns.
Slip ramps or limited access may be used with
concurrent flow or exclusive lanes. This approach pro-
vides specific locations for access. This approach also
limits flexibility for users, but enhanced safety and
enforcement.
Direct access uses flyover ramps and other ramp
designs to provide direct access to HOV lanes for park-
and-ride lots and transit centers. This approach has
higher capital costs, but provides additional travel-time
savings and is easier to enforce and enhances safety.
A few areas, such as Los Angeles, have connectors
from HOV lanes on one freeway to HOV lanes on
another freeway. These connectors have high capital
costs, but can provide significant travel-time savings
to HOVs.
Allowed Vehicles and Vehicle-Occupancy
Requirements
Some HOV facilities allow only buses. These include
busways and a few freeway lanes, such as the I-495
contraflow lane in New York/New Jersey. A few HOV
lanes are restricted to only buses and vanpools. These
include the contraflow lanes on the Long Island
Expressway and The Gowanus Expressway in New
York City. Most HOV lanes allow buses, vanpools,
and carpools. SAFETEA-LU also includes requirements
for eligible vehicles, and allows operating agencies to
allow other types of exempt vehicles, including low-
emission and energy-efficient vehicles and some types
of hybrid vehicles.
There are three commonly used vehicle-occupancy
requirements with HOV lanes. These requirements
are three people per vehicle (3+), two people per
vehicle (2+), and variable requirements (2+/3+).
A three-person per vehicle (3+) requirement is used
on the Shirley Highway in northern Virginia and the
approach to the Bay Bridge in Oakland. Most HOV
lanes that allow carpools use a 2+ requirement. The
variable requirements (2+/3+) are used on the El
Monte busway in Los Angeles and US 290 HOV
lanes in Houston.
Operating Hours
Four different hours of operation commonly used with
HOV facilities. Some HOV lanes operated 24 h a day/
7 days a week (24/7). Peak-period only – typically
morning and afternoon (in the range of 6:30–9:00 a.m.
and 3:00–6:00 p.m.) – is found on many HOV lanes.
Peak-period only/peak direction, typically morning
peak-period only, is found on some HOV lanes. The
Houston HOV system is one example of the use of
extended hours with operation hours from 5:00 a.m.
to 12:00 p.m. in the inbound direction and from 2:00 to
9:00 p.m. in the outbound direction.
Supporting Facilities and Services
Supporting facilities and programs are typically found
with HOV lanes to help encourage travelers to change
their mode. Examples of supporting facilities and
services include new and expanded bus services, park-
and-ride and park-and-pool facilities, and transit
stations. Newer or expanded express bus routes are
often provided with HOV/HOT lanes. Rideshare pro-
grams, vanpool programs, and employer-based
programs are also key elements of many successful
HOV/HOT projects.
Planning
Planning for transportation improvements, including
HOV/HOT facilities, usually occurs at different levels.
A broad regional planning effort is often undertaken
first. This level focuses on the general needs, issues, and
opportunities throughout a metropolitan area. The
outcome of this process is a long-range regional plan
that identifies the general types of facilities anticipated
in the major travel corridors and areas. Regional plans
may not define the exact type of treatment or design,
however.

These more detailed analyses are conducted in the
planning process at the corridor and facility level. Plan-
ning at this stage is much more detailed, focusing on
preliminary alternative design treatments, access
options, vehicle-occupancy levels, and other issues.
For example, a regional plan may identify the need
for an HOV/HOT facility in a specific corridor, but
the type of HOV/HOT lane and the preliminary design
would be examined in later planning phases.
Planning for HOV/HOT facilities can be thought of
as a cascading or funneling process. Planning at the
regional scale focuses on identifying an overall system
plan and the general types of facilities that may be appro-
priate. Planning at the corridor and facility level is more
detailed, resulting in projects that can be taken forward
into the design stage. Different models and analytical
tools and techniques are needed at various levels. Sketch
planning techniques, travel forecasting models, and more
detailed models and methods may be used.
Design
The design of an HOV/HOT lane is similar to the design
of a general-purpose traffic lane. There are special ele-
ments that should be considered with the design of an
HOV/HOT lane, however. Elements that may influence
design considerations of an HOV/HOT lane include the
separation between the HOV/HOT lane and the general-
purpose freeway lanes, access points and the types of
access, the toll payment method and the method for
HOV declaration or observation, the number and
nature of enforcement areas, and signs and markings.
The design of HOV/HOT lanes should follow the
standards recommended by the American Association
of State Highway and Transportation Officials
(AASHTO), the Institute of Transportation Engineers
(ITE), and the Transportation Research Board (TRB).
The design for different types of HOV facilities have
been examined by these organizations and other groups,
and are documented in different reports [5–7]. In addi-
tion, Texas [8], California [9], Washington [10], and
other states have developed design guidelines.
Enforcement
Enforcement is a key component of operating HOV/
HOT lanes. The roll of an HOV/HO enforcement
High-Occupancy Vehicle and Toll Lanes 557
program is to ensure that operating requirements,
including vehicle-occupancy levels, are maintained to
protect HOV travel-time savings, to discourage
unauthorized vehicles, and to maintain a safe operating
environment. Visible and effective enforcement pro-
motes fairness and maintains the integrity of the
HOV facility to help gain acceptance of the project
among users and nonusers. Ensuring that tolls are
paid is another key element of HOV/HOTenforcement.
Key elements of successful enforcement programs are
highlighted in this section [1].
First, the appropriate agency must have the legal
authority to issue citations for possible violations. The
appropriate agency may include the state police, local
police, and transit police. Toll authorities involved in
HOT projects must also have the legal authority to
enforce toll violations.
State legislation or local laws may be needed to
ensure that the appropriate agencies and law enforce-
ment personnel have the legal authority to issue cita-
tions for various offenses, including vehicle eligibility
requirements, vehicle-occupancy requirements, oper-
ating regulations (such as entering and exiting an HOV
lane at the appropriate places), and not paying tolls. If
new user groups (tolled vehicles, low-emission and
energy-efficient vehicles) are allowed to use an HOV/
HOT lane a check should be made to be sure the legal
authority for enforcement exists. Legislative action may
be necessary if the legal authority does not exist.
Two elements of HOV enforcement are the type of
citation issued for violating the HOV requirements and
the level of the fine. Most areas classify occupancy
violations and entering and existing the HOV lanes
illegally as a moving violation. Some areas include
demerits on a driver’s record. Fines in different states
for violating occupancy requirements and other
operating procedures range from $40 to $500 for repeat
offenses. Some areas use a graduated scale, with
fines increasing for repeat offenses, and some areas
add court costs.
Four general enforcement strategies are typically
used with HOV facilities. These strategies are routine
enforcement, special enforcement, selective enforce-
ment, and peer-enforcement.
Routine enforcement maintains the normal level of
law enforcement in the corridor or freeway, with no
extra enforcement designated for the HOV lane.
558 High-Occupancy Vehicle and Toll Lanes
Routine enforcement is used with many HOV facilities
around the country during some or all operating hours.
Advantages of this approach include no additional
funding needed and usually some level of enforcement
provided. Disadvantages include the potential for a low
level of enforcement, monitoring HOV facility not
a priority, and a potential for high violation rates.
Special enforcement includes the dedication of
additional law enforcement personnel and resources
to the HOV facility. Approaches may include assigning
a patrol car or motorcycle to an HOV lane, adding extra
patrols in a corridor, or locating enforcement person-
nel along the lane. Special enforcement may be accom-
plished by reallocating existing personnel or by adding
additional personnel during key periods. Advantages
include a more visible police presence, lower violation
rates, and enhanced public perceptions. A disadvantage
is the costs associated with extra personnel and the
need for additional funding.
Selective enforcement provides extra enforcement
personnel at specific times or in response to certain
situations. These situations may include when a new
HOV facility is open, in response to high violation
rates, when major operation changes are made, or
periodically to reinforce occupancy and other require-
ments. Advantages include higher visibility, positive
public perception, and lower violation rates. Disadvan-
tages include the cost, which requires additional
funding and may become too predictable.
The final general enforcement strategy is peer-
enforcement, which involves peer-regulation by HOV
lane users and motorists in the general-purpose lanes
by providing a telephone hotline to report potential
violators. Peer-enforcement is typically used in addi-
tion to other enforcement methods – not as the only
strategy. Peer-enforcement provides a telephone num-
ber people can call to report the license plate of
a vehicle they feel is violating the HOV requirements.
Typically, a letter is sent to the vehicle owner with
information on the correct use of the HOV/HOT
lane. License plates reported more than once typically
receive more strongly worded follow-up correspon-
dence and may be monitored by enforcement person-
nel in the field. The HERO program in Seattle is the
best example of a peer-enforcement program.
Specific enforcement techniques include the use of
stationary patrols, roving patrols, team patrols,
multipurpose patrols, electronic assistance, and cita-
tions or warnings by mail. Stationary patrols are law
enforcement personnel located at dedicated enforce-
ment areas or other areas. It is important that these
areas provide a safe location for enforcement person-
nel. The dedicated enforcement area is located just
before the exit point of some HOV/HOT lanes.
Roving patrols involve the use of law enforcement
cars or motorcycles patrolling the length or a portion of
an HOV/HOT facility. A safe area to pull vehicles over
is still needed with this approach. These enforcement
vehicles may travel in the HOV lane or the adjacent
general-purpose lanes. A safe area to pull vehicles over
and stop is still needed with this approach.
Team patrols use a combination of stationary and
roving patrols. Law enforcement personnel maintain
radio contact to identify potential violators, which may
be pulled over by downstream enforcement personnel.
Multiple enforcement points may also be used.
Multipurpose patrols may provide HOV enforce-
ment as one of many duties. These extra patrols may be
located in congested corridors to assist with incident
management and other functions.
Closed-circuit television cameras (CCTV), infrared
cameras, and photographing license plates may be used
to assist in enforcing HOV lanes. To date, no technol-
ogy is commercially available to detect the number of
individuals in a vehicle. Public and institutional accep-
tance of electronic monitoring may be an issue in some
areas.
Legal authority is needed for enforcement person-
nel to issue warnings or citations for violating HOV/
HOT requirements by mail. This approach eliminates
the need to stop a vehicle. At present, no ticketing by
mail of HOV/HOT violations is in effect in the USA.
It is important that information on enforcing
HOV/HOT facilities, especially the citation and fines
for violating requirements, be communicated to
motorists and the public. It is also important that the
courts and judicial system is informed of the HOV
penalties and the importance of the facilities to help
ensure that citations are upheld.
Public Information and Marketing
Public involvement is a key element of the transporta-
tion planning and project development process,

including planning HOV/HOT lanes. The term relates
to proactively involving the public in the discussion of
transportation plans and projects. There are specific
federal and state requirements associated with public
involvement, including those contained in the National
Environmental Policy Act (NEPA) and SAFETEA-LU.
Public and stakeholder outreach is a key element in
the planning, project development, implementation,
and operation phases of HOV/HOT projects. Public
and stakeholder outreach is focused on building under-
standing and support for the project with targeted
groups. These targeted groups might include members
of the state legislature, local elected officials, business
organizations, neighborhood groups, and other orga-
nizations. Examples of possible techniques include col-
laborative task forces and advisory committees. Public
information represents another key element in the
planning, project development, implementation, and
operation phases of HOV/HOT projects. Public infor-
mation focuses on building awareness, understanding,
and support for an HOV/HOT project among travelers
and the public. Potential techniques include printed
and electronic newsletters, press releases, Internet
sites, presentations, and meetings and workshops.
Market research is frequently used in the planning
and project development stages of HOV/HOT projects.
Market research helps test potential concepts, projects,
and messages. Examples of market research techniques
include focus groups, interviews, and mail, telephone,
and Internet surveys. Examples of market research tech-
niques used with HOT projects include focus groups in
San Diego and Minneapolis and surveys of bus riders,
carpoolers, and HOT lane users in San Diego.
Promotion and marketing is used during the
implementation and operation phases of HOV/HOT
projects. These efforts and techniques focus on pro-
moting the initial and ongoing use of the HOV/HOT
lanes. Possible marketing techniques include paid tele-
vision and radio advertisements, television and radio
public service announcements, paid advertisements in
neighborhood and city or regional newspapers, the
Internet, direct mail, service centers, and billboards.
A number of elements are typically associated with
developing a marketing program for an HOV/HOT
project. First, the target markets are identified. Exam-
ples of target markets include households in the corri-
dor, commuters in the corridor, and major employers
High-Occupancy Vehicle and Toll Lanes 559
in the corridor. The marketing concepts, messages, and
branding of the HOV/HOT project may be tested
through the use of the different market research tech-
niques described earlier. Based on the results of the
market research, the branding and messages are final-
ized. The appropriate media and approaches, which
were described previously, are identified. Media and
approaches are matched with available funding and
local perceptions to develop the marketing plan to be
implemented.
Three phases are typically associated with market-
ing HOV/HOT projects. These phases are preopening,
opening/initial operation, and ongoing operation. The
focus of the marketing program during the preopening
phase is on building awareness of the requirements for
using the HOV/HOT lanes and promoting individuals
to purchase transponders. During the opening and
initial operation of an HOV/HOT project, marketing
efforts continue to focus on explaining the use require-
ments and selling transponders. The marketing effort
during this phase also promotes the use of the HOV/
HOT lane by all groups – toll paying SOVs, carpools
and vanpools, and bus riders – and converting travelers
violating the toll or vehicle-occupancy provisions into
users. Ongoing marketing efforts focus on promoting
the use of all options, as well as transponder sales.
A variety of approaches may be used to market
HOV/HOT projects. These approaches include Inter-
net sites, direct mail, regional and neighborhood
newspaper advertisements, radio and television adver-
tising, radio and television earned media or public
service announcements, and customer service centers.
Brochures, which may be distributed at meetings,
by direct mail, or at businesses or service centers, rep-
resent another marketing approach used with many
existing HOV/HOT projects.
Case Studies
Denver
The I-25 HOV lanes were opened in the mid-1990s.
Called the Downtown Express, the HOV lanes are
7 miles in length. The barrier-separated facility
includes two lanes with shoulders on each side.
Access is provided at both ends of the lane. The lanes
operate in the inbound direction toward downtown
Denver in the morning and in the outbound direction
560 High-Occupancy Vehicle and Toll Lanes
in the afternoon. A 2+ vehicle-occupancy requirement
is used.
Although carpool, vanpool, and bus use of the
HOV lanes was good, available capacity existed. Con-
sideration of expanding the eligible user groups to
include toll paying solo drivers began in the late
1990s. Enabling legislation was approved in 1999 to
allow HOT projects in the state.
Implementation of the Express Lanes included
adding ETC and new signs along the facility. Drivers
entering the Express Lane self-declare as HOV or toll
paying vehicles by using the appropriate access lane. In
June 2006, toll paying solo drivers were allowed to use
the I-25 HOV lanes. The Express Lanes use ETC, with
preset variable pricing by time-of-day. The current fees
range from a low of $.50 on Saturdays, Sundays, and
off-peak periods to a high of $3.25 during peak times.
Solo drivers must obtain a transponder and main-
tain an active account to use the Express Lanes. The
transponders can also be used on the E-470 and the
Northwest Parkway toll facilities.
The development and operation of the HOV lanes
and the Express Lanes represent the coordinated efforts
of the Colorado Department of Transportation
(CDOT) and the Regional Transit District (RTD). The
Colorado Tolling Enterprise (CTE), a part of CDOT,
assumed operating responsibility when the Express
Lanes were initiated. The Colorado State Patrol and
the E-470 Tollway Authority are responsible for the
enforcement of the I-25 Express Lanes. Video enforce-
ment is used to identify vehicles without valid toll tags.
Traffic counts taken in November 2006, 6 months
after the implementation of the HOT project, show toll
vehicles accounting for between 28% and 32% of the
vehicles using the facility. Of the total 1,043 vehicles
using the lanes in the afternoon peak hour, 287, or
24%, were tolled vehicles, and 756, or 73%, were
HOVs. In the morning peak hour, tolled vehicles
represented 32% of the total 899 vehicles, while
HOVs accounted for 68%. The travel time for buses
and carpools has remained relatively constant [11].
Houston
The Houston metropolitan region is characterized
by low-density development, typical of southwestern
cities. In response to ongoing concerns related to traffic
congestion on the freeway system, limited available
right-of-way, and air quality, an initial 9-mile
contraflow demonstration project has evolved over an
almost 30-year period into a system that encompasses
some 100 miles of HOV lanes, numerous direct access
ramps, 28 park-and-ride lots, 4 park-and-pool lots, an
extensive network of express bus service, and
a managed lanes/HOT lane project. This system pro-
vides preferential treatment to buses, vanpools, and
carpools in the major freeway corridors.
Planning, designing, operating, and enforcing the
HOV system elements has been accomplished through
the coordinated efforts of the Texas Department of Trans-
portation (TxDOT) and the Metropolitan Transit
Authority of Harris County (METRO). Recently, Harris
County and the Harris County Toll Road Authority
(HCTRA) have joined this partnership to assist with the
development and operation of the managed lanes on I-10
West. These efforts have been coordinated with the Hous-
ton-Galveston Area Council (HGAC), the metropolitan
planning organization (MPO) for the area.
The HOV lanes are primarily one-lane, reversible,
barrier-separated facilities, located in the median of six
freeways. A short two-lane, two-direction section exists
on the Northwest (US 290) Freeway. A two-way facility,
with one lane in each direction of travel, is in operation
on the Eastex (US 59) Freeway. A concurrent flow HOV
lane is in operation on the Katy Freeway, leading to the
reversible lane. The design used for the HOV lanes was
influenced by a number of factors, including limited
right-of-way in the freeway corridors, providing a safer
operating environment using barriers, and the direc-
tional splits in the corridors.
The lanes operate in the inbound direction from
5:00 a.m. to 12:00 p.m. and in the outbound direction
from 2:00 to 9:00 p.m. The lanes are closed from 12:00
to 2:00 p.m. to reverse operations and are closed to all
traffic at other times. A 2+ vehicle-occupancy require-
ment is used on all the HOV facilities, except the
Northwest Freeway. This facility has a 3+ occupancy
requirement from 6:45 to 8:00 a.m. and 5:00 to
6:00 p.m., due to congestion occurring at the 2+ level.
The Quick Ride value pricing project operates on
these two lanes, allowing participating 2+ carpools
use of the lane for a $2.00 per trip fee.
A total of 28 park-and-ride lots and four park-and-
pool lots are located in the six corridors with HOV

lanes. The larger park-and-ride lots have direct access
to the HOV lanes and transit stations with passenger
amenities. There are spaces for between 900 and 2,500
automobiles at 19 of the lots. The number of parking
spaces at lots in each corridor range from slightly over
3,000 to almost 7,500.
The park-and-ride lots have transit stations
with covered passenger waiting areas and other
amenities. Transit centers without park-and-ride lots
or with only small lots are located at strategic transfer
points.
Direct access ramps connect the major park-and-
ride lots and transit stations to the HOV lanes. These
ramps provide travel-time savings for buses using the
HOV lanes and enhance the safe operation of both the
HOV lanes and the freeways. Use of the direct access
ramps is restricted to buses, carpools, and vanpools
during operating hours. The ramps are closed during
nonoperating periods. Carpools and vanpools can
access the ramps and the HOV lanes through the lots.
The direct access ramps provide significant travel-time
savings for buses and other HOVs. The 1990 opening
of the direct access ramp linking the Northwest Free-
way Station park-and-ride lot with the Northwest
Freeway HOV lane provided travel savings of 14 min
for vehicles entering and exiting the HOV lane. Prior to
the ramp opening, HOVs had to travel local streets,
enter the freeway, and merge across the general-
purpose lanes to enter the HOV lane. Use levels
increased after the ramp opened.
METRO provides a high level of bus service in each
corridor, with frequent trips from the major park-and-
ride lots. Over-the-road coaches are operated on many
routes, as are articulated buses. Although there is no
direct evidence linking increased ridership to use of the
coaches, surveys of bus riders indicate support for their
use and support for frequent service. The HOV lanes
and express bus services are oriented primarily in
a radial direction, with downtown Houston as the
major destination. The express bus system has evolved
over the years, however, providing service to major
activity centers such as the Texas Medical Center
(TMC), Greenway Plaza, and the Post Oak/Galleria
area. More recently, reverse commute services have
been added in some corridors, taking advantage of
buses in the general-purpose lanes deadheading back
to park-and-ride lots.
High-Occupancy Vehicle and Toll Lanes 561
METRO also provides rideshare services in the
Houston area. METRO’s RideShare program includes
a number of elements to help individuals form carpools
and vanpools. Rideshare matching services are avail-
able by telephone and online through METRO’s Inter-
net site. The METROVan program helps commuters
form vanpools and provides vans for their use.
METROVan is cosponsored by HGAC, allowing
METRO to provide vanpools outside the METRO ser-
vice area. METRO’s corporate RideSponsor program
focuses on encouraging employees to commute to
work by bus, carpools, or vanpools. The program pro-
vides computerized ridematching services, vanpools,
and employer outreach. Corporate RideSponsors are
eligible for discounted bus passes for their employees.
In 2003, some 212,079 passengers used the HOV
lanes on a daily basis. Buses carried 43,225 passengers,
vanpools accounted for 2,500 riders, carpools had
74,867 occupants, and 407 motorcycles used the
lanes daily. Morning peak hour utilization levels
range from approximately 1,000 vehicles on the Katy
Freeway HOV lane to 1,551 on the Northwest Freeway
HOV lanes.
Corresponding person volumes in the morning
peak hour average between 3,424 on the Gulf Freeway
HOV lane and 4,836 on the Northwest Freeway HOV
lane. The HOV lanes account for 40% of the morning
peak hour total person movement on three of the
freeways. The AM peak hour is defined as the hour
with the highest vehicle volumes. As a result, the peak
hour may vary by HOV lane.
The HOV lanes and direct access ramps have sig-
nificantly increased METRO bus operating speeds. The
peak hour bus operating speeds have almost doubled,
from 26 to 54 mph, resulting in significant reductions
in bus schedule times. Examples of reductions in the
morning peak hour schedule time for buses from park-
and-ride lots to downtown Houston include from 45 to
24 min from the Addicks park-and-ride lot on the Katy
Freeway HOV lane, from 40 to 25 min from the
Edgebrook park-and-ride lot on the Gulf Freeway
HOV lane, and from 50 to 30 min from the Northwest
Freeway Station park-and-ride lot on the Northwest
Freeway HOV lane.
The HOV lanes provide travel-time savings for
buses, vanpools, and carpools. Morning peak hour
travel-time savings range from approximately 2 to
562 High-Occupancy Vehicle and Toll Lanes
22 min on the different HOV lanes. The Northwest
Freeway HOV lane generally provides the largest
travel-time savings of about 22 min. The Katy Freeway
HOV lane averages between 17 and 20 min, the
Northwest Freeway 14 min, and the Gulf Freeway and
Southwest Freeway between 4 and 2 min. In addition,
the HOV lanes provide more reliable trip times to
carpoolers, vanpoolers, and bus riders.
Approximately 32,293 spaces are provided at 28
park-and-ride lots associated with the HOV lanes. An
additional 1,377 spaces are located at four park-and-
pool lots. METRO buses serve the park-and-ride lots,
while the park-and-pool lots provide staging areas for
carpools and vanpools. In 2003, the overall occupancy
levels at the individual facilities ranged from about 10%
at some park-and-pool lots to 100% at well-used park-
and-ride lots. From 1980 to 1990, the number of park-
and-ride lots doubled from 10 to 20. The number of
available spaces increased from 4,070 spaces to 12,626
spaces. Use of the lots grew from 4,070 parked vehicles
to 12,626 vehicles. As of 2003, there are 28 park-and-
ride lots, with 32,293 spaces. Approximately 54% of the
available spaces are used on a daily basis [12].
The travel-time savings and the improved trip-time
reliability have influenced commuters to change from
driving alone to taking the bus, carpooling, and van-
pooling. Periodic surveys of HOV lane users show that
between 36% and 45% of current carpoolers formerly
drove alone, while 38–46% of bus riders previously
drove alone. Surveys conducted in 1988, 1989, and
1990 indicate that the opening of the HOV lanes was
very important in their decision to ride a bus for
between 54% and 76% of the bus riders using the
Houston HOV lanes. Between 22% and 39% of the
respondents also indicated that they would not be
riding the bus if the HOV lane had not been opened.
The HOV system has resulted in an increase in
available vehicle-occupancy (AVO) levels in the corri-
dors with HOV lanes. For example, the morning peak
hour AVO increased on the North Freeway from 1.28 in
1978 before the contraflow HOV lane opened to 1.41 in
1996. The morning peak hour AVO increased on the
Northwest Freeway from 1.14 in 1987 prior to the
opening of the HOV lane to 1.36 in 1996. The 1996
morning AVO for the HOV lanes ranged from 2.6 to
3.65, compared to 1.02 to 1.12 for the general-purpose
lanes.
Periodic surveys of HOV lane users and motorists
in the general-purpose lanes included questions
designed to obtain feedback on the general perception
toward the HOV lanes and support for these facilities.
Between 40% and 81% of motorists in the general-
purpose lanes on freeways with HOV facilities and
one freeway without an HOV lane have responded
positively to these surveys that the HOV facilities are
a good transportation improvement.
A demonstration, called QuickRide, which uses an
electronic toll collection system, was implemented at
the end of January 1998 on the Katy Freeway. It was
later expanded to the Northeast Freeway. Although use
levels were modest, the demonstration was successful at
allowing an additional user group to use the HOV lanes
during the 3+ restricted period. It appeared that many
enrollees view having an electronic tag as insurance for
the occasional need and opportunity to ensure a quick
trip, but cannot use the program on a regular basis.
Plans for expanding the Katy Freeway began in the
late 1990s. The 23-mile corridor carries some 280,000
vehicles a day. The existing cross section in the most
congested section from SH 6 to I-610 includes three
general-purpose lanes in each direction, the HOV lane,
and three-lane frontage roads in each direction.
A number of alternatives were examined in the
Major Investment Study (MIS), including four
special-use lanes in the freeway median. Other options
included additional general-purpose lanes and
expanding the HOV lanes. The special-use or managed
lane option emerged from this study as the preferred
alternative. The specific operation of the managed lanes
was not defined, but user groups considered including
buses, 3+ HOVs, QuickRide HOVs, vanpools, trucks,
and long-distance travelers. Other than QuickRide par-
ticipants, tolling was not considered as an operational
strategy. The cross section for the section between SH 6
and I-610 would include a three-lane frontage road and
four main lanes in each direction of travel, and the four
managed lanes.
During the environmental impact statement (EIS)
process, the Harris County Toll Road Authority
(HCTRA) expressed interest in participating in the
managed lanes portion of the project. The EIS was
modified to include tolling. Additional public involve-
ment activities were conducted, along with a more
detailed assessment of potential toll-related issues.

The constructed facilities include four general-purpose
freeway lanes, two managed lanes in each direction of
travel. HCRTA helped fund the construction of the
facility and operates the managed lanes. Carpools, van-
pools, and buses travel for free in the managed lanes,
while SOVs can use the lanes by paying toll.
Minneapolis
Two HOV/HOT lanes are currently in operation in the
Minneapolis-St. Paul metropolitan area. HOV/HOT
lanes are in operation on I-394 to the west of down-
town Minneapolis and on I-35W to the south of down-
town Minneapolis. Both facilities were initially
implemented as HOV lanes and both were expanded
to HOV/HOT lanes to maximize and enhance mobility
in the corridors.
I-394 was constructed on the alignment of TH 12,
a four-lane signalized arterial. An interim HOV lane,
called the Sane Lane, was opened in 1985 at the begin-
ning of the construction process. The Sane Lane was
implemented for a number of reasons. First, it helped
with traffic management during the almost 7-year
construction period. Second, it introduced the HOV
lane concept to travelers in the corridor and the
metropolitan area as a whole. Third, it helped build
transit ridership and carpooling for the completed
facility.
The I-394 HOV lanes are approximately 11 miles in
length and include two different designs. A 3-mile,
two-lane, barrier-separated reversible section is located
directly to the west of downtown Minneapolis. To the
west of this segment are 7 miles of concurrent flow
HOV lanes. The reversible lanes provide direct connec-
tions with three downtown parking garages, which
include bus stops and passenger waiting areas, reduced
parking fees for carpoolers, and links to the downtown
skyway pedestrian system.
The complete I-394 HOV system was opened in
1992. Although the HOV lanes were well used, averag-
ing between 900 and 1,000 vehicles in the concurrent
flow section during the peak hour, interest in consid-
ering HOT applications emerged in the early 2000s to
help maximize the use of the lanes.
In 2003, state legislation was approved allowing the
HOT project on I-394. A task force, comprised of 22
individuals appointed by the Governor, the Lieutenant
High-Occupancy Vehicle and Toll Lanes 563
Governor, and communities in the corridor, was
formed to help oversee the project.
MnPASS was undertaken to meet a number of
objectives. These objectives included increasing the
efficiency of I-394 by increasing the person and vehi-
cle-carrying capabilities of the HOV lanes, maintaining
free-flow speeds for transit and carpools in the HOV
lanes, and improving highway and transit in the corri-
dor with project revenues. Other objectives focus on
developing ETC and advanced technologies to facilitate
dynamic pricing and in-vehicle enforcement.
MnPASS implementation activities included
restriping the concurrent flow HOV lanes to change
from unlimited to limited access, installation of the
ETC and electronic enforcement systems, and market-
ing the sale of MnPASS transponders, which are avail-
able online and at the MnPASS Customer Service
Center. The MnPASS operating hours on the concur-
rent flow segment were initially expanded to 24/7. The
hours were changed back to the peak hour, peak direc-
tion after the 24/7 operation caused traffic congestion
in the off-peak travel direction.
The sale of MnPASS transponders and use of the
lanes have grown since 2005. As of September 2011,
there are 11,890 active I-394 MnPASS accounts and
15,246 transponders. Some 9,000 transponders have
been sold. Traffic counts from mid-2006 recorded
1,756 vehicles using the concurrent flow section in the
morning peak hour. HOVs accounted for 63% of the
traffic, tolled vehicles comprised 32%, and some 5%
were toll violators.
MnPASS allows solo drivers to use the HOV lanes
on I-394 for a fee. Dynamic pricing is used, with tolls
based on the level of congestion in the HOV lanes. The
base toll is $.25 and the maximum toll is $8.00.
MnPASS, which was implemented in 2005, repre-
sents the first use of tolling in the Minneapolis-St. Paul
metropolitan area. MnPASS also represents the first
HOV/HOT project in the country on concurrent flow
HOV lanes.
Drivers must have a valid MnPASS transponder
displayed on the front windshield to use the lanes
without meeting the vehicle-occupancy requirement.
MnPASS transponders can be purchased online and at
the MnPASS Customer Service Center.
MnPASS represents a partnership of the Minnesota
Department of Transportation (Mn/DOT), Metro
564 High-Occupancy Vehicle and Toll Lanes
Transit, the Minnesota State Patrol, and local
communities.
HOV lanes on I-35W South were extended to HOT
lanes between 2009 and 2010 with funding from the US
Department of Transportation Urban Partnership
Agreement (UPA) program. With the construction of
new HOV/HOT lanes through the Crosstown Com-
mons section and the opening of the Priced Dynamic
Shoulder Lane (PDSL) approaching downtown,
16 miles of HOV/HOT lanes are in operation. As of
September 2011, there were 6,896 active I-35W
MnPASS accounts and 7,836 active transponders.
There were between 2,570 and 3,209 daily MnPASS
users on I-35 W South in September 2011. Approxi-
mately 1,000 MnPASS vehicles were using the HOV/
HOT lane during the morning peak period in the
northbound lane, along with 1,784 carpools/vanpools,
and 54 buses [13].
Future Directions
HOV/HOT facilities will continue to play an important
role in major metropolitan areas in North America.
The evolution from HOVs to HOV/HOT lanes will
continue to progress to the broader managed lanes
concept. The use of intelligent transportation systems
(ITS) and other advanced technologies will allow for
additional innovative approaches for maximizing the
use of HOV/HOT facilities.
Bibliography
Primary Literature
1. Texas Transportation Institute, Parsons Brinckerhoff Quade
and Douglas, Pacific Rim Resources (1998) NCHRP Report 414,
HOV systems manual. Transportation Research Board,
National Research Council, Washington, DC
2. Institute of Transportation Engineers (1988) The effectiveness
of high-occupancy vehicle facilities. Institute of Transportation
Engineers, Washington, DC
3. Transit Cooperative Research Program (2006) TCRP Report 95,
Chapter 2 HOV facilities, traveler response to transportation
system changes. Transportation Research Board, National
Research Council, Washington, DC
4. Turnbull K (1992) An assessment of high-occupancy vehicle
facilities in North America. Texas Transportation Institute,
College Station
5. American Association of State Highway and Transportation
Officials (AASHTO) (1992) Guide for the design of high
occupancy vehicle facilities. American Association of State
Highway and Transportation Officials, Washington, DC
6. American Association of State Highway and Transportation
Officials (1989) Roadway design guide. American Association
of State Highway and Transportation Officials, Washington, DC
7. Institute of Transportation Engineers Technical Committee
5 C-11(1992) Design features of high occupancy vehicle
lanes. Institute of Transportation Engineers, Washington, DC
8. Mounce J, Stokes R (1985) Manual for planning designing and
operating transitway facilities in Texas. Publication No. FHWA/
TX-851425-2. Texas Department of Highways and Public
Transportation, Austin
9. California Department of Transportation (1991) High occu-
pancy vehicle (HOV) guidelines for planning, design and oper-
ations. California Department of Transportation, Sacramento
10. Washington State Department of Transportation, Design
Manual – Section 1050, High-Occupancy Vehicle Priority Treat-
ment. Washington State Department of Transportation,
Olympia
11. Federal Highway Administration (2007) HOV/HOT lanes.
Federal Highway Administration, Washington, DC
12. Federal Highway Administration (2003) Houston managed
lanes case study: the evolution of the Houston HOV system.
Federal Highway Administration, Washington, DC
13. Cofiroute USA (2011) MnPASS express lanes monthly status
report. Cofiroute USA, Golden Valley
Books and Reviews
Arnold E Jr (1987) Changes in travel in the Shirley highway corridor
1983–1986. Virginia Transportation Council, Charlottesville
Beihler A (1996) The Pittsburgh busways. In: Conference proceed-
ings, 8th international conference on high-occupancy vehicle
systems, Pittsburgh. Transportation Research Board.
Unpublished
Betts S, Jacobson L, Rickman T (1983) I-5 HOV lanes: three month
report. Washington State Department of Transportation,
Olympia
Bonsall J (1993) Development of the Ottawa transitway system. In:
Conference proceedings, 6th national conference on high-
occupancy vehicle systems, Ottawa, 25–25 Oct 1993. Trans-
portation Research Record 409. Transportation Research
Board, Washington, DC
Bullard D (1991) An assessment of carpool utilization of the Katy
high-occupancy vehicle lane and characteristics of Houston’s
HOV lane users and non-users. Texas Transportation Institute,
College Station
California Department of Transportation (1992) Route 55 status
information sheet
Christiansen D, Morris D (1990) The status and effectiveness of the
Houston transitway system. Texas Transportation Institute,
College Station
Crain & Associates (1978) San Bernardino expressway bus evalua-
tion of mixed-mode operations. Southern California Associa-
tion of Governments

Curitiba’s Surface Metro (1993) The urban transport industries
report, pp 53–43
Hultgren L, Kawada K, Lawrence S (1998) San Diego’s interstate 15
value pricing project [CDROM]. In: 68th ITE annual meeting,
Toronto
Institute of Transportation Engineers (1992) More for Less with
HOV, The Benefits of High-Occupancy Vehicle Projects
Klusza R (1989) Route 55 three-year status report. California
Department of Transportation, Los Angeles
Lancaster A, Lomax T (1987) Proceedings from the second national
conference on high-occupancy vehicles lanes and transitways,
Houston
McQueen J, Levinsohn D, Waksman R, Miller G (1975) The Shirley
highway express-bus-on-freeway demonstration project:
final report. U. S. Department of Transportation,
Washington, DC
McQueen J, Levinsohn D, Waksman R, Miller G (1985) The Shirley
highway express-bus-on-freeway demonstration project: final
report. U. S. Department of Transportation, Washington, DC
Metropolitan Washington Council of Governments (1998) 1997
performance of regional high-occupancy vehicle facilities on
interstate highways in the Washington region – an analysis of
person and vehicle volumes and vehicle travel times. Metro-
politan Washington Council of Governments, Washington, DC
Paiewonsky L (1998) A new approach to HOV entry requirements:
MassHighway’s 3+/limited 2+ sticker program. TRB Pre-print
981081
Pratt R, Copple J (1981) Traveler response to transportation system
changes, 2nd edn. U.S. Department of Transportation
SRF, Inc. (1987) Technical memorandum #3 case study: I-394 HOV
lane six month data for I-394 Interim HOV lane, Wayzata
SRF, Inc. (1995) I-394 HOV lane case study: final report. Minnesota
Department of Transportation, St. Paul
Stockton B, Daniels G, Hall K, Christiansen D (1997) An evaluation
of high-occupancy vehicle lanes in Texas, 1996. Texas Trans-
portation Institute, College Station
Stockton B, McFarland F, Ogden M (1998) Feasibility of priority lane
pricing of the Katy HOV lane: feasibility assessment. Texas
Transportation Institute, College Station
Sullivan E (1998) Impacts and lessons from value pricing on the
California state route 91 express lanes [CDROM]. In: 68th ITE
annual meeting, Toronto
Sullivan E, Mastako K (1997) Impact assessment for the California
route 91 variable-toll express lanes. Transportation Research
Board Preprint No. 971046, Washington, DC
Transcore (1997) I-66 congestion management program monitor-
ing project final report. Prepared for the Virginia Department
of Transportation
Turnbull K (1988) Proceedings from the 1988 national HOV facili-
ties conference, Minneapolis
Turnbull K (1990) Proceedings from the 1990 HOV facilities con-
ference, Washington, DC, 10–12 Apr 1990. Transportation
Research Board Circular 366. Transportation Research Board,
Washington, DC
High-Occupancy Vehicle and Toll Lanes 565
Turnbull K (1991) HOV Facilities – coming of age: fifth national
high-occupancy vehicle (HOV) facilities conference. Transpor-
tation Research Board Circular 382. Transportation Research
Board, Washington, DC
Turnbull K (1991) HOV Facilities – coming of age: national conference
on HOV systems, Seattle. Transportation Research Board Circular
384. Transportation Research Board, Washington, DC
Turnbull K (1992) High-occupancy project case studies: historical
trends and project experiences. Texas Transportation Insti-
tute, College Station
Turnbull K (1992) Moving into the 21st century. In: Conference
proceedings, 6th national conference on highway-occupancy
vehicle systems, Ottawa. Transportation Research Board Cir-
cular 409. Transportation Research Board, Washington, DC
Turnbull K (1992) An assessment of high-occupancy vehicle facil-
ities in North America: executive report. Texas Transportation
Institute, College Station
Turnbull K (1996) International HOV facilities. Transportation
Research Record 1360, pp 126–137
Turnbull K (1997) HOV, HOT lanes, congestion pricing in the United
States. In: Conference proceedings, 25th European transport
forum annual meeting
Turnbull K, Capelle D (1998) Development of an HOV systems
manual. National Cooperative Highway Research Program,
Washington, DC
Turnbull K, Hubbard S (1994) HOV Systems in a new light. Trans-
portation Research Board Circular 442. Transportation
Research Board, Washington, DC
Turnbull K, Hubbard S (1994) HOV Systems in a new light. In: Confer-
ence proceedings, 7th national conference on high-occupancy
vehicle systems, Los Angeles. Transportation Research Circular
442. Transportation Research Board, Washington, DC
Turnbull K, Hall K, Ringrose M (1994) High-occupancy vehicle
treatments on toll facilities. Transportation Research Record
1446, pp 14–24
Ulberg C, Farnsworth G, Etchart D, Turnbull K, Henk R, Schrank D
(1992) I-5 north high-occupancy vehicle lane 2+ occupancy
requirement demonstration evaluation. Washington State
Transportation Center, Seattle
Urbitran, Hayden-Wegman (1997) Evaluation of the high-
occupancy vehicle (HOV) lanes on the long island expressway:
January 1997 HOV lane user and non-user survey. Urbitran,
New York
Van Luven H (1995) Implementation of HOV lanes on I-270: lessons
learned. In: Proceedings – 7th national conference on high-
occupancy vehicle systems: HOV systems in a new light.
Transportation Research Record 442, Washington, DC
Virginia Department of Transportation (1997a) I-66 HOV-2 annual
report. Commonwealth of Virginia, Richmond
Virginia Department of Transportation (1997b) I-66 HOV-2 dem-
onstration project final report. Commonwealth of Virginia,
Richmond
Washington State Department of Transportation (1985) I-5 HOV
lanes: 20-month update, Olympia
566 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
HOT Lanes/Value Pricing: Planning
and Evaluation of Multiclass Service
DOUGLASS B. LEE, JR., KENT M. HYMEL
US Department of Transportation, Volpe National
Transportation Systems Center, Cambridge, MA, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Characteristics of Managed Lanes
Evaluation of Pricing Alternatives
Future Developments
Note
Bibliography
Glossary
HOT Lane A limited access (expressway) separated
highway service that permits qualifying HOVs to
travel free, but allows non-HOVs to use the service
for a fee (toll) [1]. Also HO/T.
Managed lanes Separated lanes that are operated or
controlled actively in a way to limit congestion;
charging a toll is one way to manage and an occu-
pancy requirement is another. Other management
tools include temporary shoulder use and dynamic
signing (see http://ops.fhwa.dot.gov/freewaymgmt/
managed_lanes.htm and http://managed-lanes.
tamu.edu/).
Congestion pricing A toll or user charge on a facility
that varies by time of day in accordance with
demand, so as to maintain traffic flow on the facility
at a freely flowing speed. Flat tolls that do not vary
with demand, i.e., by time of day, do not qualify as
pricing, and are not included within the scope of
managed lanes.
HOV A high-occupancy vehicle, meaning there is at
least one passenger in addition to the driver; the
definition or criterion may require a minimum of
two passengers (HOV 3+) in order to be eligible to
use the HOV reserved lane.
Reserved lane A separated dedicated lane restricted to
one or more special purposes for which not all
vehicles are eligible. Bus lanes, HOV lanes, HOT
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
lanes, truck lanes, and express lanes are forms of
reserved lanes. Heavy vehicles may not be permitted
at all on restricted lanes.
Toll Payment by a user vehicle to access a reserved
lane; the toll may vary by time of day or with the
level of demand so as to manage congestion and
may also vary by vehicle type.
Expressway A design standard for multilane divided
highways with grade-separated interchanges. HOT
lanes are installed on mostly urban expressways.
Express lane A lane that allows some traffic to go
faster than the default alternative, by restricting
entry or skipping stops. A common usage refers to
a tolled lane on an expressway, whether HOVs are
discounted or not.
Express toll lanes (ETL) Restricted lanes that are
tolled, perhaps with discounts for HOVs; facilities
that charge discounted rates for HOVs are also
referred to as hybrid systems.
General purpose lanes (GP) Lanes available for
unrestricted traffic; also mixed flow lanes.
Slug lines Persons who line up at transit stops offering
themselves as passengers to allow autos to meet the
criterion for HOV eligibility in the same corridor.
Value pricing Application of the microeconomic the-
ory of efficient pricing to the problem of highway
congestion.
ORT Open road tolling, allowing vehicles with tran-
sponders to maintain highway speed and change
lanes while the toll is collected electronically.
VOT The value of travel time, or the opportunity cost
per hour of time spent in traveling, including
congestion.
VOR The value of reliability, which is the consistency
and predictability of travel times. Both speed and
reliability are goals of managed lanes.
VMT Vehicle miles of travel.
Definition of the Subject
Economists have recognized for a long time (and trans-
portation professionals more recently) that congestion
on highways indicates that they are being operated
inefficiently. The inefficiency stems from a failure to
vary the price that drivers face with the level of demand
for vehicle travel on the congested facility. In practice,
# 2012, DOI 10.1007/978-1-4419-0851-3
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service 567

congestion pricing on a transportation facility or ser- Introduction

vice utilizes tolls that vary either by time of day, or
HOT lanes are the most popular form of road conges-
directly with demand. Although there are many
tion pricing in the USA, but the form occurs rarely
tolled highway facilities in the USA, including bridges,
outside the USA. Other ways of controlling congestion,
tunnels, and turnpikes, (see http://www.fhwa.dot.gov/
such as charging for entry to the downtown area of
ohim/tollpage.htm), few currently vary the toll by
a city, are found in other countries but not in the USA.
time of day (see http://ceprofs.tamu.edu/
Thus HOT lanes should be seen within the context of
mburris/pricing.htm). One example is the New
a range of strategies for managing congestion by vary-
Jersey Turnpike and another is the Port Authority
ing the price of road use so as to encourage traffic to
Hudson River crossings in New York City that have
avoid times and locations where demand is high. The
recently introduced a modest variation for peak
HOT lane strategy is a blending of two strategies that
periods [2].
involve offering motorists a choice: (1) the HOV strat-
Pricing one or more lanes of a facility so as to keep
egy of offering speedier service to vehicles in which
traffic on them moving smoothly, while leaving par-
there is a minimum number of occupants, and (2) the
allel lanes unpriced, is one way to increase the price
pricing strategy of setting a toll on a facility that is high
toward a more efficient level. This strategy for con-
enough to maintain free-flowing traffic during peak
gestion pricing is the dominant paradigm in the USA,
periods, and lowering the toll at other times.
and is referred to both as HOT lanes and as value
Each management strategy has its primary features
pricing.
and numerous variations in implementation and detail,
In most instances, tolled express lanes were pre-
and consequently each strategy has its own advantages
ceded by lanes that were (or would have been) reserved
and disadvantages. Although the general method
for high-occupancy vehicles (HOVs). HOV lanes were
of raising the price during the peak – congestion
created in an attempt to induce travelers to carpool by
pricing – is clearly beneficial (if the cost of imposing
offering them a faster trip in return for the willingness
the pricing is not too large), the costs and benefits of
to coordinate travel schedules and take only one vehi-
the various incomplete or partial methods of applying
cle. The goal can also be viewed as increasing passenger
congestion pricing are not so clear. For that reason, the
throughput by getting higher-occupancy vehicles to the
theory and practical methods for evaluating HOT lanes
head of the queue [3]. Although the utilization of HOV
against other strategies for congestion management
lanes is often well below capacity, reducing the criterion
that use pricing are discussed. The issues are not fully
for high occupancy by one passenger may overshoot
resolved at either the theoretical or the practical levels,
capacity by allowing too many vehicles. The concept of
despite many strong assertions about the benefits of
selling off the extra capacity to single-occupant vehicles
some of the strategies.
(SOVs) was developed in order to “fine-tune” demand
A source of the difficulty of rigorously evaluating
and more productively utilize the full capacity of the
HOT lanes is in their nature of offering two alternative
lanes. Such lanes became high-occupancy/toll lanes, or
“classes” of service for serving the same purpose, namely,
HOT lanes.
a vehicle trip from point A to point B. Users are making
“Value Pricing” is a label that was developed for
several choices simultaneously, primarily, how many
the purpose of avoiding the term “congestion pric-
persons share the same vehicle, whether to travel during
ing,” thought at the time to create a negative impres-
the peak period in the peak direction or avoid peak
sion. Subsequently, congestion pricing has been
conditions, and whether to select the faster premium
adopted as an acceptably neutral term, and value
service or the lower-priced unrestricted service.
pricing has come to mean providing separated express
Because HOT lane management is entangled with
service for an additional fee, on the same facility as the
the currently unsettled and highly contentious policy
untolled service. Technically, such a differentiated
decisions of how to manage and pay for roads, the
alternative is a multiclass service, and it is this
future evolution of HOT lanes and related congestion
multiclass attribute that is the defining characteristic
pricing schemes is uncertain at this time in the USA.
of HOT lane service.
568 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service

Characteristics of Managed Lanes Although pricing is often classified as demand man-

The discussion that follows covers three main topics:
what are HOT lanes and other “managed” lanes, how
can we tell whether they make us better off, and how
will they evolve? The first topic – description – outlines
the characteristics of managed lanes within the urban
transportation context. The second topic – evaluation
– explains the economic theory for assessing the
benefits of the HOT lane innovation and its alterna-
tives. The third topic – future developments – specu-
lates about the likely outcomes of conflicting current
trends.
Distinguishing Features of Priced Lanes
HOV lanes, HOT lanes, and express lanes have been
generically called “managed” lanes, implying that cer-
tain attributes of the service are controlled by the pro-
vider so as to affect the quality of service. The purpose
of “lane management” is to reduce or control conges-
tion on the lanes that are managed. Pricing is one
mechanism for managing highway lanes, but occu-
pancy requirements and truck-only lanes are also con-
sidered to be tools for lane management (Collier and
Goodin [4] survey the planning, design, and operation
of managed lanes, offering several case studies).
agement, express lane management is more accurately
viewed as supply management, even to the point of
constructing additional lanes. One planning study
listed HOV lanes, HOT lanes, express toll lanes, and
truck-only toll lanes, and “special use lanes such as
express, bus-only, or truck-only” as the full range of
“management tools” [5]. To date, the multiclass service
strategy for dealing with congestion has been applied
almost exclusively in the USA (see, e.g., [6] on Euro-
pean experience).
HOT lanes have been utilized in metropolitan areas
on limited access urban expressways. Table 1 lists pro-
jects currently in operation; others are under construc-
tion or planned. An express lane can be a single lane per
direction or several; two lanes per direction is common
and also the maximum in practice.
Electronic Toll Collection Electronic toll collection
(ETC) on HOT lanes is required in practice, and can be
implemented using prepaid accounts, automatic debit
accounts, or via video license plate readers [11]. The
toll paid to use the express lanes may or may not apply
to HOVs, but the price is set on the tolled vehicles so as
to maintain free-flowing conditions on the express
lanes (the SOV share gets as high as 16% in

HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Table 1 Existing HOT/Express lanes in the
USA in 2010
Location Name Year opened Length (mile) Lanes (reversible)
Orange Country, CA SR-91 Express lanes 1995 10 4
Houston, TX QuickRide (US 290) 2000 13 1R
Houston, TX Katy freeway managed lanes 2009 13 1R
Denver, CO I-25 HOT lanes 2006 7 4
San Diego, CA I-15 Fastrak 1996 8 4R
Miami, FL I-95 HOT lanes northbound 2008 21 2
Minneapolis, MN I-394 MnPASS express lanes 2005 13 2
Minneapolis, MN I-35 MnPASS express lanes 2010 13 2
Salt Lake City, UT I-15 HOT lanes 2006 38 2
Seattle, WA SR-167 HOT lanes pilot 2008 9 2
Sources: [7–10].
The number of lanes and their reversibility often changes from one section of the facility to another
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
Minneapolis [12]). Congestion on the remaining gen-
eral purpose (GP) lanes is left to fall where it may. Users
are usually informed of the reserved lane toll near
points of entry, especially when tolls are determined
by concurrent congestion levels. Implementation of
a HOT lane facility in practice is a demanding task.
There are many reports that describe the process of
planning, outreach, consensus, physical configuration,
barrier separation, access points, operational details,
etc. Two extensive guidebooks are Parsons Brinkerhoff
[13] and PB Study Team [14].
Facilitating payment by manual toll collection,
exact change, credit card, or other procedure that
requires the vehicle to come to a full stop is very costly
and disruptive. The trend is toward open-road tolling
(ORT) in which the toll is collected electronically with-
out any slowing down, and all HOT lanes currently
operate as ORT. Mixed toll collection requires that
non-ETC payers be segregated from the main travel
stream, with enough storage space provided to serve
them off-line, and then merged back into the free-
speed stream.
Users without transponders can be billed by reading
their license plates via video camera and sending a bill
to the owner of the plate. The 407 highway around
Toronto, Canada, has been primarily ETC for over
a decade, with supplemental video tolling for the
small share of traffic without transponders [15].
The HOV Lane Precursor Tolled express lanes likely
would not have occurred in the USA without the pre-
existence of HOV lanes (although the first – SR-91 –
was built directly rather than being a conversion).
Ironically, perhaps, separate lanes for high-occupancy
vehicles was devised as an incentive mechanism that
would achieve the behavior objectives of pricing with-
out its deterrent effect.
The concept of giving priority to vehicles having
more than one passenger was implemented primarily
as a means to encourage bus transit and carpoolers over
SOV (single occupant vehicles), and thus reduce the
number of vehicles needed to carry the same number of
passengers. The physical configurations to create the
separate service range from lanes marked with
a diamond but no barrier, to concrete barrier separa-
tion, to reversible lanes with barriers, queue bypass
lanes, and separated bus-only lanes [16]. Lanes may
569
be restricted only during weekday peaks or during all
hours and days.
The potential gain from providing HOV lane ser-
vice is a reduction in the number of vehicles as a result
of ridesharing. The magnitude of the benefit is greater
if existing carpools are not reduced, if new carpools are
formed from existing peak travelers, if bus travelers do
not switch to auto, and if the increase in vehicle occu-
pancy is significant. In practice, the sources of addi-
tional occupants are same family (75%), work
colleagues (20%), and casual carpooling [17]. The last
depends heavily on the physical configuration of the
corridor and the level of transit service (“slugs” come
from transit). Thus the level of actual induced
carpooling tends to be low, and not enough to over-
come the handicaps of lower capacity overall and par-
ticularly on the HOV lane (Poole [18] notes the low
effectiveness of occupancy requirements; Poole and
Orski [19] describe the historical backlash against
HOV lanes and the origins of HOT lanes; Lee [20]
describes the loss of capacity). Because not all HOV
operators provide data on utilization rates or viola-
tion rates, actual HOV performance can be difficult to
assess [21].
The number of HOV facilities (roughly 350 in the
USA) greatly exceeds the number of HOT lane exam-
ples. Some HOV lanes are over 50 miles in length (see
[22] for a full HOV inventory and details). The large
number of existing HOV lanes indicates that the pool
of potential conversions from HOV to HOT lanes is
large. Because the user cost of assembling a carpool is
a fixed cost, per trip, the incentive is strongest for long
commuting trips (over 35 miles, according to [23]).
Conditions suitable for conversion of HOV lanes to
HOT lanes are analyzed in detail in [24].
Multiclass Service A possible outcome of applying
either the microeconomic theory of efficient pricing or
of congestion management could be a strategic deci-
sion to offer two (or more) types of service whose
primary differentiating characteristic is speed. The
criteria for eligibility for the higher quality service
could be a minimum vehicle occupancy level and/or
the payment of a toll that varies with the level of
demand.
The success and benefits of such a multiclass strat-
egy depend upon heterogeneity in the values of travel
570 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
time among the group of users at the time the service is
offered. If all vehicles have the same value of travel time
(VOT), then multiclass service may not be any better
than single class service and could be worse if there are
costs in providing the service. Hence the desirability of
offering two-class service depends upon the distribu-
tion of the values of (vehicle) time in the traffic stream.
Ideally, there would be two relatively distinct groups of
vehicles, with respect to time value. The emphasis on
valuation at the vehicle rather than the person level is
that it is the vehicle that is consuming the scarce
resource, and different occupancy levels could provide
sufficient distinction among vehicles to warrant
multiclass service even if all travelers have the same
VOT (see Both Services Priced Efficiently on p. 27).
Eligibility Requirements Use of the special lane is
controlled by vehicle occupancy requirements, price,
and other user or vehicle characteristics. Usually only
one requirement (occupancy or price) needs to be
satisfied. Other characteristics (proposals have been
made to include low-emission vehicles (LEV) and
hybrids as also deserving favorable treatment) are
unrelated to congestion management but have taken
a significant share of capacity on some facilities.
The effectiveness of HOV (untolled) lanes is difficult
to sustain because fluctuations or growth in demand
necessitate changes in the required number of vehicle
occupants in order to insure free-flow conditions on the
reserved lanes. If congestion occurs frequently with
a two-person criterion for occupancy, then the criterion
can be raised to three persons. But such a discrete change
may result in the lanes being underutilized, at least for
some period of time. In general, HOV lanes have not
been shown to be effective in increasing occupancy or
reducing congestion [25], compared to the same num-
ber of lanes in GP service. So far, HOV and HOT lanes
have been the result of new construction, which inev-
itably reduces congestion below what it would have
been without the added capacity.
Pricing refers to the application of microeconomic
concepts for reconciling supply and demand, when
normal market responses (changing price and quan-
tity) are nonexistent or too rigid (other than queuing).
For a given capacity (i.e., in the short run), price can be
set so as to clear the market without resorting to queu-
ing. Microeconomic theory requires setting price equal
to marginal cost in order to achieve efficient utilization
of the scarce resource (capacity), but “management”
can be applied more broadly to achieve other appro-
priate goals such as a level of service (LOS) standard.
Performance Objectives Because the incentive for
inducing vehicles to rideshare or pay the price depends
upon a promise of faster and more reliable service,
management of multiclass services places a primary
emphasis on ensuring that the exclusive service is
always superior. HOT lanes are sustainable in the
short run and medium run because the price on the
express lanes can be adjusted in time increments as
short as 3 min, so as to maintain a consistent level of
service (LOS) (the 3 min adjustment rate is claimed for
the MnPASS express lanes in Minneapolis [12]).
The standard for level of service (LOS) can be
defined in several ways, including the ratio of volume
to capacity or the average speed for a given type of
facility. Liu et al. [9] citing Supernak et al. [26] deter-
mine that US HOT lanes are priced to achieve 45 mph
90% of the peak period (approximately LOS C), which
is congested but still flowing smoothly (see Transpor-
tation Research Board [27]). In practice, HOT lane
prices can be adjusted upward and sometimes down-
ward to maintain service standards; for example, tolls
on the SR-91 express lanes in southern California were
actually lowered in 2009 following a recession-caused
decrease in demand [28]. Whether the tolls will be
sufficient to maintain and replace the facility over the
long run is largely unknown at present because the
history of such lanes is still fairly short (see Revenues
on p. 15 below).
Additional Costs of Multiclass Service
Creating a two-class (or more) service incurs several
costs that can be avoided or lessened if the facility is left
in a single class of service.
Safety and Enforcement With separate but unequal
services, there is an incentive to take advantage of the
faster service if that can be done without paying the
price, whether the price is the inconvenience of
carpooling or in the form of a toll. The incentive is
especially strong if the GP lanes are barely moving
while the restricted lanes are underutilized. If there
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
are no barriers between the lanes, then some drivers
may be tempted to switch lanes. Because this places fast
and slow vehicles in juxtaposition, changing lanes cre-
ates a safety problem. If the violation rates are high
(15%) for a tolled service, there is an additional incen-
tive to switch lanes, and there is a revenue loss to
the owner. A study found a 15% violation rate on a
Phoenix HOV lane that was (as is common) not barrier
separated, even with a fine of $350 [29]. Finally, if
eligibility includes a minimum number of occupants
in the vehicle (or a hybrid gasoline-electric vehicle),
enforcement may be more costly because visual verifi-
cation of the occupancy or vehicle make is required
(Poole [18] reviews the technology for occupancy ver-
ification and mentions that where HOV2 are eligible
for free travel they are a majority of vehicles).
To minimize the cost of enforcement, physical sep-
arations of some kind are needed. There is thus a trade-
off between the cost of enforcement and the cost of
physical infrastructure. Whatever the optimum bal-
ance, the net result is an increase in cost over a single
undifferentiated service.
Accidents tend to be higher on HOV/HOT facili-
ties, affected by differing traffic speeds when merging
narrower lane width and shoulder width, but not
affected by continuous versus limited access [30].
Infrastructure HOV services require shoulders, tran-
sition, or merging lanes, breakdown and emergency
services, and other features that may be needed above
those required for GP service. Some HOV lanes are
simply marked with diamonds, with no buffer or bar-
riers or pylons demarcating the lanes. Vehicles can
enter or leave at any point.
Transforming HOV into HOT additionally requires
toll collection gantries and machines, and back-office
functions related to user accounts. Toll administration
may or may not impose greater costs than normal ETC.
Electronic signs and other information systems are
used to inform users of the toll rate on the express
lanes, at the time the user needs to make a choice.
Other information may also be conveyed, such as the
days or hours of operation.
Loss of Capacity In general, the more lanes moving
in the same direction, the greater the capacity per lane.
This phenomenon occurs because vehicles can weave
571
across lanes to make use of space in front of slower or
less aggressive drivers. Hence, the disruption from
entrance and exit ramps is lessened. But with barrier-
separated lanes, such opportunities for weaving are
diminished, which results in an effective loss of capac-
ity. Rough averages for capacity and speed for different
number of contiguous lanes are shown in Table 2.
Actual capacities and speeds vary substantially from
these upper-bound estimates on capacity and speed
ranges. As for usage, many single HOV lanes rarely
get above 1,000 vehicles per hour, perhaps as a result
of the uneven demand from eligible vehicles.
One or two lanes is the norm for express lanes,
whereas GP lanes or full facility pricing can apply
across all numbers of lanes. Capacities are often depen-
dent on the sources for inflow (feeder links) and recip-
ients for outflow (exits).
Cassidy et al. [31] argue that carpool lanes can
smooth traffic flows at bottlenecks, in all lanes includ-
ing GP lanes, but these conditions apply when volume
is sufficient to a cause breakdown in flow.
Imbalanced Throughput The maximum flow on the
facility occurs when all lanes are part of the same flow.
When some lanes are allowed to go faster, the total
throughput for the two services combined is less than
for the same number of lanes without separation. This
result is due to the nonlinearity of delay: The marginal
delay from adding a vehicle to a congested stream is
HOT Lanes/Value Pricing: Planning and Evaluation of
Multiclass Service. Table 2 Estimated capacity and speed
by number of lanes in the same direction [27]
Capacity
Number (vehicles per
of lanes lane per hour) Free-flow speed (mph)
1 1,600 [17] 45–55
2 1,800 58 Goodall and Smith [12]
claim average speeds as
high as 78 mph on a two-
lane HOT lane service in
Minneapolis
3 2,050 65
4 or 2,200 70
more
572 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service

always greater than the marginal reduction in delay
from taking the vehicle from a less congestion stream.
While popular intuition is perhaps as often wrong as
right, the effect of unbalanced flows per lane is per-
ceived as making throughput worse [32].
If the objective is to minimize overall vehicle delay,
the optimal result occurs when traffic is distributed
equally across all lanes; at that point, the reassignment
of any vehicle to another lane increases delay more on
that lane than it reduces delay on the lane it left. This is
necessarily the case because the second derivative of the
delay curve is always positive. As an example, the
speed–volume relationship can be represented by
a curve of the BPR form
Fs
SðvÞ ¼
g
1 þ a vc
where S(v) = speed as a function of volume, Fs = free-
flow speed, c = capacity, and a and g are parameters
[33]. Travel time is the reciprocal of speed.
This function is used in the illustrative example
shown in Fig. 1, with a = 0.3 and g = 7.0. Consider
a two-lane facility with a fixed vehicle flow (one
direction), where the problem is how to allocate traffic
to each lane to minimize total travel time. Proceeding
from left to right in the diagram, as more traffic is
added to lane 1 the total time cost declines because
the extreme congestion in lane 2 is relieved without
heavily impacting lane 1. The lowest curve shows vehi-
cle delay for all vehicles combined. The minimum is
reached when half the traffic is assigned to each lane.
Converting time into cost, through vehicle occu-
pancy and the individual value of travel time (VOT),
produces the thick solid curve that replicates the time
curve (but higher on the graph) so long as both occu-
pancy and VOT are the same for all vehicles.
If the number of occupants and their VOT are not
uniform, lower travel cost may be obtained by sorting
users into categories of higher occupancy and higher
time value, rewarding the latter with faster travel times.
Taking the numbers shown in Table 3, and assuming
that the occupancy and time values are determined by
the lane to which the vehicle is assigned, the time cost
by the share assigned to lane 1 is shown by the heavier
dashed magenta curve. VOT per person is the same for
all travelers, but occupancy is twice as high on lane 2 as

4,000 20
occcupancy
variable with lane 18
3,500 uniform occupancy
allocation
all vehicles
16
3,000 total travel time cost (000 minutes)
total time cost of travel ($)

14

2,500 50% 56% 12

2,000 10

predetermined 56% 8
1,500
low occupancy
6
1,000
vehicle travel time 4
only (right scale)
500
2

0 –
20% 30% 40% 50% 60% 70% 80%
share of traffic in regular lane

HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 1

Total travel time cost as a function of traffic share allocated to lane 1
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service 573

on lane 1. The optimum (least total time cost) occurs means assigning high-occupancy vehicles to the slow
when about 56% of the traffic is on lane one, and the lane (more people at slower speed), so total cost rises
total cost is slightly lower than the average in part rapidly. To the left of 56%, there are fewer people in the
because higher VOT vehicles go faster and in part constant-occupancy scenario than the variable occu-
because there are fewer passengers (56% of the vehicles pancy scenario, so costs are a little lower.
have 1.2 passengers versus 50%). This curve crosses the Changing VOT has the same effect as occupancy,
uniform-traveler (dotted) curve at 50% because at that which is to make the time of some vehicles higher
point they are the same. valued than others. For a difference of a factor of 2.0
If the number of vehicles in each category is deter- in the vehicle value of time, the least-cost flow still only
mined a priori, such as in the right-hand column of varies from the equally balanced flow by less than five
Table 4, in this case at approximately 50%, the cost is percentage points.
shown by the heavy blue curve, which has a minimum The Los Angeles County MTA has announced the
between 50% and 56%. At the latter allocation, it is the intention of converting HOV lanes to HOT without
same as the previous curve in which occupancy reducing service on the GP lanes, which should not be
depended upon the lane to which the vehicle was a problem; the problem occurs if a GP lane is converted
assigned. Again, the parameters are the same at that directly to HOT. So long as the HOV lane is a new lane,
point. Assigning more than 56% to lane 1, however, the political backlash is mild compared to taking an
existing lane out of service. When the same capacity is
split into classes, flow in the GP lanes may exceed the
HOT Lanes/Value Pricing: Planning and Evaluation of breakdown level, resulting in “hypercongestion,”
Multiclass Service. Table 3 Assumed average vehicle a condition in which capacity is less than what the
occupancy (AVO) for regular GP and HOT lanes facility is capable of under ideal conditions. The extent
and magnitude of this phenomenon is open to ques-
No. tion [34], but at a network level an overloaded system
Vehicles Occupancy VOT veh.
with unbalanced flows appears to be less productive
Low occupancy/regular 1.2 $15.00 1,900 than one that is not unbalanced [35].
High occupancy/HOT 2.4 $15.00 1,500
Average (unweighted)/ 1.8 $15.00 3,400 Utilization Despite the pricing, HOT lanes may still
total be underutilized, if there is general resistance to paying
tolls. Perhaps unusual, the HOT lanes in the Seattle
area appear to be underutilized, despite congestion on
the GP lanes. The data in Table 4, inferred from volume
and occupancy detectors, indicate difficulty in achiev-
HOT Lanes/Value Pricing: Planning and Evaluation of ing full utilization of all lanes.
Multiclass Service. Table 4 Seattle SR-167 single lane Time savings on SR-167 averaged 9.1 min over 11
speed and volume (Lin et al. [9]) miles, but can get as high as 20 min. The express lanes
consistently operate over 45 mph, with a mean travel
Flow
Segment % Peak flow % Flow >1,050 vpl <1,050 time of 8.5 min and a standard deviation of 26 sec. The
No. <800 vph %Speed >55 mph vpl HOT service is relatively new, and demand is continu-
1 82 1 99
ing to grow.
Currently, the existing ten US HOT lane examples
2 31 7.2 92
are able to effectively manage the level of service on the
3 9 10 77.4 HOT lanes by changing prices dynamically or by mod-
4 21 11.1 87 ifying fixed prices based on observation of traffic flows
5 8 11.2 87.7 and patterns. Diagnostic guidance on obtaining close
to optimum utilization in HOV or HOT lanes is pro-
6 30 7.5 91.4
vided in [36].
574 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service

Demand Revealed Preference on Toll Lane Users Aggregate

demand tends to grow, due to rising income, popula-
It is well known that express lane users do not always
tion, and economic activity. Liu et al. [9] observe that
choose to pay for the faster service whenever they use
a higher-than-typical price at a given time may signal to
the facility [37]. A large share of users pay the toll two
users that the express lanes may be approaching con-
or three times a week and use the GP lanes on other
gestion due to high overall demand, hence an indica-
days, making the choice to use the express lanes trip by
tion that the GP lanes are slow. For SR-167 in Seattle,
trip. Hence the value of saving travel time varies con-
a toll of $1–2.00 for a speed increase from 30 to 60 mph
siderably for the same individuals from one day to
over 11 miles yields a time saving of around 11 min, for
another, as well as among travelers.
an implied vehicle value of time (VOT) of less than $10
Although peak travelers moving in the peak direction
per hour. The toll rate is adjusted every 5 min. A good
on the same facility are relatively affluent compared to
deal of data has been collected on user behavior regard-
the general population, they are far from identical. Occu-
ing SR-91, including interview surveys, and these data
pancy, of course, also makes a difference in the vehicle’s
have been extensively studied [41].
willingness to pay for faster speed.
Some demand studies exist for predicting the factors
Diversion For HOT lanes, most of the diversion is
that determine which class of service a driver will select
between the two classes of service. Because the lanes
on the basis of traveler attributes, but less work exists
have resulted in additional capacity when the multiclass
that examines how variations in service attributes affect
service was initiated, there has not been much need for
lane choice. Most new HOT lane users were previously
concern about diversion to other facilities.
SOVs in the GP lanes on the same facility (or HOV2s), at
If, however, existing lanes are converted to express
least in the case of conversions from HOV to HOT
or HOT lanes, the overall price on the facility will rise
service [38]. Relative to other peak travelers, express
(in dollar terms for the tolled lanes, and in delay on the
lane users have significantly higher income and educa-
GP lanes), so diversion to other facilities will need
tion than the average, even of other peak highway users.
evaluation. Depending upon the substitutability of
The small share of lower-income toll payers is likely to
nearby arterial streets – measured in travel time – tolls
be in carpools. For the most part, carpoolers use the
may need to be lowered on the express lanes in order to
express lanes because they are already carpooling; few
retain enough traffic to prevent arterials from being
carpools are formed in order to use the lanes. While
choked.
there are no surprises here, it may be reassuring that
This problem is different in kind from the consid-
HOT lanes do not cannibalize carpools or transit.
eration of delay on the GP lanes when tolls are raised on
the tolled lanes. If the facility manager is responsible for
Market Research on Express Lane Users The ameliorating delay on the GP lanes, the level of service
demand model may be approximated as this: Users on the express lanes may need to be compromised to
estimate the magnitude of the time savings and decide retain reasonable flow on the GP lanes [42, 43]. This is
whether the price is right, without changing mode or how the I-95 Express Lanes in Miami are managed,
occupancy. Other reasons as well, however, factor in. with the project manager coining the term “quality
Drivers in Minneapolis use the HOT lanes even when throughput.” Pricing both services, of course, would
there is no time advantage, perhaps by (previous) habit allow for much greater control and less compromise.
or perhaps to ensure reliability [12].
Zmud et al. [39] find toll willingness to pay is Variations in the Value of Travel Time Savings Var-
affected by income, age, trip purpose, time of day, iation in time values is widely asserted, largely on
trip distance, and time savings (time of day is, of circumstantial grounds (e.g., toll payers vary from day
course, a surrogate for demand level). to day in their choice). Patil et al. [44] use stated
Burris et al. [40] find that tolling HOVs causes preference (SP) to find a 95% confidence range of 41–
a small but significant reduction in their numbers, 79% of the wage rate, a fairly broad band. Small,
while increasing total revenue. Winston, and Yan [45] use both stated preference and
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
revealed preference data to determine that motorists
place high value on travel time and travel time reliabil-
ity, with considerable heterogeneity in both.
Travel Time Reliability Express lanes offer signifi-
cantly more reliable travel times, as well as shorter
travel times. The general pattern from user surveys
and potential user surveys is that most travelers are
occasional users who decided each day how urgent
their arrival time is [46]. Using data from SR-91 in
southern California, Brownstone and Small [47] find
high values for both time and reliability. Defining and
measuring reliability is a challenging task, and the real
or perceived benefits have not been integrated into
benefit-cost evaluations [48, 49]. Patil et al. [50] sur-
veyed travelers on the Katy Freeway in Houston regard-
ing their travel on the GPLs and the managed lanes.
One area of investigation was the travelers’ willingness
to pay for a trip on the managed lanes under unusual
situations – basically when the traveler was in more of
a hurry or pressed for time, for example, running late
for a meeting or when they see bad weather that may
cause congestion. They found the willingness to pay for
travel time savings in those instances was between two
and four times as much as regular travel time savings.
Evidence from travelers using the express lanes on
I-394 in Minnesota also points to express lane users
placing a high premium on travel time reliability. Cho
et al. [51] obtained data on travel times for the GPLs
and ELs plus toll collection data from every transaction
in 2008. Based on these data it was found that approx-
imately 50% of the paying EL users paid for travel time
savings of 1 min or less. Mostly due to these small time
savings, the average values of time were found to be
considerably higher (ranging from $73 to $116 per
hour) than what was found in the literature (often
around $10–25 per hour). This would indicate
a willingness to pay for something other than the travel
time savings offered by the lane.
Revenues
A policy that both reduces congestion and raises reve-
nues should be an attractive policy, but that has not
made congestion pricing in general widely adopted.
While some commentators express the observation
that the revenue potential seems to have overshadowed
575
the congestion management benefits, the evolution of
the HOT lanes concept from HOV to allowing a few
more vehicles to “buy in” to the service does not sug-
gest that revenue raising is the primary motivation.
Nonetheless, tolls do generate revenue, and more
under some circumstances than others. Tolls used for
congestion management vary with demand and hence
by time of day. Flat tolls that do not vary with time of
day are primarily if not exclusively for revenue gener-
ation to pay for the road. Such tolls may vary by vehicle
type to reflect costs of serving heavier or larger vehicles,
but the rates are generally not set so as to obtain
a particular mix of vehicles.
Monopoly The monopolist in microeconomic theory
faces a downward sloping demand curve (unlike a firm in
a competitive market) and sets prices that are intended to
maximize profit. The monopolist does not necessarily
have full control over the market price and does not
face totally inelastic demand, but any degree of downward
sloping demand for the firm’s output conveys some mea-
sure of monopoly power. In theory, the monopolist max-
imizes profit by setting price on the demand curve at the
quantity where marginal cost equals marginal revenue.
If profit is positive at any volume (it might not be,
depending upon the cost structure), the MC = MR
point will yield the highest profit. More generally, the
monopolist may not choose to seek the absolute max-
imum profit in the short run, perhaps lowering the
price to increase market share or provide a cushion of
loyal customers. Express lanes are typically priced so as
to reliably maintain 45 mph speeds during the peak
periods, which is normally a higher volume than the
profit maximizing volume.
The degree of monopoly power possessed by toll
roads or express lanes is controversial. There is inevi-
tably some spatial monopoly, but given the ability of
travelers to carpool, switch to transit, take an alterna-
tive route or destination, shift to an off-peak time, or
forgo the trip, the demand for any single facility is
necessarily somewhat elastic.
Exemptions A much more important factor with
respect to revenue is the level of exemptions from
paying the toll. If two-person carpools go free, it is
possible to use all the capacity for HOVs and leave no
paying customers. Adding exemptions for electric
576 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
vehicles and hybrids further erodes the revenue yield.
Broadly speaking, as the share of nonpaying vehicles
increases, the revenue production of the facility becomes
a less viable purpose for the express service [52].
The significance of revenue, then, is whether it can
pay for the operation of the facility and the collection of
the revenue. Varying the toll according to demand and
billing each vehicle separately is more costly than
collecting the same revenue through a gas tax, so the
question of whether the service is a good idea depends
upon the benefits of better congestion management in
travel time savings, reliability, and providing the option
to avoid the GP lanes, compared to the costs of creating
and maintaining a more accurate pricing system.
HOV discounts or exemptions are problematic
because compliance is hard to verify. This is true for
enforcement of all HOV lanes, but the addition of
tolling certain vehicles while others go free greatly
complicates the task.
Price Volatility
Express lanes are those which charge all users a toll,
perhaps discounted for HOVs, while HOT lanes allow
vehicles meeting the HOV criterion to travel free. SR-
91 charges HOVs half price. A problem with exempt
or discounted vehicles is that it creates uncertainty
about the potential revenue stream and amplifies any
volatility in the price per trip over the time of day or
day of the week. As the share of traffic that is insen-
sitive to price (free or discounted) becomes larger, the
remaining price-sensitive traffic has to respond pro-
portionately more to maintain the congestion at LOS
C. At the limit, when the share of price-sensitive traffic
approaches zero, the fluctuations in price needed to
maintain LOS C become infinite. Poole [53] describes
an Atlanta planning study in which the HOT facility
would generate no revenues in the peak period at some
time in the future, because the capacity would be fully
used by HOV2+ vehicles and paying vehicles would be
entirely excluded. Under such circumstances, the
investment of private capital is unlikely without
explicit guarantees. The requirements for exemption
can be increased (e.g., from 2+ to 3+), causing
a sudden drop in the toll for SOVs (and now two-
person carpools), but this volatility does not provide
confidence in the long-term stability of the operation.
Another possible source of volatility is the
misperception by some express lane users that the
price is a measure of congestion on the GP lanes, rather
than a price based on the number of express lane users.
Miami-area HOT lane users, among others, have
expressed the belief that the price is an index of con-
gestion on the free lanes [54]. While the perception has
a perverse effect on the use of price as a rationing
instrument, it indicates that express lane customers
are willing to pay a premium for guaranteed higher
quality, or more specifically, reliability. Posting the time
savings between the two services could serve as
a counter to the misperception.
Cost Recovery Because no revenues are collected
other than fuel excise taxes, HOV lanes are not held
to any financial recovery standard.
If express lanes are to be regarded as a means for
financing transportation infrastructure, the tolls will
need to be higher than currently, apply to all vehicles,
and apply to a large share of daily traffic [55]. There
may be some circumstances where the surplus revenue
after paying for operation and toll collection is large
enough to recover the capital cost of the lane and
associated infrastructure (barriers, gantries, buffers).
It seems unlikely that express lane revenues will ever
be sufficient to pay for the capital costs of the GP lanes
as well as the tolled lanes.
Hence it is misleading to propose using the surplus
revenues to pay for transit or equity-compensating
income transfers when the gross revenue is insufficient
to cover the full costs of the highway [56]. This makes
the issue of exemptions (as well as multiclass service
itself) an important factor in transportation financing,
above and beyond the short-run economic efficiency
impacts.
With the entire road priced – whether single- or
multiclass – the revenues from efficient pricing can
more than cover the long-run capital costs of the facility,
and the surplus applied to capacity expansion of the
highway, improved transit service, or equity transfers.
Dividing revenues in thirds to highway capital, transit
service, and equity transfers is referred to as the Small
rule (Small [57]), as applied to whole-facility pricing.
Capacity expansion of the highway will lower the conges-
tion toll rates, by reducing congestion, and the decision to
expand should be based on estimated benefits and costs,
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
not on the amount of revenue, although toll revenues are
a good indicator of WTP for highway investment.
Some Variations in Configuration and Operation
Numerous variations in design and operation have
been tried, for purposes of making a restricted class of
service work successfully, which is to say the service is
neither underutilized nor overutilized, and customers
are not complaining unduly. A few of the most notable
variations are described below.
Multiclass Networks If the concept of multiclass ser-
vice is expanded to the metropolitan area, two broad
strategies arise:
1. All potentially congestible expressways offer HOT
lanes along with GP lanes, and separation of the
services is maintained at interchanges with other
expressways.
2. Some portions of the expressway network are fully
priced and others are not, in a way that allows users
to reach the same destinations via tolled or untolled
paths.
The first strategy requires substantial infrastructure
to permit all choices of direction to be continuously
separated (i.e., express lane users never have to sit in
GP lane congestion). Potentially, the costs of such infra-
structure may be sufficiently large that they might out-
weigh the benefits from reduced congestion, especially
in places that are not highly congested. The second
strategy avoids many of the problems of multiclass ser-
vice, including the above infrastructure costs, but differ-
ent locations would be differentially served. Motorists
would still have a range of choice, but not as much as
when all expressways offer both GP and express services.
Shoulder Used for Travel Lane During Peak By
incorporating the shoulder as a travel lane on
a temporary basis during peak periods, an increase in
capacity can be achieved at low capital costs and an
unknown increase in safety costs.
A complementary strategy is to restripe the road for
a larger number of narrower lanes. The full 12-ft width
is not required for speeds below 50 mph, so more lanes
can be squeezed into the same roadway width [58].
Speed limits need to be lower, which actually helps to
577
smooth traffic flows during peak flows but is awkward
during off-peak periods.
Discounts for HOVs Most HOT lanes in the USA
were previously lanes restricted to high-occupancy
vehicles (HOVs), and the tolling strategy was intended
to utilize the excess capacity resulting from the discrete
nature of carpool lane eligibility criteria (two- vs three-
passenger minimums). In these cases, the tendency is to
permit the HOVs to continue to use the lanes free or at
a discount.
The effectiveness of HOV segregation as a stimulus
for reducing peak vehicle trips through higher vehicle
occupancy varies with the context. Prearranged
carpooling has been in secular decline for several
decades, but casual or opportunistic or unscheduled
carpooling (e.g., slug lines) have spontaneously
appeared when the particular conditions are suitable.
Because the HOV concept originated as a substitute for
pricing, it could be argued that keeping the HOV
preference is both redundant and a move away from
efficiency by overencouraging ridesharing. The addi-
tion of the ability to pay to use the lanes can also reduce
demand for casual carpools. The number of casual
carpoolers decreased dramatically on the Katy Freeway
once the carpool lanes were opened to paying SOVs.
This may be due both to much better travel along the
freeway (high time-value drivers no longer need occu-
pants to gain faster travel) and to passengers switching
to toll-paying SOV mode (while there is no increased
incentive to attract new volunteer passengers).
Number of Priced Lanes
The greatest penalty for reconfiguring an expressway
for multiclass service occurs when the priced service
consists of a single lane, because of the limited capacity
of a single segregated lane (see Loss of Capacity above
on p. 10). Two lanes are better because they allow for
some weaving and therefore higher flows per lane. This
suggests that a minimum of four or five lanes in the
same direction is a requirement for creating efficient
multiclass service.
Evaluation of Pricing Alternatives
Although there has been a great deal of writing, both
theoretical and empirically based, on the performance
578 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
of HOT lanes, no single analysis has taken in the full
welfare consequences of HOT lanes in comparison with
the relevant alternatives. Lee [20] urges an evaluation
framework based on welfare measures for applicable
alternative uses of the same capacity. Sullivan and
Burris [59] made pioneering efforts to differentiate
average VOT for toll payers versus non-payers. Light
[60] has constructed an abstract framework that
addresses the optimal price of one service that is priced,
the optimal capacity with an explicit distribution of
travel times and no exemptions, and no demand elas-
ticity; he uses a modification of the “two route prob-
lem.” Agency evaluations tend to focus on utilization
and revenue, and seldom attempt to perform benefit-
cost analysis [61].
Counterfactual
Many studies that assert benefits of HOV or priced
lanes are either vague about what they are being com-
pared to or pick a counterfactual that may not be the
economically efficient use of the infrastructure. HOV
lanes are judged by the difference in speed between GP
lanes and HOV lanes, rather than the change in travel
speeds for all users resulting from the dedication of one
or more lanes to restricted usage (see for example [14]
with respect to HOV lanes).
Within the context of managed lanes and highway
pricing, there is a range of alternatives that needs to be
made explicit.
Single-class
service Multiclass service
No tolls All lanes HOV lanes separated
merged into GP
Express lanes All lanes pay Express lanes pay toll
tolled the same toll
All lanes tolled Both services priced
but with different toll
rates
Efficient Congestion Pricing
To evaluate the performance of priced lanes, it is nec-
essary to review the basic theory of road pricing [62].
Without a compensating adjustment in price, the
number of vehicles using a facility when congestion
arises is greater than the number that would occur
under economic efficiency. The efficiency condition is
not met because some number of users value the travel
(to themselves) at less than the incremental cost in
delay time that they impose on the traffic stream as
a whole. The reason they choose to travel anyway is
because the user faces the average travel time cost rather
than the marginal social time cost. The marginal is
always larger than the average because the average
rises with additional traffic.
Thus, without correction, the price to users on
congested facilities is always lower than the efficient
price. The efficient price does not eliminate all conges-
tion, but it should be sufficiently high to permit
free-flowing traffic (not necessarily free-flow speed).
Maximum capacity is often defined [27] as a density
of 45 vehicles per lane per mile. The efficient volume
will be somewhere between the maximum flow and
the volume that results in free-flow speed.
Any increase in price from the unpriced situation
will result in a gain in efficiency, up to the point where
traffic speed is higher than what the users would be
willing to pay for to deter enough vehicles so that the
speed can be maintained. Express lane managers typi-
cally seek a minimum speed of 45 mph during the peak
period, which is probably close to efficient, depending
upon the VOTs of the vehicles in the stream.
The Standard Road Pricing Model The basic theory
of efficient pricing of roads applies to a single distinct
segment of road or a bridge, with a given demand
for vehicle trips (in vehicles per hour). Users bear
a group of variable costs, the most significant of
which is travel time. The facility has a flow capacity,
depending upon the number lanes, curves and grades,
and other attributes, measured in passenger car equiv-
alents per hour. As vehicle flow increases and begins to
approach capacity, travel speed declines at an acceler-
ating rate, thereby increasing travel time (the reciprocal
of speed). The relationships in their most simplified
form are displayed in Fig. 2. The vertical axis is the
generalized cost or generalized price, in which all var-
iable costs including travel time are translated into
dollars [63–65].
It is assumed that the user pays vehicle operating
costs and accident risk, and correctly anticipates the
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
generalized
cost
p1
toll
p0
traffic volume
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 2
Basic highway congestion tolling
average travel time during the period the user is on the
facility. There are no user charges or excise taxes, at least
of a nature that vary with the amount of use, or else such
fees exactly pay for the variable costs of maintaining the
road for the particular user (pavement wear, law enforce-
ment). There are no external costs that the user fails to
pay for (e.g., pollution), and no external benefits outside
of the demand curve. Fixed costs of the highway and the
vehicle are not explicitly included in this analysis.
The result of these simplifications is that the price
to the user is the same as average variable cost (AVC),
whereas the true cost to society of the marginal vehicle
is the marginal cost (MC), described heuristically as the
AVC plus the delay created for others by the presence of
the marginal vehicle. In that sense, each vehicle creates
a negative externality in the form of delay to other
vehicles. Usage of the facility occurs at an equilibrium
volume higher than that which would maximize net
benefits to society – that being where price = MC –
from the availability of the facility.
The “overusage” is the volume between Vinefficient
and Vefficient, which results in an increase in costs
represented by the area under the marginal cost curve
between the two volumes, while the incremental benefit
of this traffic is represented by the area under the
579
MC
AVC = user price
demand
vefficient vinefficient
demand curve (which is the same as the marginal social
benefits curve) between the two volumes. The “toll” is
a monetary charge levied on the vehicle, calculated to
make the price to the user (per mile or per trip) equal to
the marginal cost at the efficient equilibrium.
This inefficiency, indicated as a shaded triangular-
like area in Fig. 3, consists of two components: an
increase in consumer surplus (a positive benefit) from
the value placed on the travel by the additional vehicles,
above what they actually pay, and an increase in delay
for “old” vehicles (those that would be on the facility if
it were efficiently priced). In congested conditions the
delay greatly exceeds the consumer surplus, the differ-
ence being the area of the inefficiency triangle (also
called welfare loss). Mathematically,
Inefficiency ¼ delay savings  DCS
where the “inefficiency” is a loss or net disbenefit and
DCS = incremental consumer surplus. The data needed
to calculate these benefits consist of observed traffic
volumes and speeds, estimated delay functions, other
variable costs, and the generalized price to the user
[66]. Throughput by service and time savings are not
sufficient for evaluation, although they may be useful
performance indicators. Kidd [67] compares
580 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
generalized
cost
toll
delay savings
traffic volume
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 3
Benefits of congestion pricing
numerous configurations of HOV, HOT, BRT, and light
rail modes on performance measures emphasizing
throughput and speed, and recommends HOT lanes,
in part for their potential BRT use. Poole [68] also
emphasizes this combination.
The standard model also does not represent the
possibility for hypercongestion, where capacity is lost
by overloading the facility beyond its ideal maximum
capacity. Peak pricing can prevent this from happening,
even if the toll is well below the efficient price.
Dynamic Demand Schedule
The above assumes a given static demand curve. If
demand was always at the same level, the toll would
be constant. The nature of the congestion problem is
that demand is not constant, but instead has peaks and
valleys. A benefit of the congestion toll then becomes
the shifting of some users from peak periods into off-
peak periods during which the price (toll) is lower or
zero. In addition to the previously available user
options of higher occupancy (carpool, bus transit),
alternative mode (e.g., rail freight or transit), and for-
going the trip, the user can shift the time of the trip to
MC
efficiency
gain
ΔCS AC = user price
demand
vefficient vinefficient
an off-peak period. The more options available to the
user, generally the higher (in absolute value) is the
demand price elasticity, and hence the smaller the toll
required to achieve an efficient equilibrium.
With multiple demand periods, there is an
interdependency among them. Specifically, if the price
is raised in any period (e.g., the peak), then the demand
curve will shift outward in the other periods (e.g., the
off-peak). This effect is shown in Fig. 4. An increase in
peak price from pP1 to pP2 shifts the whole demand
curve in the off-peak market from DOP1 to DOP2; cor-
respondingly, a price decrease causes demand in the
related market to shift inward. The magnitude of the
shift is measured as a cross-elasticity.
Peak versus off-peak periods illustrate the case of
markets too closely related to be treated as separate
markets. A similar situation of markets for close sub-
stitutes arises with multiclass service. There are a number
of elasticities and cross-elasticities that result from chang-
ing prices in any one of the several closely related markets.
New users will be attracted in each of the markets, at
higher or lower occupancy levels, and others may be
deterred from the same and other markets. For some
simple purposes, the numbers of travelers, vehicles in
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
price, peak market
cost
pP2
pP1
pOP2
pOP1
Dpeak
traffic volume vP2 vP1
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 4
Peak and off-peak demand periods
the corridor, number of carpools, etc., can be approxi-
mated as remaining constant, but in general they will
change. For example, when HOT lanes constructed and
operated by private concessionaires promised guaranteed
speed at all times, bus service was planned to use the
expressways; previously the buses avoided the freeways
due to unreliability of schedules [69]. How to value such
changes makes evaluation of multiclass service difficult.
Nonoptimal Pricing
The standard model makes the convenient assumption
that average cost and user price are the same, but in
practice this only applies to travel time; for other costs
(emissions, vehicle operation) and user charges (fuel
taxes, flat tolls), any correspondence between cost and
price is coincidental. Because for HOT lane policy it is
necessary to analyze tolls separately from costs, a more
realistic representation is needed. A general framework
for evaluating transportation investment and pricing
that recognizes the price function as a separate factor is
shown in Fig. 5. In this case, price is shown as lying
below AVC. Because both price and AVC include travel
time, the two curves tend to have the same shape. In the
example, the assumption is that external costs in the
form of noise and air pollution emissions, plus agency
costs for maintenance and administration, outweigh
the variable user taxes paid by the vehicle. The intersec-
tion of the price function with the demand curve deter-
mines the initial volume v1. The average cost (including
social costs) at that volume is marked as ac1.
581
offpeak market
DOP2
DOP1
vOP1 vOP2
In reality, the price function might lie either above or
below the AVC curve.
The user impact components from Fig. 3 still apply,
but they are a little more complicated. In this example,
the context is simply two different prices, neither of
which is efficient. Although the optimum price is not
achieved by the toll, there is an efficiency improvement
from reducing volume from v1 to v2. The efficiency gain
(reduction in the efficiency loss) now consists of three
parts:
1. A reduction in excess delay (and operating costs to
the extent they increase with congestion), indicated
by the shaded rectangle
2. A loss of consumer surplus (not marked in the
diagram), which is still a triangular area lying
under the demand curve between v2 and v1
3. A reduction in net external costs that are above the
price curve, represented by a rectangle (not
marked) whose height is the difference between
price and AVC at v1 and whose length is v1 minus v2
Emissions changes can be estimated by modeling
the conditions, which translate into speed change or
“drive cycles” that alter the power load on the engine.
Congested drive cycles are much different from con-
stant speed cycles, and patterns on tolled lanes are
different from GP lanes, but the differences and mag-
nitudes are relatively small compared to time values.
Estimating changes in consumer surplus (including
the value of delay) is more difficult with multiclass
service.
582 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service

MC
price,
cost

efficiency gain
from pricing

AVC
delay and operating
cost savings
p2
ac1 user price
toll function
ac2

demand

traffic volume v2 v1

HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 5

Benefits of pricing when price is not efficient and not equal to AVC

Network Effects the price changes (whether positive or negative) is

the sum of all changes in consumer surplus. Poole
The congestion pricing model in Fig. 2 assumes implic-
[68] has stated the welfare objective in the following
itly that all facilities are optimally priced. The demand
way: “. . . a managed lane that charges all customers
interactions considered so far have been between peak
the market value of their trips (time plus reliability)
demand and off-peak demand. In a more generalized
maximizes the utility of those customers. The
system or network, each link is a separate but interre-
higher the total revenue such a lane produces, the
lated market for travel. Each traveler seeks the mini-
greater the value of the transportation benefits it is
mum path through the network to his/her destination,
providing. By contrast, the goal of maximizing the
considering operating costs and the value of time. In
number of people per lane per hour implicitly
the face of an increase in money price on a particular
values all those trips equally, and at a low value
link, the traveler may
(the extra time and convenience of being in
● Continue to travel and pay the toll (high time value) a carpool).”
● Divert to another route or path (higher time cost
for traveler, increase delay on alternative path)
● Shift to carpool or transit (perhaps offering better Related Markets
service because of reduced congestion)
The concept of a market, with supply and demand,
● Shift time to off-peak (traveler worse off than with-
and a resolution, whether a constrained or an
out the toll, but not as bad as paying the toll)
unconstrained equilibrium, is an abstraction from the
● Change in location or origin or destination, long
field of economics. Whether the concept can be applied
run or short run, or combine trips into tours (trip
to a given context, how it should be applied, and
chaining)
whether anything useful results depends upon the
● Forgo the trip
skill and insights of the analyst.
Each of these responses represents a valuation by Inherent in the idea of a market is the under-
the user of the available alternatives. The benefit of standing that markets are connected to each other:
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
a change in one market affects all other markets,
directly or indirectly. Such an understanding would
be paralyzing (nothing can be concluded unless all
markets are studied) without some principles for
bounding the scope. Three ranges of scope can be
considered:
1. Primary Market
Responses taking place within the market are
endogenous, and all relevant factors should be
resolved; other markets may be incorporated indi-
rectly via changes in consumer surplus, but are not
recognized explicitly.
2. Related Markets
Markets whose demand or costs are directly
affected by the primary market may need to be
included, such as peak and off-peak markets, as
being not autonomous (e.g., peak and off-peak
may share the same capacity).
3. Other Markets
Related markets that exhibit market failure
(monopoly, constrained pricing) may be incorporated
into the evaluation, to the extent that their inefficien-
cies are affected by pricing in the primary market.
All other markets not explicitly included are
assumed to adjust efficiently to whatever changes
occur in the primary market. When pricing a single
facility, it is assumed that all other facilities are “other”
markets, which adjust their prices and output (includ-
ing investment in capacity) to remain efficient.
Whether vehicle trips gained or lost in the primary
market are the result of diversion, occupancy shifts,
or deterred/induced travel is unimportant, or at least
ignored. Shifts between peak and off-peak, however, are
certainly relevant.
Multiclass service creates two markets that are
closely related. Similarly to peak and off-peak
demands, an increase in the price of the premium
service causes an outward shift in demand for the
standard service. If peak and off-peak are also distin-
guished, then there is a matrix of elasticities connecting
the four options. An increase in the price of the pre-
mium service in the peak, for example, will increase
demand for off-peak premium service and both stan-
dard service demand periods.
Evaluation of changes in related markets might be
approached by treating the market in which the change
583
occurs as the primary market, and changes in related
markets as relevant. Benefits would be as described
above for a single facility, plus changes in the magni-
tude of the efficiency loss in related markets. If all other
markets, whether directly related or only indirectly,
adjust efficiently, then nothing in those adjustments
need be considered in evaluating the changes in the
primary market. If, alternatively, related markets are
inefficiently priced or constrained, then the change in
inefficiency can be regarded as a benefit of the change in
the primary market [70].
An increase in price in the express lane market, for
example, would cause an increase in demand in the
substitute GP market, as shown in Fig. 6. If the related
market is priced inefficiently, and the deviation
between price and marginal cost is increased by higher
output, then the inefficiency is worsened by the price
change in the primary market. The change in efficiency
(a negative number, in this case) could be added to the
benefits of the price change.
Some Lanes Priced
If the theory shown in Fig. 2 is applied to one or several
lanes of an urban expressway, but not to all lanes, an
improvement in efficiency can be obtained even if the
pricing is not optimal, as in Fig. 5. Vehicles tolled off
the restricted lanes may take any of the options listed
above, including consolidation into more person trips
in the remaining vehicles. At least some of the vehicles,
however, will choose to not to pay the price but will use
the remaining GP lanes. This demand shift in the
closely related market, as in Fig. 4, would not be
accounted for in the benefits (disbenefits) from pricing.
One could assume that all other markets are effi-
cient and make optimum adjustments to price changes
in related markets, but this assumption is clearly vio-
lated in practice. The price is fixed at an inefficient level
(unlike the stopped clock, it is never correct), and the
capacity cannot be varied either. Inevitably, the ineffi-
ciency that is (in this case) worsened in the related
market needs to be addressed.
If a portion of the traffic is separated from the rest
and charged a price to keep it flowing at LOS C, the
price on those lanes is higher than without the pricing,
and the delay is greater on the GP lanes if the total travel
remains the same (more demand is imposed on less
584 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
generalized
cost,
price
demand
shift
inefficiency before change
in primary market
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 6
Change in inefficiency in related market
capacity). Therefore the average generalized price is
greater and the result is that the total traffic volume
will be less. Vehicle occupancy is raised by the pricing
(and not by congestion), but this reduction in vehicles
will less than offset the additional demand directed at
the GP lanes. Under a reasonable range of conditions,
then, time savings on the express lanes may be more
than the increased delay on the GP lanes (because the
average user price is higher and some efficiency is
gained) or may be less than the delay on the GP lanes
because not enough vehicles have been deterred to
offset the capacity loss from separation. This assumes,
of course, that there is no increase in the total capacity
of the facility.
All Lanes Priced Uniformly (Single Service)
The prototypical condition imagined in the standard
theory is steady flow on a single uninterrupted facility.
In this abstract market, there is a single demand curve
and single cost curve. Cost is determined primarily by
the relationship between traffic volume and delay. Mar-
ginal cost can be derived from the average flow rela-
tionship (empirically or analytically) and the optimal
toll calculated and imposed. The gains from this con-
gestion pricing is a reduction in delay for the users
choosing to the pay the toll, minus the loss of consumer
surplus suffered by the deterred vehicles, shown previ-
ously in Fig. 3. The number of travelers may stay the
MC
inefficiency after change
in primary market
user price function
demand1
demand0
v0 v1 related market output
same (or even grow) but some of them have been
inconvenienced by having to share the same vehicle.
This simplified abstract model does not say any-
thing about the distribution of travel time values
among the vehicles along the demand curve. As the
price goes up, some users are deterred by the sum of
time and money costs, but the trade-off between them
will not necessarily be the same at any point along the
demand curve. Moving up the demand curve, it is
implicitly assumed that the money price is increasing
while the delay portion of generalized price is
decreasing.
Thus the standard model is robust for the mecha-
nism it uses (a single price for all users), but does not
acknowledge that conditions may be considerably
more complicated.
Both Services Priced Efficiently
The fundamental question in regard to multiclass ser-
vice is whether there are market segments with the
demand for the service that are sufficiently distinct as
to warrant separate tailored services along with the
additional costs that differentiation entails. In the case
of highway travel, the question is what distribution of
travel time values among the vehicles potentially
demanding service justifies differentiating the demand
into two or more classes and charging different prices.
Clearly, charging a price for both services is a gain in
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
benefits over charging for only one service, but are the
optimal prices charged for the two classes different
enough to justify the loss in capacity?
There seems to be little analysis or evidence on this
problem. Many analysts assert that the distribution of
VOTs is wide, but how wide in truth and what width is
needed to take the extra step of multiclass service is not
yet solved.
To evaluate a multiclass service is to address the
problem of allocating a fixed amount of capacity to
two classes of users, each with its own demand
(reflecting differing preferences for travel time, occu-
pancy, and other attributes) and pricing structure. The
resulting equilibrium is then compared to the base case.
First-best evaluation is a comparison of alternatives
all of whom are efficiently designed and priced.
Second-best evaluation occurs when at least one con-
straint is imposed on at least one of the alternatives,
and the net benefits of one alternative versus another
are estimated.
A first-best evaluation of multiclass highway service
would compare an efficiently priced single facility to
two efficient service alternatives. Total capacity would
be fixed, the two services would be designed with
demand attributes that maximized the net benefits to
users, including pricing at marginal cost. The conclu-
sion as to whether two-class service was better than
a single class would derive from the net social benefits
(user benefits minus implementation costs) for each
alternative.
A hypothetical example is shown in Fig. 7. A single-
class service is portrayed on the left, and a two-class
single market standard service
MC
demand
D1
v Vs
HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service. Figure 7
Benefits of two-class service relative to single service
585
service in the two markets on the right. Each market is
resolved efficiently at p = MC, and the net benefits of
each – the area above the MC curve and below demand,
up to the equilibrium volume – are shaded for each
service. In the single-class service, a dashed demand
line indicates the separation of demand into those who
will choose each one when it is offered.
Once the two types of service are offered, each has
a separate demand curve (related in that a change in
price in one of the markets will shift demand in the
other; thus each demand curve is based on a given price
in the other market). By separating the two market
segments, the attributes of each service can be tailored
to its customers. The two demand curves are not nec-
essarily a partition of the single-service demand: new
users may be induced by the existence of, especially, the
premium service. With differing prices, movements
along the demand curves may result in greater or lesser
total volume, i.e., the sum of v1 and v2 may be larger or
smaller than V.
The problem is greatly complicated by the need to
distinguish different values of time. The standard effi-
cient pricing diagram assumes that time and monetary
costs can be combined on a single scale, implying that
the ratio of time to dollars – the value of travel time – is
fixed. Thus the two-dimensional diagrams in Fig. 7 are
only slices through a three-dimensional space in which
demand is represented as a surface that is a function of
both time and money. If a third dimension to the
diagram is imagined, the standard service is a slice
through the low-time-value end of the diagram while
the premium service is a slice at the high VOT end.
premium service
MC1 MC2
D1 D2
v1 v2
586 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
The single market is a slice through the middle, or at
average VOT.
In the particular representation in Fig. 7, the gen-
eralized price to the user in both markets is below what
it was in the combined market because the money price
is lower and the time cost is less highly valued. Hence,
total net benefits are higher for multiclass than for
single service. For the premium service, the price is
lower because the higher money price is more than
offset by the faster travel. For the GP lanes, travel
times are worse than in the combined market, but are
valued less highly than the reduced price. These results
are not true in general, but occur only if the service
characteristics and prices very closely match the
demands of each group. Otherwise, the generalized
prices combining toll and travel time will be close to
the single market price or above it, and the multiclass
service will offer no advantage over the efficiently
priced single-class service. Keep in mind that all ser-
vices – combined or separated – are priced efficiently in
this comparison.
The provision of two classes of service creates a net
benefit if the sum of the benefits for standard and
premium services exceeds the benefit of the single ser-
vice by enough to offset the costs of providing two
services. If all travelers are homogeneous in their pref-
erences, then there are no benefits from multiclass
service because all users want the same thing. This
fact and several other aspects of the problem are
addressed by Small and Yan [71]. Thus the two market
segments must have distinctly different demand curves,
and the service classes must have attributes that sepa-
rate the demand segments. The benefit of multiple
services relative to a single service depends upon how
different the market segments are, and how well the
service designs appeal to and satisfy those differences.
The theoretical basis for this evaluation has been most
completely developed by Small et al. [45], but they did
not incorporate any penalties for partitioning the facil-
ity into service classes.
Empirically, attempting to measure the benefits by
valuing the attributes (such as travel time differences) is
not likely to be useful, because less easily observed
differences are too hard to measure or observe (e.g.,
the cost of schedule shift, or coordinating a carpool).
Revealed preference can show demand quantities and
generalized price at equilibrium, but the remainder of
the demand curve is a problem, as are the cross-
elasticities.
A paper from the operations research literature [72]
describes a model for allocating customers to two clas-
ses of service according to their willingness to pay for
service time. The model determines the capacity given
to each service and the profit-maximizing prices
charged. Possibly a model of this form could fit the
multiclass service optimization problem.
Equity
Any form of pricing of highways raises numerous
claims of inequitability. The concept of HOT lanes
generates additional objections beyond those that
occur in the context of “simple” (single class) tolling
and congestion pricing because there is a side-by-side
comparison. People observe that, among toll payers,
the burden of payment declines as a share of income as
income rises, so pricing is regressive. Others observe
that middle and higher income users are the majority
among payers, so they are getting the benefits, and that
is regressive. These both reveal an incomplete under-
standing of distributional equity.
Lexus Lanes Vertical equity impacts (impacts on the
distribution of income) can be addressed objectively
and quantitatively, although the data are coarse at best.
Among travelers, peak highway users have higher aver-
age incomes than both transit users and off-peak users.
Where express lanes are available, income levels of toll
payers are usually higher than non-payers (Sullivan and
Burris [59]; FHWA [73]). As the income of toll payers
decreases, vehicle occupancy usually goes up. Hence
the accusation that express toll lanes disproportion-
ately benefit higher income users. The counter anec-
dote to the Lexus lane barb is the observation that
working parents use the express lanes to pick up their
children from daycare. Despite the expressions of con-
cern, equity discomfort tends to subside after initial
reaction to proposed HOT lane conversions, and has
not so far been an impediment to implementation [55].
Unlike tax instruments, both the costs and benefits
of paying for a market good are borne directly by the
buyer. The buyer weighs the costs and benefits to him-
self/herself and makes a choice. If the consumer
chooses to buy, then the benefits outweigh the cost,
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
possibly by a large amount but a magnitude not appar-
ent to the outside observer. If there are no external
costs and no subsidies, then the social benefit of
the good or service is at least as great as its costs, and
neither consumers nor taxpayers are disadvantaged.
The existence of the good may divert consumption
from other goods, but there are no inherent equity
impacts, i.e., income is not being transferred without
compensation. The above notwithstanding, “some
argue the HOT lanes unfairly favor those wealthy
enough to afford high tolls” [74].
This leaves the only complaint being that the under-
lying distribution of income is unfair. The solution,
then, is a “Robin Hood” income transfer that
rearranges income solely according to income, not
according to road usage or any other criterion.
The poor will then have the wherewithal to purchase
peak SOV travel, or use the income for some other
purpose.
Perceptions of equity are difficult to address in
rational terms. Resentment against those who choose
to buy their way out of congestion is not really any
different from resentment against those who buy lux-
ury yachts and large mansions, except that the cost of
buying out of congestion is far less.
Earned Credits In an effort to improve the public
acceptability of tolls, it has been proposed that drivers
earn credits by using the unpriced lanes. The credits
could then be applied to use of the express lanes [75].
Other proposals have been offered [76]. No projects
have so far been implemented.
Generally credits are earned for purchasing some-
thing, as opposed to not purchasing. Thus credits
would be awarded for paying the tolls for some period
of time or amount, and a bonus paid to the user.
The purpose is to create brand loyalty, which is not
relevant here.
Hence the purpose is based on equity concerns.
Sometimes it is vertical equity that is of concern, in
which case the credits should be awarded to all persons
or households solely on the basis of income, and
unrelated to whether the recipients actually drive in
the peak period or would like to. Rewarding drivers
for using the GP lanes does not seem to accomplish any
equity goal and is perverse from an efficiency stand-
point (it encourages peak usage).
587
However credits are awarded, they constitute
a financial transfer that reduces the net revenues from
the tolls. Other losses of revenue include collection
costs, operating subsidies for transit, and matched
reductions in unrelated taxes such as property taxes.
If the proceeds are rebated to users of the facility
because of usage, they have a negative efficiency impact
as well as an equity impact.
Land Use Impacts The rent gradient model of urban
form relates the rate of decline in land value with
distance from a point of central access to the unit cost
of transportation. In this model, a reduction in the
price of transportation to the user tends to flatten the
rent gradient, causing land use activities to spread out
at lower densities. Other things being equal, an increase
in capacity reduces the travel time from delay and leads
to dispersal of economic activity. Correspondingly, an
increase in prices – e.g., through higher prices to users –
should result in more compact development. While
whole-facility pricing shifts the relative spatial prefer-
ences of households, on average the price is higher with
efficient pricing. Thus the standard congestion pricing
model applied to an urban network will tend to steepen
the rent gradient.
HOT lane pricing does not lead to unambiguous
results when interpreted via the rent model. While the
explicit money price is greater on the tolled lanes, delay
is greatly reduced; the net result, however, is an increase
in price – for the same capacity. HOT lanes have never,
however, been taken from existing capacity. The one
attempt to do so – the infamous Santa Monica freeway
diamond lane in Los Angeles – was a political disaster
sufficiently painful to ensure that removing existing
lanes has not been attempted again, and extensive
outreach is undertaken before any kind of reserved
lane is implemented [77]. Even if the amount of
capacity added is less than what would be added by
all GP lanes, the result is nonetheless to increase
capacity and lower the cost of travel, spreading out
the urban area.
Future Developments
Although the mileage of highways in “pure” HOV
facilities vastly exceeds the tolled mileage, there are
few plans to add HOV-only capacity; treatments
588 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
being considered include general capacity expansion
and tolling [16].
DKS Associates [10] describes activities in ten
metro areas, most of which do not have congestion
pricing currently, and some that do are not included
in the survey. The report also describes how HOT lanes
can be incorporated into the regional planning process,
and offers lessons learned regarding implementation
from each of the areas.
Demonstration Effect
For at least some analysts, the HOT lane was a way to
leverage the perception that HOV lanes were
underutilized into a demonstration that pricing
works. In this scenario, the mechanism of pricing
could substitute for the awkwardness of verifying occu-
pancy, and the occupancy requirement could then be
dropped because the pricing would have the same
effect. With pricing established as a management tool,
all lanes could be priced and the physical separation no
longer necessary. All links of the network (or the arte-
rial network) could be readily tolled, and traffic flow
greatly improved. With no other evidence of political
willingness to try congestion pricing, HOT lanes was
the only game.
In this way, the expansion of the concept could
encompass all roads, or at least urban highways. Fiel-
ding and Klein [78] were early promoters of the spread
of congestion pricing by imitation. The next level in
this scenario is the congestion pricing of whole facilities
in a strategic pattern. The planned pricing of the SR 520
bridge in the Seattle area will be the first example of
such whole-facility congestion pricing in the USA, if
and when it occurs.
Evolution of Multiclass Networks
Poole [79] has described a scenario in which single-lane
HOT lanes phase out the HOV exemption or discount,
because the exemptions prevent the revenue streams
from being large enough to cover capital costs. These
more-or-less fully priced one-lane highways are then
successful enough to construct the ramps and flyovers
that permit the express service to be continuous when
one freeway intersects another. With this success as
a model, the one-lane priced highways can be expanded
to two lanes in each direction, resulting in a reasonable
enough level of capacity and associated throughput to
justify the separation.
An alternative scenario has emerged. Travelers are
greatly enamored of the freedom to choose between
tolled and untolled lanes, and HOT lanes are being
extended into separated networks. The expense of the
infrastructure for maintaining separation at inter-
changes has not deterred investment in HOT lanes.
Although the benefits in time savings may not warrant
the construction costs in many circumstances, such
calculations do not seem to have been done and there
is not an obvious political market for them.
From HOV to BRT The San Francisco Bay Area plans
to evolve from HOV lanes to a single-lane HOT lane
network with physical separation at interchanges, and
from there to a tolled network that is heavily used by
express buses [80]. The region plans to add capacity by
utilizing shoulders and relaxing freeway design stan-
dards. HOVs will remain free or discounted. Realiza-
tion of this vision still implies a continuing primary
contribution to transportation funding from the fed-
eral government.
From HOV to HOT Texas seems to be moving in the
direction of two-class service with express lanes having
at least two lanes (generally permanent) in each direc-
tion on both service classes [41]. This will entail con-
siderable expenditures for physically separated
infrastructure, yet HOVs appear to remain free and
BRT does not carry a large share of passengers. The
Texas DOT has produced plans for public–private part-
nerships to build and finance major multimodal
intercity corridors, but there has been some political
backlash from the legislature.
From HOV to Full Facility Tolling The Puget Sound
region has an extensive existing HOV network, some of
which the region expects to convert to HOTalong the way
to creating a network of fully tolled expressways with no
differentiated classes of service and no exemptions or
discounts [81]. While this end state is the most likely
to be sustainable in the long run, the path for moving to
there from HOVs is not presently spelled out.
From HOV to HOT and Truck Lanes The Reason
Foundation has proposed an evolutionary path that
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
leads from express lanes to multiclass service in which
all services are priced and truck-only lanes are
a candidate service [79]. Poole claims that these char-
acteristics will emerge once federal subsidies are limited
to highways of national interest and local regions take
responsibility for most of the transportation infrastruc-
ture, including its funding. The resolution will be
a system paid for primarily by users.
Inevitably there are limits to the expansion of
simple (i.e., unpriced) HOT lanes in metropolitan
areas. Swisher et al. [82] portray various longitudinal
scenarios showing how express lane viability (i.e., nei-
ther too much nor too little traffic) can be retained by
pricing selected users, including carpools. The ideal
timing for introducing occupancy and pricing policy
changes depends upon the growth of demand in the
corridor.
At present, the prospects of universal road charges
(VMT charges graduated by congestion and other
costs) appear uncertain and not imminent, although
there are some indications that the eroding contribu-
tion of fuel excise taxes and the lack of other revenue
sources are improving the climate for tolling [83]. It
may be that the inefficiencies of congestion will be
tolerated in many locations, as well as the inefficiencies
of multiclass services.
Public–Private Partnerships The USA is beginning
to explore various arrangements that allow for more
flexible management of the highway asset procurement
process. Design-build construction and its extension
into operation and financing have been used in some
instances to create transportation infrastructure. Avail-
ability payments can allow the public sector owner to
postpone payments to the builder until the facility is in
operation.
If a project can cover its operating costs from tolls,
then the possibility exists for letting a private firm
operate the road and collect the tolls, potentially also
setting the toll rates. One of a planned series of HOT
lanes in the Washington, DC area is being constructed
via a public–private partnership between the State of
Virginia, the Federal Highway Administration, and
Transurban-Fluor [84, 85]. The first modern era US
toll lane facility was financed and constructed by
a private firm, although it was eventually necessary
for the State of California to buy out the private firm.
589
SR-91 opened in 1995 and was repurchased in 2002 by
the Orange County Transportation Authority [86]. The
logic of letting some vehicles pay extra for speed creates
an opportunity for letting the private sector front the
funds to construct the additional lanes. Planned toll
lanes in Sacramento may be funded in this way [87].
Using General Revenues The alternative to toll rev-
enues for financing is general fund revenues. Many
states and localities have passed referenda to pay for
transportation with sales taxes. The federal government
has transferred $30 billion into the Highway Trust Fund
to keep it solvent. Perhaps because most voters see
themselves as highway users, they favor spreading the
costs over more payers than just users. This scenario
maintains strong controls at the federal level over the
raising of revenues and how they are spent, with Con-
gress making decisions based on political balancing.
Whether the path ahead lies through extended
multiclass service, private equity, and political devolu-
tion, or major general fund support for transportation
is subject to conjecture and, we hope, debate.
Note
Information contained in this chapter consists of the
judgment of the authors and does not necessarily rep-
resent the policy of the Volpe Center or the U.S. DOT.
Bibliography
1. Federal highway administration (2008) Managed lanes: a
primer. US DOT/FHWA, Washington, DC. http://ops.fhwa.dot.
gov/freewaymgmt/pubs.htm#mngd_lns_hov. Accessed 27
June 2011
2. Port authority of New York and New Jersey (2010) Bridges and
tunnels, tolls, PANYNJ. http://www.panynj.gov/bridges-
tunnels/tolls.html. Accessed Aug 2010
3. Small KA, Gomez-Ibanez JA (1998) Road pricing for congestion
management: the transition from theory to policy. In:
Button KJ, Verhoef ET (eds) Road pricing, traffic congestion
and the environment: issues of efficiency and social feasibility.
Edward Elgar, Cheltenham, pp 213–246
4. Collier T, Goodin G (2004) Managed lanes: a cross-cutting
study. Prepared for Federal Highway Administration, Texas
Transportation Institute, College Station
5. Sallach M, Vu P, Zhan YJ (2010) Corridor level analysis of
potential Managed lanes investment on I-75 South in Atlanta,
Georgia. Paper for TRB, Georgia State Road and Tollway
Authority, Atlanta
590 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
6. Mirshahi M, Obenberger J et al. (2007) Active traffic manage-
ment: the next step in congestion management. Prepared for
DOT/FHWA, US, Washington, DC. http://international.fhwa.
dot.gov/pubs/pl07012/. Accessed 27 June 2011
7. Federal highway administration (2010) Value pricing pro-
jects involving tolls. US DOT/FHWA. http://ops.fhwa.dot.
gov/tolling_pricing/value_pricing/projects/involving_tolls/
priced_lanes/index.htm. Accessed Aug 2010
8. Martin PT, Vladisavljevic I, Yusufzyanova D (2007) The I-15
Express lanes evaluation. Prepared for Utah Department of
Transportation, Salt Lake City, University of Utah. http://
www.udot.utah.gov/expresslanes/dld/U%20of%20U%20
Express%20Lanes%20Evaluation%20November%202007.pdf.
Accessed 27 June 2011
9. Liu X, Zhang G, Wu YJ, Wang Y (2009) Analyzing system
performance for Washington state route 167 high occupancy
toll (HOT) operations. Prepared for TRB, University of
Washington, Seattle
10. DKS Associates (2009) A domestic scan of congestion pricing
and managed lanes. Prepared for Administration, DKS Associ-
ates in association with PBSJ and Jack Faucett Associates,
Federal Highway, Oakland
11. Federal Highway Administration (2008) Technologies
that enable congestion pricing: a primer, US DOT/FHWA,
Washington, DC. http://ops.fhwa.dot.gov/tolling_pricing/
value_pricing/publications.htm. Accessed 27 June 2011
12. Goodall N, Smith BL (2009) What drives single occupant trav-
eler decisions in HOT lanes? : an investigation using archived
traffic and tolling data from the MnPASS Express Lanes. Paper
for TRB, University of Virginia, Charlottesville
13. Brinckerhof P (2003) A guide for HOT lane development. Pre-
pared for Federal Highway Administration, Parsons
Brinkerhoff with Texas Transportation Institute in partnership
with US DOT/FHWA, Washington, DC
14. PB Study Team (2002) HOV performance program: evaluation
report, Los Angeles county metropolitan transportation
authority, Los Angeles
15. 407 ETR (2010) ETR 407 express toll route, 407 ETR Concession
Company Limited, Toronto. http://www5.407etr.com.
Accessed 27 June 2011
16. Fuhs C, Obenberger J (2002) HOV facility development:
a review of national trends. Prepared for TRB, Parsons
Brinkerhoff, Houston
17. Burris MW, Lipnicky K (2009) HOV or general purpose lanes?
Public Works Manag Policy 14(2):130–147
18. Poole Jr RW (2008) The HOT topic: enforcement in the fast
lanes. Tolltrans, pp 44–49
19. Poole Jr RW, Kenneth Orski C (1999) Building a case for hot
lanes: a new approach to reducing urban highway congestion.
Reason Public Policy Institute, Los Angeles
20. Lee DB Jr (2008) Toward the evaluation of value pricing.
Transport Res Record 2079:71–78
21. Chang M, Wiegmann J, Smith A, Bilotto C (2008) A review of
HOV lane performance and policy options in the United States.
Prepared for FHWA, Booz Allen & Hamilton and HNTB, McClean
22. Booz Allen Hamilton, and HNTB (2008) A compendium of
existing HOV lane facilities in the United States. Prepared for
Federal Highway Administration, BAH and HNTB, McClean
23. Cervero R (1999) Reviving HOV lanes. Transport Quart 53(4):67–81
24. Sas M, Carison S, Kim E, Quant M (2007) Considerations for
high occupancy vehicle (HOV) to high occupancy toll (HOT)
lane conversions guidebook. Prepared for Federal Highway
Administration, HNTB/Booz Allen Hamilton, McClean
25. Plotz J, Konduri KC, Pendyala RM (2009) To what extent can
HOV lanes reduce vehicle trips and congestion?: An explor-
atory analysis using national statistics on HOV mode use.
Paper for TRB, Arizona State University, Tempe
26. Supernak J, Kaschade C, Steffey D (2003) Dynamic pricing on
I-15 in San Diego: impact on travel time and its reliability.
Transport Res Record 1839:45–54
27. Transportation Research Board (2000) Highway capacity
manual: HCM 2000. TRB, Washington, DC
28. Barboza T (2009) O.C. Toll road price drops to attract com-
muters. Los Angeles times, 27 Mar 2009
29. Arizona Republic (2007) More Arizona drivers disobeying HOV
rules. Arizona Republic, 31 Oct 2007
30. Jang K, Chung K, Ragland DR, Chan CY (2009) Safety perfor-
mance of high-occupancy-vehicle facilities. Transport Res
Record 2099:132–140
31. Cassidy MJ, Daganzo CF, Jang K, Chung K (2006) Empirical
reassessment of traffic of traffic operations: freeway Bottle-
necks and the case for HOV lanes. Institute of Transportation
Studies, Berkeley
32. Anderson T (2007) Car-Pool lanes may be switched to toll
lanes. Los Angeles Daily Breeze, 30 Nov 2007
33. AASHTO (2003) A manual of user benefit analysis for high-
ways. American Association of State Highway and Transporta-
tion Officials, Washington, DC
34. Cassidy MJ, Bertini RL (1999) Some traffic features at freeway
Bottlenecks. Transport Res (B) 33:25–42
35. Jung S, Wunderlich K (2008) Roadway network productivity
assessment: system-wide analysis under variant travel
demand. Prepared for Federal Highway Administration,
Noblis, Falls Church. http://ops.fhwa.dot.gov/tolling_pricing/
value_pricing/publications.htm. Accessed 27 June 2011
36. Booz Allen Hamilton, HNTB (2008) Policy options evaluation
tool for managed lanes (POET-ML) users guide and method-
ology description. Prepared for Federal highway Administra-
tion, BAH and HNTB, McLean
37. Yan J, Small KA, Sullivan EC (2002) Choice models of route,
occupancy, and time of day, with value-priced tolls. Transport
Res Record 1812:69–77
38. Burris MW, Goel R (2009) Mode choice due to HOV to HOT
conversions. Paper for TRB, Texas A&M University, College Station
39. Zmud J, Bradley M, Douma F, Simek C (2007) Attitudes and
willingness to pay for tolled facilities: a panel survey evalua-
tion. Transport Res Record 1996:58–65
40. Burris MW, Ungemah DH, Mahlawat M, Pannu MS (2009) Inves-
tigating the impact of tolls on high-occupancy-vehicle lanes
using managed lanes. Transport Res Record 2099:113–122
 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
41. Sullivan E (2000) Evaluating the impacts of the SR 91
variable toll express lane facility. Prepared for California
Department of Transportation, Cal Poly State University,
San Luis Obispo. http://ceenve3.civeng.calpoly.edu/sullivan/
SR91/final_rpt/finalrep_full.pdf. Accessed 27 June 2011
42. Regan ED (2010) Pricing trade-offs: revenue vs. traffic manage-
ment. Presentation to the National Road Pricing
Conference, Houston TX, Wilbur Smith Associates, Atlanta.
http://tti.tamu.edu/conferences/nrp10/program/. Accessed
27 June 2011
43. MacGregor ME (2010) How revenue and traffic management
are combined in the North Tarrant Expressway (NTE) managed
lanes project in Fort Worth, TX. Presentation to the National
Road Pricing Conference, Houston TX, Texas Department
of Transportation, Dallas. http://tti.tamu.edu/conferences/
nrp10/program/. Accessed 27 June 2011
44. Patil S, Burris M, Shaw D (2009) A comparison of survey
designs for stated choice models: an application to commut-
ing on managed lanes. Prepared for TRB, Texas A&M Univer-
sity, College Station
45. Small KA, Winston C, Yan J (2006) Differentiated road pricing,
express lanes, and carpools: exploiting heterogeneous prefer-
ences in policy design. Brookings – Wharton Papers on Urban
Affairs, Washington, DC, pp 53–96
46. Weiss E (2008) Will drivers pay to hurry up and wait? Washing-
ton Post, 25 Feb 2008
47. Brownstone K, Small KA (2005) Valuing time and reliability:
assessing the evidence from road pricing demonstrations.
Transport Res (A) 39:279–293
48. Joint Transport Research Centre (2009) Improving reliability
on surface transport networks, OECD/International Transport
Forum, Paris, France
49. Lam CT, Small KA (2001) The value of time and reliability:
measurement from a value pricing experiment. Transport
Res (E) 37:231–251
50. Patil S, Burris M, Shaw D, Concas S (2010) The value of travel
time savings on managed lanes for urgent situations. Pre-
pared for TRB, Texas A&M University, College Station
51. Cho Y, Goel R, Gupta P, Bogonko G, Burris M (2011) What are
I-394 HOT lane drivers paying for?. Prepared for TRB, Texas
A&M University, College Station
52. Poole Jr RW (2006) Managed lanes: what should we maximize?
Surface Transport Innovations, Dec 2006
53. Poole Jr RW (2006) Transportation policy review: competing
models of HOT lanes. Public Works Financing, July–Aug 2006,
pp 25–26
54. Stone R (2010) High tolls lure drivers to I-95’s pay lanes, Miami
Herald, 31 Mar 2010
55. Bhatt K, Higgins T, Berg JT (2008) Value pricing pilot program:
lessons learned. Prepared for Federal Highway Administra-
tion, K.T. Analytics and Cambridge Systematics, Washington,
DC. http://ops.fhwa.dot.gov/tolling_pricing/value_pricing/
publications.htm. Accessed 27 June 2011
56. Cuff D (2009) MTC approves plan for toll lanes, Contra Costa
Times, 23 Apr 2009
591
57. Small KA (1992) Using the revenues from congestion pricing.
Transportation 19(4):313–333
58. Ng CF, Small KA (2008) Tradeoffs among free-flow speeds,
capacity, cost, and environmental footprint in highway
design. University of California, Irvine
59. Sullivan E, Burris M (2006) Benefit-cost analysis of variable
pricing projects: SR-91 express lanes. J Transport Engin
132(3):191–198
60. Light T (2009) Optimal highway design and user welfare under
value pricing. J Urban Econ 66(2):116–124
61. Loudon WR, Synn J, Miller H (2010) Consideration of congestion
pricing and managed lanes in the metropolitan transportation
planning process. Prepared for TRB, DKS Associates, Oakland
62. Federal Highway Administration (2008) Economics: pricing,
demand, and economic efficiency, a primer. US DOT/FHWA,
Washington, DC
63. Mohring H (1976) Transportation economics. Ballinger,
Cambridge, MA
64. Vickrey WS (1963) Pricing in urban and suburban transport.
Am Econ Rev 53:452–465
65. Walters AA (1968) The economics of road user charges.
International Bank for Reconstruction and Development,
Washington, DC
66. Mohring H (2006) Congested roads: an economic analysis with
twin cities’ illustrations. In: Roth G (ed) Street smart. Transac-
tion Publishers for the Independent Institute, New Brunswick,
pp 141–170
67. Kidd B (2006) Multimodal optimization of urban freeway cor-
ridors. Prepared for Arizona Department of Transportation,
Lee Engineering, Phoenix
68. Poole Jr RW (2010) Managed lanes: networks vs. individual
facilities. Prepared for TRB, Reason Foundation, Plantation
69. Craig T (2007) Deals clinched on HOT lanes, Washington Post,
21 Dec 2007
70. Verhoef ET, Nijkamp P, Rietveld P (1996) Second-best conges-
tion pricing. J Urban Econ 40(3):279–302
71. Small KA, Yan J (2001) The value of “value pricing” of roads:
second-best pricing and product differentiation. J Urban Econ
49:310–336
72. Celik S, Maglaras C (2005) Dynamic pricing and leadtime quo-
tation for a multi-class make-to-order queue. Columbia Uni-
versity Business School, New York
73. Federal Highway Administration (2008) Income-based
equity impacts of congestion pricing: a primer, US DOT/FHWA
Washington, DC. http://ops.fhwa.dot.gov/tolling_pricing/
value_pricing/publications.htm. Accessed 27 June 2011
74. Cassidy WB (2010) Congestion pricing advances in virginia.
J Commerce, May 27. http://www.joc.com/government-
regulation/congestion-pricing-advances-northern-virginia
75. Buckeye KR, DeCorla-Souza P, Lari A, Aultman S (2009) User
perceptions of fee lane concepts in Minnesota. Minnesota
Department of Transportation, St. Paul
76. Gulipalli P, Kalmanje S, Kockelman K (2008) Credit-based con-
gestion pricing: expert expectations and guidelines for appli-
cation. J Transport Res Forum 47(2):5–20
592 HOT Lanes/Value Pricing: Planning and Evaluation of Multiclass Service
77. Billheimer JW (1978) The Santa Monica freeway
diamond lanes: evaluation overview. Transport Res Record
663:8–16
78. Fielding GJ, Klein DB (1993) “High occupancy/toll lanes: phas-
ing in congestion pricing a lane at a time”, Reason Foundation
Policy Study 170. Reason Foundation, Santa Barbara
79. Poole Jr RW (2010) Pricing and Fiscal implications of different
types of HOT Lanes. Presentation to the National Road
Pricing Conference, Houston TX, Reason Foundation, Santa
Monica, CA. http://tti.tamu.edu/conferences/nrp10/program/.
Accessed 27 June 2011
80. Klein L (2010) 2010 bay area express lane network. Presentation
to the National Road Pricing Conference, Metropolitan Trans-
portation Commission, Houston, Oakland. http://tti.tamu.edu/
conferences/nrp10/program/. Accessed 27 June 2011
81. Stone C (2010) SR 520 Lake Washington Project: Seattle’s
urban partnership agreement. Presentation to the National
Road Pricing Conference, Houston, Washington State Depart-
ment of Transportation, Olympia. http://tti.tamu.edu/confer-
ences/nrp10/program/. Accessed 27 June 2011
82. Swisher M, Eisele WL, Ungemah D, and Goodin GD (2003) Life-
cycle graphical representation of Managed HOV lane evolu-
tion. In: 11th International HOV conference, Seattle
83. Copeland L (2010) States turn to tolls to fund roads. USA
Today, 2 Aug 2010
84. Adams T (2010)Toll concession model for the I-495 beltway
HOT lanes. Presentation to the national road pricing confer-
ence, Transurban, Houston, Washington, DC 2–4 June 2010.
http://tti.tamu.edu/conferences/nrp10/program/). Accessed
27 June 2011
85. Weiss EM (2008) Money spent vs. time saved debated
for HOT lanes. Washington Post, Aug 25. http://www.
washingtonpost.com/wp-dyn/content/article/2008/08/24/
AR2008082402559.html
86. Miles M (2007) Tolling & value pricing: the California experi-
ence. In: ITE – Technical Conference 2007, California Depart-
ment of Transportation, San Diego
87. Turner M (2008) Officials look to toll lanes to solve traffic,
funding issues. Sacramento Bus J, Feb 22. http://www.
bizjournals.com/sacramento/stories/2008/02/25/story1.html
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
Hybrid Electric and Hydraulic
Technology Applications in Off-Road
Vehicles
M. ABUL MASRUR1, VIJAY K. GARG2
1 lates one form of power (DC or AC) into another.
College of Engineering & Science, University of
Detroit Mercy, Detroit, MI, USA
2 highly compressed mixture of air and spray of fuel
Dublin, OH, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Off-Road Vehicles
Diesel Electric Locomotives
Hybrid Hydraulic Vehicles
Future Directions
Appendix
Bibliography
Glossary
Plug-in hybrid A plug-in hybrid is a regular hybrid
vehicle that has a large high-capacity battery bank
that can be charged by plugging in to normal
household outlet. While standard hybrids require
a combination of regenerative braking and energy
from the engine to recharge the batteries, plug-ins
can essentially operate as electric vehicles with an
internal combustion engine backup. This is also
known as PHEVs (Plug-in Hybrid Electric Vehicles)
Alternator An electromechanical device used to trans-
late mechanical power into AC electrical power.
Battery state of charge SOC or state of charge is the
equivalent of a fuel gauge for a battery, that is,
a measure of how much energy is left in the battery
with respect to its full rated capacity. Although many
times SOC is related directly by measuring the
terminal voltage, SOC cannot be accurately deter-
mined in general by voltage measurement, because
the terminal voltage of a battery may stay substan-
tially constant until it is completely discharged. SOC
is also highly dependent on temperature.
Complex hybrid vehicles In this system, in addition
to the ICE and the motor, there is an additional
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
593
motor/generator set (along with a power electronic
converter) which can be used to charge the battery
as needed. This can continue charging the battery
while the main motor is driving the wheels.
Converter A generic term for a system which trans-
Diesel IC engine An IC engine that uses the heat of
(diesel) near the end of the compression stroke to
create the combustion.
Drive cycle It is a relationship of either speed versus
time or power versus time, for a particular vehicle
which is driven in a given road profile. The road
profile can be city streets, highways, steep grade,
or some chosen mixed combination of these or
other profiles, which will be predefined.
Fuel cell It is a stationary device used to directly
translate chemical energy into electrical energy.
Generator It is the generic name used to imply some-
thing that translates mechanical energy into electri-
cal energy.
Hybrid electric vehicle (HEV) A vehicle which uses
more than one source (two being most common)
of propulsion, one being the internal combustion
engine, and the other uses an electric motor driven
by either a generator run by the IC engine or
a combination of generator and a battery which can
be charged by the generator to provide the electricity,
or it can be a fuel cell instead of a battery or gener-
ator, which can directly run the motor.
Internal combustion engine (ICE) ICE is the one in
which combustion of the fuel (e.g., gasoline) takes
place internally inside a cylinder or other enclosed
space using spark ignition or compression ignition
to some combustible gas, which will expand and
produce high pressure that will produce mechanical
power. ICE is thus a converter of chemical energy in
mechanical energy.
Inverter A system which converts DC into AC.
Microgrid If an HEV is provided with an additional
converter, which can translate the battery power
into a utility level AC power, then it can be
interfaced with either the main utility grid directly
or with other HEVs or other smaller subsystems
using utility level AC voltage. An interfaced set of
vehicles, other utility level AC voltages, and utility
# 2012, DOI 10.1007/978-1-4419-0851-3
594 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
grid, all or part of these entities taken together can
be termed a microgrid.
Off-road vehicle An off-road vehicle is a vehicle,
which is capable of driving on and off paved or
gravel surface. Vehicles that do not travel public
streets or highways are generally termed off-
highway vehicles, for example, tractors, forklifts,
cranes, backhoes, bulldozers, and golf carts.
Parallel hybrid vehicles In parallel hybrid, the IC
engine and the electric motor, both are mechani-
cally connected to the wheel. The ICE is also used to
charge the batteries.
Power electronics It is a system using solid state high-
power and high-voltage electronic switches to turn
on/off a voltage or current, which together with
other elements like inductance, capacitance to con-
vert DC, AC, etc., into another form of controlled
power.
Rectifier A system which converts AC into DC.
Series hybrid vehicles It is a hybrid vehicle, in which
the electric motor (or motors) is the only unit
which is used to run the wheels. The IC engine
charges the batteries by means of a generator.
Ultracapacitor A capacitor that has much greater
energy density and power per pound than regular
capacitors. It uses chemicals, but the energy is
stored in the electrostatic form and, unlike
a battery, the energy is not stored through the
chemical reaction and no change in chemicals take
place while storing energy. It is also sometimes
referred to as “supercapacitor.”
Definition of the Subject and Its Importance
Hybrid electric vehicle (HEV), as discussed elsewhere
in this entry, basically involves hybridization or mixing
of power and propulsion sources and/or loads, leading
to the achievement of better overall system efficiency
and accompanying pollution control. In short, it hinges
on the principle that not every “type of ” power source
(which, e.g., could be IC engine, battery, generator,
etc.) is always the best to provide power to every type
of load demand, which can vary over time. So, the idea
is to use the best source as the load changes. Here, the
word “best” is defined in terms of efficiency, perfor-
mance, and similar attributes. The technology evolved
in different forms over the years. The main motivation
behind the technology in the recent past has been due
to cost of fossil fuel, that is, gasoline and diesel in
particular, which showed an ever-increasing trend.
But additional benefit like less pollution has also
accompanied the technology. HEV technology has
already been in existence for use in diesel electric loco-
motives for quite a while. There the intent was to get
very high torque to move the vehicle from rest, by using
electric motor instead of a mechanical system (with
engine and transmission directly propelling the
wheels), thus avoiding the use of bulky mechanical
transmission system. At this time, the technology,
which is highly interdisciplinary in nature, requires
further improvement in the areas of battery, power
electronics, and motor. To this end better materials
are being investigated to develop lightweight and high
temperature withstanding systems. The importance of
the technology is obviously evaluated mainly by how
much it can help with fuel economy, followed by pol-
lution control, and then the performance, weight, and
size. Depending on the application, one or more of
these criteria can be of importance to one user (e.g.,
small-size automobiles), whereas as another set of
criteria can be of importance to another user (e.g.,
heavy construction and off-road vehicles). Obviously,
there is much to gain from the development of this
important technology and the future seems very prom-
ising at this time with significant technical and social
ramifications.
Introduction
There are many HEV (hybrid electric vehicle) applica-
tions which include both vehicular and electrically
operated nonpropulsion-type military usages. Vehicle
applications are directly related to vehicular propul-
sion. However, interfacing of the vehicular electrical
system to create a microgrid to supply/receive power
to/from utility grid is gaining popularity. Multiple
HEVs, with proper interfacing mechanisms, can create
utility-level voltages to run various stationary equip-
ments. This can be done by running the HEV battery
(and if necessary, the IC engines [ICE] in the HEV). In
another application, several HEVs, along with an
industrial/utility power system, can form a microgrid
environment and help enhance the overall power sys-
tem robustness and availability. Hybrid vehicles can be
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
IC engine Fuel cell
Microgrid (could
Vehicle # 1
be a utility grid)
Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 1
A generic hybrid power system
interfaced with a utility grid or a microgrid, where
a number of vehicles could be supplying power
through some common bus. It is, in principle, possible
to have an IC engine–based generator, and fuel cell,
vehicles, all properly synchronized and then exchang-
ing power. A system-level diagram of such a situation is
shown in Fig. 1. However, in this entry only off-road
vehicles/machines applications will be discussed.
One of the most important reasons for considering
HEV for vehicular applications is the cost of the fuel,
reduced emissions, and acoustic noise. Several different
architectures have been implemented in HEV,
depending on the application. For example, a series
HEV (SHEV) architecture can be implemented by
using several hub motors, with one motor on each
wheel, or alternatively, there could be one motor
per axle for propulsion. This can take advantage of
redundancy in case one of the motors fails. In this
case, the vehicle can run in a gracefully degradable
mode, with somewhat lesser performance and drive
to safety as needed. A brief introduction of architec-
tures used in vehicular applications is given in the
following section.
Series, Parallel, and Complex Architectures for
Ground Vehicle Applications
For the sake of continuity, various types of automotive
HEV architectures are discussed in this section briefly.
HEV architectures used in off-road vehicles are similar
to automotive HEV architectures, except the engine
technology. Automotive HEVs use gasoline IC engines
in general, while off-road vehicles in general use diesel-
based engines. (In this chapter, IC engine implies reg-
ular automotive gasoline engine unless explicitly
595
Battery Ultracapacitor
Loads
Vehicle # 2
indicated otherwise.) A very informative research
paper [1] has made a comparison between the series
and parallel HEV architectures (Fig. 2a, b) with a regu-
lar IC engine vehicle. From the reference cited above
[1], it has been found that IC engine vehicles are
slightly lighter than parallel HEVs. However, the series
hybrid vehicle was found to be relatively heavier. If a
nonhybrid IC engine vehicle is considered a baseline,
then one can compare the other architectures as fol-
lows. To match the performance of the baseline vehicle,
obviously it is necessary to make the electric motor in
a Series HEV (SHEV) to have the same size as the IC
engine. The authors believe that comparison of vehicle
weight and volume of an IC engine vehicle and “series
HEV” should be based on drive cycle, and the
average energy efficiency of the propulsion system,
that is, the comparison should be done under similar
circumstances in terms of its operation and type of
engine used.
The peak power can be handled by the battery,
which should be charged within its bounds. The
authors recommend simulation and experimental
studies of different drive cycles and then coming up
with a worst case scenario to help decision making. The
battery and other storage devices play an important
role in decision making, depending upon whether
they can provide the maximum power needed or not.
Only if the battery is kept under completely floating
condition all the time, and the power from the ICE and
generator is fed to the motor in parallel with the bat-
tery, will require the generator to be sized equal to the
maximum power demand. The intent of the battery or
any peak power source is to address the maximum
power demand in SHEV, and hence ICE and generator
size in SHEV could be reduced. However, the size of
596 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Fuel

ECU ICE Gear Generator Motor Geartrain Wheel

HEV Power Battery Power

controller electronics electronics

Gen. control Motor

control
a Series HEV

Fuel

ECU ICE Motor Geartrain Wheel

HEV Battery Power

controller electronics

Motor
control
b Parallel HEV

Fuel

Motor /
ECU ICE Gear Generator Motor Geartrain Wheel

HEV Power Battery Power

controller electronics electronics

Motor / Gen.
control
Motor
Complex HEV control
c

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 2

(a) Series HEV configuration, (b) parallel HEV configuration, (c) complex HEV configuration (Courtesy: Pubs-Permissions
@ieee.org)

ICE and the generator in SHEV will definitely depend
on the average power demand by the vehicle based over
the expected drive cycle in which it will operate. In
a parallel HEV, the size of the electric motor could be
less than half the size of SHEV motor.
The reference [1] assigns the balance of power to
the ICE in the Parallel HEV (PHEV). The mechanism
to split the power assignment to the motor and the ICE
in the parallel HEV assumes that the original ICE in the
baseline vehicle was chosen based on the maximum
power needed under the worst possible drive cycle
scenario. There is also no separate generator in the
PHEV in the reference cited above. However, in some
architectures, like Toyota Prius and some others as well,
there is a motor/generator that is separate from the
main propulsion motor, which can be used as genera-
tor and starter as well.
The reason for having this generator is highly
dependent on the strategy for control. Under certain
conditions of the battery state of charge (SOC) and
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
power demand in the propulsion, it may be necessary
to provide additional charge to the battery, through
this generator.
The battery, or any other storage elements like the
ultracapacitor, is an important part of the SHEV or
PHEV. In specifying batteries, normally voltage and
amp-hrs are used as metric. But in PHEV and SHEV
applications, it is also necessary to know the current (or
equivalent power) rating to satisfy the maximum cur-
rent (or equivalent power) demand by the propulsion
motor. The current (and power) demand from the
propulsion motor can be met by the generator and
the battery working together. As noted earlier, quite
often the battery is used to meet some high transient
(or peak) current demand from the motor. Hence, it is
very important to know the battery current rating, both
continuous and instantaneous, in addition to its volt-
age and amp-hrs. All these issues require a vital deci-
sion-making process in developing the design
specifications for an HEV. The discussion above applies
to both gasoline- and diesel-based HEV for better fuel
economy and other benefits, such as reduced
emissions.
Another question that requires a serious discussion
is the choice of HEV architecture, that is, whether it
should be series, parallel, or complex (Fig. 2a–c). The
question can be best answered depending on what is the
priority – is it fuel economy, performance, size or
weight, reliability, or emissions, which are briefly
answered below.
● If fuel economy is of utmost priority, then it should
be a PHEV.
● On the other hand, if the performance of the vehicle
is important (i.e., power output), then SHEV is a
better choice. This might help reduce transmission
package as well as it makes the system capable of
providing high power on demand quickly. This is
due to the fact that motor reacts to control the
commands faster than the ICE.
● If size and lightweight are of concern due to the
vehicle requirements, then PHEV will be a better
choice than SHEV. In general, SHEV is to some
extent heavier than PHEV, hence this fact might
also affect fuel economy slightly.
● Reliability is not to be underestimated in both PHEV
and SHEV. However, reliability considerations might
597
be better addressed in general by using PHEV,
because of redundancy in propulsion due to ICE
and the motor. However, PHEV control system is
more complex, in addition to complexity in its
mechanical coupling. A thorough study of failure
modes of these devices is needed before one can
conclusively make a decision on reliability matters.
● Another issue accompanying reliability in vehicular
systems is maintenance. It appears that mainte-
nance of an SHEV is somewhat simpler due to
several reasons. Its control and mechanical linkage,
unlike PHEV, are simple. If there are hub motors in
its propulsion, they can be rather quickly replaced
in case of failure.
● Off-road and military applications of these vehicles
require device ruggedness. This implies that both
mechanical and electrical as well as other design
aspects like high temperature extremes (both hot
and cold) must be considered. In general, high
temperature and mechanical vibrations can nega-
tively impact items like power electronics and bat-
teries. This means the devices can get physically
larger and bulky to provide for appropriate cooling
needs, if performance is to be achieved at these
extreme conditions.
● The above discussion indicates that there are pros
and cons of PHEV and SHEV. Having said that, it
should be noted that the decision regarding the
choice between PHEV and SHEV could be very
much application and drive cycle dependent. Final
decision on the above requirements is quite complex,
and systematic trade-off studies are required for all
requirements like performance, fuel economy, reli-
ability, etc., to reach most optimum decision.
Off-Road Vehicles
Success in hybrid electric vehicles in automotive appli-
cations has led certain other vehicular areas to consider
this technology as well. These include ships and air-
crafts, which are nonground vehicles. Diesel electric
locomotives have already been using the technology
in a slightly different form and these will be discussed
as well. Off-road vehicles are ground vehicles for heavy-
duty applications. These vehicles, in general, do not use
normal roads for their operation. This entry specifically
focuses on heavy-duty applications of off-road
598 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

vehicles. Examples of such vehicles are construction

vehicles and equipments (e.g., mining vehicles/trucks,
refuse trucks, agricultural vehicles like tractors, etc.)
and some military vehicles. Off-road vehicles, in gen-
eral, operate at low speeds compared to regular passen-
ger vehicles and use diesel engines, instead of gasoline
IC engines. Use of diesel engines has certain advantages
compared to gasoline IC engines as discussed in the
following.
● The compression ratio of diesel engine can be much
higher than the gasoline, for example, it can be
around 15–20% compared to 9–12% for the gaso-
line engine. The higher compression ratio necessi-
tates the engine to be heavier, which helps in
construction to sustain the cylinder pressure.
● At low speeds diesel engines are more efficient than
gasoline IC engines and are better in terms of fuel
economy due to the absence of throttle vales, and
avoid parasitic losses. This leads to smaller-sized
(HP ratings) engines, specifically at lower speeds.
● Other benefits of diesel engines include lower
greenhouse gas emissions such as CO2.
● Low speed characteristics of diesel engines are bet-
ter compared to gasoline engine, in terms of starting
torque generation.
● Diesel engine has less components due to the
absence of ignition system, and hence leads to
higher overall system reliability.
A comparison of gasoline and diesel engine speed–
power characteristics is shown in Fig. 3. It is apparent
that at lower speed range diesel engine can provide
higher power compared to a gasoline engine of com-
parable size. This makes it suitable for high-power
low-speed type of operation, which is more consistent
with off-road vehicles.
The drive cycles and speed-torque demands of these
off-road vehicles can be quite different as compared to
regular automobiles. Off-road mining/trucks and con-
struction vehicles operate in rough terrain and at low
speeds or under stationary conditions. Therefore, high
power and torque are required due to the nature of
these vehicle operations. In order to supply high torque
and power requirements from a diesel engine, it is
necessary to have a large gear box and transmission
system, etc. In addition, the overall efficiency of diesel
engine is still low (of the order of 40%, but still may be
Diesel
Power
Gasoline
Speed
Hybrid Electric and Hydraulic Technology Applications
in Off-Road Vehicles. Figure 3
Comparison of gasoline and diesel engine curves
higher than gasoline engine) over the complete drive
cycles (or torque speed demand of the load).
To alleviate these issues, hydraulic or electric hybrid
propulsion systems are being developed and have been
used lately. Hydraulic systems operate by continuously
running the engine, which is a low-torque high-speed
system relative to load. The engine is used to drive
a hydraulic pump which can drive a hydraulic motor.
Through the use of an incompressible fluid one can
achieve very high pressure, which can be applied at the
load drive end. However, the hydraulic systems, in
general, can cause problems. Examples of such prob-
lems are complex plumbing system with pipes, which,
under rough terrain conditions, can cause plumbing
system failures. This may be due to fluid leaks, leading
to degraded performance and eventually hydraulic sys-
tem failure. Other disadvantage of hydraulic systems is
the parasitic losses.
Based on the above discussion of diesel engines and
their use in off-road vehicles, it can be concluded that
the merits of diesel engine combined with electric pro-
pulsion (diesel HEV) can lead to most fuel-efficient
vehicles, subject to the specific application needs.
Hence, for the same performance, and particularly
during acceleration, it is beneficial to use diesel hybrid,
within the same package size. This is especially benefi-
cial in a parallel configuration, when both the engine
and the electric propulsion are required to play an
important role to achieve optimum performance in
terms of power and acceleration. The issue of
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
performance is inherently better addressed in series
hybrids, since propulsion is done only by electric
motor.
In HEV, energy storage and electric propulsion
motor systems offer very high efficiency of the order
of 75% and above. In present HEV, engine is used to
charge battery and the propulsion torque/power can be
shared between the electric propulsion and the engine.
Hence, lately various heavy off-road vehicle manufac-
turers of mining and construction equipments are pay-
ing serious attention to transform their system to
electric hybrid type of propulsion. These systems offer
clean environmental advantages in addition to the fuel
economy. It must be noted that hybrid electric off-road
vehicles (HEV) are different than the regular automo-
tive HEV, and do not use normal roads for their oper-
ation. Mining vehicles, agricultural vehicles like
tractors, and some military vehicles also fall in this
category. Off-road vehicles will be discussed further in
the following section, which have quite different drive
cycles and speed-torque demands, as opposed to
a regular automobile.
Off-Road HEV
In the preceding sections off-road heavy vehicular
applications like mining and construction were men-
tioned. These vehicles require large transmission sys-
tem when equipped with only a diesel engine drive,
which has relatively low overall efficiency. These issues
can be circumvented through HEV system having over-
all efficiencies of the order of 75%. In an off-road HEV
a diesel engine is used to drive a generator or alternator,
which in general will generate a variable speed (hence
variable frequency) variable voltage (amplitude) sys-
tem. This variable frequency and variable voltage is
then converted to a DC voltage of constant value
using an AC to DC converter/rectifier. The constant
IC engine
Alternator / Rectifier /
filter
generator
Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 4
System-level architecture of a diesel electric HEV
599
DC voltage is then converted to a three-phase AC
system (inverter) to drive a three-phase AC motor.
The inverter can be electronically controlled to deliver
desired three-phase voltage (amplitude, frequency, and
phase) to drive an electric motor to handle the vehicle
load. It should be noted that in the system just
described, one can avoid the use of a battery. Although
battery is a vital storage element in a regular HEV, in
very large off-road vehicles, the battery system can be
avoided to reduce cost and package size. In addition, in
certain types of vehicles sufficient energy is not avail-
able for storage during the vehicle operation. Such
a system is very similar to a diesel electric locomotive
system.
Various architectures of the off-road and other con-
struction vehicle systems are possible. Two possible
systems are shown below. In this architecture the IC
diesel engine, in general, is operated at its optimal
speed to achieve maximum efficiency. IC engine drives
an alternator/generator, which generates a variable
three-phase AC voltage (amplitude and frequency).
This is rectified and a DC is produced. The DC power
is then fed to the inverter to produce desired voltage
(amplitude and frequency) to drive a propulsion
motor. In general, for very heavy mining applications,
induction motors are found to be more robust and
suitable in rough terrain/environment. The motor is
then used to drive the vehicle wheels. Generally, large
off-road vehicles have two axles and two motors (one
on each axle) for propulsion.
In the architecture shown in Fig. 4, the use of
battery is avoided. In principle, a battery could be
used for energy storage in vehicles, but the size of the
battery in that case would be large. In some applica-
tions an ultracapacitor bank can also be used. The
mechanical energy, during slowing down of the vehicle
or some of its movable parts, can be stored in the
ultracapacitor as electrical energy. The ultracaps
Inverter Motor Wheel
600 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
normally have high specific power and low specific
energy, that is, it can provide very high burst of power
input to a device for a short duration. During the slow
down, the controller commands the ultracapacitor,
inverter, and the alternator/rectifier to coordinate
properly, so that energy flows through the motor and
inverter (with the motor operating as a generator) into
the ultracapacitor. The stored energy can be utilized to
accelerate the vehicle, if so needed at a later time, to
save energy and thus leading to better system efficiency
and fuel economy. An architecture for such a system
noted above is shown in Fig. 5. In this architecture to
drive the motor, the DC power could be provided by
the ultracapacitor or the IC engine/rectifier. The system
will have the necessary switches to select either the
ultracapacitor or the alternator/rectifier or both.
It will be interesting to visit some typical sizes
involved in such off-road vehicles discussed above.
For example [2–4]:
● A Caterpillar engine model Cat 3524B EUI has gross
power 2,648 kW (3,550 HP), machine gross oper-
ating weight 623,690 kg (1,375,000 lb).
● A particular Komatsu has gross horsepower
2,611 kW (3,500 HP), and weight of 505,755 kg
(1,115,000 lb).
● A particular Liebherr has gross horsepower @
1,800 rpm of 722 kW (3,650 HP), 592 t (British)/
652.5 t (US).
The sizes of the tires of these vehicles are in general
of the order of 4 m, as is evident from the specifications.
This indicates the huge size of these vehicles, some
pictures of which are given in Fig. 6.
All the above vehicles, that is, Caterpillar, Komatsu,
and Liebherr, are transitioning to hybrid (HEV) power
IC engine
Rectifier /
Alternator
filter
Controller
Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 5
System-level architecture of a diesel electric HEV with ultracapacitor
train. Appendix 1 provides additional specifications
from the Komatsu information sheet [3] for vehicle
shown in Fig. 6b.
The traction motors in both Komatsu and Liebherr
are induction motors. The alternators are normally
brushless. The gear ratio, reduction from motor to
wheel, is of the order of 32–37. The gross engine horse-
power of the vehicles is about 2.6–2.7 MW. The inverter
used to drive the motor generally uses IGBT for
switches. Assuming alternator and inverter efficiencies
of the order of 0.95 each, the inverter power output (or
motor power input) is about 2.35–2.44 MW. So it is
reasonable to assume that each induction motor size has
to be of the order of 1.2–1.25 MW (assuming two
motors for the system, with one at each axle). The
vehicle at maximum speed amounts to a wheel revolu-
tion of about 84 rpm (assuming a tire diameter of about
4 m). With a gear ratio of about 32.62 the motor speed
should be about 2,700 rpm. Nominally, this leads to
about 3,000 rpm, which, at 50 Hz, leads to a four-pole
induction motor. Four-pole induction motor is nor-
mally a common standard for such applications.
Another application of HEV in current off-road
vehicles involves construction vehicles such as excava-
tors and scrapers (also known as “track”-type tractor).
It is expected that an array of other off-road vehicles
may transition to HEV technology, once the customers
start accepting the advantages offered by this
technology.
Presently in excavators the usage of HEV concept is
used in its arm movement, and is not used in vehicle
propulsion. The excavator arms need brief burst of
high power followed by low power return and then
slow down its speed to stop the arm. Such an applica-
tion is very well catered by using a diesel engine driving
Inverter Motor Wheel
Ultracapacitor
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles 601

a b

c Liebherr T282 B

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 6

Typical mining vehicles: (a) Caterpillar, (b) Komatsu, (c) Liebherr (Courtesy: Caterpillar http://catsays.blogspot.com/2005/
01/caterpillar-797b-mining-truck.html, http://www.cat.com/contact. Komatsu http://www.komatsuamerica.com/default.
aspx?p=equipment&f1=view&prdt_id=947, http://www.komatsuamerica.com/contact. Wikimedia http://commons.
wikimedia.org/wiki/File:Liebherr_T282.jpg [2–4])

a generator to produce electricity to drive the excavator
arm motor, which is an electric drive. The possible
architecture is shown in Fig. 7.
This architecture is basically the same as the HEV
architecture shown previously for the mining truck
(or hauler) propulsion. Only difference in this case is
that the propulsion size is much smaller. It has been
claimed that the Komatsu hybrid excavator leads to an
additional fuel economy of order of 25% and that some
users have achieved fuel economy enhancement as high
as 41% for specific applications. A picture of the
Komatsu PC 200–8 hybrid excavator is shown in Fig. 8
[5].
Presently, excavators are considered to be good
candidates for the installation of energy storage
devices such as battery or ultracapacitors. This is per-
haps due to size of the machines which require smaller
and lower cost storage devices as compared to mining
trucks. Some manufacturers prefer to use battery
instead of ultracapacitor for the purpose of storage.
An example is the New Holland’s excavator, which uses
a 36 HP diesel engine, 20 kW generator, and a 288
V lithium battery. The choice of storage is highly
dependent on exact usage and duty cycle. If it is just
a high-power excavator, then ultracapacitor seems to
be a suitable candidate. However, if in addition to
excavator application, there are other auxiliary equip-
ments, which call for energy that may not be immedi-
ately available from an ultracapacitor, a battery may be
used. The duty cycle in terms of torque requirement,
duration of the torque, auxiliary equipments, cost,
size, and similar factors all together leads to the deci-
sion process in terms of what will be best, that is,
ultracapacitor or battery.
602 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Diesel

Rectifier / Excavator
Alternator Inverter Motor
filter arm coupling

Controller Ultracapacitor

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 7

System-level architecture of an excavator with hybrid electric technology and using ultracapacitor

Electric power
assistance at the
The upper structure time of engine
turning acceleration

Electric motor for Power generation

the turntable of the motor

upper structure
Collects energy when turning
slows down in loading
operation
Engine
Inverter
Utilizes electricity discharged
Capacitor from the capacitors when the
Efficiency and engine accelerates
instantaneously store and
discharge electric energy

Komatsu Hybrid System

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 8

Komatsu excavator system (Courtesy: Komatsu news item http://www.komatsu.com/CompanyInfo/press/
2008051315113604588.html [5])

Track-Type Tractor Scraper/Bulldozer Another
application of off-road construction and agricultural
equipment is a bulldozer or scraper. This was first
introduced by Caterpillar in late 2009 [11] and should
be available in 2010. The structure of the vehicle, the
D7E model is shown in Figs. 9 and 10 [11, 12].
The propulsion is achieved by a diesel electric sys-
tem with a net power of 175 kW. Here a diesel engine
drives a generator. The AC generator and propulsion
module voltage is 480 V. The electrical power from the
generator is processed by an inverter, which drives
a motor. D7E has three-phase induction motors for
its drive [13]. It has been indicated [11] that the cost
of this model is 20% more than its previous D6 and D8
models. This 20%, which is about $100,000, is claimed
to be recoverable in 2 years due to fuel savings of about
10–30%. Performance of this is also claimed to be
better than its previous versions. While gas mileage is
of importance, in this type of machines the efficiency is
measured through how much material can be moved
per gallon of fuel, because these machines are not
predominantly vehicles in the sense of traveling dis-
tances. In terms of materials moved, it is apparently
10–30% more efficient than its nonhybrid equivalent.
This machine does not have battery for energy storage,
and the power from the generator is directly processed
by the inverter and transferred to the motor. The
machine can apparently save about 6 gal of fuel over
an 8-h period, which can be significant if it is used
continuously throughout the year.
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles 603

Modular electric
air conditioner
Electric cables

Generator
Power inverter

Acessory power
converter

Final drive

Electric motors

Dedicated steering pump

Beltless engine

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 9

Caterpillar D7E propulsion system (Courtesy: Caterpillar Information Brochure Web article, http://gas2.org/2009/09/18/
caterpillar-builds-worlds-first-hybrid-bulldozer/, Web article, http://www.cleanmpg.com/forums/showthread.php?
t=33855, http://www.cat.com/contact [11, 12])

Off-Road Military Vehicles for logistic vehicles; hence the impact of fuel economy
is even higher, compared to commercial vehicles.
In addition to commercial off-road vehicles, military
Hybridization can also lead to redundancy (particu-
application also includes similar vehicles. These vehi-
larly for parallel architecture with both IC engine and
cles, in principle, have similar requirements.
electric propulsion), which can have positive impact
Several pictures of off-road military vehicles are
on vehicle survivability in case of partial failure. It is
shown in Fig. 11 as illustration and to help readers get
also important to note that military environment can
visual perspectives of these machines [14–20].
involve a wide temperature range and high NVH
Figures above show vehicles which are regular pro-
(noise, vibration, harshness). Power electronics and
pulsion type, and could be enhanced by HEV technology.
electric motors have to withstand such situations.
From the various types of vehicles shown in the
These are important enabling technologies which
pictures above, it must be noted that these vehicles
have to undergo further transformation to make
have different operating requirements in terms of
HEV technology to gain wide acceptance in military
their drive cycles and power when compared to regular
applications.
automotive applications.
Some vehicles run long distances, and some have
Summary of Off-Road HEV
extreme power demand and performance needs, such
as survivability and fuel economy. Introduction of Regarding the rating of the motor for the mining
hybrid propulsion can lead to fuel economy, which truck – nominal DC link rating of these motors can
has direct and indirect benefits. Direct benefits come be around 2,600 V, leading to about 1,600 V line-to-line
in the form of fuel economy, which leads to cost rms input to the three-phase motor. A 1.2 MW motor
savings. However, for military applications, carrying corresponds to about 541 A AC rms, assuming a power
of fuel to the field involves additional cost and need factor of about 0.8. In this case the DC link current
604 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Power inverter

Accessory
power inverter

Electric
cables

AC generator

Propulsion
module

Two ac electric
Final drive motors for
propulsion
235 hp, 9.3
liter engine

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 10

Details of the diesel electric power train organization in Caterpillar D7E (Courtesy: Caterpillar Information Brochure Web
article, http://gas2.org/2009/09/18/caterpillar-builds-worlds-first-hybrid-bulldozer/, Web article, http://www.cleanmpg.
com/forums/showthread.php?t=33855, http://www.cat.com/contact [11, 12])

for just one motor will be 1.2 MW/2,600 = 461 A DC
per motor. All these are assuming no additional loss in
the system and the numbers are merely giving an idea
of the voltage and current values. Some systems [6] use
3.3 kV IGBT technology to drive these motors, which is
consistent with the above ratings indicated for the
motor. For the excavator example above the numbers
are much lower. As per preceeding paragraph, at
288 V DC, the DC link current will be about 70 A,
the three-phase line-to-line rms for the motor will
be about 0.612  288 = 176 V, motor current will be
about 82 A.
Diesel Electric Locomotives
Diesel engine–based locomotives are commonly used
nowadays. In diesel electric locomotives the propulsion
system is electric as explained through the block dia-
gram in Fig. 12. There are advantages why diesel electric
system is used, as explained in sections “Introduction”
and “Series, Parallel, and Complex Architectures for
Ground Vehicle Applications” (i.e., if the propulsion
was purely mechanical, then a big-sized transmis-
sion system with gearbox etc. will be needed to
create the needed torque). A second advantage in
using electric propulsion system in locomotives is
that the electric propulsion system provides contin-
uously variable speed system. This is particularly
advantageous in improving the transmission effi-
ciency of short haul trains where speed fluctuations
are encountered frequently. Otherwise, a diesel
engine with a finite number of gears is required to
be operated at different speeds other than the most
optimum speed at which the efficiency is highest. In
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles 605

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 11

(a) Stryker (http://commons.wikimedia.org/wiki/File:14_stryker.jpg) (b) HEMTT (http://commons.wikimedia.org/wiki/File:
HEMTT.jpg) (c) Bradley (http://commons.wikimedia.org/wiki/File:Two_M-3_Bradleys.jpg) (d) Abrams (http://commons.
wikimedia.org/wiki/File:M1A1_Abrams_Tank_in_Camp_Fallujah.JPEG) (e) Fennek (European) (http://commons.
wikimedia.org/wiki/File:BundeswehrFennek.jpg) (f) MRAP (http://commons.wikimedia.org/wiki/File:
FBI_Mine_Resistant_Ambush_vehicle.jpg) (g) Gladiator (http://commons.wikimedia.org/wiki/File:Gladiator_RSTA.jpg)
Courtesy of the websites indicated in the above references [14–20]

this system (HEV), the diesel engine in general runs
at the most optimal speed and drives the generator
to produce electricity, which is used to run
a traction motor mechanically connected to run
the wheels as shown in Fig. 9. With the advent of
power electronics, this system is rather easy to realize,
using reliable and efficient traction motors, which can
be induction or synchronous motor – either perma-
nent magnet or field-excited types. With this back-
ground, let us look at the architecture of the diesel
electric system shown in Fig. 12.
In general, the traction motors are placed on each
axle of the locomotive to drive the wheel pairs. These
traction motors used in heavy locomotives are AC
induction motors. The squirrel cage induction
motors are a very reliable device for such applica-
tions, due to their robust designs and structures and
resilient to temperature conditions, unlike perma-
nent motors.
A picture of a diesel electric locomotive is shown in
Fig. 13 [7].
Its diesel engine is rated at 4,000 HP and power
at wheels is 3,350 HP. Six motors are used in this
engine, with one per axle. The motors are four-pole
squirrel cage three-phase induction motors, with
a maximum voltage of 2,030 V and 433 kW
606 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Diesel Main Main Traction

generator set rectifier inverter motors

Inverter to
Auxiliary
supply Rectifier
battery
auxiliary
power

HEP Motor
inverter blowers

Coach Air
loads compressor

Locomotive
cooling fans

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 12

System-level architecture of a diesel electric locomotive

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 13

A large diesel electric locomotive (Courtesy: Siemens Technical Information Brochure http://www.siemens.cz/extra/msv/
cd/TS/Vozidla/Diesel-Electric_Loco_SD70MAC_EN.pdf, contact.mobility@siemens.com [7])

continuous power rating. The net motor power
rating is 2,598 kW, or about 3,500 HP. The maxi-
mum speed of the motor is 3,220 rpm, and there is
a gear ratio of 85:16 between the motor and the
wheel. Although specific power density of induction
motors can be somewhat lower than permanent
magnet motors, in locomotive applications, where
the system package size may not be an important
consideration (as in small passenger cars), this may
be an ideal choice. As shown in block diagrams, in
general, no traction system batteries are presently
used in locomotive applications. To start the diesel
engine, either electric, hydraulic, or pneumatic
starting mechanism can be used. In case of electrical
start, a small starter battery is needed.
Recent advances in technology, for capturing regener-
ative energy, involve some storage battery, ultracapacitor,
or a combination of both. Even a flywheel storage unit,
can be considered for regeneration. Regeneration can help
improve fuel economy and is of more importance for
short haul trains and may not be as important for a long
haul train.
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles 607

Ultracapacitor
Propulsion battery

Diesel Main Main Traction

generator set rectifier inverter motors

Auxiliary
Inverter to Auxiliary
system
supply rectifier
battery
auxiliary
power

Motor
HEP blowers
inverter

Air
compressor
Coach
loads

Locomotive
cooling fans

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 14

System-level architecture of a diesel electric locomotive including the usage of ultracapacitor

The architecture of a locomotive system capable
of regeneration is shown in Fig. 14. The system is
essentially the same in principle as the one in Fig. 12,
except that it has a propulsion battery and/or
ultracapacitor, which can provide energy to the pro-
pulsion motor.
Hybrid Hydraulic Vehicles
Nonelectric hybrid vehicles, specifically hybrid hydrau-
lic vehicles (HHV), are being considered by the techni-
cal community as viable and can sometimes be more
beneficial than an electric hybrid vehicle.
Energy Storage
As compared to electric energy storage system, for
example, battery in HEV, in a hydraulic system the
energy is stored in the form of compressed gas in
a cylinder. To pressurize a gas, one needs power and
energy, which comes from the IC engine to activate the
hydraulic pump. While extracting the energy one can
use a hydraulic motor. In other words, the hydraulic
pump is analogous to an electrical generator, since the
latter converts mechanical power into electrical,
whereas hydraulic pump converts mechanical power
into fluid power. Similarly, hydraulic motor is analo-
gous to an electric motor, and the pressurized gas in the
cylinder is analogous to a battery. The efficiency of the
hydraulic pump, motor, and storage is around 90%.
To illustrate the hydraulic hybrid system, one can
compare Figs. 15 and 16.
For the hybrid hydraulic vehicle architecture shown
above, the correspondence with its electrical system
counterpart is shown in Table 1.
Figure 17 shows a full series hybrid hydraulic power
train configuration [8], which corresponds with the
system-level diagram shown in Fig. 16. This includes
a high-pressure (HP) accumulator (Appendix 2)
containing benign gas such as nitrogen, which can be
compressed or expanded by means of incompressible
hydraulic fluid acting through some piston type of
608 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Additional mechanical
device

Fuel

ECU ICE Gear Motor/alternator Motor Gear/tran Wheel

saxle
Parallel HEV
propulsion HEV Power Propulsion Power
control electronics battery electronics

Motor control

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 15

System-level diagram of HEV

Additional mechanical
device

Fuel

Gear/tran Wheel
ECU ICE Gear Hydraulic pump Hydraulic motor
saxle

Parallel HHV
propulsion
HHV Valves Accumulator Valves
control

Hydraulic
motor control

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 16

System-level diagram of HHV

mechanism. The pressure in HP cylinder and the pipe-
line can be as high as 3,000–5,000 psi, whereas the low-
pressure cylinder and the corresponding pipeline
containing incompressible fluid can be of the order of
a few hundred psi. The IC engine drives a pump, which
converts the pressure from low to high as the fluid is
moved from LP to HP cylinder. To drive the vehicle
wheels, the high-pressure gas from the HP cylinder may
be allowed to expand and the hydraulic fluid is moved
through the motor drive assembly. The hydraulic
motor takes the high-pressure fluid in its input port,
and converts it to mechanical power at the wheels.
When the fluid has passed through the hydraulic
motor, its pressure drops and the low-pressure fluid is
transferred to the LP cylinder.
The amount of energy stored in the hydraulic accu-
mulator system is quite low in terms of kW-h/kg. As
a comparison, the specific energy density in hydraulic
accumulator can be of the order of 1.9 w-h/kg [Ref
EPA], whereas a battery can have specific energy den-
sity of 30 w-h/kg. However, the specific power density
of hydraulic system can be of the order of 2,500 W/kg,
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles 609

whereas the electrical system–specific power density engine, the power generated will be used effectively in
(e.g., a battery) can be about 650 W/kg. Therefore, propulsion. However, if the energy coming from the IC
the hydraulic hybrid system can be more suitable for engine is greater than the propulsion load, the excess
high-power and relatively low-energy applications, energy will go to the accumulator for storage. The energy
particularly where short bursts of high-power acceler- stored in the hydraulic accumulator is rather low, hence
ation and deceleration are involved. this option cannot continue too long, since excessive
It is evident from the earlier discussions that an IC gas pressure in the accumulator will then be needed,
engine can run at optimum efficiency to drive the hybrid which can be difficult to sustain. If the HP accumulator
hydraulic system to deliver the energy to the wheels. cannot store any additional energy, then it will be
A few scenarios can take place here. If the energy needed necessary to change the engine operating point, so as
by the vehicle propulsion matches the energy from the IC to match the propulsion needs. Alternatively, it may be
possible to stop the engine for a short while and use the
energy from the hydraulic accumulator, and then start
Hybrid Electric and Hydraulic Technology Applications
the engine again. As long as the average energy con-
in Off-Road Vehicles. Table 1 Equivalence of major
sumption by the motor is same as the average energy
components between hybrid electric and hybrid hydraulic
delivered by the engine, the process is sustainable, with
vehicular systems
transients absorbed by the accumulator. If there is no
energy storage accumulator in the system, then in prin-
Electrical system ciple it is possible to keep the engine running at an
component Hydraulic system component
optimal speed-torque point. In that case it will be
Alternator Hydraulic pump necessary to switch the hydraulic valves on and off in
Electric motor Hydraulic motor a way such that the average fluid flow corresponds to
Battery Hydraulic accumulator the average power and energy used by the hydraulic
motor. By following this process, it may not be neces-
HEV controller Hybrid hydraulic vehicle (HHV)
controller
sary to perform any stop-and-go operation by the
engine. But this will lead to a pulsating torque at the
Power electronic Hydraulic valve system
motor shaft output, whose average value will corre-
system
spond to the average load demand.

HP accumulator

Hydraulic
drive
assemble

LP reservoir

Engine pump/motor

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 17

Series HHV system architecture (Courtesy: EPA Kargul, J.J. “Hydraulic hybrids – demonstration for port yard hostlers,” in
EPA presentation, www.epa.gov/otaq/technology, July 11, 2007 [8])
610 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Regenerating mode
Propulsion mode

82% Efficiency Pump

Vehicle
Shaft
braking
92 100 kinetic
Accumulator energy
(92%)
100
Wheel

Hydraulic
motor
Reservoir Shaft

90 82
(98%) (91%)
Wheel

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 18

Regenerative braking efficiency in HHV (Courtesy: EPA US EPA, “Clean Automotive Technology. . . Innovation that Works,”
in www.epa.gov/otaq/technology World’s First Full Hydraulic Hybrid SUV, Presented at 2004 SAE World Congress [8]]

One of the major advantages of using hydraulic
system for propulsion is the fact that it can eliminate
the need for a massive conventional transmission sys-
tem. Also, hydraulic fluid can be more easily moved
from one point to another without the need for an
elaborate transmission gearbox etc. However, this con-
cept of transmission can be used in a regular nonhybrid
vehicle as well, where only the transmission system
could be hydraulic. But as described in the previous
paragraph, to be able to drive an IC engine at its most
optimum point and at the same time eliminate torque-
speed pulsations at the load will need some sort of
storage, which can be an accumulator. Without
a storage, it is, in principle, possible to realize the
optimum operating point of the engine by switching
the valves on and off to match the average load
demand, but the output torque-speed at the load will
then be pulsating, which can be smoothed by using
a flywheel etc.
Regenerative Braking in Hybrid Hydraulic Vehicles
Hybrid hydraulic vehicles offer the benefit of regener-
ation and the ability to capture the regenerative energy
(which would have been otherwise wasted) during
slowing down due to braking and later use it to
accelerate again. The hydraulic system–specific energy
density or W-h/kg is relatively low compared to
a battery. However, hydraulic system can be quite ade-
quate for storing the regenerated energy during brak-
ing, since power is high but energy involved is typically
low in this situation. Figure 18 shows the regeneration
process [8].
In Fig. 18, the given efficiency of both the hydraulic
pump and the motor are little over 90%. The efficiency
of the accumulator (HP and LP together) is about 98%.
Hence, the efficiency of the whole regeneration process
is 82%.
One advantage of the hybrid hydraulic system is the
maturity of the hydraulic system technology. Compo-
nents used in a hydraulic system, shown in the figure
above, have very high efficiency. Figure 19 shows a typ-
ical hydraulic motor/pumps (hydraulic pump and
hydraulic motor are similar in appearance and their
speed-torque characteristics) used in hybrid hydraulic
vehicles. This figure shows a bent-axis hydraulic motor,
which is from a Bosch-Rexroth specification sheet [9].
For this motor the nominal gas pressure at the entrance
can be over 5,000 psi, maximum speed can be
4,500 rpm, and torque @ 5,100 psi can be 371 lb-ft.
The length of the unit excluding the geared shaft is
about 5.15 in.
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
Hybrid Electric and Hydraulic Technology Applications
in Off-Road Vehicles. Figure 19
Bent-axis hydraulic pump/motor (Bosch-Rexroth)
(Courtesy: Bosch-Rexroth Rexroth – Bosch Group, “Axial
Piston Fixed Displacment Motor AA2FM (A2FM),” in
Product brochure from Bosch. https://www.boschrexroth.
com/various/utilities/contact/index.jsp?
cont_country=5&oid=3901 [9])
Future Directions
All people (technical, management, and nontechnical
alike) involved with Advanced Heavy Hybrid Propul-
sion Systems need to work on developing and vali-
dating cost-effective, advanced next-generation
heavy hybrid components and systems, which at the
vehicle level could enable a power train efficiency
increase as high as 100% relative to today’s conven-
tional power train technology. It is important to focus
on the development and improvement of various
interdisciplinary technologies, systems, and tools,
such as:
● Improved diesel engine concepts: Advanced and novel
concepts in the areas of hybrid diesel engines, fuel-
injection systems, combustion- and emission-
control technologies, lightweight materials, partic-
ularly for diesel engine structure, to withstand high
pressure, and other novel hybrid diesel engine con-
cepts with the potential of increased fuel efficiency
and reduced emissions.
● Low-cost, lightweight, high temperature
withstandable power electronics: Heavy hybrid
vehicles will require low-cost, lightweight,
611
simpler power electronic systems that can be
easily integrated into off-road vehicular system
architectures. It will also call for power electronics
that can withstand very high temperature
environment.
● Energy conversion for auxiliary power and energy
storage technologies and systems: Although auxiliary
systems are not used for propulsion, newer vehicles
have significant amount of nonpropulsion power
demand. Hence, new technologies and concepts in
areas like integrated thermoelectric systems, photo-
voltaic, and even wind energy type of technologies
are important.
● Energy storage technology: Battery, ultracapacitor,
and flywheels need significant development to sup-
port heavy vehicle hybridization.
● Advanced transmission and power train compo-
nents and systems: Better continuously variable
transmissions (CVTs) technology and their
integration into heavy hybrid applications are of
importance, which can help the hybridization
process.
● Advanced materials for hybrid power trains and vehi-
cle systems: High-temperature materials, composites,
and various polymer materials that lead to enhanced
energy use and management, and reduced compo-
nent and system weight and volume.
● Rare-earth free materials: One important area that
needs to be attended to is the development of
electric motor and machines technology, partic-
ularly those involving permanent magnet
machines, without using rare-earth materials.
This can help reduce the use of strategic
elements, and also can help achieving high
temperature withstanding ability by these electro-
mechanical devices. In fact, rare-earth free mate-
rials are important for some advanced battery
systems, hence this area will help with battery
technology.
Appendix
Appendix 1: Specifications of Komatsu Mining Truck
Information given below is taken from the Komatsu
data sheet [3]. Although exact details on the systems are
normally proprietary information of the manufac-
turers, some commonality can be noticed between
612 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles

Outer cylinder

V
V
Bladder
B Diaphragm
D

a b

Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles. Figure 20

(a) Bladder accumulator (b) Diaphragm accumulator

these systems and some reasonable idea about the sizes whereby energy is stored. Piston separates the liquid
of the components can be obtained, based on the infor- from the gas.
mation in these publicly available specification sheets. In the diaphragm accumulator a diaphragm B is
pushed by some liquid from the bottom, which then
AC current changes the volume of the gas in the upper part of B,
Alternator GTA-39
that is, between the diaphragm and the cylinder itself.
In a bladder accumulator, the bladder B contains the
Dual impeller in-line 453 m3/min 16,000 cfm gas which can be compressed. When the liquid enters
blower
the cylinder from the bottom, the bladder can expand
Control AC Torque Control System or contract, thus energy can be stored in the gas when
Motorized wheels GDY106 AC Induction Traction compressed and taken out from the gas when
Motors expanded. The energy is transferred through the
Ratio 32.62:1 fluid itself.
Speed (maximum) 64.5 km/h 40 mph

Bibliography
Appendix 2: Accumulators
Accumulators basically consist of cylinders which have
valves to let the gas in or out with appropriate safety
mechanisms. Other types of accumulators are the
bladder or diaphragm types as shown in Fig. 20a, b.
Bladder-type accumulator consists of a flexible
bladder submerged in an incompressible fluid.
Inside the bladder there is a gas, which can be
pressurized to compress or depressurized to expand.
There are several working mechanisms for the
accumulators, for example, in the hydropneumatic
scheme the liquid pushes a piston, which then
pushes a gas stored in the cylinder and compresses it,
1. Ucarol H, Kaypmaz A, Tuncay RN, Tur O (2005) A performance
comparison study among conventional, series hybrid and
parallel hybrid vehicles. web article. http://www.emo.org.tr/
ekler/c792a8279211dec_ek.pdf
2. Caterpillar. http://catsays.blogspot.com/2005/01/caterpillar-
797b-mining-truck.html
3. Komatsu product brochure. http://www.komatsuamerica.
com/default.aspx?p=equipment&f1=view&prdt_id=947
4. Wikimedia http://commons.wikimedia.org/wiki/File:Liebherr_
T282.jpg (also see Liebherr product brochure, http://www.
liebherr.com/ME/en-GB/products_me.wfw/id-14270-0
5. Komatsu product brochure. http://www.komatsu.com/
CompanyInfo/press/2008051315113604588.html
6. Brown GN, Ebacher BJ, Koellner WG, Increased productivity
with AC drives for mining excavators und haul trucks. http://
info.industry.siemens.com/data/presse/docs/2050e.pdf
 Hybrid Electric and Hydraulic Technology Applications in Off-Road Vehicles
7. http://www.siemens.cz/extra/msv/cd/TS/Vozidla/Diesel-
Electric_Loco_SD70MAC_EN.pdf
8. Kargul JJ (2007) Hydraulic hybrids – demonstration for port
yard hostlers, in EPA presentation. www.epa.gov/otaq/tech-
nology, Accessed 11 July 2007
9. Bosch Rexroth product brochure, Bent axis hydraulic motor,
Rexroth – Bosch Group, Axial Piston Fixed Displacement Motor
AA2FM (A2FM), Bosch. https://www.boschrexroth.com/vari-
ous/utilities/contact/index.jsp?cont_country=5&oid=3901
10. Olaer product brochure, Diaphragm and bladder accumula-
tors, Olaer Inc., “Bladder and diaphragm accumulator
product brochure, including High-pressure diaphragm accu-
mulator conforming to EC regulations”, ELM Series. http://
www.olaer.fr/en/recherche.php?idr=13, http://www.olaer.fr/
en/recherche.php?idr=11
11. http://gas2.org/2009/09/18/caterpillar-builds-worlds-first-
hybrid-bulldozer/
613
12. http://www.cleanmpg.com/forums/showthread.php?t=33855
13. Kwak S, Kim T, Park G (2010) Phase-redundant-based
reliable direct AC/AC converter drive for series hybrid off-
highway heavy electric vehicles. IEEE Trans Veh Technol
59(6):2674–2688
14. http://commons.wikimedia.org/wiki/File:14_stryker.jpg (Stryker)
15. http://commons.wikimedia.org/wiki/File:HEMTT.jpg (HEMTT)
16. http://commons.wikimedia.org/wiki/File:Two_M-3_Bradleys.
jpg (Bradley)
17. http://commons.wikimedia.org/wiki/File:M1A1_Abrams_Tank_
in_Camp_Fallujah.JPEG (Abrams)
18. http://commons.wikimedia.org/wiki/File:BundeswehrFennek.
jpg (Fennek - European)
19. http://commons.wikimedia.org/wiki/File:FBI_Mine_Resistant_
Ambush_vehicle.jpg (MRAP)
20. http://commons.wikimedia.org/wiki/File:Gladiator_RSTA.jpg
(Gladiator)
614 Hybrid Energy Storage Systems for Vehicle Applications

Hybrid Energy Storage Systems for have different characteristics, may result in a high-
performance energy storage system.
Vehicle Applications
MEHRDAD EHSANI
Introduction
Department of Electrical & Computer Engineering,
Texas A&M University, College Station, TX, USA It is widely agreed that vehicle electrification will lead to
revolutionary improvements on vehicle performance,
Article Outline energy resource conservation, and pollution emissions.
High-quality energy storage system is one of the most
Glossary crucial components. Unfortunately, the currently avail-
Definition of the Subject able chemical, electrical, and mechanical energy storage
Introduction systems, including various kinds of chemical batteries,
Electric Load Characteristics ultracapacitors, and flywheels, cannot meet these
Candidates of Vehicle Energy Storage requirements alone. In this entry, the possibility of com-
Hybrid Energy Storage posing a high-energy, high-power hybrid energy storage
Battery/Ultracapacitor System system is presented based on the analysis of inherent
Flywheel/Ultracapacitor System characteristics of different energy storage methods. The
Future Directions basic components in this system are chemical batteries,
Bibliography ultracapacitors, and flywheels (mechanical batteries).
Glossary The objective is to fully utilize the advantages and
suppress the disadvantages of each component to com-
Electric load A device or system connected to the pose a compact energy storage system with high energy
electrical power circuit for the purpose of demand- capacity and high power capability [3, 4, 6, 10].
ing power.
Energy storage A device or system capable of storing
energy in one of many physical forms. Electric Load Characteristics
Hybrid A combination of two or more items sharing The electric load in a hybrid vehicle comprises of trac-
a common function. tion load and non-traction load. Regarding traction
Hybrid energy storage A combination of two or more load, the energy storage is only responsible to supply
energy storage devices with complimentary a intermittent peak power which may be from a few
capabilities. seconds, such as in hard acceleration, steep hill
Non-traction load Power demand for all purposes climbing, obstacle negotiation, etc., to several minutes,
other than traction. such as in cross-country operation, medium hill
Traction load Power demand for the purpose of pro- climbing, etc. The vehicle in the former operation
pelling the vehicle. needs the energy storage to supply peak power in
a few seconds with high magnitude, whereas in the
Definition of the Subject
latter situation, the vehicle needs the energy storage
The electric load of a vehicle can be decomposed into device to supply power in several minutes with
two components – static and dynamic load. The static medium magnitude. In normal road operation, the
component is slowly varying power with limited mag- primary power source (engine or fuel cell) should be
nitude, whereas the dynamic load is fast varying power able to produce sufficient power to support the vehicle
with large magnitude. The energy storage system, operation and, at the same time, produce additional
accordingly, comprises of two basic elements. One is power to charge the energy storage device to recoup its
energy source to support the static load and other is past discharge [7].
a power source to support the dynamic load. A smart Similar to the traction motor load, the non-traction
combination of the available energy storages, which electric load can also be characterized by continuous

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3

power demand and intermittent peak power demand.
Non-traction electric load is the total of electric loads
that are used to serve the vehicle operation and auxiliaries.
For a conventional vehicle, the non-traction electric loads
consist of continuous loads, prolonged loads, and inter-
mittent loads. The continuous loads, includes engine
ignition, fuel injection, instrument panel, fuel pump,
etc. These may need around 200 W. The prolonged
loads, mainly including all the lights, may need about
260 W. Intermittent loads, mainly including, heater,
front and rear wipers, brake light, electric windows, radi-
ator cooling fan, heater blower motor, rear fog light,
electric seat movement, electric mirror, etc., may need
a total power of 1,700 W. The total of the non-traction
electric peak power could reach about 2,140 W. However,
the average is about 610 W. It is certain that more non-
traction electric loads will be added in near future vehicles.
They may include direct electrically driven power steering,
direct electrically driven air conditioning, direct electri-
cally engine cooling pump, electrically actuated braking,
electrically driven active suspension, electrically driven
engine valve actuator, etc. The total peaking power for
the future electric loads may reach 20–25 kW while the
average power may be only about 5 kW, one fourth or one
fifth of the peak power [10].
The average of the electric loads needs the energy
storage device to possess sufficient energy to support
the vehicle’s long-time operation, and the peak of the
electric loads needs the energy storage device to be
capable of delivering sufficient power in a short time
period (usually less than 1 min).
Candidates of Vehicle Energy Storage
As mentioned above, the basic requirement for
vehicle energy storage device is to have sufficient energy
and also be able to deliver high power for a short time
period. With the present technology, chemical batteries,
ultracapacitors, and flywheel system (mechanical battery)
are the main candidates for the vehicle energy storage
device.
Chemical Batteries
The chemical battery is an energy storage device that
stores energy in the chemical form and exchanges its
energy with outside devices in electric form. Basically,
a cell of battery has one positive electrode, which
Hybrid Energy Storage Systems for Vehicle Applications 615
accepted electrons, one negative electrode, which
gives out electrons and an electrolyte, which conducts
ions and isolates electrons between the electrodes. The
battery operation (discharging and charging) follows
principles of electrochemistry and thermodynamics
[2, 6, 9, 13].
Theoretical Specific Energy of Chemical Batteries
The amount of energy stored in a battery is completely
determined by the amount of active reactants inside the
battery, or more precisely, the free energy, DG, of the
chemical reaction. The commonly used terms to mea-
sure the energy capacity of a battery are Watt-hours per
kilogram (Wh/kg) called specific energy capacity and
Watt-hours per liter (Wh/L) called energy density. The
theoretical specific energy (maximum available) can be
expressed as
DG
Especific ¼ P ðWh=kgÞ; ð1Þ
3:6 Mi
where DG is the free energy of the reaction in a battery
cell and SMi is the sum of the molecular weight of
the individual species involved in the battery reaction.
The free energy, DG of a reaction is closely associated
with the number of electrons transferred in the reaction
and the electric potential (voltage), Vr , by
DG ¼ n F Vr ; ð2Þ
where n is the number of electrons transferred in the
reaction and F is the Faraday constant, 96,495 ampere-
seconds per gram-equivalent. Thus, the specific energy
can be written as
Vr F n
Especific ¼ P ðWh=kgÞ: ð3Þ
3:6 Mi
From theoretical electrochemistry and thermody-
namics, the reaction potential of a battery cell can be
described by
 
RT Pðactivities of products
Vr ¼ Vr0  ln ;
nF Pðactivities of reactants
ð4Þ
0
where Vr is a standard voltage at standard conditions
when the battery is fully charged, R is gas constant,
8314 J/ K. mole and T is absolute temperature. The
value in square brackets is nearly unity when the
616 Hybrid Energy Storage Systems for Vehicle Applications

battery is charged, increasing with the process of
battery discharging. When the reaction reaches
equilibrium, Vr becomes zero, and the battery is fully
discharged.
Practical Specific Energy of Chemical Batteries
Generally, practical batteries have much lower specific
energies than their theoretical values. Figure 1 shows an
example of the weight distribution of the components
of a lead-acid EV battery with a specific energy of
45 Wh/kg at the C5/5 rate. In this example, only
about 25% of the total weight of the battery is useful,
that is, the practical specific energy is only about 25%
of the theoretical one. Table 1 shows the data for some
existing batteries.
resistance. With the battery model as shown in
Fig. 2, the maximum power that the battery can supply
to the load is
V02
Ppeak ¼ ; ð5Þ
4ðRc þ Rint Þ
where Rc is the conductor resistance (Ohmic resistance)
and Rint is the internal resistance caused by chemical
reaction.
Internal resistance Rint represents the voltage drop,
DV, which is associated with the battery current. The
voltage drop DV, termed overpotential in battery ter-
minology, includes two components: one is caused by
reaction activity DVA and the other caused by electro-
lyte concentration DVC. General expressions of DVA
and DVC are

Specific Power of Chemical Batteries DVA ¼ a þ b log I ð6Þ

Specific power is the maximum power, per unit and

battery weight, that the battery can produce in  
RT I
a short period. The specific power of a chemical DVC ¼  ln 1  ; ð7Þ
nF IL
battery depends mostly on the battery internal
where a and b are constants, R is gas constant, 8314
J/  K. mole, T is the absolute temperature, n is the
number of electrons transferred in the reaction, F is
the Faraday constant, 96,495 ampere-seconds per
gram-equivalent, and IL is the limit current. Accurate
Pb + PbO2 + H2SO4 determination of battery resistance by analysis is
Cover
Container (Reacted) difficult. Specific powers are usually obtained by
Separator measurement.
Table 1 shows the status of battery systems poten-
tially available for EV [13]. It can be seen that although
Top Lead Pb + PbO2 + H2SO4 specific energies are high in advanced batteries, the
(Unreacted)
specific powers show no significant improvement.
About 300 W/kg might be the optimistic estimate.
Current Recent Li-ion high-power and high-energy batteries
collectors
Water in have been reported with as high as 4,000 and
electrolyte
150 Wh/kg, respectively. However, their practicality
and cost need much improvement.

Energy Loss and Thermal Management

Hybrid Energy Storage Systems for Vehicle
Applications. Figure 1
Weight distribution of the components of a lead-acid
EV battery with a specific energy of 45 Wh/kg at
the C5/5 rate [13]
The presence of internal resistance in batteries causes
energy loss in the battery discharging and charging
process. This dissipated energy causes the battery tem-
perature to rise, shortening the battery life. The energy
loss can be expressed as
 Hybrid Energy Storage Systems for Vehicle Applications 617

Hybrid Energy Storage Systems for Vehicle Applications. Table 1 Status of battery systems for EV
Specific energy (Wh/kg) Peak power (W/kg) Energy effi.(%) Cycle life Cost ($/kW.h)
Acidic aqueous solution
Lead-acid 35–50 150–400 >80 500–1,000 120–150
Alkaline aqueous solution 75 800 250–350
Nickel/cadmium 40–60 80–150 65 1,500– 200–400
2,000
Nickel/iron 50–60 80–150 70 300 100–300
Nickel/zinc 55–75 170–260 70 750–1,200+ 200–350
Ni-metal-hydride 70–90 200–300 <50 ? ?
Aluminum/air 200–300 90 60 500+ 50
Iron/air 80–120 90 60 600+ 90–120
Zinc/air 100–200 30–80
Flow
Zinc/bromine 70–85 90–110 65–75 500–2,000 200–250
Vanadium redox 20–30 110 75–85 - 400–450
Molten salt
Sodium/Sulfur 150–240 230 85 800+ 250–450
Sodium/Ni/Cl 90–120 130–160 80 1,200+ 230–345
Li-ion-sulfur (FeS) 100–130 150–250 80 1,000+ 110
Organic/lithium
Lithium-ion 80–130 200–300 >95 1,000+ 200
Li-ion high power a
85–95 4,000 – – –
Li-ion high energy a
135–150 600 – – –
a
Reported by SAFT [14]

Rc I Eloss ¼ Rbatt I 2 ; ð8Þ

where Rbatt = Rc + Rint is the battery internal resistance
Rint(I) and I is the battery current. The battery resistance is
RLoad
usually related to the battery temperature and battery
state-of-charge. Low temperature usually leads to high
V0 VLoad battery resistance because of less reaction activities of
the reactants. Thermal management is usually needed
to maintain the battery in an appropriate temperature
range. Discharging at low state-of-charge and charging
Hybrid Energy Storage Systems for Vehicle at high state-of-charge can also lead to high battery
Applications. Figure 2 resistance because of the nonuniformity in the concen-
Battery circuit model tration of the reactants. Thus, the battery should be
618 Hybrid Energy Storage Systems for Vehicle Applications

managed to operate in a middle range of state-of- Hybrid Energy Storage Systems for Vehicle
charge, such as 40–60%. Furthermore, high charging Applications. Table 2 Characteristic data of an
and discharging current should be avoided so that the ultracapacitor
energy loss in battery can be reduced (refer to Eq. 8), Rated capacitance [F] 145
and consequently, the thermal management can be (DCCa, 25 C)
simplified. Capacitance tolerance [%] 20
Rated voltage [V] 42
Ultracapacitors Surge voltage [V] 50
The ultracapacitor is another candidate for energy Max. series resistance, [mO] 10
storage in the hybrid vehicle. The ultracapacitor is ESR (DCC, 25 C)
characterized by high specific power, high efficiency, Specific power [W/kg] 2,900
excellent temperature adaptability, and long service density (42 V)
life. However, it suffers from very limited specific Max. current [A] 600
energy. The characteristics of a 42-V ultracapacitor Max. stored energy [J] 128,000
pack specifically designed for HEV are listed in Table 2
Specific energy [Wh/kg] 2.3
[11]. Reference [8] has reportedly developed 2700-F density (42 V)
ultracapacitors with 2.5 V of cell voltage, 2.55 kW/kg
Max. leakage current [mA] 30
of specific power, and 3.23 Wh/kg of specific energy. It
(12 h, 25 C)
is believed that the unit power cost of the ultracapacitor
is lower than that of batteries. However, due to its low Weight [kg] 15
specific energy and its high dependence of the state-of- Volume [L] 22
charge on voltage, it is difficult to use ultracapacitors 
Operating [ C] 35 to +65
alone as the energy storage on hybrid vehicles. The temperature
most promising approach is to hybridize the Storage temperature [ C] 35 to +65
ultracapacitors with other energy storage devices, 
Life time (25 C) [y] 10 DC <20% of initial
such as batteries or flywheels as will be discussed in value, ESR <200% of
following sections. initial value
Cyclability 500,000, DC <20% of
Flywheel (25 C, I = 20 A) initial value, ESR
<200% of initial value
Flywheel is an energy storage device which directly a
DCC discharging at constant current
stores mechanical energy in an extra high-speed rotat- Source: Ref. [11]
ing wheel. Because of the difficulty of using the
mechanical energy of the flywheel directly for traction,
a high-speed electric machine is usually connected to
the flywheel to constitute a system, called “mechanical Energy Capacity of Flywheel System
battery,” as shown in Fig. 3. The flywheel can be The energy stored in a high-speed flywheel is
charged and discharged through the electric machine. expressed as
The major advantages of flywheel system are their
1
relative high energy capacity and low environmental Ef ¼ Jf o2f ; ð9Þ
dependence, especially temperature. However, high 2
speeds require high-strength materials, special design where Jf is the moment of inertia of the flywheel in
bearings (mostly magnetic bearings), and high vacuum kgm2 and of is the angular velocity of the flywheel in
housing to reduce the aerodynamic drag. These rad/s. High energy capacity demands high speed (usu-
requirements may cause technical difficulties in their ally 50,000–60,000 rpm) and high moment of inertia
application to vehicles. from the flywheel.
 Hybrid Energy Storage Systems for Vehicle Applications 619

I The theoretical specific energy of a flywheel can be

Power calculated by Eqs. 9 and 10 for the geometrically
electronics V
specified flywheel in Fig. 4. Figure 6 shows the theoret-
ical specific energy of the flywheel versus the flywheel
geometric size r1 and flywheel angular speed. This
Axle figure indicates that with increasing flywheel size, r1,
and angular speed, the specific energy can be increased
Housing Rotor significantly. For a midsize vehicle, r1 = 035 m and
flywheel Electric Stator
rpm = 50000 may be good estimates. In this case, the
machine theoretical specific energy of flywheel will reach about
300 W/kg.
Hybrid Energy Storage Systems for Vehicle
In practice, the effective mass used to store energy
Applications. Figure 3
in a flywheel system only takes part of the total system
Configuration of a flywheel system (mechanical battery)
mass. The ineffective masses, such as the housing, the
stator of the electric machine, the bearings, etc., and the
less effective mass, such as the rotor of the electric
b1
machine, may make up to over half of the total system
mass. Thus, the practical specific energy of a flywheel
A system is much lower than its theoretical value and may
r2 = 0.65r1
r3 = 0.1r1
be in the range of 100–150 Wh/kg, which is in the same
b2
b1 = 0.5r1
range as that of advanced chemical batteries, such as
r1
B b2 =0.15r1 NiMH and lithium batteries, but much high than that
r2
of the traditional batteries, such as lead-acid and
nickel/cadmium batteries.
r3

Power Capability of Flywheel System

Hybrid Energy Storage Systems for Vehicle
Applications. Figure 4
Geometry of a flywheel
The moment of inertia of a flywheel per unit fly-
wheel mass (weight) depends exclusively on its geo-
metric design (refer to Fig. 4), as expressed by
 
Jf 1 b1 ðr14  r24 Þ þ b2 ðr24  r34 Þ
¼ : ð10Þ
Mf 2 b1 ðr12  r22 Þ þ b2 ðr22  r32 Þ
The effective approach to increase the Jf /Mf is to
increase b1 and r1. That is, to put as much mass as far
away from its rotating center as possible. This is
usually limited by the rotor material strength. So
high-strength materials are usually employed in
high-speed flywheels. Figure 5 shows the spatial size
of a flywheel, which is represented by r1, versus the
moment of inertia of the flywheel per unit mass,
using the geometric numbers in Fig. 4.
The power of a flywheel system can be expressed as
dEf dof
Pf ¼ ¼ Jf o f ; ð11Þ
dt dt
do
where Tm ¼ Jf dt f is the torque of the electric
machine to accelerate (charging) or decelerate the
flywheel (discharging). Thus, the power capability
of the flywheel system is completely determined
by the power capability of its electric machine.
The electric machine used in the flywheel system
should be have high speed, high efficiency, light weight,
and small volume. Further, in order to simplify the
control of the vehicle electric loads, such as traction
motors, actuators, etc., the terminal voltage of the
flywheel system should be maintained approximately
constant in the intended operating speed range of the
flywheel. Electric machine theory indicates that varying
the machine speed while keeping a constant terminal
voltage can be accomplished by controlling its
620 Hybrid Energy Storage Systems for Vehicle Applications

0.16

0.14

0.12

Jf / M, kgm2 / kg
0.1

0.08

0.06

0.04

0.02

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
r1, m

Hybrid Energy Storage Systems for Vehicle Applications. Figure 5

Outer radius of the flywheel versus the moment inertia of the flywheel per unit mass

900

800 rpm=60,000
Theoretical specific energy, Wh / kg

700 rpm=55,000
rpm=50,000
600 rpm=45,000
rpm=40,000
500 rpm=35,000
rpm=30,000
400 rpm=25,000
rpm=20,000
300 rpm=15,000
rpm=10,000
200

100

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
r1, m

Hybrid Energy Storage Systems for Vehicle Applications. Figure 6

Theoretical specific energy of flywheel versus flywheel geometry size, r1

magnetic field. Thus, wide speed operating range with Referring to the motors that are appropriate for elec-
constant voltage requires the machine to have wide tric and hybrid electric vehicles, the induction motor
“field weakening” capability and a long constant (IM), permanent magnet (PM), brushless dc motor
power speed range. (BLDC), and switched reluctance motor (SRM)

seem to be possible candidates for application to the
flywheel system. The induction motor is the most
mature technology among these motors. Its maximum
speed can be extended up 3–4 times of its base speed by
using field oriented control technology [1]. The BLDC
is well known for its lightweight, small volume, and
high efficiency due to its permanent magnet field.
However, the extent of its speed range beyond its
base speed is very limited due to its limited capability
of the field weakening, resulting in a limited constant
voltage speed range. The SRM is well known for its
simple structure, straightforward control, excellent
reliability, high speed, and long maximum speed
beyond its base speed. These inherent properties
make the SRM an attractive candidate for application
to the flywheels system, notwithstanding its relative low
torque density, relative to the BLDC. Based on our
analysis and development of traction motors for EV
and HEV [12], at Texas A&M University, it is believed
that the specific power of the overall flywheel system
will not be much more than 500 W/kg. This is similar
to that of advanced lead-acid batteries and less than
that of advanced high-power batteries, such as lithium
batteries and ultracapacitors.
Efficiency of Flywheel System
Energy losses in the flywheel system are not mainly
mechanical, such as friction and aerodynamic drag,
but also electrical losses in its electric machine. With
high vacuum inside the housing of the flywheel system
and with magnetic bearings, it is believed that mechan-
ical efficiency may reach 98–99%. The electric efficiency
mostly depends on the electric motor/generator, whose
efficiency may be around 85–90%. Thus, the total effi-
ciency of the flywheel system may be around about 83–
87%, which is similar to that of advanced high-efficiency
batteries and higher than that of conventional batteries,
but much less than that of ultracapacitors.
Hybrid Energy Storage
As discussed above, chemical batteries, ultracapacitors,
and flywheels each have their specific operating char-
acteristics. The chemical battery and flywheel are more
for high energy storage and deliver their energy with
medium power for a relatively long period of time.
Whereas, the ultracapacitor is more for high-power
Hybrid Energy Storage Systems for Vehicle Applications 621
source and delivers its high power in a relatively short
period of time. It is possible, by combining two of these
basic energy storage technologies together to constitute
a high-energy and high-power storage system. This is
the hybrid energy storage system as is conceptually
shown in Fig. 7.
Battery/Ultracapacitor System
Passive Interconnection
The most straightforward approach is to directly con-
nect ultracapacitors to batteries as shown in Fig. 8. This
configuration has the simplest structure and no control
unit. The behavior of the ultracapacitors is more like
a current filter. Thus, the high-peak battery currents are
leveled, as illustrated in Figs. 9 and 10. In Fig. 9, a step
load current is applied to the storage system. First, this
step current is in the ultracapacitors and then the load
current gradually transfers from the ultracapacitors to
the batteries due to the voltage drop in the
ultracapacitors. When the load current disappears, the
ultracapacitors are charged back up by the batteries
automatically. Figure 10 shows the simulation results
of the battery current and the ultracapacitor current
while a random load current applied.
The leveled battery current, due to the
ultracapacitors, can have many benefits for the batte-
ries. These include smaller battery pack, higher battery
operating efficiency, easier battery thermal manage-
ment, and longer battery life. The limited variation of
hybridized battery terminal voltage can also ease the
control of the connected load power converter. With
proper design of all the components, this simple hybrid
structure may yield a high-power, high-energy, and
high-efficiency energy storage system.
Active Interconnection
Clearly, in the passively connected battery/ultracapacitor
system, the power flow between the battery pack, the
ultracapacitor pack, and the load cannot be managed.
Consequently, the high power benefits of the
ultracapacitors may not be best utilized. Figure 11
shows the concept of an actively controlled battery/
ultracapacitor system. In this system, a power electronic
regulator is used to manage the power flow between the
battery pack, the ultracapacitor pack, and the load.
622 Hybrid Energy Storage Systems for Vehicle Applications

Low power demand

High specific
Power
energy Load
converter
source

High specific
power source

High power demand

High specific
Power
energy Load
converter
source

High specific
power source

Negative power
High specific
Power
energy Load
converter
source

Primary power flow

High specific
Secondary power flow
power source

Hybrid Energy Storage Systems for Vehicle Applications. Figure 7

Conceptual operation of the hybrid energy storage

IB IC
+ +
IL
. . . . . . .

. . . . . . .

Battery Power
Ultracap Load
pack converter
pack

− −

Hybrid Energy Storage Systems for Vehicle Applications. Figure 8

Passively connected batteries/ultracapacitors system
 Hybrid Energy Storage Systems for Vehicle Applications 623

100

Current,Ampere
Load current Battery current
50

–50 Ultracapacitor
current
–100
0 5 10 15 20 25 30 35 40 45 50
Time, Sec
110
Battery and ultracapacitor
105
Voltage,V

100

95
Battery only
90
0 5 10 15 20 25 30 35 40 45 50
Time, Sec

Hybrid Energy Storage Systems for Vehicle Applications. Figure 9

Battery current and ultracapacitor current with a step load current input

300
Battery current
250
Ultracapacitor current
200
150
Current, A

100
50
0
-50
-100
-150
-200
0 200 400 600 800 1000 1200 1400
Time, Sec.

Hybrid Energy Storage Systems for Vehicle Applications. Figure 10

Battery current leveled by passively connected ultracapacitors

Basically, the power conditioning operation can be
divided into three different modes: (1) ultracapacitor
peaking operation for high power demand,
(2) ultracapacitor charging from batteries, and (3) bat-
tery alone operation. These operation modes are
implemented by a central control unit. The central con-
trol unit commands the power electronics, according to
the control strategy and the received signals from the
current and voltages sensors, as shown in Fig. 11. The
control objectives are (1) to meet the load power
requirement, (2) to keep the battery current in
a preset band, and (3) to keep the battery state-of-
charge, SOC, in its safe region (e.g., 0.4–0.6), where
the battery efficiency are usually optimized.
624 Hybrid Energy Storage Systems for Vehicle Applications

SOCb =
f(Vb,Ib)
SOCb

Central

Voltage

Current
control
SOCu = SOCu
unit
f(Vu)

Load current
Control
signals
Batteries

Current
Voltage

To electric
Ultracapacitors Power loads
electronics

Hybrid Energy Storage Systems for Vehicle Applications. Figure 11

Conceptual illustration of the actively controlled battery/ultracapacitor system

IF IL

IC
Power
...

electronics V

Axle

Housing
Rotor

Flywheel Stator
Electric
machine

Hybrid Energy Storage Systems for Vehicle Applications. Figure 12

Passively connected flywheel/ultracapacitor system

This system can fully use the high power property of
the ultracapacitors, resulting in a smaller battery pack.
Further, the actively controlled battery current can lead
to more efficient battery operation and easier thermal
management.
Flywheel/Ultracapacitor System
As discussed in the flywheel section, the flywheel sys-
tem, similar to batteries, can be categorized as an
energy intensive storage device. This is mainly due to
its power capability being constrained by its electric

SOCb =
f(Vb,Ib)
Voltage
Current
Flywheel system
Power
electronics
Current
Voltage
Ultracapacitors
Hybrid Energy Storage Systems for Vehicle Applications. Figure 13
Actively controlled flywheel/ultracapacitor system
machine. Integration of the high-power ultracapacitors
with the high-energy flywheel system may result in
a high-power, high-energy and high-efficiency storage
system.
As before, the ultracapacitors can be either passively
connected to the flywheel system, as shown in Fig. 12,
or through an electronic power management unit, as
shown in Fig. 13. The passively connected flywheel/
ultracapacitor system may be expected to have
a performance similar to that of the passively
connected battery/ultracapacitor system, as shown in
Figs. 9 and 10. The actively controlled flywheel/
ultracapacitor system can be expected to have the
advantage of a smaller flywheel electric machine, stable
operation, and high efficiency.
Compared with battery/ultracapacitor hybrid sys-
tem, flywheel/ultracapacitor system would have the
additional advantage of extended service life, wide
temperature tolerance, and lower maintenance. These
advantages are very important in vehicle applications.
Future Directions
The vehicle energy storage should be able to supply
sufficient energy and power to meet both the steady
Hybrid Energy Storage Systems for Vehicle Applications 625
SOCb
Central
SOCu control
SOCu =
unit
f(Vu)
Load current
Control
signals
Power
electronics
and dynamic load requirements. Thus, high specific
energy and high specific power are necessary to achieve
compact vehicle energy storage. Chemical batteries and
mechanical flywheels can be categorized as energy
sources and ultracapacitors as power sources. By hybrid-
izing these storage devices, a high-energy and high-
power-energy storage system may be developed. In the
passively connected battery/ultracapacitor hybrid sys-
tem, the battery peak currents can be significantly mod-
erated. Consequently, the battery efficiency, service life,
and thermal management can be significantly improved.
In the actively controlled battery/ultracapacitor hybrid
system, the power flow between the batteries,
ultracapacitor and loads can be actively controlled.
Therefore, the operation of the whole system can be
better optimized. Flywheel/ultracapacitor hybrid sys-
tems, either passively connected or actively controlled,
can have the advantages of compact design, high effi-
ciency, extended service life, wide temperature tolerance,
and low maintenance requirements. These advantages
are very important for advanced vehicles.
Although all of the above systems have been used
individually in EV and HEV designs, the hybridized
concepts, described here remain to be applied to the
commercial EV and HEV products of the future. As the
626 Hybrid Energy Storage Systems for Vehicle Applications
ultracapacitor technology and applications become
more mature, their use in vehicle-hybrid-energy storage
applications is likely to spread. However, as the power
capability of chemical batteries continues to improve,
the use of hybrid energy storage becomes less imperative.
The eventual outcome of the above competing develop-
ments is not clear at this time. However, hybrid energy
storage may find uses in advanced vehicles in one form
or another for traction and other loads.
Bibliography
1. Chan CC, Chau KT (2001) Modern electric vehicle technology.
Oxford University Press, New York
2. Crompton TR (1996) Battery reference book, 2nd edn. Society
of Automotive Engineering
3. Delphi Automotive Company, http://www.delphi.com/news/
pressReleases/pr11607-04112002
4. Ehsani M, Gao Y, Butler KL (1999) Application of electrically
peaking hybrid (ELPH) propulsion system to a full size passen-
ger car with simulated design verification. IEEE Trans Veh
Technol 40:6
5. Ehsani M et al (2003) Investigation of high-energy and high-
power storage systems for military vehicle applications.
SAE-FTT 2003
6. Gao Y, Ehsani M (2002) An investigation of battery technolo-
gies for the army’s hybrid vehicle application. In: Vehicular
technology conference, 2002. Proceedings VTC 2002-Fall.
2002 IEEE 56th, vol 3, pp 1505–1509
7. Kajs J (2001) Combat vehicle mobility requirements. Seminar
at Texas A&M University
8. Maxwell Technologies, http://www.maxwell.com
9. Messerle HK (1969) Energy conversion statics. Academic,
New York
10. Miller JM, Emadi A, Rajarathnam AV, Ehsani M (1999) Current
status and future trends in more electric car power systems. In:
Proceedings of the IEEE vehicular technology conference
Houston
11. Montena, Montena components. http://www.montena.com
12. Rahman Z, Butler KL, Ehsani M (2000) Effect of extended-
speed, constant-power operation of electric drives on
the design and performance of EV-HEV propulsion system.
SAE future car congress, Apr 2000, Crystal City, Paper No.
2000-01-01557
13. Rand DAJ, Woods R, Dell RM (1998) Batteries for electric vehi-
cles. Society of Automotive Engineering
14. SAFT Batteries, http://www.saftbatteries.com/automotive/uk/
f/f.htm
15. ThermoAnalytics, http://www.thermoanalytics.com/support/
publications/batterytypesdoc.html

Intelligent Vehicles Technology,
Introduction
FEI-YUE WANG
Institute of Automation, Chinese Academy of Sciences,
Beijing, China
The basic goal of intelligent vehicle research is to
provide a safe and comfortable ride for drivers and
passengers. To reach this goal, an intelligent vehicle
needs to dynamically sense the surrounding environ-
ment, receive the correct command from drivers, and
make appropriate movements in real time.
In the 15 chapters in this book, 11 address how to
sense the environment via vision sensors, 3 study how to
monitor drivers’ status and understand their commands,
and 1 discusses vehicle motion control. The basis and
content of these chapters are described below.
To obtain richer and more accurate visual informa-
tion, researchers are trying to design more powerful
vision sensors. One important direction combines
different vision sensors to achieve better image quality.
For example, how to obtain true color night vision
video was studied in the chapter “▶ True Color Night
Vision Video Systems in Intelligent Vehicles.” The
proposed night vision cameras sense the near-infrared
radiation of the car’s own headlights to extend the
range of forward vision. The near-infrared and visible
spectrum can then be combined into natural-looking
color images for the driver to read.
Adjusting the focus of high-resolution video
cameras and directing them to the regions of interest
can provide better image data of certain objects.
“▶ Active Multifocal Vision System, Adaptive Control
of ” presents methods for obtaining active pointing
and gaze stabilization camera systems via an adaptive
multi-focal vision system. Sliding mode control has been
shown to be an effective control strategy to reach this goal.
Stereo vision sensors have become more popular
because they can recover more three-dimensional
information about nearby objects via stereo video
data. In “▶ 3D Pose Estimation of Vehicles Using Ste-
reo Camera,” neighboring vehicles pose a detection
problem and are discussed as a special application
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Intelligent Vehicles Technology, Introduction 627
of the 3D stereo vision technique. The common
characteristic points clustering problem is discussed
and a model-based scene flow method is applied to
determine the positions and instantaneous motion
states of the vehicles.
“▶ Dynamic Environment Sensing Using an Intel-
ligent Vehicle” studies the 3D environment reconstruc-
tion problem. Not only moving vehicles but also
stationary objects (e.g., trees, walls, etc.) are recognized
and modeled in such applications. Similarly, “▶ Driver
Assistance Systems, Automatic Detection and Site
Mapping” discusses how to reconstruct objects in
the environment, particularly when driving in
constructing sites.
LIDAR (Light Detection And Ranging) sensor is
another frequently used vision sensor and often serves
as a supplement to ordinary vision sensors. It can
measure the distance between objects and the vehicle
and provides depth information that cannot be
directly collected via ordinary vision sensors. In
“▶ Vehicle Detection, Tightly Coupled LIDAR and
Computer Vision Integration for,” a calibration algo-
rithm was developed to map the geometry transforma-
tion collected via a multi-plane LIDAR system into
camera vision data. Both simulations and real-world
experiments have been carried out to show its
reliability. “▶ Active Pedestrian Protection System,
Scenario-Driven Search Method for” studies pedes-
trian detection problem via the fusion of LIDAR and
camera systems.
Another type of approach combines camera vision
systems and radar systems. Some related discussion is
presented in “▶ Night Vision Pedestrian Warning in
Intelligent Vehicles.” The pedestrian detection problem
is studied as an example to prove the merit of radar/
camera fusion systems.
Since occlusion is generally unavoidable, the vision
capability of a single vehicle is always limited. One
straightforward way to solve this difficulty is to employ
vision information that is collected from different
vehicles in a small area. Clearly, this objective requires
us to share the understandings of surrounding
environments in an efficient way. As an early attempt,
“▶ Cooperative Group of Vehicles and Dangerous
Situations, Recognition of ” discusses how to define
# 2012, DOI 10.1007/978-1-4419-0851-3
628 Intelligent Vehicles Technology, Introduction
the objects around several vehicles and share their
position information.
The final chapter in vision sensory field, “▶ Driving
under Reduced Visibility Conditions for Older Adults,”
discusses how to enhance the gathered vision informa-
tion in reduced visible conditions and transform it suf-
ficiently for older adults. This chapter can be viewed as
an interdisciplinary study that crosses the boundaries of
vision sensor system design and driver assistance
systems.
The first chapter on driver assistance, “▶ Driver
Assistance System, Biologically inspired,” is hard to
categorize. It discusses how drivers pay attention to
certain changes in vision and how we can learn from
drivers’ response. It points out that advanced driver
assistance systems may be designed to alert drivers in
way that fit their biological characteristics. However,
a detailed notification method is yet to be developed.
“▶ Driver Behavior at Intersections” provides
a relatively mature approach that identifies drivers’
turning features at intersections based on the sampled
vehicle motion records. Following a stylized procedure,
it first introduces how to gather the required driving
data. Then a decision model of drivers’ side turn behav-
ior is proposed and the gathered driving data are used
to calibrate the model. This chapter can be viewed as
a representative of related reports.
The final chapter on driver assistance is “▶ Driver
Characteristics Based on Driver Behavior.” It presents
a classical method to model driver characteristics, the
driver behavior questionnaire. If well designed, a driver
behavior questionnaire can tell us how drivers make
their decisions under certain conditions. This method
is not novel but may still useful to driver modeling.
The only chapter on vehicle motion control
is “▶ Unscented Kalman filter in Intelligent Vehicles.”
It covers vehicle tire friction monitoring. Because
a vehicle’s motion is primarily determined by the
friction forces transferred from the road via the tires,
tire friction features receive attention. This chapter
illustrates how researchers consider the connection
between vehicle motion dynamics and tire friction
features. It also presents application of the Unscented
Kalman Filter (UKF) to realize tire feature monitoring.
There is an additional chapter related to intelligent
vehicles. As mentioned above, sharing information
between vehicles and infrastructures is a hot topic in
recent studies. In “▶ Vehicular ad hoc networks,
enhanced GPSR and Beacon-assist Geographic
Forwarding in,” utilization of real-time vehicle location
information to accelerate vehicle ad hoc communica-
tion is discussed.

Internal Combustion Engines,
Alternative Fuels for
THOMAS WALLNER1, SCOTT A. MIERS2
1
Energy Systems Division, Argonne National
Laboratory, Argonne, IL, USA
2
Mechanical Engineering, Michigan Technological
University, Houghton, MI, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
Limitations
Legislation
Overview of Alternative Fuels
Fuel Properties
Alternative Fuels for Spark-Ignition (SI) Engines
Alternative Fuels for Compression-Ignition (CI)
Engines
Future Directions
Bibliography
Glossary
Advanced biofuel Renewable fuel not produced from
food crops such as cellulosic biofuel and biomass-
based diesel.
Alternative fuel A fuel that can serve or be used in
place of another or an unconventional fuel choice.
Diesel gallon equivalent (DGE) The amount of alter-
native fuel required to match the energy content of
1 gal of diesel fuel.
Gasoline gallon equivalent (GGE) The amount
of alternative fuel required to match the energy
content of 1 gal of gasoline.
Mixture calorific value The amount of energy
contained per volume of fresh charge typically at
stoichiometric conditions that can be introduced
into the cylinders of an internal combustion engine.
Renewable fuel A fuel created from resources that
are never used up or can be replaced by new growth.
Definition of the Subject and Its Importance
Alternative fuels for internal combustion engines
include a wide range of liquid and gaseous chemicals
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Internal Combustion Engines, Alternative Fuels for 629
for both spark-ignition (SI) and compression-ignition
(CI) applications. Alternative in this context describes
a fuel that can be used in place of a conventional fuel
such as gasoline or diesel. Alternative fuels are not
equivalent but to a wide extent include renewable
fuels since renewable only suggests a fuel can be created
from resources that are never used up or can be
replaced by new growth. The most common alternative
fuels for SI engines are ethanol in blends with gasoline
as well as compressed natural gas (CNG), liquefied
natural gas (LNG), and liquefied petroleum gas
(LPG). Biodiesel, synthetic diesel, green diesel, and DME
are the most common alternative fuels for CI engines. In
light of increasing energy demand worldwide, limited fossil
fuel reserves partially located in politically unstable regions,
growing environmental awareness and concern related to
air pollutants as well as greenhouse gas emissions, alterna-
tive fuels have transitioned into social and political focus.
Legislation and mandates were put in place specifying
required amounts of alternative fuels in the fuel mix in
Europe, Asia, and the United States. This article highlights
some of the dominant alternative fuels, their market share,
properties as well as their implication on engine and
vehicle design, engine efficiency, and emissions.
Introduction
Although gasoline and diesel are nowadays considered
conventional fuels, early engine concepts were designed
to use other fuels. Combustion engines date back as early
as 1807 with François Isaac de Rivaz of Switzerland and
his combustion engine that used a mixture of hydrogen
and oxygen for fuel [1]. The idea of alternative fuels is as
old as engines themselves with Rudolf Diesel including
considerations for use of gaseous, liquid and solid fuels
in “Theory and Construction of a Rational Heat
Motor” in 1894 [2]. The first mass produced vehicle,
Ford Model T, was designed to run on ethanol.
Economic growth and prosperity are closely linked
with availability and cost of energy. A large portion of
worldwide energy demand and the majority of energy
used for transportation are covered through fossil
resources. The two most dominant fuels for transporta-
tion applications are gasoline in spark-ignition engines
and diesel in compression-ignition engines. Announce-
ments of newly discovered crude oil reserves and
# 2012, DOI 10.1007/978-1-4419-0851-3
630 Internal Combustion Engines, Alternative Fuels for
uncertainty on size of known reserves result in a large
error band in prediction of the time span in which fossil
fuels will be depleted. However, the fact that fossil fuels
are finite is indisputable and with increasing worldwide
energy demand the need for viable and sustainable
alternative fuels becomes ever more pressing.
Today more than ever, society is concerned with the
availability of energy which has become a matter of
national security. The transportation sector alone
accounted for 70% of the total US oil consumption in
2007. Foreign sources continue to outpace domestically
produced fuel in the United States as well as Europe
and Asia, leading to increased dependence on oil from
politically unstable regions. Alternatives, either
from domestic reserves or locally produced have been
identified as a critical factor in ensuring national
security and also as a means to stabilizing fuel prices
in times of fluctuating crude oil prices.
As third-world countries develop and further
growth of other countries continues, the amount of
transportation-borne pollution increases. This devel-
opment poses a threat due to local pollution resulting
from emissions of regulated constituents and air toxics
as well as the global impact of increased greenhouse gas
emissions. Alternative fuels can play a major role
in reducing regulated emissions through cleaner
combustion, and renewable fuels in particular can
help reduce greenhouse gas emissions.
There are several potential solutions to these issues
from reduced consumption, increased production,
and/or alternative sources of energy. It is this last
option that is the focus of this article and how alterna-
tive fuels and their properties impact engine as well as
vehicle operation.
Limitations
This article is focused on technical implications of alter-
native fuels on engines and vehicles, their performance
and emissions characteristics with particular focus on
automotive applications. The characteristics of the most
dominant alternative fuels are defined and compared to
conventional gasoline and diesel fuel. The discussion on
engine and fuel efficiency is strictly limited to the effects
of alternative fuels on engine performance. Differences
of alternative fuels in production pathways, production
efficiency and challenges for their distribution are not
included in this article. The interested reader is referred
to [3] for detailed well-to-wheel analysis of alternative
fuel pathways.
Before an attempt is made to classify alternative fuels
it is important to define alternative fuels. The word
alternative in the context of fuels suggests that a fuel
can serve or be used in place of another and might also
suggest an unconventional choice. However, alternative
fuels are not to be confused with renewable fuels since
renewable suggests that the resources used to create the
fuel are never used up or can be replaced by new growth.
Alternative fuels and renewable fuels are neither mutu-
ally exclusive nor mutually inclusive; although the vast
majority of renewable fuels are also considered alterna-
tive fuels, by far not all alternative fuels are renewable.
Methanol is just one example for a fuel that is renewable
if produced, e.g., from biomass, and since not widely
used also an alternative fuel. Ethanol is a renewable fuel
if produced from corn or sugarcane, but since it is
commonly used it may or may not be considered an
alternative fuel in the sense of an unconventional choice.
For our discussion ethanol is still considered an alterna-
tive since it is used in place of other fuels. Natural gas is
just one example for an alternative fuel that, if fossil
reserves are used, is not a renewable fuel.
Legislation
Recently introduced legislation worldwide mandates
the increased use of alternative fuels. Directive 2009/
28/EC of the European Parliament and of the Council
of 23 April 2009 on the promotion of the use of energy
from renewable sources and amending and subse-
quently repealing Directives 2001/77/EC and 2003/30/
EC states the need to increase energy efficiency as part
of the triple goal of the “20-20-20” initiative for 2020,
which means a saving of 20% of the European Union’s
primary energy consumption and greenhouse gas
emissions, as well as the inclusion of 20% of renewable
energies in energy consumption.
Japan issued a New National Energy Strategy
in 2007 targeting to expand the domestic biofuel
production to cover approximately 10% of the fuel
consumption by 2030 mainly through increased
bioethanol and biodiesel utilization [4].
US legislation through the Renewable Fuel Stan-
dard (RFS) mandates a fourfold increase in renewable

40
35
Renewable fuel production [Billion gallons]
30
25
20
15
10
Internal Combustion Engines, Alternative Fuels for. Figure 1
Mandated biofuel production in the United States
fuel production from approximately 9 billion gallons in
2008 to 36 billion gallons in 2022. The RFS also limits
the amount of conventional biofuels to 15 billion
gallons [5]. Figure 1 shows the mandated biofuels
production in the United States which regulates
contribution by conventional biofuels, advanced
cellulosic and non-cellulosic and biomass-based diesel.
Despite the rapid growth, US biofuels consumption
remains a small contributor to US motor fuels,
comprising about 4.3% of total transportation fuel
consumption (on an energy-equivalent basis) in 2009
which is expected to grow to about 7% with the
mandated 36 billion gallons by 2022 [6].
Overview of Alternative Fuels
Alternative fuels for internal combustion engines
include fuels used in blends with conventional fuels
such as ethanol which is used in low level gasoline/
ethanol blends (e.g., E10) or biodiesel in low level
blends with diesel (e.g., B5) as well as fuels requiring
a dedicated fuel system design such as compressed
natural gas (CNG) or liquefied petroleum gas (LPG).
The following chapters focus on these mainstream
Internal Combustion Engines, Alternative Fuels for 631
Biomass-based Diesel
Advanced (non-cellulosic) biofuels
Advanced (cellulosic) biofuels
Conventional biofuels
Year
alternative fuels, including their market share, proper-
ties, and impact on engine efficiency and emissions as
well as other renewable fuel options with considerable
market share either globally or in local markets.
Classification of Alternative Fuels
As mentioned earlier, different alternative fuels require
more or less significant changes to vehicle fuel system
design, engine calibration, and safety equipment. Some
of these changes will be addressed in detail when
discussing individual alternative fuels. In general,
a classification can include dedicated vehicles, flexible
fuel vehicles, and multi-fuel vehicles. Dedicated
vehicles only operate on a single (alternative) fuel and
are therefore typically purpose-built and optimized for
this specific fuel. Dedicated vehicles are typically built
for applications with fuels that cannot be stored in
conventional fuel tanks, such as compressed gaseous
fuels or liquefied fuels. On the other hand, blended
operation describes usage of two or more fuels of the
same type (liquid or gaseous) at varying mixture ratios.
Vehicles capable of operating on varying blends of
gasoline and ethanol are commonly referred to as
632 Internal Combustion Engines, Alternative Fuels for

FlexFuel vehicles. Finally, multi-fuel vehicles are
defined here as vehicles that can operate on two
or more fuels and feature multiple fuel storage and
delivery systems.
According to their most common use, alternative
fuels can also be classified as alternatives for spark-
ignition engines or compression-ignition engines.
This very generic classification is used for the following
discussion although there are examples of alternative
fuels that can be used in both applications. Gasoline is
used as the baseline fuel for spark-ignition engine
applications. For compression-ignition engine applica-
tions diesel is used as the reference.
Market Share of Alternative Fuels
Worldwide approximately 5% of the automobile fleet
are alternative fuel or advanced technology vehicles [7].
These vehicles can be grouped into dedicated vehicles
that are designed and built for a certain type of fuel on
the one hand and vehicles that are capable of operating
on alternative or conventional fuels on the other hand.
Total vehicle population is an estimated 884 million
vehicles which includes all types of vehicles such as
light-duty, medium-duty, and heavy-duty vehicles
as well as busses.
Considering all vehicle types CNG is the dominant
alternative fuel for dedicated vehicles with a share of
1.27% which amounts to a total of more than 11
million NG vehicles worldwide. However, distribution
and share of CNG vehicles is not uniform and strongly
depends on local fuel supplies. Figure 2 shows the
absolute number of vehicles for all countries with
a total CNG vehicle population in excess of 10,000
units. In addition the fraction of natural gas vehicles
compared to the countries’ total vehicle population is
presented. The leading nations in terms of total CNG
population are Pakistan (2.25 million units), Argentina
(1.8 million units), Iran (1.7 million units), and Brazil
(1.6 million units). Following India with approxi-
mately 700,000 units is Italy with 677,000 units as
the largest European CNG vehicle fleet. The CNG
vehicle fleet in the United States is approximately
100,000 vehicles, which accounts for only 0.04% of
the entire vehicle population. The number of CNG
vehicles available in the United States dropped from

2.5 100

2 80 Fraction of NG vehicles of total vehicle count [%]

Number of NG vehicles
Number of vehicles [Mio]

Fraction of total vehicles

1.5 60

1 40

0.5 20

0 0
Pakistan
Argentina
Iran
Brazil
India
Italy
China
Colombia
Ukraine
Banglad…
Thailand
Bolivia
Egypt
Armenia
Russia
USA
Germany
Peru
Bulgaria
Uzbekistan
Malaysia
Japan
Korea
Sweden
Myanmar
Venezuela
France
Canada
Tajikistan

Internal Combustion Engines, Alternative Fuels for. Figure 2

NG vehicles worldwide (Based on data supplied via NGVA Europe and the GVR)

eight models in MY2004, to five in MY2005, and one in
MY2006 [8].
In terms of market share of alternative fuel technol-
ogy, CNG is only exceeded by ethanol and gasoline/
ethanol flexible fuel (FlexFuel) vehicles. FlexFuel
vehicles are vehicles that can run on gasoline and any
gasoline/ethanol blend up to 85 vol % (E85). Since
their introduction in 1979, more than 12.5 million
ethanol and FlexFuel vehicles have been sold in Brazil
[9]. In the United States, the FlexFuel fleet in 2009
accounted for approximately 8.35 million vehicles
[10]. Although the number of FlexFuel vehicles exceeds
that of natural gas–fueled vehicles, the total number of
vehicles likely operated on ethanol is lower than that of
CNG. In the United States, the number of vehicles
actually operated on E85 is only 450,000 [11]. Figure 3
shows the development of alternative fuel vehicles in
use in the United States since 1995. While LPG was the
dominant alternative fuel in 1995 with more than
170,000 units, this number dropped to approximately
150,000 units in 2008. In the same time frame, the
number of CNG vehicles increased from approximately
50,000 units in 1995 to more than 110,000 units in
2008. As mentioned earlier, the number of E85 vehicles
800,000
700,000
600,000
H2 LNG
Number of vehicles [-]
500,000 CNG LPG
400,000
300,000
200,000
100,000
Internal Combustion Engines, Alternative Fuels for. Figure 3
Alternative fuel vehicles in use in the USA [11]
Internal Combustion Engines, Alternative Fuels for 633
that are actually fueled with gasoline/ethanol blends
was around 450,000 units in 2008, still making it the
dominant alternative fuel in the United States. Metha-
nol as an alternative fuel was fairly popular in the 1990s
with a peak of more than 21,000 units in 1997.
Although the number of hydrogen powered vehicles
has been steadily increasing since 2003, their market
share with slightly over 300 vehicles in 2008 was less
than 0.05% of all alternative fuel vehicles.
The United States consumed approximately 43
billion gallons of petroleum diesel fuel in 2005. Europe
consumed 59.4 billion gallons, and Asia (excluding the
middle east) consumed 48.5 billion gallons in 2005. In
2009, Europe produced the largest percentage of world-
wide biodiesel followed by the United States and Asia
Pacific. Figure 4 shows the distribution of biodiesel
production for EU member states from 1998 to 2010
[12]. Overall biodiesel production has been declining
in Europe since 2001, while it is increasing in the
United States and Asia Pacific.
Since 2000, US production of biodiesel has
increased by a factor of 350, as shown in Fig. 5.
Government-imposed incentives combined with the
requirements of the Renewable Fuel Standard have
M85
E85
634 Internal Combustion Engines, Alternative Fuels for

10,000

9,000

Biodiesel production [1000 tons] 8,000

7,000

6,000 Germany France

Spain Italy
5,000
Other EU Total EU

4,000

3,000

2,000

1,000

0
2002 2003 2004 2005 2006 2007 2008 2009

Internal Combustion Engines, Alternative Fuels for. Figure 4

EU biodiesel production [12, 13]

800

700
Estimated US Biodiesel Production [Mio Gal]

600

500

400

300

200

100

0
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Internal Combustion Engines, Alternative Fuels for. Figure 5

Estimated US Biodiesel Production by year [14]

been the primary reasons behind the expansion of
biodiesel production and associated utilization in the
United States.
In 2009, the United States produced 17.7% of the
world’s biodiesel, making it the second largest producer
behind Europe. The biodiesel market is expected to
grow in the United States over the next several years
and is predicted to reach
6,500 million liters (1,717
million gallons) by 2020. Biodiesel is considered
a biomass-based diesel source, and a minimum of
3.79 billion liters (1 billion gallons) must be produced
by 2012. This is a 50% increase over 2010 production.
The biodiesel market in Germany is not anticipating
growth but rather a decline in production. Taxes
have made the biodiesel too expensive for consumers
combined with export subsidies for US producers.
The European Union has imposed import tariffs on
US biodiesel to promote and protect European
production.
Fuel Properties
An understanding of alternative fuels and their impact
on vehicle design as well as engine performance and
emissions characteristics requires knowledge of fuel
properties. For this purpose the fuel properties of the
most common liquid as well as gaseous fuels are
discussed in detail and compared to conventional
fuels as a baseline. The alternative fuels are grouped
into alcohol fuels including methanol, ethanol as well
as iso-butanol as the most promising butanol isomer;
liquefied fuels including liquefied natural gas (LNG)
and liquefied petroleum gas (LPG), and gaseous fuels
including methane/CNG and hydrogen. Since these
fuels are mainly used in spark-ignition (SI) engines,
gasoline as the most common SI engine fuel is used as
the baseline for comparison. Biodiesel and synthetic
diesel fuel produced through the Fischer–Tropsch pro-
cess, green diesel as well as dimethyl ether (DME) are
compared to diesel fuel for CI engine applications.
In order to compare the energy content of alterna-
tive fuels to a conventional fuel, gasoline gallon equiv-
alent (GGE) is introduced. GGE refers to the amount of
alternative fuel required to match the energy content of
1 gal of gasoline. Although this measure is somewhat
controversial due to the variability in the energy con-
tent of gasoline as well as alternative fuels [15], it is used
Internal Combustion Engines, Alternative Fuels for 635
here to simplify the discussion regarding energy con-
tent of alternative fuels. In a similar way, diesel gallon
equivalent (DGE) is defined as the amount of alterna-
tive fuel required to match the energy content of 1 gal
of diesel fuel. GGE is used for alternative fuels that are
mainly used in spark-ignition engines and therefore
displace gasoline whereas DGE is used for alternative
fuels used in compression-ignition engines with the
potential to substitute diesel fuel. The conversion factor
between GGE and DGE is approximately 0.88.
Alcohol Fuels
Table 1 summarizes the fuel properties of the most
common alcohol fuels and compares them to gasoline
as a baseline. Characterized by a hydroxyl functional
group (-OH) bound to a carbon atom the most com-
mon alcohols are methanol, with only a single carbon
atom, ethanol (C2) as well as butanol, a four carbon
alcohol. The negative impact of the oxygen in the
hydroxyl group on the energy content decreases with
increasing carbon count. While the lower heating value
of methanol is less than 47% of gasoline, it increases
with carbon count to approximately 63% for ethanol
and 78% for butanol. Since the density of these fuels is
in a similar range, the ratio of volumetric energy con-
tent is similar to the lower heating value with
a significant advantage of gasoline compared to the
alcohols in general and the short chain alcohols in
particular. The lower volumetric energy content is
also reflected in the GGE values for alcohol fuels.
Approximately 2 gal of methanol contain the same
energy as 1.5 gal of ethanol, 1.2 gal of butanol, or
1 gal of gasoline. A major advantage of these most
common alcohol fuels compared to gasoline is their
higher knock resistance expressed as octane number,
which is around 100 for methanol as well as ethanol
and 98 for iso-butanol compared to approximately 90
for gasoline. Although concerns have been voiced that
laboratory engine Research and Motor octane rating
procedures are not suitable for use with neat oxygen-
ates [16], the presented values are still indicators to
evaluate knock resistance. The difference in oxygen
content is once again apparent when comparing the
stoichiometric air demand. It increases from 6.45 for
methanol to 9 for ethanol and 11.2 for butanol com-
pared to 14.7 for gasoline. Finally, the solubility of
636 Internal Combustion Engines, Alternative Fuels for

Internal Combustion Engines, Alternative Fuels for. Table 1 Comparison of fuel properties – gasoline versus common
alcohol fuels [16, 18, 19]

Parameter Unit Gasoline Methanol Ethanol iso-Butanol

Chemical formula C4 – C12 CH3OH C2H5OH C4H9OH
Composition (C, H, O) wt.% 86, 14, 0 37.5, 12.6, 49.9 52, 13, 35 65, 13.5, 21.5
Lower heating value MJ/kg 42.7 20 26.9 33.1
3
Density kg/m 720–780 792 789 805
Volumetric energy content MJ/l 31.6 15.8 21.2 26.5
Vol. energy content relative to gasoline % 100 50 67 84
GGE gal 1 2.01 1.5 1.19
Research octane number (RON) – 92–98 106–114 107–111 105
Motor octane number (MON) – 80–90 91–95 89–94 91
AKI 90 100 100 98
Stoichiometric air/fuel ratio – 14.7 6.45 9.0 11.2

Solubility in water @ 20 C ml/100 ml H2O <0.1 Miscible Miscible 7.6

Boiling temperature C 25–215 64.7 78.4 107–108

Flashpoint C –43 12 17 28
Ignition limits vol % 1.0–7.6 4–75 3.5–15 1.2–10.9
l 0.4–1.4
Latent heat of vaporization MJ/kg 0.305 1.103 0.84 0.69

ethanol in water is known to present a challenge for Liquefied Fuels

transporting gasoline/ethanol blends in pipelines [17].
Table 2 compares the properties of liquefied natural gas
A critical factor for liquid fuels is their evaporation
(LNG) and liquefied petroleum gas (LPG) to those of
characteristics. These characteristics affect engine
gasoline. Natural gas, after it has been cooled down to
behavior, in particular at cold start, as well as evapora-
its liquid state at a temperature of –161 C, is referred
tive vehicle emissions. Vapor pressure curves for alco-
to as liquefied natural gas (LNG). Since the liquefaction
hol fuels as a function of blend level are shown in Fig. 6
requires the removal of certain components that could
based on data from Anderson et al. [20]. The vapor
form solids during the process, LNG typically contains
pressure of alcohols is lower than that of gasoline.
approximately 95% methane [23]. Liquefied petroleum
When alcohols are mixed with gasoline, they form an
gas (LPG), also called autogas, is a mixture of propane
azeotropic mixture with some components of the gas-
(C3H8) and butane (C4H10) with changing composi-
oline which results in an asymmetric relation between
tion depending on the region as well as seasonal differ-
vapor pressure of the blended fuel and the volumetric
ences. In the United States, Canada, and Japan, LPG
blend ratio [21]. Evaporation of ethanol produces 4.6
consists mainly of propane, whereas the butane content
times as much cooling of the charge as iso-octane,
in other areas of the world can be significant and
a major component of gasoline, while methanol pro-
typically increases for countries with warmer climate
duces 8.4 times as much cooling. Charge cooling
(butane content in Greece is 80%). The actual compo-
reduces the tendency of the end gas to autoignite and
sition of LPG significantly affects its properties. Figure 7
therefore, further decreases the tendency for combus-
shows the vapor pressure of propane and butane, the
tion knock [22].
 Internal Combustion Engines, Alternative Fuels for 637

90

80

70

60
Vapor pressure [kPa]

50

40
Methanol
30 Ethanol
iso-Butanol
20

10

0
0 20 40 60 80 100
Alcohol content [Vol-%]

Internal Combustion Engines, Alternative Fuels for. Figure 6

Vapor pressure of alcohol fuels blends based on data from [20]

Internal Combustion Engines, Alternative Fuels for. Table 2 Comparison of fuel properties of gasoline with LNG and
LPG [18]

Parameter Unit Gasoline LNG LPG (USA)

Chemical formula C4 – C12 CH4 (typ. > 95%) C3H8 (maj) C4H10 (min.)
Lower heating value MJ/kg 42.7 46.7 46
3
Density kg/m 715 – 765 430–470 508
Volumetric energy content MJ/l 31.6 21 23.4
Vol. energy content relative to gasoline % 100 66 74
GGE gal 1 1.5 1.35
Octane number ((R + M)/2) – 90 120 105
Stoichiometric air/fuel ratio – 14.7 17.2 15.8

Boiling temperature C 25–215 –162 –42
Physical state for storage – Liquid Cryogenic liquid Pressurized liquid
Ignition limits vol % 1.0–7.6 5.3–15 2.1–9.5
l 0.4–1.4 0.7–2.1 0.4–2

two main constituents of LPG as a function of temper- is due to the lower density of the two alternative fuels
ature. Although the lower heating value of both LNG compared to gasoline and results in a volumetric energy
and LPG is higher than that of gasoline, the volumetric content of LNG of only 66% and 74% for LPG com-
energy content of both fuels is significantly lower. This pared to gasoline. The resulting GGE values of LNG and
638 Internal Combustion Engines, Alternative Fuels for
16
14
12
Vapor pressure [bar]
10
8 Propane
Butane
6 Butane
4 Carbon dioxide
2 Oxygen
0
–40 –30 –20 –10 0 10
Temperature [°C]
Internal Combustion Engines, Alternative Fuels for.
Figure 7
Vapor pressure as a function of temperature for propane
and butane
LPG are 1.5 and 1.35, respectively. The octane rating of
LNG is approximately 120 and significantly higher than
gasoline (90) and LPG with an octane rating of 105.
Gaseous Fuels
As mentioned earlier, LNG typically contains more
than 95% methane (CH4) since other impurities are
removed for the liquefaction process. Although com-
pressed natural gas (CNG) also mainly contains meth-
ane, its range of composition is wider than that of LNG
and varies geographically. In addition to methane it
typically includes significant amounts of ethane, pro-
pane, and butane as well as other constituents in
smaller amounts, as shown in Table 3.
Given that methane is by far the dominating con-
stituent of CNG, the comparison of fuel properties of
gaseous fuels shown in Table 4 uses the properties of
methane as representative for CNG. Although not
widely used for transportation applications at this
point, the fuel properties of hydrogen as another prom-
ising alternative fuel are included in this comparison.
The mass-specific lower heating value increases from
gasoline to methane to hydrogen starting at 43.5–
50 MJ/kg and 120 MJ/kg respectively. It is interesting
to note that the lower heating value increases with
increasing hydrogen content in the mass-specific fuel
composition. However, since both methane and hydro-
gen are gaseous at ambient conditions with a density
Internal Combustion Engines, Alternative Fuels for.
Table 3 Typical composition of natural gas [24]
Methane CH4 70–90%
Ethane C2H6 0–20%
Propane C3H8
C4H10
CO2 0–8%
O2 0–0.2%
Nitrogen N2 0–5%
20 30 40
Hydrogen sulfide H2 S 0–5%
Rare gases A, He, Ne, Xe Trace
that is several orders of magnitude lower than that
of gasoline, it takes 912 gal of methane or 3,075 gal of
hydrogen to achieve the energy equivalent of 1 gal of
gasoline. The comparison is more favorable on a weight
basis, suggesting that 1 kg of hydrogen contains the
same energy as 2.5 kg of methane and 2.86 kg of
gasoline.
Biodiesel and Synthetic Diesel
The average properties of petroleum diesel fuel and
biodiesel are shown in Table 5. In addition, to accen-
tuate the differences in properties between biodiesel
and other alternative diesel fuels, the properties of
a typical synthetic diesel fuel produced through the
Fischer–Tropsch process are included in Table 5 as well.
Diesel fuel properties are controlled by ASTM
D975. Hundred percent biodiesel has been specified
as an alternative fuel to diesel and has its own standard,
ASTM 6751. For blends between 6% and 20% biodie-
sel, a third standard is used, ASTM D7467. Before neat
biodiesel can be blended with petroleum diesel, both
fuels must meet their respective ASTM standards and
then be analyzed as a blend using D7467 if the biodiesel
content is between 6 and 20 vol %.
For example, if the cetane number falls too low,
it can lead to increased ignition delay, combustion
pressure, and noise, along with increased smoke during
cold start.
Increased viscosity has been shown to increase fuel
spray penetration resulting in cylinder wall wetting and
 Internal Combustion Engines, Alternative Fuels for 639

Internal Combustion Engines, Alternative Fuels for. Table 4 Comparison of fuel properties of gasoline and gaseous
fuels (methane and hydrogen) [18, 19]

Parameter Unit Gasoline Methane Hydrogen

Chemical formula C4 – C12 CH4 H2
Composition (C, H, O) wt.% 86, 14, 0 75, 25, 0 0, 100, 0
Lower heating value MJ/kg 43.5 50 120
3
Density gaseous kg/m – 0.72 0.089
liquid 730–780 430–470 71
GGE (by volume) gal 1 912 3075
GGE (by weight) kg 2.86 2.5 1
Stoichiometric air demand – 14.7 17.2 34.3

Boiling temperature C 25–215 –162 –253
Ignition limits vol % 1.0–7.6 5–15 4–75
l 0.4–1.4 0.7–2.1 0.2–10
Minimum ignition energy mJ 0.24 0.29 0.02

Self-ignition temperature C approximately 350 595 585
Diffusion coefficient mm/s – 1.9  10 –6
8.5  10–6
Quenching distance Mm 2 2.03 0.64
Laminar flame speed cm/s 40–80 40 200

Internal Combustion Engines, Alternative Fuels for. Table 5 Fuel properties for diesel, biodiesel, and synthetic diesel
[25, 27]

Parameter Unit Diesel Biodiesel Synthetic diesel

Chemical formula C8–C25 C12 – C22
Composition (C, H, O) wt% 87, 13, 0 77, 12, 11 85–87, 13–15, 0
Lower heating value MJ/kg 42–44 36–38 43.9
3
Density kg/m 810–860 870–895 770–780
Volumetric energy content MJ/l 34–37.8 31–34 33.8–34.2
Vol. energy content relative to diesel % 100 93 94.4–95.5
DGE gal 1 1.074 1.05–1.06
Cetane number – 40–55 45–65 73–80
Viscosity @ 40 C mm /s 2
2.8–5.0 2–6 2.0–3.5

Flash point C 60–78 100–170 59–109

Cloud point C –24 – –10 –5–5 –19–0
Stoichiometric air/fuel ratio – 14.7 13.8

Boiling range/temperature C 180–340 315–350 160–350
Ignition limits vol % 0.6–6.5 0.3–10
Lubricity Baseline Higher Lower
640 Internal Combustion Engines, Alternative Fuels for
impingement. Biodiesel has a slightly higher viscosity
than petroleum diesel, while synthetic diesel is similar
to petroleum diesel. In addition, the atomization of the
fuel is reduced as fuel viscosity is increased. Too low of
viscosity will increase pump wear due to reduced
lubricity. Biodiesel has been utilized as a lubrication
additive recently.
Sulfur content has previously been utilized for
lubrication in fuel injection systems but increased
emissions controls have forced the reduction of sulfur
in diesel fuel. The sulfur can be converted into
sulfur dioxide and sulfur trioxide, poisoning the cata-
lyst and sensors in the exhaust stream. In addition,
small droplets of sulfuric acid and other sulfates can
become nucleation sites for particulate matter emis-
sions. In the United States, the sulfur content of
on-road diesel fuel was mandated to not exceed
15 ppm sulfur as of 2006. Biodiesel and synthetic diesel
do not contain sulfur.
The cloud point is the temperature at which crys-
tals begin to form in the fuel and prevent the flow of
fuel through filters and pumps. A significant challenge
for biodiesel fuels is overcoming the significant
increase in cloud point temperature compared to
petroleum diesel fuel. Additives, tank heaters, and
various blends of petroleum and biodiesel have all
been experimented with to improve the cold-weather
operation of biodiesel.
Synthetic diesel fuel whether produced from bio-
mass (BTL), coal (CTL), or natural gas (GTL) does not
vary widely in properties based on the feedstock due to
the consistent processing techniques employed. If the
Fischer–Tropsch process is employed, the feedstock is
first converted to a synthesis gas (syngas) consisting of
hydrogen and carbon monoxide. This process involves
combustion of the feedstock in a reduced oxygen envi-
ronment. Once the hydrogen and carbon monoxide are
produced, the effect of the feedstock composition is
removed from the final fuel properties. The remaining
process to produce liquid fuel (Fischer–Tropsch) basi-
cally receives the same input regardless of feedstock.
The other method of producing synthetic diesel fuel is
to first convert the biomass to bio-oil through pyrolysis
(zero oxygen environment) and then convert the bio-
oil to fuel. Again, the pyrolysis step removes the effect
of feedstock composition, thus producing a consistent
fuel. Both processes produce a high quality alternative
diesel fuel with high cetane, good cold-flow and oxida-
tive stability properties, and ultra low sulfur content.
The Fischer–Tropsch process was first commercialized
by Sasol in South Africa. They produce over 165,000
barrels/day of transportation fuel from coal. However,
the process is on the order of three times more
expensive to produce fuel than conventional petroleum
production processes [28].
Green Diesel and DME
Hydrogenated-derived renewable diesel fuel or “green”
diesel is an alternative compression-ignition engine
fuel that is produced by utilizing the conventional
processing method for petroleum diesel fuel, fractional
distillation. The process is feedstock tolerant and unlike
transesterification, the properties of the fuel do not
change significantly when the feedstock varies from
vegetable oil to waste grease. However, there are signif-
icant differences in properties compared to biodiesel
and even petroleum diesel. Green diesel has a higher
cetane number, typically 75–90 compared to 40–50 for
petroleum diesel. The pour point and energy content
can be nearly the same as petroleum diesel, alleviating
the negative aspects of biodiesel. Green diesel is not
oxygenated so the oxidative stability is very good. The
process of producing green diesel is more energy inten-
sive and requires higher temperatures and pressures
compared to transesterification. The majority of
research is currently focused on increasing the output
of fuel and scaling up production facilities.
Dimethyl ether (DME) is a two-carbon, oxygenated
molecule that has unique properties in the liquid state
as an alternative diesel fuel. DME is primarily produced
by converting hydrocarbons such as natural gas to
synthesis gas (syngas). The syngas is then converted to
methanol in the presence of a copper-based catalyst. An
additional dehydration process of methanol using
a silica-alumina catalyst results in the production of
DME. This two-step process is the most widely utilized
currently, but a one-step production process is cur-
rently being researched. Because DME is in the vapor
state at standard temperature and pressure, it must be
compressed, like liquid propane, and stored in tanks
for transportation and consumption. Table 6 shows
some of the primary differences between green diesel
and DME compared to petroleum diesel fuel.
 Internal Combustion Engines, Alternative Fuels for 641

Internal Combustion Engines, Alternative Fuels for. Table 6 Green diesel and DME fuel properties compared to
petroleum diesel [26–29]

Unit Petroleum diesel Green diesel DME

Chemical formula C8–C25 C2H6O
Lower heating value MJ/kg 42–44 44 28
 3
Density (+20 C) kg/m 810–860 780 675
Volumetric energy content MJ/l 34–37.8 34.3 18.9
Vol. energy content relative to diesel % 100 94 52
DGE – 1 1.06 1.92
Cetane number – 40–55 98–99 >65
 2
Viscosity @ 40 C mm /s 2.8–5.0 3.0–3.5 0.21

Flash point C 52 –41

Auto ignition temperature C 250 350

Cloud point C –24 – –10 –15 n/a
Lubricity Baseline Similar to baseline Poor
Stoichiometric air/fuel ratio – 14.7 8.99
Ignition limits vol % 0.6–6.5 3.4–28

Effect of Feedstock Composition on Biodiesel Pro-
perties Unlike ethanol whose properties are relatively
unaffected by the feedstock due to the processing tech-
nique employed (distillation), biodiesel properties have
been shown to vary widely depending on the feedstock
properties (fatty acid profile) and catalyst employed
(typically methanol or ethanol) during the fuel
production process.
The fatty acid profile of the feedstock directly influ-
ences the properties of the biodiesel. The primary fea-
tures of interest for the fatty acid include chain length,
degree of unsaturation, and branching of the chain.
The properties of the biodiesel that are most directly
impacted by changes in the fatty acid profile include
cetane number, heat of combustion, cold-flow, oxida-
tive stability, viscosity, and lubricity. Table 7 shows the
fatty acid profiles for the primary feedstock options in
biodiesel and the wt% of each fatty acid contained in
the feedstock. Note that C18:1 refers to an 18 carbon
chain molecule with one double bond. Increasing
numbers of double bonds reduce the saturation of the
molecule. If no double bonds exist, the molecule is said
to be fully saturated.
It is well known that cetane number decreases with
decreasing chain length and increasing branching.
As well, oxidative stability increases with degree of
saturation. Table 8 shows the range of cetane number,
heat of combustion, cloud point or cold filter plugging
point (CFPP), and kinematic viscosity for several bio-
diesel fuels.
Figure 8 shows the effect of 33 different feedstocks
on fuel oxidative stability, measured in hours. This data
shows that as the number of double bonds increases
(increased degree of unsaturation), the oxidative sta-
bility decreases significantly. A test of iodine number of
the biodiesel provides a measure of the degree of
unsaturation of the fuel.
One of the significant advantages of biodiesel is the
ability to improve the lubricity of the fuel with only
a small quantity of ester. As shown in Fig. 9, the lubricity
improves significantly for the first 1% of methyl ester
added to the baseline diesel fuel. Lubricity is measured
using the ASTM D6078 Scuffing Load Ball-on-Cylinder
Lubricity Evaluator (SLBOCLE) test rig. The higher the
value, the higher the load before scuffing occurs which
translates to a fuel with improved lubricity.
642 Internal Combustion Engines, Alternative Fuels for

Internal Combustion Engines, Alternative Fuels for. Table 7 Fatty acid profile for the most common biodiesel feed-
stocks [30, 31]

Vegetable oil C16:0 C18:0 C18:1 C18:2 C18:3

Castor 1 1 3 4
Coconut 7.5–10.5 1–3.5 5–8 1–2.6
Cottonseed 22–26 2–3 15–22 47–58
Palm 40–47 3–6 36–44 6–12
Peanut 6–14 2–6 36.4–67.1 13–43
Rapeseed (Canola) 2–6 4–6 52–65 18–25 10–11
Soybean 10–12 3–5 18–26 49–57 6–9
Sunflower 5–7 3–6 14–40 48–74

Internal Combustion Engines, Alternative Fuels for. Table 8 Fuel properties of biodiesel fuels from various
feedstocks [31]
Cetane
Vegetable oil number Heat of combustion [KJ/kg] Cloud point or CFPP [ C] Kinematic viscosity [40 C, mm2/s]
Coconut ethyl 67.4 38,158 5 3.08
Cottonseed 51.2 –5 (pour point) 6.8 (21 C)
Palm ethyl 56.2 39,070 8 4.5 (37.8 C)
Rapeseed 48–56 37,300–39,870 –3; CFPP –6 4.53
(Canola)
Soybean 48–56 39,720–40,080 from –2 to 3 4.0– 4.3
Sunflower 54–58 38,100–38,472 0–1.5 4.39

The impact of biodiesel on cetane number can be
clearly identified in Fig. 10. As a general rule regardless
of feedstock, increasing quantity of biodiesel volume
increases the cetane number and improves the over-
all combustion quality. The data shown in Fig. 10 uti-
lized a baseline petroleum diesel fuel meeting ASTM
D975 specifications.
Comparison of Fuels Based on Mixture Calorific
Value
In addition to the fuel properties highlighted in
Tables 1–9, the mixture calorific value for different
fuels and mixture formation strategies can be calculated
and presents a measure for the volume-specific energy
content of fuel/air mixtures. It specifies the amount of
energy contained per volume of fresh charge typically at
stoichiometric conditions that can be introduced into
the cylinders of an internal combustion engine. The
mixture calorific value is not only used to describe the
fuel properties of a specific fuel and mixture formation
strategy, but also allows estimation of the theoretical
power density of a specific alternative fuel. The brake
mean effective pressure (BMEP) or engine torque can be
calculated based on the volumetric efficiency (VE) of a
specific engine, the brake thermal efficiency (BTE), and
the mixture calorific value (HG). Under the (theoreti-
cal) assumption that volumetric efficiency and brake
thermal efficiency remain unchanged when switching
from one fuel to another, any change in mixture
 Internal Combustion Engines, Alternative Fuels for 643

36

16
Oxidation stability [h]

12

4
3 h min.

0
Perilla Seed
Choice White Grease
Linseed
Palm
Fish
Evening Primrose
Rice Bran
Tung
Hemp
Sunflower
Used Cooking Oil
Castor
Mustard
Hepar, Low IV
Camelina
Hepar, High IV
Beef Tallow
Borage
Soybean
Corn, Distiller’s
Jatropha
Moringa oleifera
Yellow Grease
Neem
Canola
Coffee
Algae 1
Cuphea viscosissima
Lesquerella fendleri
Poultry Fat
Algae 2
Babassu
Coconut
Internal Combustion Engines, Alternative Fuels for. Figure 8
Dependence of biodiesel oxidative stability on feedstock [32]

7,000

6,000

5,000
SLBOCLE [grams]

4,000
Soy
Canola
3,000
Lard
Edible tallow
2,000 High FFA yellow grease

1,000

0
0% 1% 2% 3% 4% 5%
Methyl ester quantity [%]

Internal Combustion Engines, Alternative Fuels for. Figure 9

Lubricity improvement with methyl ester quantity [33]
644 Internal Combustion Engines, Alternative Fuels for

66

62 Soy
Canola
Lard
Cetane number [-]

58 Edible tallow
High FFA yellow grease

54

50

46
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Methyl ester quantity [%]

Internal Combustion Engines, Alternative Fuels for. Figure 10

Effect of methyl ester on cetane number [33]

Internal Combustion Engines, Alternative Fuels for. Table 9 Comparison of methyl ester and ethyl ester properties with
a soybean feedstock
Kinematic
Acid Iodine Peroxide Glycerol- free/ Water/ Cetane Density Oxygen viscosity
Ester No. No. No. Glycerol-bound Sediment No. g/cm3 wt% 40 C 100 C
Methyl soy 0.15 121 340 0.007/0.223 0 52.3 0.8836 11.44 4.03 1.64
Ethyl soy 3.02 122 123 0.003/0.031 0 47.3 0.8817 11.55 4.33 1.74

calorific value is directly reflected in corresponding fuel/air mixture (rMixture), relative air/fuel ratio (l),
change in (maximum) brake mean effective pressure, and stoichiometric air demand (LSt) and is defined as:
in other words, maximum torque. LHVrMixture
HG ¼ ð2Þ
BMEP ¼ VE  BTE  HG ð1Þ lLSt þ 1
For air-aspirating operation the density of air (rAir)
For mixture-aspirating engines (carbureted or port
is used instead of the fuel/air mixture density and the
injected/external mixture formation) the mixture cal-
value is defined as:
orific value is defined relative to 1 m3 of mixture and
LHVrAir
for air-aspirating engines (direct injection/internal H G ¼ ð3Þ
mixture formation) to 1 m3 of air. The mixture calorific lLSt
value for mixture-aspirating operation is calculated Figure 11 shows the mixture calorific value for
based on lower heating value (LHV), density of the various alternative fuels both for air-aspirating and
 Internal Combustion Engines, Alternative Fuels for 645

4.8

Mixture calorific value HG, HG [MJ/m3] 4.6

Air-aspirating
4.4
Mixture-aspirating
–

4.2
Methanol
4
Ethanol Butanol Hydrogen
3.8 Diesel
Gasoline
Propane
3.6
Methane
Biodiesel
3.4

3.2

3
5 10 15 20 25 30 35
Stoichiometric air/fuel ratio [-]

Internal Combustion Engines, Alternative Fuels for. Figure 11

Mixture calorific value for conventional and alternative fuel options

mixture-aspirating operation as a function of stoichio- considered an alternative fuel. The main production
metric air/fuel ratio. To achieve high power density, the routes for methanol all use natural gas as a feedstock.
mixture calorific value should be as high as possible. Methanol can be used to make methyl tertiary-butyl
With liquid fuels the difference in the mixture calorific ether (MTBE), an oxygenate that was used in blends
value between air- and mixture-aspirating operation is with gasoline as an octane booster. MTBE production
minimal since the liquid fuel due to its high density and use has declined in recent years because it has been
takes up only a small volume. However, for gaseous found to contaminate ground water. As an engine
fuels the difference between the mixture calorific value fuel, methanol has similar chemical and physical
in air-aspirating and mixture-aspirating operation characteristics to ethanol.
increases significantly with decreasing gas density A limited study of low level methanol gasoline
(rPropane = 1.83 kg/m3, rMethane = 0.72 kg/m3, rHydrogen blends up to 15 vol % on a four-stroke, single-cylinder,
= 0.089 kg/m3). The impact is most pronounced for variable compression ratio engine with a displacement
hydrogen as a fuel, because it displaces approximately volume of 582 cm3 suggests an approximately 10%
30% of the aspirated air at stoichiometric conditions increase in torque at the highest blend level which is
with external mixture formation. Even for liquid fuels attributed to the improved volumetric efficiency due to
there is noticeable difference in mixture calorific the higher heat of vaporization [34]. An experimental
values that should be considered when comparing evaluation of engine cold start behavior with gasoline
alternatives. as well as 10 vol % and 30 vol % blends of methanol in
gasoline on a small three-cylinder engine came to the
conclusion that the addition of methanol improves
Alternative Fuels for Spark-Ignition (SI) Engines combustion stability, indicated mean effective pressure
(torque), and reduces misfires. In addition a 70%
Alcohol Fuels (Ethanol, Butanol, Methanol)
reduction in CO emissions as well as a 40% reduction
Methanol As per the Energy Policy Act of 1992, in HC emissions were also observed. These positive
Methanol (CH3OH), also known as wood alcohol, is effects of methanol addition were attributed to higher
646 Internal Combustion Engines, Alternative Fuels for
vapor pressure, lower boiling temperature as well as the
high oxygen content of methanol [35]. A detailed ther-
modynamic evaluation of alcohol fuels including
their effect on engine efficiency and NOx emissions
concluded that the use of methanol could result in
increase in engine brake thermal efficiency of up to
1% regardless of engine load and speed. This increase
was attributed to slightly reduced pumping losses
because of the increased amount of fuel vapor at
constant load compared to the iso-octane baseline.
The study further found that a 25% NOx emissions
reduction can be expected with methanol compared
to the baseline fuel which is mainly caused by reduced
process temperatures [36]. The high octane rating
of methanol also makes it well suited for knock-free
operation at high compression ratios. A study that was
performed on single-cylinder, four-stroke, naturally
aspirated, high-compression direct-injection stratified
charge spark-ignition engine revealed the potential of
methanol for high engine efficiencies. Performing
optimization of injection timing and spark timing at
the research engine with a compression ratio of 16:1 at
full load conditions at 1,600 RPM resulted in
a maximum indicated efficiency of approximately
51% [37].
Due to challenges with methanol such as toxicity as
well as safety related issues due to invisible methanol
flames widespread use of methanol as a transportation
fuel is rather unlikely.
Ethanol Ethanol is probably the most widely used
liquid alternative fuel and is typically used in blends
with gasoline. The two main variables when using
gasoline/ethanol blends are the blend level and whether
hydrous or anhydrous ethanol is used. Typically gaso-
line is blended with anhydrous ethanol whereas neat
ethanol when used in vehicles is hydrous (E100).
Although attempts are made to get higher blends
approved, the most commonly used blend throughout
the world is E10, a blend of 90 vol % gasoline with 10
vol % anhydrous ethanol. US legislation limits the
amount of oxygen content in blends of gasoline with
alcohol fuels. In 1991, the maximum oxygen content
was increased from 2 wt.% to 2.7 wt.% for blends of
aliphatic alcohols and/or ethers excluding methanol
[38]. To ensure sufficient gasoline base was available
for ethanol blending, the EPA also ruled that gasoline
containing up to 2 vol % of MTBE could subsequently
be blended with 10 vol % of ethanol [39]. Only recently
EPA granted a partial waiver to allow gasoline that
contains greater than 10 vol % ethanol and up to 15
vol % ethanol (E15) for use in model year 2007
and newer light-duty motor vehicles, which includes
passenger cars, light-duty trucks, and sport utility
vehicles (SUV) [40].
Several markets also promote higher ethanol
blends, with E85 being the most prominent one with
existing distribution networks in several countries
including the United States and Sweden.
Due to its large domestic production capabilities of
ethanol from sugarcane Brazil has a special role in the
worldwide ethanol market with a distribution network
for both, Type C gasoline as well as neat hydrous
ethanol (
5 vol % water content). Type C Gasoline
describes a 25  1 vol % anhydrous ethanol blend with
gasoline [41] and is referred to as gasohol in Brazil.
A critical aspect potentially limiting the maximum
blend level of any alcohol fuel with gasoline is the vapor
pressure of the fuel blend since it is critical for
cold startability. Figure 6 shows the vapor pressure for
several alcohol fuels as a function of volumetric blend
level. Vapor pressures below 45 kPa [42] are known to
potentially cause cold start problems which effectively
limits the maximum ethanol blend level without taking
additional measures to approximately 75% during
winter months. Other limitations when increasing the
amount of ethanol in the fuel might be inferred due to
incompatibility of the blend with metals, plastics, and
elastomers in the fuel system. Despite these direct
effects, higher ethanol blends potentially also limit the
use of existing fuel infrastructure, in particular pipe-
lines [17]; however, these limitations are beyond the
focus of this article.
A more detailed overview of vehicle components
that are affected by ethanol addition as a function of
blend level are shown in Fig. 12 [43]. Blend levels below
5 vol % do not require any changes to the vehicle
regardless of vehicle age. Typically, modifications are
also not required at blend levels up to 10 vol % for
most vehicles less than 15–20 years of age. Only older
vehicles might require changes to the engine carbure-
tors. Although studies suggest that most fuel system
components are not negatively affected at a blend level
of 20 vol % [44–46], modifications to the fuel system
 Internal Combustion Engines, Alternative Fuels for 647

Vehicle component

Fuel pressure device

Evaporative system

Catalytic converter

Cold start system

Exhaust system
Intake manifold
Ignition system
Fuel injection

Basic engine
Carburetor

Fuel pump

Fuel filter

Fuel tank

Motor oil
≤ 5% For any vehicle
Ethanol

5–10% For vehicles up to15–20 years old

blend

10–25%
25–85% For specially designed vehicles
≥ 85%

- Not necessary - Probably necessary

Internal Combustion Engines, Alternative Fuels for. Figure 12

Required vehicle modifications with ethanol blend level (Based on [43])

including injectors, pumps, regulators and filters, the direct fuel injection concluded that at full load at
ignition system as well as the catalytic converters are 2,000 RPM a 9% improvement in efficiency as well as
common for blend levels up to 25 vol %. At higher maximum torque could be achieved while reducing
blend levels up to 85 vol % additional modifications exhaust gas temperatures by 60 C [48]. The Saab
include the base engine, engine oil as well as intake and Biopower, a turbocharged sedan built to operate on
exhaust system. Only for blend ratios beyond 85 vol % gasoline as well as E85 utilizes the advantageous fuel
and neat ethanol, a separate cold start system might properties of the ethanol blend. Rated at a maximum
be required. power of 110 kW at 5,500 RPM and a peak torque of
As mentioned earlier, specific engine designs are 240 Nm at 1,800 RPM with gasoline, the same engine
typically only found for engines that are intended to achieves a maximum power of 132 kW at 5,500 RPM
run on higher ethanol blends in excess of 25 vol %. and a peak torque of 280 Nm at 2,400 RPM with E85
Experimental results with port fuel injection on a 1.8 L [49]. With a compression ratio of 9.3:1 the 2.0 L engine
four-cylinder engine converted to single cylinder oper- takes advantage of the higher octane rating of the
ation revealed the efficiency and emissions potential of ethanol blend by operating the turbo engine at higher
high ethanol blends. In particular at moderate engine combustion pressures, producing more power without
speed of 1,500 RPM torque could be improved signif- risk of engine knock. Independent vehicle dynamome-
icantly due to the increased knock resistance resulting ter testing of the Saab Biopower showed that the vehicle
in a 20% increase with E100 compared to gasoline, acceleration time with E85 was 1 s faster than with
while improving indicated thermal efficiency from gasoline, and that the car also met US Tier 2, Bin 5
32% to 40%. At the same time a reduction in oxides emissions levels on both gasoline and E85, which is
of nitrogen emissions by approximately 125 ppm significant since Europe, where the car is certified,
together with a 2 C reduction in exhaust temperature does not require emissions certification on E85 and
per 10% increase in liquid volume fraction of ethanol applicable Euro 4 emissions standards are less strin-
in the fuel could be observed which was attributed to gent. In addition, a detailed exhaust speciation revealed
lower adiabatic flame temperatures [47]. A theoretical that ethanol and aldehyde emissions were higher on
as well as experimental study comparing performance, E85 while hydrocarbon-based hazardous air pollutants
efficiency, and emissions of gasoline and E85 with were higher on gasoline [50]. A detailed study of the
648 Internal Combustion Engines, Alternative Fuels for
impact of alcohol blends on regulated emissions and air
toxics with direct injection and ethanol blend levels of
up to 50 vol % also concluded that ethanol addition
resulted in an increase in aldehyde emissions, in par-
ticular acetaldehyde, whereas formaldehyde emissions
remained almost constant with ethanol addition. The
same study also concluded that with E50 oxide of
nitrogen emissions were reduced by approximately
15% compared to gasoline operation, which was attrib-
uted to a temperature reduction as a result of direct
injection and the increased in-cylinder cooling effect of
ethanol compared to gasoline. Specific emissions
trends for oxides of nitrogen as well as relevant air
toxics at a typical part load engine operating point
are summarized in Fig. 13 [51]. These trends hold
true independent of engine load, the magnitude
changes slightly depending on engine speed and load
conditions.
A study targeted at designing an engine specifically
for neat ethanol application suggested that a combina-
tion of direct injection and turbocharging at increased
compression ratios is ideally suited for operation on
E100. The proposed direct injection, turbo charged
180
1,500 RPM
2.62 bar BMEP
160
140
Emissions relative to gasoline [%]
120
100
80
60
40
20
0
Oxides of nitrogen (NOx)
Internal Combustion Engines, Alternative Fuels for. Figure 13
Influence of ethanol content on oxides of nitrogen and air toxics [51]
engine differs in the direct injector operating pressure,
piston shape and compression ratio to efficiently run
with E100 and gasoline. The E100 version operates at
an increased injection pressure (300 bar vs 200 bar) and
an increased compression ratio (13:1 vs 9:1). Improve-
ments in power and torque vary from 20% to 28% over
the range of engine speed, while the fuel conversion
efficiency increases 17% to 23% with a peak efficiency
of approximately 40% [52]. The report acknowledges
that cold start issues exist for E100 engines and indi-
cates that fuel is injected after intake valve closing
which helps to insure complete vaporization of the
ethanol and prevents or minimizes wall wetting, how-
ever, no specific measures to ensure proper cold start
behavior are discussed.
Measures to mitigate the cold start issue with E100
include fuel choice, e.g., E85 where the 15 vol % of
highly volatile gasoline are typically sufficient to
improve cold startability, sometimes combined with
intentional overfueling to further improve the cold
start behavior. However, it would be preferable to use
neat ethanol to fully utilize its fuel properties and also
avoid overfueling in order to reduce cold start
Gasoline
E10
E50
Formaldehyde (CH2O) Acetaldehyde (C2H4O)

emissions. Technical solutions that have been devel-
oped to mitigate the cold start problem include heated
fuel injectors [53], heated fuel-rails as well as installa-
tion of a second tank, filled with highly volatile gasoline
to run the engine for a few seconds before switching to
ethanol [54].
Ethanol currently is one of the most dominant alter-
native fuels which is mainly due to its favorable proper-
ties and production pathways. The properties and
challenges of ethanol as a fuel are well understood and
most technical hurdles for use of ethanol in internal
combustion engines have been overcome. It is likely
that ethanol will remain a dominant alternative fuel for
both, low and mid-level blends with gasoline for the
majority of applications as well as high level blends and
neat ethanol for local markets. Further development is
expected in dedicated engines for high level blends and
neat ethanol, however, the main area for advancement
will have to be in renewable production pathways which
create ethanol from biomass rather than food crops.
Butanol Butanol, a four carbon alcohol that exists in
several isomers, is currently under investigation as
a potential alternative fuel. Butanol is particularly
attractive compared to ethanol due to its higher volu-
metric energy content and its lower affinity to water
(see Table 1). In a joint venture BP and DuPont
founded Butamax™ Advanced Biofuels LLC to market
and promote biobutanol, which consists mainly of iso-
butanol. In addition to 135 cars, including 1995 – 2009
model years, that have been tested to date with 1.5
million miles driven, a retail demonstration was com-
pleted in 2009 in the UK. A total of 10 million liters of
biobutanol were blended and supplied to ten retail sites
as EN228 compliant gasoline. Over the course of the
demonstration approximately 250,000 vehicle fills were
performed and 80 million miles driven suggesting
biobutanol can be treated as a normal fuel component
and used without special procedures [55]. Since buta-
nol has a lower oxygen content than ethanol, a 16 vol %
blend of butanol with gasoline contains the same
amount of oxygen, approximately 3.7 wt.%, than a 10
vol % blend of ethanol with gasoline (E10).
In addition to vehicle tests performed and publi-
cized by BP and DuPont, few engine performance and
efficiency tests have been performed on iso-butanol as
well as other butanol isomers. A comparative study
Internal Combustion Engines, Alternative Fuels for 649
performed with gasoline as well as 10 vol % ethanol
in gasoline (E10) and 10 vol % n-butanol in gasoline on
a 2.2 L Direct-injection engine suggests that engine
efficiency and emissions characteristics are almost
identical throughout a majority of the engine operating
regime. Apparent differences were only found at the
highest speed and load conditions which could be
attributed to differences in the knock resistance of the
different fuels [56]. Consequently the study was
extended to include iso-butanol in addition to
n-butanol and blend levels were increased up to 85
vol % with similar findings as for the low level blends.
Even at the highest blend levels for ethanol, n-butanol,
and iso-butanol, the engine efficiency and emissions
characteristics are almost identical to the gasoline base-
line. However, at high load conditions the engine could
be run with significantly earlier spark timing with eth-
anol and iso-butanol compared to n-butanol and gas-
oline. The earlier combustion phasing due to the
increased knock resistance of ethanol and iso-butanol
compared to gasoline resulted in an increase in brake
thermal efficiency in excess of 1% [57]. Due to its lower
octane rating n-butanol does not exhibit the advan-
tages in terms of knock resistance that ethanol and
iso-butanol show compared to gasoline. In fact, n-
butanol was found to behave very similar to regular
gasoline in terms of combustion knock [58]. A study of
gasoline-butanol blends with up to 80 vol % alcohol
content on a port injected spark-ignition engine
suggests that the emissions characteristics remain
similar with increasing blend ratio, however, improved
combustion stability with increasing butanol was
observed which could be attributed to shorter ignition
delays [59]. An issue that was raised specifically with
the use of alcohol fuels is the amount of aldehydes and
other air toxics as a result of the combustion. A study of
regulated and non-regulated emissions as a result of
combustion of various blends of gasoline with ethanol,
n-butanol, and iso-butanol suggests that aside from
a consistent reduction in engine-out emissions of
oxides of nitrogen regardless of blending agent, both
formaldehyde and acetaldehyde emissions increased
with addition of butanol to the fuel blend, whereas
formaldehyde did not increase significantly with addi-
tion of ethanol. Figure 14 shows oxides of nitrogen
as well as formaldehyde and acetaldehyde emissions
for n-butanol as well as iso-Butanol gasoline blends at
650 Internal Combustion Engines, Alternative Fuels for

250
1,500 RPM
2.62 bar BMEP

Emissions relative to gasoline [%] 200

Gasoline
n-But16
iso-But16
150
n-But83
iso-But83

100

50

0
Oxides of nitrogen (NOx) Formaldehyde (CH2O) Acetaldehyde (C2H4O)

Internal Combustion Engines, Alternative Fuels for. Figure 14

Influence of butanol content on oxides of nitrogen and air toxics [51]

blend levels of 16 and 83 vol % [51]. Since the load
point is identical and the butanol blends with 16, and
83 vol % contain the same amount of oxygen as ethanol
gasoline blends at 10 and 50 vol %, these results are
directly comparable to those for ethanol blends shown
in Fig. 13. Although the increase in air toxics in engine-
out emissions is noticeable, further studies will have to
clarify the effect on tailpipe emissions.
Butanol appears to be a promising alternative fuel
with certain fuel properties superior to ethanol such as
volumetric energy content and affinity to water. Several
studies suggest that butanol is well suited as a fuel for
spark-ignition internal combustion engines. However,
butanol nowadays is produced mainly as an industrial
chemical rather than a transportation fuel. Since feed-
stock and production pathways for butanol are similar to
ethanol, the higher alcohol also faces the same challenges
in terms of economy of production from celluloses.
Liquefied Fuels (LPG, LNG)
Liquefied Petroleum Gas At normal ambient pres-
sure the boiling point of methane is –162 C, propane is
at –42 C, isobutane –10 C, and n-butane 1 C. Special
measures have to be taken to keep liquefied fuels
from evaporating at ambient conditions. As mentioned
earlier, LPG composition differs depending on regional
supplies which affect vapor pressure and storage
conditions (see Fig. 7). Typically LPG is stored in
a steel or composite tank at moderate pressure around
10 bar. The two most common mixture formation strat-
egies for LPG are vapor fumigation and LPG injection
into the intake. More recently work has also been
reported on LPG direct injection. For vapor fumigation
the LPG supply to the engine is typically controlled
by a regulator or vaporizer, followed by a mixer in the
intake manifold. For a typical vaporizer system a power
density that is up to 15% lower than a comparable gaso-
line engine can be expected [60, 61]. LPG injection
systems supply the liquid phase to the intake manifold,
where a phase change occurs once the fuel is injected.
Since this phase change results in increased charge
density due to the cooling effect of the evaporation,
LPG injection systems are superior compared to fumi-
gation systems in terms of power density and also
reduce the tendency for backfiring [62]. An increase

120
115
Peak torque relative to gasoline [%]
110
105
100
95
90
85
80
2,300
Internal Combustion Engines, Alternative Fuels for. Figure 15
Relative peak torque of LPG mixture formation concepts compared to gasoline (Based on [61])
in power density of LPG injection systems of approxi-
mately 5–10% compared to the gasoline baseline has
been reported [60, 61]. Most recently LPG engines
have also been operated with direct injection, and
a performance advantage in excess of 10% compared
to a gasoline baseline has been reported [61]. Figure 15
shows the maximum engine torque as a function of
engine speed for gaseous LPG port injection, liquid
LPG port injection, and LPG direct injection relative
to gasoline.
A comparative emissions study of a bi-fuel vehicle
capable of operating on gasoline or LPG using
a vaporizer system showed significant emissions advan-
tages for LPG. Depending on the drive cycle
a reduction in CO emissions of 10–30%, 30–50% in
HC, and 40–77% in NOx was reported [63]. The same
study also concluded that CO2 emissions were reduced
by approximately 10% when operated on LPG com-
pared to gasoline regardless of drive cycle. Given that
the CO2 emissions factor per unit energy for LPG is
approximately 12% lower than that of gasoline [64],
the previously mentioned reduction in CO2 emissions
suggests that the two fuels are equivalent in fuel econ-
omy. Due to the rapid vaporization of LPG a significant
advantage also in terms of particulate emissions can be
realized. A study of combustion and emissions
Internal Combustion Engines, Alternative Fuels for 651
Gaseous LPG PFI
Liquid LPG PFI
LPG DI
2,400 2,500 2,600 2,700 2,800 2,900 3,000
Engine speed [RPM]
characteristics with direct injection of LPG and gaso-
line concluded that the particulate emissions in LPG
operation were lower by a factor of 100 compared to
gasoline [65]. Finally, the higher octane rating of LPG
compared to gasoline indicates that a dedicated LPG
engine could possibly run at a higher, more efficient
compression ratio. A computational study that focused
on developing an engine design for direct injection of
CNG and LPG proposed 14.2:1 as a suitable compres-
sion ratio for LPG operation [66]. Since the fuel prop-
erties of propane were used for the study, the actual
compression ratio for real world LPG with a higher
butane content might be lower. Nonetheless the pro-
posed compression ratio is still significantly higher
than that of typical gasoline engines with compression
ratios below 10.5:1 for port injection and less than 12:1
for gasoline direct injection [67].
Liquefied petrol gas (LPG) is well established as an
alternative fuel in certain markets and the combustion
properties of LPG are well understood. However, the
market share compared to mainstream fuels is small in
most markets limiting dedicated vehicle development.
Besides the challenges due to the liquefied state of the
fuel, additional issues with infrastructure limitations
and safety concerns appear to further limit the interest
in large scale deployment.
652 Internal Combustion Engines, Alternative Fuels for
Liquefied Natural Gas LNG is stored on-board of
vehicles as a cryogenic liquid in super-insulated storage
tanks. The typical operating pressure of these tanks is in
the range of approximately 5 bar but can reach up to
15 bar. Although LNG is stored in its liquid form, it is
typically evaporated before use. The quality of the
evaporated gas differs depending on whether natural
boil-off gas or forced boil-off gas is used. The gaseous
phase in the top portion of LNG tanks is called natural
boil-off gas, consists mainly of methane and some
nitrogen, and has a high knock resistance. If LNG is
extracted from the liquid phase and evaporated sepa-
rately, it is considered forced boil-off gas which could
have a different knock resistance than the natural boil-
off gas. In addition to the potential difference in knock
resistance the lower heating value of natural boil-off
gas is approximately 33–35 MJ/m3 and significantly
lower than that of forced boil-off gas with an LHV of
38–39 MJ/m3 [68].
Regardless of which type of gas from an LNG tank is
used, the fuel is delivered to the engine in its gaseous
state. Therefore, an engine combustion system designed
and marketed for operation on LNG is in fact designed to
run on natural gas or methane and does not necessarily
differ from CNG engines [69, 70]. There are, as outlined
above, significant differences in the fuel storage, supply,
and conditioning systems. For engine design, effi-
ciency, performance, and emissions the same consider-
ations as for compressed natural gas apply.
Gaseous Fuels (Natural Gas, Hydrogen)
Compressed Natural Gas Compressed natural gas is
typically stored on-board a vehicle in pressure cylin-
ders, with single steel cylinders mainly used in the past
which are more and more replaced by several smaller
composite cylinders. The gas is typically stored at pres-
sures around 200–270 bar with the additional cylinders
adding around 60 kg of extra weight to the average
vehicle [71].
The first CNG model of the Honda Civic released in
1998 had a natural gas storage capacity of approxi-
mately 8 GGE at a fuel economy of 24 MPG City and
34 MPG Highway [72]. The 2011 model still has the
same storage capacity (7.8 GGE) at reported fuel econ-
omy numbers of 24/36 MPG allowing for a vehicle
range of up to 450 km (280 miles) [73].
Vehicle range and weight of the fuel storage system
are challenges for natural gas vehicles. At constant
range a gaseous storage system operating at 200 bar
requires approximately four times the storage volume
and outweighs the gasoline system by a factor of 3.
Lightweight materials, increased storage pressures,
and improved engine efficiencies are expected to
improve the volumetric ratio to 1:2.9 and the gravi-
metric ratio to 1:1.4 with a vehicle range of up to
600 km [74].
The dominant mixture formation strategy for nat-
ural gas in automotive applications is intake manifold
injection. The higher knock resistance of natural gas
compared to gasoline allows for dedicated NG engines
to operate at higher compression ratios resulting in
higher thermal efficiencies on the one hand and higher
NOx emissions on the other. The lower mixture calo-
rific value of natural gas with external mixture forma-
tion (see Fig. 11) compared to gasoline causes reduced
power density with the gaseous fuel which can be
further compromised when a lean burn strategy is
employed. The increased NOx levels with NG also
warrant the use of exhaust gas recirculation (EGR),
however, EGR is limited due to combustion instabil-
ities and misfires which occur at increased rates.
The reduced power density with injection into the
intake manifold can be compensated with
supercharging or turbocharging with intercooling.
More recently, attempts have also been made to employ
direct injection (DI) of natural gas but efforts are
hampered since high pressure gaseous injection hard-
ware is not available on the open market [75].
A comparative study of gasoline and CNG on
a four-cylinder spark-ignition engine showed signifi-
cant differences between the two fuels at identical oper-
ating conditions. At wide open throttle operation at
speeds between 1,500 and 5,000 RPM the engine pro-
duced between 8% and 16% less torque with sequential
CNG injection relative to gasoline port fuel injection
which was mainly attributed to the reduced mixture
calorific value. The increased reduction in engine
torque at higher speeds was further attributed to the
lower flame speeds of CNG compared to gasoline.
Similar results in terms of reduction in maximum
torque had been reported on the same engine when
operated with a mixer type CNG system instead of the
sequential fuel injection. However, the fuel conversion

efficiency with intake manifold injection was reported
to be an average of 13% higher compared to gasoline
whereas the carburetor system only showed an average
3% improvement over the gasoline baseline. Both the
carburetor and the sequential injection system showed
emissions reductions of up to 80% for CO, 8–20% for
CO2, and an average reduction in hydrocarbons of
approximately 50%, while NOx emissions were
increased by around 33% with the carburetor system
compared to gasoline and no NOx emissions were
reported for the sequential injection system [76, 77].
The previously mentioned results are representative
of stoichiometric engine operation just like conventional
gasoline engines. However, the relatively wide ignition
limits of natural gas suggest that a lean burn strategy
which is beneficial for engine efficiency be applied.
Experiments on a 1.3 L four-cylinder engine with
one cylinder operated with port fuel injection of natu-
ral gas showed that lean operation with relative air/fuel
ratios up to 1.3 is feasible without a dramatic decrease
in combustion stability. The lean burn strategy also
resulted in an up to 75% reduction in NOx emissions
compared to the stoichiometric case with only
a moderate increase in HC and CO emissions [78].
Modern stationary natural gas engines operate at
relative air/fuel ratios around 1.7 which is close to
the misfire region which results in a further
increase in engine efficiency while minimizing NOx
emissions [79].
Further extension of the lean combustion limits can
be achieved with stratification of the fuel/air mixture
which is frequently accomplished by employing direct
injection. Research on a single-cylinder 0.9 L engine
with direct injection during the compression stroke
suggests that the lean limit could be expanded to
approximately 1.8 also identifying the injection timing
as a critical parameter to influence combustion and
emissions characteristics [80]. A comparison of early
injection, late injection, and split injection on a 1 L
single-cylinder research engine further showed that
stratified mixtures resulting from late and split injec-
tion can extend the lean limit up to relative air/fuel
ratios in excess of 2.5, eliminating NOx emissions.
However, decreased combustion stabilities and expo-
nential increase in hydrocarbon emissions were
reported for these conditions [81]. Finally, hydrogen
addition has also been examined as a tool to extend the
Internal Combustion Engines, Alternative Fuels for 653
lean limit of natural gas engines. Results on a 1.6 L
single cylinder engine with hydrogen addition rates of
up to 19% showed an extension of the lean limit from
a relative air/fuel ratio around 1.8 without hydrogen
addition to almost 2 with 19% hydrogen. A simulta-
neous increase in indicated efficiency which can be
attributed to the increased burn rate with hydrogen
addition was also reported [82].
Energy security and domestic production in addi-
tion to natural gas being considered a clean-burning
fuel are the main drivers for renewed interest in natural
gas in the United States [83]. Natural gas has been
widely used for stationary power supply at outstanding
engine efficiencies and emissions levels. However, most
natural gas vehicle applications rely on conversion
from conventional gasoline engines for light-duty
applications or diesel natural gas dual fuel applications
for heavy duty applications. Future development for
natural gas applications is expected to result in dedi-
cated engines optimized to cater to the specific prop-
erties of the gaseous fuels. In terms of storage
technology both compressed natural gas as well as
liquefied natural gas storage will have their market
share since either one is well suited for certain
applications.
Hydrogen Although the situation for hydrogen in
terms of fuel storage is similar to that of natural gas
with compressed and cryogenic storage as the two
dominant techniques, hydrogen vehicles are not typi-
cally classified based on their storage system. Since, as
for natural gas, the fuel supply to the engine is almost
exclusively in gaseous form, engine technology does
not depend on the storage system. Attempts have
been made to utilize the low temperature of gaseous
hydrogen from liquid storage to increase the power
density of hydrogen port injection engines [84].
Hydrogen storage is a significant challenge for the
development and viability of hydrogen-powered
vehicles. On-board hydrogen storage in the range of
approximately 5–13 kg is required to enable a driving
range of greater than 300 miles for the full platform of
light-duty automotive vehicles using fuel cell power
plants or hydrogen internal combustion engines. Cur-
rent on-board hydrogen storage approaches involve
compressed hydrogen gas tanks, liquid hydrogen
tanks, cryogenic compressed hydrogen, metal hydrides,
654 Internal Combustion Engines, Alternative Fuels for
high-surface-area adsorbents, and chemical hydrogen
storage materials. Storage as a gas or liquid or storage in
metal hydrides or high-surface-area adsorbents consti-
tutes “reversible” on-board hydrogen storage systems
because hydrogen regeneration or refill can take place
on-board the vehicle. For chemical hydrogen storage
approaches (such as a chemical reaction on-board the
vehicle to produce hydrogen), hydrogen regeneration is
not possible on-board the vehicle and thus, these spent
materials must be removed from the vehicle and
regenerated off-board. On-board hydrogen storage sys-
tem performance targets were developed through the
FreedomCAR and Fuel Partnership, a collaboration
among DOE and the US Council for Automotive
Research, and the major energy and utility companies.
The targets developed are system-level targets and are
customer-driven based on achieving similar perfor-
mance and cost levels as competitive vehicles.
The storage system includes the tank, storage media,
safety system, valves, regulators, piping, mounting
brackets, insulation, added cooling capacity, and any
other balance-of-plant components. In order to
achieve system-level capacities of 1.8 kWh/kg system
(5.5 wt.% hydrogen) and 1.3 kWh/L (0.040 kg hydro-
gen/L) in 2015 and the ultimate targets of 2.5 kWh/kg
system (7.5 wt.% hydrogen) and 2.3 kWh/L (0.070 kg
hydrogen/L), the gravimetric and volumetric capacities
of the material alone must clearly be higher than the
system-level targets [85, 86].
A primary classification of mixture formation strat-
egies of hydrogen engines can be done based on the
location of mixture formation or the location of the
hydrogen dosing devices. External mixture formation
refers to concepts in which hydrogen and air are mixed
outside the combustion chamber, whereas internal
mixture formation refers to concepts with hydrogen
being introduced directly into the combustion cham-
ber. Some researchers have also proposed combined
concepts with a combination of external and internal
mixture formation [87]. As indicated in Fig. 11, the
mixture formation strategy, especially in hydrogen
operation, has significant impact on the theoretical
power output of the engine. The dramatic difference
in theoretical power output is mainly caused by the low
density of hydrogen, resulting in a significant decrease
in mixture density when external mixture formation is
employed [88].
10,000
8,000
NOx emissions [ppm]
6,000
4,000
2,000
0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Fuel to air equivalence ratio Φ [–]
Internal Combustion Engines, Alternative Fuels for.
Figure 16
NOx emissions trends for hydrogen fuelled engines
A study of hydrogen mixture formation concepts in
comparison to gasoline performed on an automotive
single-cylinder engine confirmed the superior power
density with hydrogen direct injection. A 15%
improvement in torque at 2,800 RPM with hydrogen
direct injection compared to gasoline port injection
could be achieved; the relative torque improvement of
direct injection compared to hydrogen port injection
exceeded 75% [89].
In terms of engine efficiency hydrogen internal
combustion engines have been shown to be superior
to other, more conventional fuels. Peak efficiency num-
bers reported by several research groups around the
globe suggest that brake thermal efficiencies in the
range of 45% can be achieved with hydrogen direct
injection [90–93].
Since hydrogen is the only fuel that does not con-
tain any carbon, its combustion does not create any
carbon-based emissions constituents except for traces
that are expected to result from lube oil combustion.
The major challenge for hydrogen combustion engines
is their NOx emissions which are a result of high tem-
peratures during the combustion process (Fig. 16).
A trade-off between relative air/fuel ratio and NOx
emissions for homogeneous hydrogen combustion
resulting from port injection or early direct injection
has been reported by several researchers [89, 94]. Com-
bustion of lean hydrogen–air mixtures with fuel-to-air
equivalence ratios of less than 0.5 (l > 2) results in

extremely low NOx emissions. Due to the excess air
available in the combustion chamber, the combustion
temperatures do not exceed the NOx critical value of
approximately 1,800 K [94]. Exceeding the NOx critical
equivalence ratio results in an exponential increase in
oxides of nitrogen emissions, which peaks around
a fuel-to-air equivalence ratio of 0.75 (l
1.3). At
stoichiometric conditions, the NOx emissions are at
around 1/3 of the peak value. The highest burned gas
temperatures in hydrogen operation occur around a fuel-
to-air equivalence ratio near 1.1, but at this equivalence
ratio oxygen concentration is low so the NOx concentra-
tion does not peak there [95]. As the mixture gets
leaner, increasing oxygen concentrations initially offset
the falling gas temperatures, and NOx emissions peak
around a fuel-to-air equivalence ratio of 0.75 (l 1.3).
Since hydrogen engines, due to the efficiency advan-
tages, are generally operated with fuel lean conditions,
the above mentioned NOx emissions trade-off with
air/fuel ratio also limits the power output.
If NOx emissions have to be avoided, hydrogen
engine operation with port injection is frequently lim-
ited to air/fuel ratios below 0.5 and supercharging is
employed to increase power density [96, 97]. With
hydrogen direct injection the mixture homogeneity
can be influenced through tailored design of injection
strategy including injector location and nozzle design
as well as injection strategy. Parameter studies with
hydrogen direct injection revealed that the NOx emis-
sions characteristics significantly depend on the injec-
tion strategy. A consequent optimization of injection
timing with conventional hydrogen direct injection
allowed a NOx tailpipe emissions reduction below the
level of a comparable gasoline engine; with a multiple
injection approach NOx emissions could be further
reduced to a level of less than 25% compared to
single-injection strategies [98].
A recently released hydrogen internal combustion
engine vehicle has been pushing the envelope in terms
of emissions levels achievable with combustion
engines. A BMW Hydrogen 7 Mono-Fuel demonstra-
tion vehicle that was tested for fuel economy as well
as emissions on the Federal Test Procedure FTP-75
cold-start test as well as the highway test achieved emis-
sions levels that were only 3.9% of the Super Ultra Low
Emissions Vehicle (SULEV) standard for nitric oxide
(NOx) and 0.3% for carbon monoxide (CO) emissions.
Internal Combustion Engines, Alternative Fuels for 655
For non-methane hydrocarbon (NMHC) emissions
the cycle-averaged emissions are actually 0 g/mile,
which require the car to actively reduce emissions
compared to the ambient concentration [99].
The properties of hydrogen make it well suited for
internal combustion applications with excellent engine
efficiency potential and inherently low emissions
signatures for carbon based emissions. Further
development of injection equipment will be required
for production applications of advanced mixture
formation concepts. The main challenges for a wide-
spread utilization of hydrogen in transportation
remain the lack of a fuel infrastructure as well as
sufficient on-board storage density.
Alternative Fuels for Compression-Ignition (CI)
Engines
Operation of compression-ignition engines on non-
petroleum fuel is not a new concept. In fact, Rudolf
Diesel, one of the pioneers of the diesel engine, origi-
nally designed the compression-ignition engine to run
on a variety of fuels, including straight vegetable oil.
In 1900, at the World Exhibition in Paris, he demon-
strated his engine running on pure peanut oil.
Operation of today’s modern, highly sophisticated
compression-ignition engines on straight vegetable oil
is not recommended due to the higher viscosity and
increased contaminants in the oil. While short-term
operation may be possible, long-term effects include
build-up of engine deposits, ring sticking, lube oil
gelling, and other significant maintenance problems
that can reduce engine life.
The primary alternative fuels for compression-
ignition engines include biodiesel and synthetic diesel.
Emerging fuel technologies include hydrogenated-derived
diesel fuel (“green” diesel) and dimethyl ether (DME).
Biodiesel
Biodiesel is the most widely used alternative diesel fuel
in the world. Biodiesel is defined as, “a fuel comprised
of mono-alkyl esters of long chain fatty acids derived
from vegetable oils or animal fats, designated B100, and
meeting the requirements of ASTM D 6751” [100].
The properties of biodiesel can vary widely, depending
on the feedstock and processing technique employed.
656 Internal Combustion Engines, Alternative Fuels for
Biodiesel is produced from a fat or oil by reacting it
with an alcohol like methanol or ethanol in the pres-
ence of a catalyst such as sodium or potassium hydrox-
ide. The transesterification process produces esters
(biodiesel) and glycerin. Typically, the alcohol is sup-
plied in excess to ensure a quick reaction and it can be
reused. The processing temperatures are around 65 C
and the pressures rarely exceed 140 kPa. The conversion
efficiency, from oil to methyl ester exceeds 98%. If the
alcohol is methanol, then the biodiesel is called a fatty
acid methyl ester (FAME). If the alcohol is ethanol,
then the biodiesel is called a fatty acid ethyl ester
(FAEE). The two “biodiesels” have significantly differ-
ent properties, as shown in Table 9 for a biodiesel
derived from a soybean feedstock. As the properties of
biodiesel change, there is a direct impact on fuel econ-
omy, emissions, performance, and engine efficiency.
Experimental testing has shown that particulate
matter, CO, and HC emissions can be significantly
reduced with relatively low levels of biodiesel blend,
while NOx emissions have been shown to increase.
Figure 17 shows a typical trend in emissions as the
quantity of biodiesel is increased.
20
10
0
–10
Change in emissions [%]
–20
–30
–40
–50
–60
–70
–80
0 20
Biodiesel content [Vol-%]
Internal Combustion Engines, Alternative Fuels for. Figure 17
Impact of emissions on biodiesel percentage [101]
One theory as to why NOx increases with increasing
biodiesel content relates to the reduction in PM. As the
combustion event occurs, the formation of soot and
particulate matter is reduced due to the presence of
oxygen in the fuel. Typically, this mass in the cylinder
would increase the specific heat of the burned-gas
mixture and lower the overall temperature. With the
reduction in particulate matter mass, the temperature
rises and the production of NOx increases [102].
A comparison of unregulated, gaseous toxic emis-
sions for two biodiesel blend ratios is shown in Fig. 18.
Except for Styrene and Toluene, all toxic emissions
levels reduced as the biodiesel blend quantity increased
from 20% to 100%. Even for the 20% blend compared
to a base petroleum diesel fuel, all but three toxic
emissions constituents decreased measurably.
Because of the lower energy density compared to
petroleum diesel fuel, biodiesel is associated with
a decrease in fuel economy. As the blend volume
increases, the impact on fuel economy becomes greater.
One would expect up to a 15% reduction in fuel econ-
omy due to the reduction in lower heating value. To
maintain equivalent power, additional fuel must be
NOx
PM
CO
HC
40 60 80 100
 Internal Combustion Engines, Alternative Fuels for 657

80%

Average change compared to base fuel [%]

60% 20% biodiesel
100% biodiesel
40%

20%

0%

–20%

–40%

–60%

–80%
Acetaldehyde

Acrolein

Benzene

1,3-Butadiene

Ethylbenzene

Formaldehyde

n-Hexane

Naphthalene

Styrene

Toluene

Xylene
Internal Combustion Engines, Alternative Fuels for. Figure 18
Comparison of gaseous, toxic emissions at two biodiesel blend ratios [103]

injected when operating on biodiesel. This is achieved this fuel, the properties are very similar to petroleum
through an increase in injection pressure or injection diesel and it blends well with conventional fuel. The
duration. It is advantageous to increase injection pres- fuels are straight chain paraffinic hydrocarbons that do
sure to not only increase the fuel delivery rate but also not contain aromatics, oxygen, or sulfur. Neste claims
improve atomization of the liquid fuel. However, field that their green diesel, named NExBTL results in an
data has not shown the significant reduction in fuel 18% reduction in NOx and almost a 30% reduction
economy as expected. The oxygenated fuel is also asso- in PM, compared to petroleum diesel fuel [105]. This
ciated with an increase in combustion efficiency up to data was the result of a delayed injection event
5% at blend levels below 25% [104]. The positive from a lower bulk modulus compared to petroleum
impact on combustion efficiency reduces as the blend diesel. It has been found that the bulk modulus on
level increases beyond 30%. The higher cetane number pump-line-nozzle injection systems is a sensitive
of biodiesel reduces the ignition delay, resulting in parameter in affecting the start of injection. Therefore,
reduced noise and vibration from the engine [104]. each fuel system needs to be custom tuned for different
This also reduces the premixed combustion event of alternative fuels to ensure optimum performance and
diesel operation, known to have a significant influence minimal emissions. This requires advanced sensors and
on the production of NOx emissions. significantly more involved calibration to enable flex-
fuel operation in the field.
In another study using a common-rail injection
Hydrogenated-Derived Renewable Fuel,
system, the impact of 30% and 100% HVO on emis-
Hydrogenated Vegetable Oil (HVO), or Green Diesel
sions and fuel consumption was studied. Figure 19
Green diesel refers to the diesel-like fuel product pro- shows the significant reduction in emissions and even
duced from renewable feedstocks utilizing the conven- gravimetric fuel consumption when no changes were
tional distillation process for petroleum fuel. Because made to the injection system. Additional benefits were
conventional distillation methods are used to produce realized by modifying the injection timing depending
658 Internal Combustion Engines, Alternative Fuels for
THC
10
5 30% HVO / 70% Diesel
petroleum-based diesel fuel [%] 0
Relative change compared to
–5
–10
–15
–20
–25
–30
–35
–40
Internal Combustion Engines, Alternative Fuels for. Figure 19
Impact of green diesel (HVO) on emissions and fuel consumption [106]
on the parameter of interest. For example, when spe-
cific fuel consumption (SFC) was held constant for all
three fuels, the 100% HVO reduced NOx by 16%
and smoke by 23% compared to petroleum diesel
fuel [106].
Synthetic Diesel
Synthetic diesel fuel is characterized by high cetane
number, low pour point, low sulfur content, and high
energy density. These characteristics provide a high
quality alternative diesel fuel that is capable of blending
with petroleum diesel at any ratio. In one study utiliz-
ing a synthetic diesel fuel derived from biomass feed-
stock (BTL), emissions and fuel consumption were
positively impacted when operating the engine on the
BTL fuel. The engine was a 1.7 L, four-cylinder,
direct injected diesel (Euro II) with intercooled
turbocharging. A single speed/load point was utilized
for all the testing, which simulated a typical cruise
point for this engine. The fuel was produced by Choren
Industries under the brand name SunDiesel. The
impact on NOx and BSEC is shown in Fig. 20. Brake
specific energy consumption (BSEC) was utilized
instead of brake specific fuel consumption (BSFC)
SFC SFC
NOX CO FSN (gravimetric) (volumetric)
100% HVO
due to the varying energy densities between the base-
line D2 diesel fuel and SunDiesel. Depending on the
parameter of concern, a significant reduction in NOx
emissions at the same BSEC level was achievable with
the synthetic diesel fuel [107].
A significant reduction in the number of soot par-
ticles in the exhaust stream was measured when the
engine was operated on SunDiesel, as shown in Fig. 21.
There is also a slight shift toward smaller particles for
the synthetic diesel fuel compared to the petroleum
based fuel. This modification of the particulate matter
size and quantity can have implications on the partic-
ulate filter efficiency and regeneration.
In a comprehensive look at GTL, BTL, and CTL
fuels produced using the Fischer–Tropsch process, Gill
et al. [108] concluded that the higher cetane number
and lower aromatic content were the primary fuel
properties that directly affected the emissions. The
higher cetane number reduced the ignition delay, lead-
ing to a reduction in NOx emissions. Typically, injec-
tion timing is retarded to reduce NOx emissions but at
the expense of fuel consumption (higher BSFC). With
a higher cetane number and thus lower ignition delay,
the injection timing can be delayed with almost no
impact on the BSFC. The shorter ignition delay also
 Internal Combustion Engines, Alternative Fuels for 659

12,000

3°BTDC
BSEC [kJ/kW-hr] 11,500

SunDiesel
ULSD
11,000

10,500

12°BTDC
10,000
2 3 4 5 6
NOx [g/kW-hr]

Internal Combustion Engines, Alternative Fuels for. Figure 20

Brake-specific energy consumption versus specific NOx emissions for D2 and SunDiesel [107]

20,000
Speed: 2,500 rpm
3° Load: 50% rated
EGR: 10%
16,000
Sundiesel
6°
D2
Number of particles [–]

12,000
9°
3°

8,000
12°
6°

9°
4,000

12°
0
10 100 1000
Mobility diameter [nm]

Internal Combustion Engines, Alternative Fuels for. Figure 21

Soot particle size and distribution for D2 and SunDiesel [107]
660 Internal Combustion Engines, Alternative Fuels for

provides additional time for the oxidation of soot par- DME has poor lubricity and high compressibility com-
ticles and subsequently reduced the engine-out partic- pared to diesel fuel and thus requires modification to
ulate matter. The reduced aromatic content of the XTL the fuel delivery and storage system. To achieve com-
fuels reduced the formation of soot by removing soot parable mileage to a petroleum diesel vehicle, the fuel
nucleation sites [108]. EGR is another effective method storage capacity for a DME-powered vehicle would
of reducing NOx emissions but typically increases the have to be doubled [110].
PM production. XTL fuels with their lower aromatic As shown in Fig. 22, DME has shown significant
content have a higher tolerance to EGR, producing reductions in HC, CO, and NOx emissions compared
less PM emission for the same EGR rate as petroleum to petroleum diesel. Smoke emissions for DME were
diesel fuel. too low to be measured. Of interest is the associated
reduction in BSFC, though perhaps a BSEC compari-
son would be more meaningful due to differences in
Dimethyl Ether (DME)
energy density.
DME is characterized by high cetane (>55) and low DME can be produced from the conversion of nat-
emissions characteristics. The oxygenated fuel has been ural gas, coal, oil residues, and bio-mass [109]. DME is
shown to reduce PM, NOx, and THC emissions com- an effective energy carrier and reforming of DME to
pared to petroleum diesel fuel and engine noise is produce hydrogen-rich fuel-cell streams is currently
typically reduced as well. The absence of carbon– being researched. In fact, hydrogen yields have been
carbon bonds combined with high oxygen content found to be equivalent to methanol at comparable
(
35 wt.%) and rapid evaporation are the primary operating temperatures [110].
reasons for the emissions reduction with DME. This When comparing alternative fuels on a well-to-
makes it an attractive alternative fuel for compression- wheels analysis for efficiency and GHG production,
ignition engines and provides additional means to DME combined with existing engine technology is
reduce NOx emissions through injection timing with- quite favorable. For example, compression-ignition
out a significant increase in PM emissions. However, engines operating on DME has the highest well-to-

6
Road load emission and fuel consumption

3 Petroleum diesel
DME w/oxidation catalyst

0
] ] ] ] ]
-h -h -h SN -h
W kW kW F kW
[kg
/ k
[g/ [g/ e[ [g/
ok x
FC HC CO Sm NO
BS

Internal Combustion Engines, Alternative Fuels for. Figure 22

Road load emissions and fuel consumption for DME and conventional petroleum diesel [110]

wheel efficiency of all non-petroleum–based fuel utiliz-
ing conventional, hybrid-electric, and fuel processor
fuel cell technologies (excluding natural gas) [110].
When compared against existing engine technology,
DME produces the lowest amount of well-to-wheel
GHG emissions compared to FT diesel, biodiesel,
methanol, and ethanol [110].
Alcohol/Diesel Blends
Various drivers including energy security and potential
for emissions reduction of diesel engines have triggered
attempts to blend diesel with alcohol fuels. A variety
of factors including blend stability, viscosity and lubric-
ity, materials compatibility, energy content, cetane
number, safety and biodegradability have to be consid-
ered if alcohol is to be added to conventional diesel fuel
for engine applications. Although all factors require
attention, the safety aspect is particularly emphasized
at this point. Diesel fuel has a flashpoint of 64 C and is
therefore classified as a Class II or combustible liquid
according to NFPA guidelines. Alcohol fuels as well as
gasoline with flashpoints below 37.8 C (see Table 1) are
classified as Class I or flammable liquids requiring
more stringent storage requirements such as greater
distance in location of storage tanks from property
lines, buildings, and other tanks. Studies have shown
that blends of 10, 15, and 20 vol % of ethanol in diesel
exhibit combustion safety characteristics essentially
identical to those for pure ethanol. Essentially any
blend of alcohol with diesel fuel has to be stored and
handled like gasoline rather than diesel [111, 112].
Blends of diesel and ethanol, also referred to as
E-Diesel, generally show increased specific fuel con-
sumption due to the lower energy content of ethanol
compared to diesel. However, engine efficiencies
remain constant or improve slightly with addition of
ethanol. It is also accepted that the addition of ethanol
to diesel fuel has a beneficial effect in reducing the PM
emissions at least. The amount of improvement varies
from engine to engine and also within the working
range of the engine itself [113]. A heat release and
emissions study of 5, 10, and 15 vol % ethanol/diesel
blends also concluded that soot, NOx, and CO
emissions decreased with increasing blend levels
while hydrocarbon emissions showed an opposite
trend [114].
Internal Combustion Engines, Alternative Fuels for 661
A comparative study of methanol diesel and ethanol
diesel blends at blend levels of 5 and 10 vol % on a single-
cylinder engine concluded that trends for both alcohols
were similar. The results suggest a consistent reduction
in soot and CO emissions as well as increased specific
fuel consumption with alcohol addition. However, the
reported trends of decreasing HC emissions and increas-
ing NOx emissions with addition of ethanol and meth-
anol are inconsistent with earlier studies. All reported
trends did not show any significant difference between
addition of ethanol and methanol [115].
A study was performed using a four-cylinder
Mercedes Benz C220 turbo-diesel vehicle and compar-
ing the emissions for conventional diesel fuel, 20 vol %
and 40 vol % n-butanol. Tests were performed for the
cold-start UDDS, hot-start UDDS, and HWFET drive
cycles as well as several steady-state points. The results
showed that for the urban drive cycle, both total hydro-
carbon (THC) and carbon monoxide (CO) emissions
increased as larger quantities of butanol were added to
the diesel fuel. Oxides of nitrogen (NOx) were not
significantly affected by the 20% butanol blend and
decreased with the 40% butanol blend. Drivability of
the vehicle decreased noticeably for the 40% butanol
blend, especially for the cold-start urban drive cycle.
Fuel consumption increased and thus fuel economy
decreased (mpg) as the blend ratio of butanol
increased, as shown in Fig. 23 due primarily to the
lower energy density of butanol compared to diesel.
For the steady state tests, a significant reduction in
filter smoke number (FSN) with increasing butanol
quantity was observed, as shown in Fig. 24 [116].
A performance and emissions study performed on
a 6.0 L bus engine at several steady state operating points
reached similar conclusions. At the tested blend levels of
8 and 16 vol % of n-butanol with diesel, smoke, NOx,
and CO emissions were consistently reduced with
increasing butanol content. At the same time increasing
butanol content also resulted in increased HC emis-
sions as well as increased specific fuel consumption
despite a simultaneous increase in engine efficiency
[117].
Although the technical advantages of alcohol addi-
tion to diesel have been demonstrated other limiting
factors such as blend stability, materials compatibility
and, especially, safety have to be critically evaluated.
Given that even addition of small amounts of alcohol
662 Internal Combustion Engines, Alternative Fuels for

60

55
0% butanol
50
20% butanol
45
40% butanol
40
Fuel economy [MPG]

35

30

25

20

15

10

0
UDDS cold-start UDDS hot-start HWFET

Internal Combustion Engines, Alternative Fuels for. Figure 23

Impact on fuel economy for butanol/diesel blends over three drive cycles [116]

2.5
0% butanol
20% butanol
2
40% butanol
Smoke [FSN]

1.5

0.5

0
0 2 4 6 8 10
Road load grade [%]

Internal Combustion Engines, Alternative Fuels for. Figure 24

Effect of n-butanol in diesel fuel on smoke emissions [116]

changes the flammability dramatically raises the ques-
tion on whether a transition to alcohol diesel blends is
economically viable.
Future Directions
Although alternative fuels can be grouped according to
their fuel properties or suitability for certain engine types,
their diversity requires dedicated engines to fully utilize
their unique characteristics. Due to the wide range of
engine applications with their specific requirements such
as emissions limits or efficiency targets and the variety of
alternative fuel options with specific advantages and lim-
itations it is likely that a future transportation scenario
will rely on an increased range of fuels. Meeting current
and future engine and vehicle efficiency as well as emis-
sions targets will likely also require advanced combustion
concepts to be implemented which in turn might benefit
from introduction of alternative fuels. To this extent
a stronger link between engine development and fuel
development is desirable which should result in opti-
mized utilization of certain fuel properties in dedicated
engines. This would put an end to the currently applied
approach of developing fuels that best mimic the prop-
erties of gasoline or diesel which themselves are far from
optimal for certain engine concepts.
Further, an isolated evaluation of fuels based on
their efficiency potential or emissions characteristics
will likely yield a suboptimal solution. Assessment of
alternative fuel options must include technical merits
and downsides as well as economical aspects. While
technical considerations have been discussed here, an
economical assessment including the entire vehicle and
fuel life cycle known as “well-to-wheel” analysis has to
be performed. These assessments have to include fac-
tors such as fuel production, storage and distribution
options as well as secondary factors such as water use,
recycling, or land usage. Furthermore, findings and
recommendations of such studies might vary dramat-
ically depending on regional factors such as climate as
well as typical driver behavior. To this extent a link
between engine and fuel related aspects and vehicle
related aspects will have to be established which will
clarify the impact of engine and fuel choices for differ-
ent powertrain configurations such as hybrid vehicles.
Overall improving energy security while reducing
harmful pollutants as well as greenhouse gas emissions
Internal Combustion Engines, Alternative Fuels for 663
will require a combination of improved transportation
efficiency and increased use of alternative fuels. Any
efficiency benefit a new engine technology can provide
is also desirable when burning alternative fuels along
the lines of “Using alternative fuels is desirable, having
to use less even more so.”
Bibliography
Primary Literature
1. Dutton K (2006) A brief history of the car. New Ideas 1
2. Diesel R (1894) Theory and construction of a rotational heat
motor. Spon & Chamberlain, London
3. MacLean H, Lave LB (2003) Evaluating automobile fuel/propulsion
system technologies. Prog Energy Combust Sci 29:1–69
4. Tanaka R (2007) Biofuels in Japan. Report by the British
Embassy in Tokyo
5. Section 201-202 Renewable Fuel Standard (RFS) Energy Inde-
pendence and Security Act of 2007 (Pub.L. 110-140, originally
named the CLEAN Energy Act of 2007)
6. Schnepf R, Yacobucci B (2010) Renewable Fuel Standard (RFS):
Overview and issues. CRS Report for Congress, Order Code
R40155
7. Sperling D, Gordon D (2009) Two billion cars: driving toward
sustainability. Oxford University Press, New York, pp 42–43.
ISBN 978-0-19-537664-7
8. Yacobucci BD (2008) Natural gas passenger vehicles: availabil-
ity, cost, and performance. CRS Report for Congress. Order
Code RS22971
9. Brazilian Automotive Industry Association – ANFAVEA,
Brazilian Automotive Industry Yearbook 2010. Available online
at http://anfavea2010.virapagina.com.br/anfavea2010/
10. Energy Information Administration (2008) Annual survey of
alternative fuel vehicle suppliers and users, “as reported in”
Alternatives to traditional Transportation fuels 1998–2008
reports (Tables 14 or S1 depending on year of report). www.
eia.doe.gov/cneaf/alternate/page/atftables/afv-atf2008.pdf
11. US Energy Information Administration (EIA) (Apr 2010) EIA’s
Alternatives to traditional transportation fuels, Table V1. www.
eia.doe.gov/cneaf/alternate/page/atftables/afv-atf2008.pdf
12. European Biodiesel Board (2009) http://www.ebb-eu.org/
stats.php
13. International Energy Agency (IEA) Statistics division (2007)
Energy balances of OECD countries (2008 edition)–Extended
balances and energy balances of Non-OECD countries (2007
edition)–Extended balances. IEA, Paris. http://data.iea.org/
ieastore/default.asp
14. http://www.biodiesel.org/buyingbiodiesel/plants/
15. Boyd RA (2009) Proposed method of sale and quality specifi-
cation for hydrogen vehicle fuel. Summary of current informa-
tion. Standards Fuel Specifications Subcommittee (FSS). U.S.
National Work Group for the Development of Commercial
Hydrogen Measurement
664 Internal Combustion Engines, Alternative Fuels for
16. American Petroleum Institute (API) (2001) Alcohols and ethers,
Publication No. 4261, 3rd edn. API, Washington, DC
17. Whims J (2002) Pipelines considerations for ethanol, Agricul-
tural marketing resource center. Kansas State University
18. Perry RH, Green DW (1999) Perry’s chemical engineers’ hand-
book. McGraw Hill, Malaysia
19. Heywood JB (1988) Internal combustion engine fundamen-
tals. McGraw-Hill, New York
20. Andersen V, Anderson JE, Wallington TJ, Mueller SA, Nielsen
OJ (2010) Vapor pressures of alcohol-gasoline blends. Energy
Fuels 24:3647–3654
21. The Royal Society (2008) Sustainable biofuels: prospects and
challenges. The Royal Society, London. ISBN 978 0 85403 662 2
22. Bradley D (2009) Combustion and the design of future engine
fuels. Proc. IMechE Part C: J. Mech Eng Sci 223:JMES1519. doi:
10.1243/09544062JMES1519
23. Foss M (2007) Introduction to LNG. An overview on liquefied
natural gas (LNG), its properties, organization of the LNG
industry and safety considerations
24. http://www.naturalgas.org/overview/background.asp
25. Lapuerta M, Armas O, Rodrıguez-Fernandez J (2008) Effect of
biodiesel fuels on diesel engine emissions. Prog Energy
Combust Sci 34:198–223
26. Teng H, McCandless JC, Schneyer JB (2004) Thermodynamic
properties of dimethyl ether – an alternative fuel for
compression-ignition engines. SAE Technical Paper 2004-01-
0093
27. Tilli A, Kaario O, Imperato M, Larmi M (2009) Fuel injection
system simulation with renewable diesel fuels. SAE Technical
paper 2009-24-0105
28. Teng H, McCandless JC, Schneyer JB (2001) Thermochemical
characteristics of dimethyl ether — an alternative fuel for
compression-ignition engines. SAE Technical Paper 2001-01-
0154
29. Teng H, McCandless JC, Schneyer JB (2002) Viscosity and
lubricity of (Liquid) dimethyl ether – an alternative fuel for
compression-ignition engines. SAE Technical Paper
2002-01-0862
30. Knothe G (2008) “Designer” biodiesel: optimizing fatty ester
composition to improve fuel properties. Energy Fuels
22:1358–1364
31. Knothe G (2005) Dependence of biodiesel fuel properties on
the structure of fatty acid alkyl esters. Fuel Process Technol
86:1059–1070
32. Sanford SD et al (2009) Feedstock and biodiesel characteristics
report, Renewable Energy Group, Inc. www.regfuel.com
33. Kinast JA (2003) Production of biodiesel from multiple feed-
stocks and properties of biodiesel and biodiesel/diesel blends,
NREL/SR-510-31460, March 2003
34. Abu-Zaid M, Badran O, Yamin J (2004) Effect of methanol
addition on the performance of spark ignition engines. Energy
Fuels 18:312–315
35. Song R, Hu T, Liu S, Liang X (2008) Combustion characteristics
of SI engine fueled with methanol gasoline blends during cold
start. Front Energy Power Eng China 2(4):395–400
36. Caton JA (2009) A thermodynamic evaluation of the use of
alcohol fuels in a spark-ignition engine. SAE Paper No. 2009-
01-2621
37. Li J, Gong C, Su Y, Dou H, Liu X (2010) Effect of injection and
ignition timings on performance and emissions from a spark-
ignition engine fueled with methanol. Fuel 89:3919–3925
38. Environmental Protection Agency. Regulation of fuels and fuel
additives; definition of substantially similar. [FRL-3856-9]
39. http://www.faqs.org/faqs/autos/gasoline-faq/part1/section-4.
html
40. http://www.epa.gov/otaq/regs/fuels/additive/e15/420f10054.
htm
41. Ré-Poppi N, Almeida FFP, Cardoso CAL, Raposo JL Jr, Viana LH,
Silva TQ, Souza JLC, Ferreira VS (2009) Screening analysis of
type C Brazilian gasoline by gas chromatography – Flame
ionization detector. Fuel 88:418–423
42. Bennett J (2007) Bioethanol in road transport fuels. Presenta-
tion at the Royal Society ‘International Biofuels Opportunities’
conference and workshop, London
43. Joseph Jr H (2007) The vehicle adaptation to ethanol fuel.
Presentation at the Royal Society ‘International Biofuels
Opportunities’ conference and workshop, London
44. Jones B, Mead G, Steevens P, Timanus M (2008) The Effects of
E20 on metals used in automotive fuel system components.
Report by the Minnesota Center for automotive research at
Minnesota State University, Mankato
45. Jones B, Mead G, Steevens P (2008) The Effects of E20 on
plastic automotive fuel system components. Report by the
Minnesota center for automotive research at Minnesota
State University, Mankato
46. Jones B, Mead G, Steevens P, Connors C (2008) The Effects of
E20 on elastomers used in automotive fuel system compo-
nents. Report by the Minnesota center for automotive
research at Minnesota State University, Mankat
47. Kar K, Cheng W, Ishii K (2009) Effects of ethanol content on
gasohol PFI engine wide-open-throttle operation. SAE Paper
No. 2009-01-1907
48. Grabner P, Eichlseder H, Eckhard G (2010) Potential of E85
direct injection for passenger car application. SAE Paper No.
2010-01-2086
49. Saab engine specifications (2008) http://dyc.saab-web.com/
main/GLOBAL/en/model/95/techspecs.shtml. Accessed 31
May 2011
50. West B, López A, Theiss T, Graves R, Storey J, Lewis S (2007)
Fuel economy and emissions of the ethanol-optimized Saab
9-5 biopower. SAE Paper No. 2007-01-3994
51. Wallner T, Frazee R (2010) Study of regulated and non-
regulated emissions from combustion of gasoline, alcohol
fuels and their blends in a DI-SI engine. SAE Paper No.
2010-01-1571
52. Boretti A (2010) Analysis of design of pure ethanol engines.
SAE Paper No. 2010-01-1453
53. Kabasin D, Hoyer K, Kazour J, Lamers R, Hurter T (2009)
Heated injectors for ethanol cold starts. SAE Paper No. 2009-
01-0615

54. Jeuland N, Montagne X, Gautrot X (2004) Potentiality of etha-
nol as a fuel for dedicated engine. Oil & Gas Sci Technol Rev IFP
59(6):559–570
55. Schubert A (2010) Expanding the role of biofuels through
development of advanced BioButanol. Biofuels Workshop.
SAE International Fuels & Lubricants Meeting. Rio de Janeiro
56. Wallner T, Miers S, McConnell S (2009) A comparison of etha-
nol and butanol as an oxygenate and their effect on efficiency,
combustion performance and emissions of a direct-injected 4-
cylinder engine. J Engin Gas Turbines Power 131(3):032802–
032809
57. Cooney C, Wallner T, McConnell S, Gillen J, Abell C, Miers
S (2009) Effects of blending gasoline with ethanol and butanol
on engine efficiency and emissions. In: ASME Spring technical
conference. Milwaukee/WI. ASME Paper No. ICES2009-76155
58. Szwaja S, Naber JD (2010) Combustion of n-butanol in a spark-
ignition IC engine. Fuel 89:1573–1582
59. Dernotte J, Mounaim-Rousselle C, Halter F, Seers P (2010)
Evaluation of Butanol–gasoline blends in a port fuel-injection,
Spark-ignition engine. Oil & Gas Science and Technology – Rev
IFP 65(2):345–351
60. http://www.lpgli.com/features.html
61. Li X, Yang L, Pang M, Liang X (2010) Effect of LPG injection
methods on engine performance. Adv Mater Res 97–
101:2279–2282
62. LPG Autogas (2005) Expansion Issue 1
63. Saraf RR, Thipse SS, Saxena PK (2009) Comparative emission
analysis of gasoline/LPG automotive bifuel engine. Int
J Environ Sci Eng 1:4
64. Energy Information Administration, Documentation for Emis-
sions of Greenhouse Gases in the U.S. 2005, DOE/EIA-0638
(2005), Oct 2007
65. Oh S, Lee S, Choi Y, Kang K, Cho J, Cha K (2010) Combustion
and emission characteristics in a direct injection LPG/Gasoline
spark ignition engine. SAE Paper No. 2010-01-1461
66. Boretti AA, Watson HC (2009) Development of a direct injec-
tion high flexibility CNG/LPG spark ignition engine. SAE Paper
No. 2009-01-1969
67. Heywood JB, Welling OZ (2009) Trends in performance char-
acteristics of modern automobile SI and diesel engines. SAE
Paper No. 2009-01-1892
68. CIMAC Working Group “Gas Engines” (2008) Information
about the use of LNG as engine fuel. CIMAC position paper
69. Westport Innovations Inc (2010) Direct injection natural gas
demonstration project. Westport GX Geavy-Duty LNG engine,
Canada
70. Mattas G, Thijssen B (2004) Dual-fuel diesel engines for LNG
carriers. Marine News 1
71. ETSAP (2010) Energy Technology Systems Analysis
Programme. Automotive LPG and Natural gas engines.
http://www.etsap.org
72. NREL (1999) Honda civic dedicated CNG Sedan. Fact Sheet
73. Honda website. http://automobiles.honda.com
74. Thien U (2008) Widening the driving range of NGV ‘s. Presen-
tation 6th European Forum Gas, Bratislava
Internal Combustion Engines, Alternative Fuels for 665
75. Korakianitis T, Namasivayam AM, Crookes RJ (2011) Natural-
gas fueled spark-ignition (SI) and compression-ignition (CI)
engine performance and emissions. Prog Energy Combust
Sci 37(1):89–112
76. Aslam MU, Masjuki HH, Kalam MA, Abdesselam H, Mahlia TMI,
Amalina MA (2006) An experimental investigation of CNG as
an alternative fuel for a retrofitted gasoline vehicle. Fuel
85:717–724
77. Geok HH, Mohamad TI, Abdullah S, Ali Y, Shamsudeen A, Adril
E (2009) Experimental investigation of performance and emis-
sion of a sequential port injection natural gas engine. Eur J Sci
Res 30(2):204–214, ISSN: 1450-216X
78. Cho HM, He B-Q (2008) Combustion and emissions character-
istics of a lean burn natural gas engine. Int J Automot Technol
9(4):415–422
79. Packham K (2007) Lean-burn engine technology increases
efficiency, reduces NOx emissions. Power topic #7009. Techni-
cal information from cummins power generation
80. Zeng K, Huang Z, Liu B, Liu L, Jiang D, Ren Y, Wang J (2006)
Combustion characteristics of a direct-injection natural gas
engine under various fuel injection timings. Appl Therm Eng
26:806–813
81. Goto Y, Sato Y, Narusawa K (1998) Combustion and emissions
characteristics in a direct injection natural gas engine using
multiple stage injection. Proceedings of the International
Symposium on Diagnostics and Modeling of Combustion in
Internal Combustion Engines (COMODIA), 1998. Japan Society
of Mechanical Engineers, pp 543–548
82. Tunestål P, Christensen M, Einewall P, Andersson T, Johansson
B (2002) Hydrogen addition for improved lean burn capability
of slow and fast burning natural gas combustion chambers.
SAE Paper No 2002-01-2686
83. ASTM Standardization News (2010) The renewed promise of
natural gas. ASTM Committee D03 and natural gas standards
by Adele Bassett. May/June 2010
84. Hallmannsegger M, Fickel H-C (2004) The mixture formation
process of an internal combustion engine for zero CO2-
emission vehicles fueled with cryogenic hydrogen. In: IFP
International Conference, Rueil-Malmaison, France
85. DOE EERE website. http://www1.eere.energy.gov/
hydrogenandfuelcells/storage/current_technology.html?m=1&
86. Wallner T (2009) Opportunities and risks for hydrogen internal
combustion engines in the United States. In: Conference
proceedings “Der Arbeitsprozess der Verbrennungskraft-
maschine.” Verlag d. Technischen Universität Graz. ISBN 978-3-
85125-068-8
87. Rottengruber H, Berckmueller M, Elsaesser G, Brehm N,
Schwarz C (2004) Operation strategies for hydrogen engines
with high power density and high efficiency. In: 15th annual U.
S. hydrogen conference, Los Angeles, CA
88. Verhelst S, Wallner T (2009) Hydrogen-fueled internal combus-
tion engines (H2ICEs). Prog Energy Combust Sci 35:490–527
89. Eichlseder H, Wallner T, Freymann R, Ringler J (2003) The
potential of hydrogen internal combustion engines in
a future mobility scenario. SAE Paper No. 2003-01-2267
666 Internal Combustion Engines, Alternative Fuels for
90. Welch A, Mumford D, Munshi S, Holbery J, Boyer B, Younkins
M, Jung H (2008) Challenges in developing hydrogen direct
injection technology for internal combustion engines. SAE
Paper No. 2008-01-2379
91. Heindl R, Eichlseder H, Spuller C, Gerbig F, Heller K (2009)
New and innovative combustion systems for the H2-ICE:
compression ignition and combined processes. SAE Paper
No. 2009-01-1421
92. Obermair H, Scarcelli R, Waller T (2010) Efficiency improved
combustion system for hydrogen direct injection operation.
SAE Paper No. 2010-01-2170
93. Tanno S, Ito Y, Michikawauchi R, Nakamura M, Tomita H
(2010) High-efficiency and Low-NOx hydrogen combustion
by high pressure direct injection. SAE Paper No. 2010-01-
2173
94. Tang X, Stockhausen WF, Kabat DM, Natkin RJ, Heffel JW
(2002) FordP hydrogen engine dynamometer development.
SAE Paper No. 2002-01-0242
95. Salimi F, Shamekhi AH, Pourkhesalian AM (2009) Role of
mixture richness, spark and valve timing in hydrogen-fuelled
engine performance and emission. Int J Hydrogen Energy
34:3922–3929
96. ETEC hydrogen internal combustion engine full-size pickup
truck conversion. Hydrogen ICE truck brochure
97. Wallner T, Lohse-Busch H, Shidore N (2008) Operating strat-
egy for a hydrogen engine for improved drive-cycle effi-
ciency and emissions behavior. Int J Hydrogen Energy
34:4617–4625
98. Wimmer A, Wallner T, Ringler J, Gerbig F (2005) H2-direct
injection – a highly promising combustion concept. SAE
Paper No. 2005-01-0108
99. Wallner T, Lohse-Busch H, Gurski S, Duoba M, Thiel W, Martin
D, Korn T (2008) Fuel economy and emissions evaluation of
a BMW hydrogen 7 mono-fuel demonstration vehicle. Int
J Hydrogen Energy 33:7607–7618
100. http://www.biodiesel.org/resources/biodiesel_basics/
101. United States Environmental Agency (2002) A comprehen-
sive analysis of biodiesel impacts on exhaust emissions.
EPA420-P-02-001, Oct 2002
102. Mueller CJ, Boehman AL, Martin GC (2009) An experimental
investigation of the origin of increased NOx emissions when
fueling a heavy-duty compression-ignition engine with soy
biodiesel. SAE Int J Fuels Lubr 2:789–816
103. Graboski et al (2003) The effect of biodiesel composition on
engine emissions from a DDC series 60 engine. NREL/SR-
510-31461
104. Agarwal AK (2006) Biofuels (alcohols and biodiesel) applica-
tions as fuels for internal combustion engines. Prog Energy
Combust Sci 33:233–271
105. Kuronen M, Mikkonen S, Aakko P, Murtonen T (2007)
Hydrotreated vegetable oil as fuel for heavy duty diesel
engines. SAE Technical Paper 2007-01-4031
106. Aatola H, Larmi M, Sarjovaara T, Mikkonen S (2008)
Hydrotreated Vegetable Oil (HVO) as a renewable diesel
fuel: trade-off between NOx, particulate emission, and fuel
consumption of a heavy duty engine. SAE Technical Paper
2008-01-2500
107. Miers SA, Ng H, Ciatti SA, Stork K (2005) Emissions, perfor-
mance, and In-cylinder combustion analysis in a light-duty
diesel engine operating on a Fischer-Tropsch, Biomass-to-
Liquid Fuel. SAE Technical Paper 2005-01-367
108. Gill SS et al (2010) Combustion characteristics and emissions
of Fischer Tropsch diesel fuels in IC engines. Progr Energy
Combust Sci 31:466–487. doi:10.1016/j.pecs.2010.09.001
109. Arcoumanis C, Bae C, Crookes R, Kinoshita E (2008) The
potential of di-methyl ether (DME) as an alternative fuel for
compression-ignition engines: a review. Fuel 87:1014–1030
110. Semelsberger TA, Borup RL, Greene HL (2006) Dimethyl ether
(DME) as an alternative fuel. J Power Sour 156:497–511
111. McCormick RL, Parish R (2001) Advanced petroleum based
fuels Program and renewable diesel program milestone
report: technical barriers to the use of ethanol in diesel fuel.
NREL/MP-540-32674
112. Waterland LR, Venkatesh S, Unnasch S (2003) Safety and
performance assessment of ethanol/diesel blends (E-diesel).
Subcontractor Report NREL/SR-540-34817
113. Hansen AC, Zhang Q, Lyne PWL (2005) Ethanol - diesel fuel
blends – a review. Bioresour Technol 96:277–285
114. Rakopoulos CD, Antonopoulos KA, Rakopoulos DC
(2007) Experimental heat release analysis and emissions of
a HSDI diesel engine fueled with ethanol–diesel fuel blends.
Energy 32:1791–1808
115. Sayin C (2010) Engine performance and exhaust gas emis-
sions of methanol and ethanol–diesel blends. Fuel 89:3410–
3415
116. Miers S, Carlson R, Ng H, McConnell S, Wallner T, LeFeber J
(2008) Drive cycle analysis of butanol/diesel blends in a light-
duty vehicle. SAE Paper No. 2008-01-2381
117. Rakopoulos DC, Rakopoulos CD, Hountalas DT, Kakaras EC,
Giakoumis EG, Papagiannakis RG (2010) Investigation of the
performance and emissions of bus engine operating on
butanol/diesel fuel blends. Fuel 89:2781–2790
Books and Reviews
Agarwal AK (2007) Biofuels (alcohols and biodiesel) applications as
fuels for internal combustion engines. Prog Energy Combust
Sci 33:233–271
Eichlseder H, Klell M (2008) Hydrogen in automotive engineering
(in German, Wasserstoff in der Fahrzeugtechnik). Vieweg +
Teubner, Wiesbaden. ISBN 978-9-8348-0478-5
Huo H, Wu Y, Wang M (2009) Total versus urban: well-to-wheels
assessment of criteria pollutant emissions from various
vehicle/fuel systems. Atmos Environ 43:1796–1804
MacLeana HL, Lave LB (2003) Evaluating automobile fuel/propulsion
system technologies. Prog Energy Combust Sci 29:1–69
Pischinger R, Klell M, Sams T (2002) Thermodynamics of internal
combustion engines (in German: Thermodynamik der
Verbrennungskraftmaschine). Springer, Wien. ISBN 3-211-
83679-9
 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and
Abroad, Examination of History and
Innovations
JOSÉ I. FARRÁN
Adavant Consulting, San Francisco, CA, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Pedestrian Conflicts
Future Directions
Bibliography
Glossary
Bedstead barrier A type of pedestrian-light rail cross-
ing control device whose design prevents a pedes-
trian from walking straight onto the tracks by
forcing them to make one or more turns immedi-
ately ahead of the tracks.
Channelization device A physical element that limits
pedestrian movements to within the preferred
pathway and that guides pedestrians to cross the
LRT tracks at the designated crossing location.
Dynamic envelope The clearance on either side of
a light rail vehicle required for the overhang
resulting from any combination of loading, lateral
motion, or suspension failure.
Flashing light signal assembly A warning device
consisting of two red signal indications arranged
horizontally that are activated to flash alternately
when a light rail vehicle is approaching or present at
a highway-rail at-grade crossing.
Light rail transit (LRT) A mode of transit service (also
called streetcar, tramway, or trolley) operating pas-
senger rail cars singly (or in short, usually two-car
or three-car, trains) on fixed rails in right-of-way
that is often separated from other traffic for part or
much of the way.
Light rail vehicles (LRVs) Vehicles typically driven
electrically with power being drawn from an over-
head electric line via a trolley or a pantograph
driven by an operator on board the vehicle and
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
667
may have either high platform loading or low-
level boarding using steps.
Off-quadrant The location at a light rail crossing, from
the perspective of an approaching motorist, on the
far side of the tracks and same side of the roadway.
Warning devices for pedestrians at highway-rail
crossings are typically absent in the off-quadrant.
Pavement markings Markings set into the surface of,
applied upon, or attached to the pavement for the
purpose of regulating, warning, or guiding traffic.
Pedestrian A person who travels on foot or who uses
assistive devices, such as a wheelchair, for mobility.
Pedestrian automatic gate A type automatic gate at
a highway-rail at-grade crossing that descends when
activated by a light rail vehicle blocking the pedes-
trian path. The gate is typically delineated with red
and white diagonal bars along its length and may
include one small red light at the tip, which is
illuminated when the gate is activated.
Pedestrian-light rail at-grade crossing The general
area where a pedestrian pathway and a light rail
vehicle cross at the same level, within which are
included the light rail tracks, pathway, design fea-
tures, and traffic control devices for those traversing
the area.
Positive control device An element that provides
a temporary or permanent physical barrier between
the light rail tracks and locations where pedestrians
can safely queue.
Public transportation (also called transit, public
transit, or mass transit) Transportation by
a conveyance that provides regular and continuing
general or special transportation to the public, but
not including school buses, charter, or sightseeing
service.
Right-of-way (r-o-w) A strip of land devoted to light
rail transportation purposes.
Traffic control devices All signs, signals, markings,
and other devices used to regulate, warn, or guide
motor vehicle, pedestrian, or bicycle traffic.
Swing gate A self-closing, non-locked, fence-type gate
designed to swing open away from the track area
and return to the closed position upon release.
Z-crossing A type of pedestrian-light rail crossing
control device whose design forces a pedestrian to
make a 90-degree turn and walk parallel to and
# 2012, DOI 10.1007/978-1-4419-0851-3
668 Light Rail Transit in the US and Abroad, Examination of History and Innovations
facing oncoming light rail vehicles immediately
ahead of the tracks.
Definition of the Subject
Descendants of the streetcar, light rail vehicles (LRVs)
have their own distinct characteristics, including a
broad range of operating environments (both on-street
and in semi-exclusive or exclusive rights-of-way) and
a wide extent of typical operating speeds (from 15 to 60
mph or 10 to 40 km/h). This flexibility, coupled with
passenger attraction and capacity of the vehicles, has
made light rail an increasingly viable public transpor-
tation option.
Accommodating light rail transit (LRT) in these
multiple environments while maintaining or improv-
ing safety has led transportation planners and engi-
neers to utilize a variety of innovative traffic control
practices. These practices are used to manage
conflicting movements with LRVs and to provide
motorists, as well as pedestrians and bicyclists with
better information upon which to base their decisions
and have evolved over time in response to changes in
the way that LRVs operate (higher speeds, more silent
operation) and pedestrians behave (increased inatten-
tiveness and more distractions). Since increased LRT
safety is a worldwide need, traffic engineers and trans-
portation planners in the United States and Europe
have implemented and adjusted over the years
a number of practices to minimize conflicts between
LRVs and pedestrians.
This entry examines and describes the observations
resulting from the author’s research and firsthand
experience in traffic engineering practices in the United
States and Europe related to solving LRT conflicts with
pedestrians. Those practices and traffic control devices
that are most innovative and with the greatest potential
for implementation in new or expanding LRT systems
are graphically identified.
Introduction
A key strength of LRT is its ability to integrate well into
an urban environment sharing the public right-of-way
with motorists, bicyclists, and pedestrians. It is this
characteristic that makes LRT successful in those
urban corridors where a high-capacity mode of trans-
portation is required to operate well in both an
exclusive and a shared right-of-way. Although LRT is
an inherently safe mode of transportation, accidents
sometimes do occur, negatively affecting not only
the parties involved in the accident, but also the entire
operation of the LRT system and the reputation of the
transit property. Furthermore, safety concerns, partic-
ularly with pedestrians, are always a subject of great
debate between residents and transit planning agencies,
in those communities where a new LRT system or an
LRT line extension is being proposed.
The interactions between pedestrians and LRVs are
substantially different from those between motorists
and LRVs. In general, motorists tend to be more
aware of their environment, while pedestrians, travel-
ing in the relatively safe venue of protected sidewalk
areas, do not routinely share the same continuous
attentive edge.
When crossing the travel path of motor vehicles or
LRVs, pedestrians are supposed to shift their attention
to a similar state of awareness as that exhibited by
motorists or LRV operators. However, this shift does
not always occur. Pedestrian inattentiveness is exacer-
bated when interacting with modern LRVs, which are
nearly silent. Moreover, unlike motor vehicles, LRVs
cannot swerve or stop quickly enough to compensate
for pedestrians who are errant or disobedient of traffic
control devices.
According to past research conducted by the author
[1], incidents between pedestrians and LRVs are the
least prevalent type of LRT-related collisions,
representing approximately 10% of the total, but are
the most severe and account for at least 50% of all
fatalities resulting from such incidents. Additional
research conducted for Transit Cooperative Research
Board [2] further validated this finding.
Table 1 summarizes the total number of collisions
with injuries and fatalities that have occurred at five
LRT systems operating in the Western United States
(Los Angeles, Portland, San Francisco, and Santa
Clara) over the past few years. According to Table 1,
more than three fourths of those involved in
a pedestrian collision with an LRV are killed, compared
to less than one fifth of those involved in a motor
vehicle collision.
Similar, more extensive, geographical data provides
similar conclusions. Table 2 presents the aggregate
number of collisions by type that have occurred at
 Light Rail Transit in the US and Abroad, Examination of History and Innovations 669

most of the LRT systems in the United States from 2002 of three persons are killed for every ten that are
to 2007. involved in a pedestrian collision with an LRV, as
According to Table 2, LRV collisions with pedes- opposed to three persons killed for every 100 that are
trians represent approximately one fifth of the total but involved in a motor vehicle collision with an LRV. It is
result in about two thirds of all fatalities resulting from clear that finding an effective means of addressing and
those collisions. Also according to Table 2, an average minimizing pedestrian-LRV collisions is critically
important to the operations of LRT systems in the
United States and abroad.
Light Rail Transit in the US and Abroad, Examination of The interactions between pedestrians and LRVs are
History and Innovations. Table 1 Aggregate number of substantially different from those between motorists
LRV collisions by typea and LRVs. In general, motorists tend to be more
Resulting in Resulting in aware of their environment, while pedestrians, travel-
LRV collisions with injuries fatalities ing in the relatively safe venue of protected sidewalk
Motor vehicles 184 53% 19 19% areas, do not routinely share the same continuous
Pedestrians 115 33% 74 76%
attentive edge. Moreover, unlike motor vehicles, LRVs
cannot swerve or stop quickly enough to compensate
Other 50 14% 5 5%
for pedestrians who are errant or disobedient of traffic
Total 349 100% 98 100% control devices.
a
Provided by LRT agencies in Los Angeles (1990–2006), Portland Possible strategies to minimize LRV/pedestrian
(1994–2007), San Francisco (2006–2007), and San Jose (1987–2007) conflicts are summarized in Table 3. These solutions

Light Rail Transit in the US and Abroad, Examination of History and Innovations. Table 2 Aggregate number of LRV
collisions by typea 2002–2007
LRV collisions with Resulting in injuries Resulting in fatalities Property damage only Total
Motor vehicles 272 66% 17 24% 278 93% 567 72%
Pedestrians 117 28% 47 67% 4 1% 168 21%
Other 26 6% 6 9% 17 6% 49 6%
Total 415 100% 70 100% 299 100% 784 100%
a
Reported to the Federal Transit Administration by the LRT agencies in Baltimore, Boston, Buffalo, Cleveland, Dallas, Denver, Houston,
Los Angeles, Memphis, Minneapolis, New Jersey, New Orleans, Philadelphia, Pittsburgh, Portland, Sacramento, Salt Lake City, San Diego,
San Francisco, San Jose, Seattle, St. Louis, and Tampa

Light Rail Transit in the US and Abroad, Examination of History and Innovations. Table 3 Possible solutions to
observed pedestrian conflicts

Situation Possible solution

Low level of pedestrian
awareness
Poor definition of pedestrian
pathways
Cross-street access
● Install LRV-activated signs and signals
● Install LRV-specific passive signage at pedestrian eye level
● Delineate dynamic envelope by contrasting pavement color and/or texture or paint
● Construct physical barriers (fence, bollards with chains, landscape)
● Provide adequate storage/queuing/refuge space
● Install positive pedestrian control devices (automatic gates, swing gates, bedstead
barriers, or Z-crossings)
670 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 1
LRV-activated internally illuminated warning sign (urban
environment, median operation) Salt Lake City (Utah)
address the types of situations identified above, and
underscore the need for establishing a set of principles
and guidelines to achieve greater uniformity and con-
sistency in implementing a safer LRT system.
The following sections of this entry provide
a summary of innovative pedestrian control designs
and practices that have been applied successfully to
light rail system in the United States and Europe to
address the types of conflicting motor vehicle and
pedestrian movements with LRVs described above.
Pedestrian Conflicts
The design of a pedestrian-light rail at-grade crossing is
only effective if pedestrians actually cross at the
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 2
LRV-activated internally illuminated warning sign (urban
environment, median operation) Valencia (Spain)
designated point by taking a path that allows them
clear observation of the warning devices. Devices to
control pedestrian travel across LRT tracks and warn
about the presence or approach of an LRV can be
grouped under three major categories [3]:
● Regulatory and warning devices (both passive and
active)
● Delineation markings
● Positive control devices
This section describes some of the most innovative
traffic control devices included under each category and
provides graphic examples of their application at several
LRT systems in the United States and Europe.
 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 3
LRV-activated internally illuminated warning sign (urban
environment, median operation) Bilbao (Spain)
Regulatory and Warning Devices
Pedestrian crossing regulatory and warning devices
include signs, pedestrian signals, flashing light signals,
and audio devices. Signs used at LRT crossings are
typically fixed standard signs, such as the railroad
cross buck or the LRV symbol, or LRV-activated inter-
nally illuminated depicting words or graphic symbols.
LRV-activated internally illuminated signs warn pedes-
trians of the increased risk associated with violating the
crossing in the presence of an LRV. Figure 1 shows an
example of an LRV-activated internally illuminated
warning sign in Salt Lake City, used in combination
with a pedestrian traffic signal. Figures 2 and 3 show
similar internally illuminated signs in Valencia and
Bilbao, depicting two variations of the international
671
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 4
LRV-activated internally illuminated warning sign with
directional arrow (urban environment, median operation)
Portland (Oregon)
LRT crossing warning sign. These signs are typically
installed where substantial pedestrian presence is
expected or where additional information is needed
due to complicated track or pedestrian crossing
geometry.
Figure 4 shows an innovative internally illuminated
sign in Portland depicting a side view of an LRV plus an
arrow representing its direction of travel. LRV-activated
internally illuminated signs are also used to warn pedes-
trians of additional LRVs approaching the crossing.
Figure 5 shows the use of an LRV-activated internally
illuminated SECOND TRAIN APPROACHING sign in
San Jose, used in combination with a conventional
flashing light signal assembly.
672 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 5
LRV-activated SECOND TRAIN APPROACHING warning sign
(suburban environment, separate r-o-w) San Jose
(California)
Supplemental warning plaques are also used
extensively in United States and European cities eval-
uated in this paper. Figure 6 shows an example of
a supplemental plaque used in combination with
a pedestrian traffic signal head and an LRV-activated
internally illuminated warning sign in San Francisco.
Figure 7 shows a similar example used in the Barcelona
LRT system, while Fig. 8 shows an installation in
Bordeaux where the plaque is combined with an LRV
warning sign at a pedestrian-only crossing. The intent
of the signs and plaques is to clearly alert pedestrians
about the possibility of LRVs approaching the crossings
from both sides, and are typically installed at all pedes-
trian crossings.
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 6
LRV-activated internally illuminated warning sign with
supplemental plaque (urban environment, median
operation) San Francisco (California)
Pedestrian crosswalk signals in most cities are sim-
ilar to those installed at non-LRV crosswalks, with
internally illuminated heads depicting the pedestrian
WALK and DON’T WALK symbols. The city
of Bordeaux, however, has installed LRV-specific regu-
latory signals at key locations with heavy pedestrian
usage, such as the pedestrian mall in the old town area
(Fig. 9), to emphasize the impending arrival of a train
and raise the pedestrian awareness about the prohibition
to cross the street. The signal is normally blanked out
and becomes activated by the approaching LRV.
Flashing red light signals used at pedestrian-LRV
crossings have the same design and type of operation as
those used in conventional highway-rail grade
 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 7
Two-way LRV warning sign with LRV symbol used with
a pedestrian signal (urban environment, median
operation) Barcelona (Spain)
crossings, consisting of two red light units mounted on
a horizontal line flashing alternately to indicate the
approach or presence of an LRV. Figure 10 shows
a pedestrian flashing red light signal assembly in San
Jose, used in combination with an LRV warning sign
and a supplemental plaque.
Delineation Markings
Delineation markings direct pedestrians to cross LRT
tracks at a designated location. Pedestrian path delin-
eation can be accomplished with line striping, differ-
ential pavement color or texture, contrasting surface
materials or landscaping. Delineation is also used to
673
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 8
Two-way LRV warning sign with LRV symbol at an
unsignalized crossing (urban environment, median
operation) Bordeaux (France)
mark the edge of the dynamic envelope of the LRV, that
is, the clearance on either side of an LRV required
for the overhang resulting from any combination of
loading, lateral motion, or suspension failure. Figure 11
shows a combination of line striping, contrasting color
and differential pavement texture to delineate a pedes-
trian crossing in Dallas. Figures 12 and 13 depict a
similar combination found in Barcelona and Bilbao,
respectively, used to delineate the LRV dynamic
envelope.
In addition to the dynamic envelope delineations
shown in Fig. 12, arrow striping indicating the direc-
tion that LRVs typically traverse the crossing is also
674 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 9
LRV-activated pedestrian signal (urban environment,
pedestrian mall operation) Bordeaux (France)
used extensively in Barcelona at all pedestrian crossings
to warn pedestrians and help them look in the most
appropriate direction (the opposite from where the
arrow is pointing) before they walk onto the track
area. The arrow in combination with the LRV symbol
is striped between the two rails for a given LRV direc-
tion immediately downstream of the pedestrian path-
way. A single arrow is used where LRVs typically
operate in a single direction (Fig. 14), while two arrows
are used when LRVs typically operate two-way in
a single track (Fig. 15).
As an alternative to the previous European mark-
ings, LRT systems in the United States, such as those in
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 10
Pedestrian flashing red light signals (suburban
environment, separate r-o-w) San Jose (California)
Salt Lake City and San Francisco, have implemented
pavement markings at the crosswalk edge with the
word LOOK and two arrow heads pointing both ways,
either adjacent to the tracks or on the sidewalk, as
shown in Figs. 16 and 17.
Positive Control Devices
Positive control devices provide a physical barrier
between the outer edge of the LRV dynamic envelope
and the area where it is safe for pedestrians to congre-
gate. These devices can be fixed or moveable.
Fixed barriers are used to restrict pedestrian move-
ments as they approach an LRT crossing and lead
 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 11
Pedestrian-LRV crossing delineation (urban environment,
separate r-o-w) Dallas (Texas)
them toward a designated crossing location. They
include fencing, railing, bollards, and chains or wire
strung between posts. Fixed barriers, such as those
shown in Fig. 18 for the Dallas LRT system, are used to
reinforce delineation markings and to increase aware-
ness of the potential presence of an LRV at those loca-
tions where a more strict control of pedestrian flows
approaching a crossing may be necessary for safety
considerations.
Fixed barriers are also used to configure pedestrian-
only crossings of the LRT tracks. The most common
types of fixed-barrier crossings are Z-crossings and
bedstead barrier crossings.
675
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 12
LRV dynamic envelope delineation (urban environment,
median operation) Barcelona (Spain)
The Z-crossing is designed to turn pedestrians
toward the direction of approaching LRVs before they
cross each track, forcing them to take note of potential
danger. They are typically used at mid-block (away
from intersections) pedestrian-only crossings. Bed-
stead barrier crossings are similar to Z-crossings, but
their design is more compact so that they can be used in
tight urban spaces. The barriers are placed in an offset,
maze-like manner that requires pedestrians moving
across the LRT tracks to turn toward the direction
of approaching LRVs before they cross each track.
Both Z-crossings and bedstead barrier crossings can
also be used in combination with other devices such
676 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 13
LRV dynamic envelope delineation (urban environment,
median operation) Bilbao (Spain)
as signals or automatic gates. Figure 19 depicts
a Z-crossing in San Francisco while Fig. 20 shows
a bedstead barrier crossing in Bordeaux, both used in
combination with pedestrian signals and supplemental
warning plaques.
Moveable barriers are used to restrict pedestrian
movements as they are about to enter into the dynamic
envelope of the LRV. The barriers can be automatically
activated and deactivated by an LRV as it approaches
and departs the crossing, or they can be operated man-
ually by the pedestrian. The two most common types of
moveable barriers at LRV crossings are the swing gate
and the pedestrian automatic gate.
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 14
LRV directional striping on a one-way track (urban
environment, median operation) Barcelona (Spain)
A swing gate is a manual device that requires
a pedestrian to pull the gate in order to enter the
crossing and to push the gate to exit the protected
track area. Operating the gate alerts pedestrians to the
presence of the LRT crossing, forces them to pause, and
prevents them from running freely across the LRT
tracks. Swing gates do not lock when closed and are
designed to return to the closed position by gravity
after passage of the pedestrian.
Swing gates are often used in conjunction
with channelization barriers, LRV-activated internally
illuminated signs, flashing light signals, and audible
warning devices, as shown in Figs. 21 (San Jose) and
 Light Rail Transit in the US and Abroad, Examination of History and Innovations 677

Light Rail Transit in the US and Abroad, Examination of History and Innovations. Figure 15
LRV directional striping on a two-way track (urban environment, median operation) Barcelona (Spain)

Light Rail Transit in the US and Abroad, Examination of History and Innovations. Figure 16
LOOK pavement marking (urban environment, median operation) Salt Lake City (Utah)
678 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 17
LOOK pavement marking (urban environment, median
operation) San Francisco (California)
22 (Los Angeles), in order to achieve a higher level of
positive pedestrian control at the crossing.
Pedestrian automatic gates are configured and
operate much the same as vehicular gates, except that
the arms are shorter and may include one or three red
lights at the center or the tip that flash when the gate is
activated. When used at pedestrian-rail at-grade cross-
ings, the gate arm is located approximately 3 feet above
the pathway when in the horizontal position.
At those locations where there is a concern about
pedestrians ducking under the arm when the gate is
closed, a secondary arm may be installed to hang below
the primary arm, as shown in Fig. 23 for the Dallas LRT
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 18
Pedestrian delineation with bollards and chains (urban
environment, separate r-o-w) Dallas (Texas)
system. Pedestrian automatic gates are used to physi-
cally discourage pedestrians from crossing the LRT
tracks when an LRV approaches the crossing, particu-
larly at locations with inadequate stopping sight
distance.
Because pedestrians may approach the crossing
from both ways, the preferred method for pedestrian
automatic gate installation is to provide them at
both sides of the crossings (in all four quadrants
when installed at motor vehicle and pedestrian
grade crossings). Establishing clearly marked pedes-
trian safety zones and escape paths minimizes the
possibility of trapping pedestrians in the LRT
 Light Rail Transit in the US and Abroad, Examination of History and Innovations 679

Light Rail Transit in the US and Abroad, Examination of History and Innovations. Figure 19
Pedestrian Z-Crossing configuration used in combination with a pedestrian signal (urban environment, median operation)
San Francisco (California)

Light Rail Transit in the US and Abroad, Examination of History and Innovations. Figure 20
Bedstead barrier at mid-block crossing used in combination with pedestrian signal (urban environment, median
operation) Bordeaux (France)
680 Light Rail Transit in the US and Abroad, Examination of History and Innovations
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 21
Pedestrian swing gate installation (suburban environment,
separate r-o-w) San Jose (California)
right-of-way when pedestrian automatic gates are
used at both ends of the crossing. An emergency
exit swing gate can be used for this purpose, desig-
nated for use only as an escape route for a pedes-
trian that may remain between the track and a
lowered automatic pedestrian gate. Figure 24 presents
an example of the placement of emergency exit swing
gates at the Los Angeles LRT system.
Future Directions
The previous sections have shown that although
accidents between pedestrians and LRVs are the
least common type of LRT accident, they are by far
Light Rail Transit in the US and Abroad, Examination of
History and Innovations. Figure 22
Pedestrian swing gate installation (urban environment,
separate r-o-w) Los Angeles (California)
the most severe. A means of addressing and mini-
mizing this type of accidents is by installing devices
that convey to pedestrians the higher level of risk
associated with violating a closed LRV crossing. The
types of devices used to control pedestrian travel
across the LRT tracks and warn them about the
presence of an approaching LRV include regulatory
and warning devices (both passive and active),
delineation markings, and positive control devices.
These pedestrian control devices are typically
implemented in combination in order to achieve
a higher level of pedestrian control at the LRT
crossing.
 Light Rail Transit in the US and Abroad, Examination of History and Innovations 681

Light Rail Transit in the US and Abroad, Examination of

History and Innovations. Figure 23
Pedestrian automatic gate with secondary arm (suburban
environment, separate r-o-w) Dallas (Texas)

Light Rail Transit in the US and Abroad, Examination of

History and Innovations. Figure 24
Bibliography
Pedestrian automatic gate with push gate as escape path
Primary Literature (urban environment, separate r-o-w) Los Angeles
(California)
1. Korve HW, Farrán JI, Mansel DM, Levinson HS, Chira-Chavala T,
Ragland DR (1996) Integration of light rail transit into city
streets. In: TCRP 17, Transportation Research Board,
Washington, DC
Transportation, Federal Transit Administration, Washington,
2. Cleghorn D, Clavelle A, Boone J, Masliah M, Levinson HS
DC, pp 1–69
(2009) Improving pedestrian and motorist safety along light
Korve Engineering, ATS Consulting, LLC, Fidell Associates, CERS,
rail alignments. In: TCRP 137, Transportation Research Board,
and Bear Consulting (2007) Audible signals for pedestrian
Washington, DC
safety in LRT environments. Research Results Digest 84, Trans-
3. Korve HW, Ogden BD, Siques JT, Mansel DM, Richards HA,
portation Research Board, Washington, DC
Gilbert S, Boni E, Butchko M, Stutts JC, Hughes RG
State of California Department of Transportation (2006) California
(2001) Light rail service: pedestrian and vehicular safety.
manual on uniform traffic control devices for streets and
In: TCRP 69, Transportation Research Board, Washington, DC
highways, 2006 edn. Part 10: Light rail transit grade crossings.
Business, Transportation and Housing Agency Department of
Books and Reviews, Monographs Transportation, CA, pp 10A-1–10D-12
State U.S. Department of Transportation – Federal Highway
Farradyne PB (2002) Final report for the second train warning sign Administration/Rail Grade Crossing Technical Working Group
demonstration project on the Los Angeles metro blue line. In: (2002) Guidance on traffic control at highway-rail grade cross-
FTA-CA-26-7017-01, Los Angeles, CA, US Department of ings. FHWA, Washington, DC
682 Light Rail Transit in the US and Abroad, Examination of History and Innovations
U.S. Department of Transportation – Federal Highway Administra-
tion (2009) Manual on uniform traffic control devices for
streets and highways, 2009 edn. Part 8: traffic control for
railroad and light rail transit grade crossings. Business, Trans-
portation and Housing Agency Department of Transportation,
CA, pp 747–788
U.S. Department of Transportation – Federal Highway
Administration – Office of Safety Design (2007) Railroad-
highway at-grade crossing handbook, revised 2nd edn.
Federal Highway Administration – Office of Safety Design,
Washington, DC
Collections
Boorse JW (2003) Delineation of the dynamic envelope of light rail
transit cars and trains. In: TRB 2003 Annual Meeting, Transpor-
tation Research Board, Washington, DC
Currie G, Smith P (2005) An innovative design for safe and acces-
sible light rail/tram stops suitable for streetcar style conditions
In: TRB 2005 Annual Meeting, Transportation Research Board,
Washington, DC, pp 1–15
Farrán JI (1998) The light rail transit safety experience in North
America. In: Urban transport IV – urban transport and
the environment for the 21st century. WIT Press, Computa-
tional Mechanics Publications, Ashurst, Southampton, UK,
pp 43–52
Farrán JI (2002) The pedestrian paradigm: controlling pedestrians
across LRT tracks. In: Commuter rail/transit conference pro-
ceedings. Boca Raton, FL, pp 1–9
Farrán JI (2007) Innovative traffic control practices for light rail
transit: a European perspective. In: Joint international light
rail conference: a World of applications and opportunities,
Transportation Research E-C112, TRB, Washington, DC,
pp 87–106
Irwin D (2003) Safety criteria for light rail pedestrian crossing.
In: Proceedings 9th national light rail transit conference,
American Public Transportation Association, TRB, Washington,
DC, pp 266–288
Korve HW, Farrán JI, Mansel DM (1995) Pedestrian control systems
for light rail transit. In: Seventh national conference on
light rail transit: building on success–learning from experi-
ence, Compendium of Technical Papers, Baltimore, MD,
pp 91–102
Journal Articles
Farrán JI (2006) Pedestrian and motor vehicle traffic control prac-
tices for LRT – Innovations in the new Barcelona LRT system,
TRB (2006). Trans Res Record 1955:56–61
Siques JT (201) Pedestrian warning and control devices, guidelines,
and case studies TRB (2001). Trans Res Record 1762:18–24
 Light Rail Transit, Shared Infrastructural Issues 683

Light Rail Transit, Shared Definition of the Subject

Infrastructural Issues Interest in light rail transit (LRT), as a viable urban
transportation system, has been growing worldwide
RONGFANG RACHEL LIU
and new LRT systems are appearing in urban areas of
Department of Civil and Environmental Engineering,
many countries. One advantage of LRT is its ability to
New Jersey Institute of Technology, Newark, NJ, USA
use existing railroad infrastructure to reduce construc-
tion cost and environmental impact. The majority of
LRT systems in the USA share right-of-way, or utilize
Article Outline
the same tracks but temporally separated from the
Glossary freight railroads, while commingled or simultaneous
Definition of the Subject operation has been common practice for many
Introduction countries in Asia and Europe. With the success of
Historical Background LRT applications in recent decades, service providers
Current Operating Scenarios in the USA are looking for more space and infrastructure for
Demand for Commingled Operation expansion. Interest in applying overseas commingled
Challenges Faced by the Attempts experience in the USA has been growing.
Potential Solutions Current Federal Railroad Administration (FRA)
Future Directions policy, dominated by safety concerns, does not allow
Acronyms commingling of LRT and freight rail operations within
Bibliography the same time period. The general practice is that light
rail agencies own or control the track system, operate
transit service as the host, and lease or permit freight
Glossary
operations in off-peak periods. The majority of the
Commingled operation Transit trains and freight freight rail is local deliveries to businesses involving
trains are separated by headway intervals measured relatively short, slow-moving trains.
in minutes in an operating schedule. Most of the new light rail services operate through
Light rail transit Also called streetcar, tramway, or central cities on city streets, but new light rail services
trolley, is a system with passenger rail cars operating tend to be extended into the suburbs on private
singly or in short, usually two-car, trains on fixed right-of-way, with longer station spacing. Though the
rails in right-of-way that is not separated from commingled operation between LRT and freight
other traffic for much of the way. railroad is a practice adopted in Europe, it is currently
Shared corridor Transit and freight operators share a not permitted in the USA. However, with the technical
transportation corridor, but tracks are separated by innovations in rolling stock and control systems, more
at least 25 ft and no more than 200 ft. LRTservice providers are looking at the potential appli-
Shared right-of-way The transit vehicles run on cability of European experience in the USA, and the
separate tracks, but track centers are less than feasibility of commingled operation between LRT and
25 ft, that is, separation between the centerline of freight railroad is increasingly being discussed.
the freight track and the centerline of the passenger According to recent studies [1–3], FRA defined
track is less than 25 ft. three types of shared-use transit and freight operations,
Shared track Heavy or LRT vehicles operate on the which are shared track, shared right-of-way (ROW),
same tracks used by freight trains. and shared corridor.
Temporal separation Both transit and freight trains Shared Track: Heavy or LRT vehicles operate on the
utilize the same track but are separated by time same tracks used by freight trains. FRA regulations
windows. govern this type of operation, in which time separation,

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
684 Light Rail Transit, Shared Infrastructural Issues

that is, no simultaneous operation of freight trains and 2000 450

Passenger miles
LRVs, is required in most cases. 1800 400
Passenger Trips
Commingled Operations (Mixed Operation): Transit 1600

Passenger miles (millions)

350

Passenger trips (millions)

trains and freight trains are separated by headway 1400
300
intervals measured in minutes in an operating sched-
1200
ule. Though this approach occurs in Europe and Japan 250
1000
for all sorts of transit operations, it is only approved for 200
commuter services governed by rigid FRA rules in the 800
150
USA. Joint operation between LRT and freight railroad 600
operation is not allowed [1]. 400 100
Temporally Separated Operations (Time-Separated 200 50
Operation): Both transit and freight trains utilize the
0 0
same track but are separated by time windows. In

North America, the common practice is to operate
transit during the day and freight at night, generally
in the early morning hours [1].
Shared Right-of-Way (ROW): In this case, the tran-
sit vehicles run on separate tracks, but track centers are
less than 25 ft, that is, separation between the centerline
of the freight track and the centerline of the passenger
track is less than 25 ft. FRA requires railway mainte-
nance workers to observe specific safety precautions
when multiple main tracks are adjacent.
Shared Corridor: Transit and freight operators share
a transportation corridor, but tracks are separated by at
least 25 ft and no more than 200 ft. FRA believes that
intrusion by derailed freight or transit cars onto
a parallel railroad track is unlikely beyond 200 ft, so
does not consider separations of more than 200 ft to be
shared corridor for purposes of safety regulation.
Commingled operation, for the purpose of this
entry, is defined as “simultaneous train operation on
shared track by railroad trains and by light rail transit
vehicles that are not fully FRA compliant.” Shared right-
of-way, similar to FRA’s definition “adjacent tracks,” is
“the arrangement that the track center spacing between
the freight track and adjacent LRT track is less than 25 ft.”
Shared corridor is “the arrangement that the track center
spacing between the freight track and LRT track is more
than 25 ft but less than 200 ft.”
Seeking a better understanding of the current inter-
action practices between rail transit and freight rail-
roads, the first three sections of this entry presents the
historical background and the general landscape of
shared-use practices in the USA and overseas, which
are largely based on a survey of transit and freight
railroad interaction in North America and extensive
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
Year
Light Rail Transit, Shared Infrastructural Issues. Figure 1
Historical Growth of LRT Usages Source: Based on Data
from APTA, 2009
research and communication with their counterparts
overseas. The fourth and fifth sections of this entry
discuss potential demands and challenges faced by
transit service providers and freight railroads when
and if commingled operation is permitted in the
USA. Potential solutions and future directions of
research are included in the last two sections.
Introduction
Congestion on highways and urban streets has stimu-
lated renewed interest in rail transit in a large number
of US cities. Data compiled by the American Public
Transportation Association (APTA) reveal that, over
the past three decades, total passenger-miles carried
by US public transit almost doubled, from 30 billion
in 1997 to 53 billion in 2007 [4]. Among the modes of
rail transit, the number of unlinked passenger trips by
Light Rail Transit (LRT) quadrupled, from 103 million
in 1977 to 419 million in 2007, as shown in Fig. 1. The
explosive growth of LRT usage has largely been pro-
pelled by the renaissance of LRT in the USA after the
long decline of streetcar and interurban systems
between 1920 and 1960.
An attractive feature of LRT is that it can use some
of the existing real estate of the general railroad system,
which significantly reduces the cost of construction and

environmental impacts by eliminating the need to
build a new rail corridor. However, physical incompat-
ibility, disparate operating practices, and safety regula-
tions have led to separating the two operations
temporally. To better understand the interaction
between light rail transit and freight railroad, this
entry highlights the difference between LRT and freight
railroad operations and critical issues in shared use of
railroad infrastructure. There are major structural dif-
ferences between the Federal Railroad Administration
(FRA)-compliant cars and light rail vehicles (LRV),
therefore FRA requires that, to minimize or eliminate
the risk of collision between the two types of equip-
ment, complete physical or, at least, temporal
separation of their systems must be established.
A few LRT service providers, reaping the benefit of
newly developed LRT, and at the same time facing the
capacity constraints for more LRT services, are willing
to challenge the temporal separation rules. The idea of
commingled operation may not be totally irresponsible
thinking, given the fact that a number of commingled
operations do exist in Europe and Asia [5].
Historical Background
A brief review of historical development of LRT and
shared-use issues may help to understand the origins of
the conflicts and expedite the pace to derive solutions.
Congestion along highways and urban streets in a large
number of American cities has stimulated renewed
interest in guideway transit, especially light rail transit
(LRT). The interest in commingling freight and pas-
senger trains intensifies as potential LRT projects
appear on the drawing board or join the existing transit
systems in service. The inevitable interaction between
railroad and transit stirs up great concern about the
safety of these services when they share common track,
right-of-way, or corridors. It also raises issues of track
capacity, and multiple institutional issues relating to
ownership of the right-of-way, control, access, liability,
and similar matters.
The modern-day light rail transit (LRT) can easily
trace its origin back to horse-drawn wagons or omni-
buses in the nineteenth century. Its closest relatives or
direct ancestors are “streetcar,” “tram,” “cable car,” or
“trolley,” some of which are still operating today. The
terms are often used interchangeably, even by
Light Rail Transit, Shared Infrastructural Issues 685
professional transportation planners and engineers. In
the USA, ridership data and operating statistics of
streetcars, trams, and trolleys have all been aggregated
under the LRT category and can be traced back to 1902
[6]. Figure 2 demonstrates the intensity of streetcar
operation in Newark New Jersey during the early
twentieth century, and is representative of the USA
during that period.
There are a large number of books, papers, and web
sites documenting the historical demise of the guide-
way transit industry in the USA and around the world
[7–9]. Some of them have developed theories and
hypotheses on the reasons underlying the dramatic
mid-century drop in transit usage and others have
presented vivid images, such as the one documented
in Fig. 3. Limited by the length, this entry has no choice
but to leave the colorful stories, sometimes romantic
connotations, of streetcars or trams to focus on the
modern-day LRT, which is the key stakeholder in the
shared use of railroad infrastructure.
Along with increasing gasoline prices and roadway
congestion in the 1970s, there came increasing doubt
that a single mode or technology, even as flexible and
comfortable as the automobile, could effectively serve
the broad range of activity patterns and travel demand in
large metropolitan areas. Facing the dilemma of
extinction or renewal and growth, the lonely survivors
of a half-dozen streetcar or tram systems in America, as
shown in Fig. 4, became the focus of the transportation
professionals. The first national conference on light rail
transit, organized by the Transportation Research Board,
National Research Council, was held in Philadelphia in
1975, which may serve as a symbol that LRT develop-
ment in America was at that time opening a new page.
Joining the renaissance of LRT development, San
Diego, CA; Buffalo, NY; Portland, OR; San Jose, CA;
and Sacramento, CA established new LRT services in
the 1980s. Of these, San Diego LRT initiated shared use
of railroad track with freight operations. Their negoti-
ations with freight railroad and navigation of federal
regulations will be highlighted in the following section.
According to the 2009 Public Transportation Fact
Book published by APTA, the latest data available, by
2007 there were 33 LRT operations, with 1,810 vehicles
available for peak service in the USA. The total LRT track
miles reached 1,517 miles, including 144.1 miles in con-
struction and design and 561.7 mile in future planning as
686 Light Rail Transit, Shared Infrastructural Issues

Light Rail Transit, Shared Infrastructural Issues. Figure 2

Streetcar Operation in the early twentieth century in Newark, NJ (Source: North Jersey Chapter, National Railway Historical
Society)

Light Rail Transit, Shared Infrastructural Issues. Figure 3

Pacific Electric Railway cars piled atop one another at junkyard on Terminal Island, California, 1956 Source: Los Angeles
Times Photographic Archive, UCLA Library

of September 1, 2008. The total number of unlinked
passenger trips by LRT had reached 419 million, still
a far cry from 13 billion during its heyday in the 1920s,
but a fourfold growth from 103 million, the darkest days
for LRT in 1977.
As defined by the Transportation Research Board
and adapted by the US Department of Transportation
[10], modern-day light rail transit is “a metropolitan
electric railway system characterized by its ability to
operate single cars or short trains along exclusive
 Major North America Streetcar Systems - until 2020

Toronto, Canada
Newark, NJ
Pittsburgh, PA
Philadelphia, PA
Boston, MA - Green Line
New Orlean, LA
San Francisco, CA
Vancouver, Canada
Montreal, Canada
Edmonton, Canada
Ottawa, Canada
Kenosha, WI
Richmond, VA
Salt Lake City, UT
Galveston, TX
Dallas, TX
Houston, TX
Portland, OR
Oklahoma City, OK
Charlotte, NC
New York City, NY - Brooklyn
Buffalo, NY
Minneapolis, MN
Detroit, MI
Indianapolis, IN
Chicago, IL
Atlanta, GA
Tampa, FL
Washington, D.C.
Denver, CO
San Jose, CA
San Diego, CA
San Pedro, CA
Sacramento, CA
Los Angeles, CA
Little Rock, AR
Tucson, AZ
Phoenix, AZ

1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020

Light Rail Transit, Shared Infrastructural Issues. Figure 4

The Timeline of Major Streetcar Systems in the North America
Light Rail Transit, Shared Infrastructural Issues
687
688 Light Rail Transit, Shared Infrastructural Issues
rights-of-way at ground level, on aerial structures, in
subways, or occasionally, in streets and to board
and discharge passengers at track or car floor level.”
Operating along the old tram or streetcar alignment or
built along brand new corridors, the modern-day LRT
has made great strides in rolling stock technology,
which leads to higher efficiency, quick operating
responsiveness, and high speed, especially when sepa-
rate rights-of-way are used.
If the definition by TRB and USDOT has empha-
sized the operating environment and characteristics of
LRT, the definition by America Public Transportation
Association (APTA) has focused on the capacity.
It states that light rail transit is “an electric railway
with a ‘light volume’ traffic capacity compared to
heavy rail. Light rail may use shared or exclusive
rights-of-way, high or low platform loading and
multi-car trains or single cars.”
Joint operation of passenger and freight railroad
services is not a revolutionary concept. It has a long
tradition in the USA over the last century and is widely
practiced in Europe, Asia, and other parts of world
[11–13]. However, current Federal Railroad Adminis-
tration (FRA) policy dictates that freight and LRT
trains using the same track must be separated by time
of day, primarily because light rail vehicles (LRVs) do
not comply with FRA crashworthiness regulations. On
the other hand, applications of commingled operations
of freight and LRT have been growing overseas, notably
in Europe and Asia, where light rail services share track
with conventional freight and/or passenger rail services
without time-of-day separation.
The commingled operation of freight and LRT,
given the growing interest in providing similar services
in the USA, has caught the attention of transit opera-
tors, government agencies, and the research commu-
nity. This entry highlights the concerted effort by the
FRA, Federal Transit Administration (FTA), and Fed-
eral Highway Administration (FHWA), to explore the
innovation concept and feasibility of integrated
rail operations of short-line or lightly used freight
railroads and LRT.
On July 10, 2000, FRA and FTA jointly issued the
statement entitled “Joint Statement of Agency Policy
Concerning Shared Use of the General Railroad System
by Conventional Railroads and Light Rail Transit Sys-
tems” (Joint Policy Statement) [14]. The Joint Policy
Statement explains how the two agencies will coordi-
nate their safety authority with regard to the sharing of
track by rail transit and conventional railroads. On the
same date, FRA issued its statement entitled “State-
ment of Agency Policy Concerning Jurisdiction over
the Safety of Railroad Passenger Operations and
Waivers Related to Shared Use of the Tracks of the
General Railroad System by Light Rail and Conven-
tional Equipment” (Policy Statement). The Policy
Statement gives policy guidance and explains the legal
basis for safety jurisdiction and agency policy on
exercising the same in a shared-use context.
The Joint Policy Statement is general and proce-
dural in nature, while FRA’s Policy Statement contains
more substantive guidance and practical details.
According to FRA, transit operations that will share
use of the general railroad system trackage must either
comply with the FRA’s safety rules or seek a waiver.
FRA will be more likely to waive many of its rules where
“temporal separation” is in effect and the petitioner
demonstrates that its system safety program and state
safety oversight program address relevant safety issues.
If the transit and conventional railroads operate simul-
taneously on shared tracks, the petitioners will
face a steep burden of demonstrating that extraordi-
nary safety measures are taken to adequately minimize
the likelihood of a collision between conventional
and light rail equipment. Where a rail transit line
shares only a common right-of-way with
a general system railroad line, FRA does not consider
that situation to involve shared use, and a waiver is not
necessary.
The primary reason that collision safety is a major
issue is that the structural strength of typical light rail
equipment is well below the requirements for rail pas-
senger vehicles, as specified in 49CFR Part 238, refer-
ence to the latest edition of FRA safety regulations in
CFR Parts 200–299. A manuscript by a FRA Staff [15]
provides an overview of the policy statement with
respect to shared track and limited connections. A few
tables outlining the likely treatments of various subject
rules are given and examples include track, structure,
signaling; motive power and equipment; and operating
practices. These examples provide helpful guidance for
transit operators interested in applying for waivers.
A large portion of the entry is devoted to shared
rights-of-way and limited connection categories.

The author made the following statement on the shared
track and commingled operations:
" For joint simultaneous operations on shared track, the
petitioner will face a steep burden of demonstrating
measures taken to reduce likelihood of collision between
light rail and conventional equipment. Fully integrated
joint operations, such as the Karlsruhe, Germany
system, include advanced train control systems required
for all equipment. Accordingly, transit operator will
be required to observe the same rules and procedures
as the general system, i.e., FRA is less likely to
waive rules such as engineer certification for light rail
operators.
In another elaboration of the Joint Statement, Sheys
and Spear [16] summarized FRA’s treatment of shared-
use safety issues in petitions filed by Utah Transit
Authority (UTA), New Jersey Transit (NJ TRANSIT),
and Santa Clara Valley Transportation Authority
(VTA). UTA sought a waiver from the injury report
requirements, Hours of Service Act record-keeping
regulations, and the requirement for flagging
a crossing in the event of a grade-crossing activation
failure. FRA granted all UTA’s requested waivers con-
ditionally, except for part of the Hours of Service Act
record-keeping. NJ TRANSITsought a waiver of several
FRA rules related to window glazing and emergency
window exits. FRA granted the petitions but required
NJ TRANSIT to conduct a public education and aware-
ness campaign to familiarize residents of the area with
the new service on the line. VTA’s petition related to
a 7.6 mile extension, 1.6 miles of which involve tempo-
rally separated transit and freight use of a rail corridor,
but only extremely infrequently, that is, three or four
times per year. This case indicates that FRA is likely to
treat shared track/temporal separation cases involving
very infrequent usage differently than those involving
routine freight usages.
The American Public Transportation Association
(APTA) formed a Shared-Use Working Group (SWG)
in 1999 when the Federal Railroad Administration
and Federal Transit Administration were developing
the Joint Policy Statement and Final Rule. The purpose
of the SWG was to press for consistent and flexible
regulation of transit operations in corridors shared
with the general railroad system. The SWG has
brought together general managers of transit agencies,
Light Rail Transit, Shared Infrastructural Issues 689
federal officials, and technical specialists for a series of
energetic, well-attended meetings and forums. The
SWG is currently cochaired by John Inglish of the
Utah Transit Authority, and Thomas Hickey of Hous-
ton Metro.
There were also a few studies addressing the subject
of commingled operation of freight and LRT since the
late 1990s, among which, three Transit Cooperative
Research Program (TCRP) studies, one FRA-
sponsored study, and their related publications are
most notable and are summarized below.
The Transportation Research Board commissioned
a concerted research effort to explore various issues
associated with joint operations between light rail tran-
sit and freight railroads, driven by the large numbers of
urban areas in the USA that were contemplating new or
expanded rail transit services, especially where active
railroad trackage provides an opportunity for potential
cost-effective joint use.
The first TCRP research [13] was conducted and
completed before the publication of the FRA/FTA Joint
Statement. It reviewed a number of joint operations
domestically and overseas and summarized four cate-
gories of critical issues related to commingled opera-
tions: regulation, operation, infrastructure, and
vehicles. The study pointed out that one should be
very careful when talking about transferring overseas
experiences to the USA. There are a number of different
circumstances, such as traffic mix, labor agreements,
and safety features that make simple replication of
these systems in the USA very difficult. Even the sam-
pling operations in Germany and Japan maybe very
different: while in Germany one can find light rail,
passenger railroad, and freight railroad sharing track,
in Japan these examples are very rare. If it occurs, track
sharing is most often by passenger railroads and transit
operators. Therefore, all the factors mentioned above
vary from case to case and must be studied indepen-
dently. The TCRP research also recognized that differ-
ences and variations in standards among rail systems in
the USA, even in neighboring metropolitan areas, make
joint operation even harder to accomplish. The inter-
change of rolling stock is also very hard to achieve due
to all the institutional and technology requirements,
even though the use of one another’s rolling stock on
different commuter lines does occur for various
reasons.
690 Light Rail Transit, Shared Infrastructural Issues
In order to recognize the FRA and FTA joint state-
ment of policy concerning shared use, the principal
authors of TCRP Report 52 produced another TCRP
report [5]. After a brief review and summary of
FRA/FTA joint statement and TCRP Report 52, the
author proceeded to provide some additional informa-
tion on Karlsruhe and other track-sharing locations
that had occurred since the publication of TCRP
Report 52. The detailed descriptions of various loca-
tions in Germany are very useful to researchers to
identify potential models/aspects for application in
North America. The article clearly identified that
Karlsruhe may be an ideal model but not for direct
application in North America. It also urges others to
examine some less intense, simple, and straightforward
interactions, such as those at Durren or Zwickau, for
more transferable applications in the USA.
The article also emphasized the validity of major
issues identified in TCRP Report 52. It included
a matrix of interest groups or stakeholders and the
intensity of their interest in the shared-use issues. The
author provided some key issues and policy sugges-
tions for further research in four large areas: regulation,
operation, physical plant and infrastructure, and vehi-
cle. A few key questions raised in this article are directly
relevant to the understanding of shared-use issues:
● Is joint use a sound policy worth advancing?
● What risk mitigation measures can be applied to
shared track to reduce the risk to tolerable levels?
● Does joint use constitute a hazard to public safety?
● What are the quantifiable risks and liabilities arising
from various degrees of joint use?
As an introduction of “shared-use operations” to
the general transportation community and a freestand-
ing statement of current conditions, Phraner [17]
presented another report at the 8th Joint Conference
on LRT in Dallas, which provided a definition and then
current applications of joint use. The article also listed
motivations and deterrents to shared use.
Another study conducted by the Volpe National
Transportation Systems Center [18] for FTA focused
on safety and insurance issues. Shared-track operations
involve some unique risks, and the liabilities associated
with these risks can affect whether such services are
offered and, if offered, their operating plans and costs.
Insurance is used to mitigate or manage such risks and
insurance premiums can be a significant cost item for
transit agencies. The authors explored the issues of
liability and insurance that arise when operating on
track shared between light rail transit and freight rail-
roads. The study included six shared-track operations.
The transit agencies that currently operate light rail
transit on the same track as freight railroads are: San
Diego Trolley, Inc. (SDTI), Utah Transit Authority
(UTA), New Jersey Transit (NJ TRANSIT), which has
two services, the River Line and Newark City Subway,
and the Maryland Transit Administration (MTA).
The most salient risk involved in shared-track oper-
ations is that of a collision between a light rail transit
vehicle and a freight railroad vehicle. None of the
shared-use agreements specifically addresses the possi-
bility of this kind of risk, since it has never occurred.
Other risks include one party damaging the other
party’s property. Track ownership seems to have
a significant effect on risk. Part of the reason that
freight railroads agree to sell the ownership of track to
transit operators is to avoid risk. Some insurance
industry officials believe it would be useful to have
a study that could quantify the perceived risk of shared
operations. However, there has been no published risk
assessment performed specifically for the case of shared
track between light rail transit and freight railroads
until the two recent FRA studies, which will be
addressed in the following section.
Freight railroads generally try to minimize their risk
for anything related to passenger transit. Based on the
six case studies, there are three basic methods for
assigning liability: by fault; by financial threshold; and
by incident circumstances. By fault is the simplest way
to address the issue. However, when the fault for an
incident is not clear or if multiple parties are involved,
the process of fault determination can be lengthy.
Financial thresholds are easier to determine. In many
cases cutoff points assign liability. Liability assignment
by circumstance describes various situations and
assigns liability based on who is involved and the type
of damage is caused. When liability is assigned by
financial threshold or incident circumstances, one
party is held harmless irrespective of fault. In these
cases, there is an implicit lack of incentive for safe
behavior.
According to the study, there is no instance where
separate liability insurance for the shared-track

operations exists because the safety rules and proce-
dures are viewed as effective in mitigating possible
risks. A table included in the manuscript gives an over-
view of the various insurance levels and premiums for
each of the case studies. Insurance level varies by agency
from $35 million to $250 million and annual premiums
ranged from $150,000 to $4.2 million.
The primary objective of another FRA study [19]
was to explore the feasibility of concurrent operations
of compliant and noncompliant vehicles on the same
track, concentrating on identifying and evaluating risk
mitigation measures that would meet FRA’s require-
ment of providing “equivalent safety.” The study was
deliberately constrained to existing noncompliant
vehicle types proposed for LRT services in North Amer-
ica, and sharing track with low-density local freight
service to best represent the current interest.
The study summarized principal vehicle character-
istics and physical plant for five potential shared-track
operations, which include: Baltimore LRT in Mary-
land, SPRINTER DMU from Oceanside to Escondido
in California, the River Line in southern New Jersey,
San Diego Trolley, and TRAX light rail in Salt Lake City,
Utah. The areas of noncompliance, issues related to
shared use of LRT and freight railroad, such as railroad
operating and maintenance practices, equipment, and
operating environment, are discussed.
After furnishing a table summarizing the effects of
noncompliance on accident likelihood and severity, the
report provided a risk analysis of a representative
noncompliant passenger rail operation with concur-
rent local rail freight service. The primary area of con-
cern with commingled operation is the number and
severity of collisions between noncompliant vehicles
and conventional freight cars and locomotives. The
risk analysis results showed that the incremental risk
from shared operation could be mitigated by practical
risk mitigation measures, provided that freight opera-
tions are confined to a small number of trips during
off-peak hours. Mitigation measures include adding
a second track to a single-track system, and applying
one of several alternative train control systems to
reduce the risk of a collision.
A recent draft implementation strategy developed
by Tereschuck [20] outlines the core operational
requirements for retention of an equivalent level of
safety for commingled operation of freight and LRT
Light Rail Transit, Shared Infrastructural Issues 691
services. Based on previous TCRP research and Federal
fact-finding mission trips to Europe, the author
highlighted five essential operating characteristics con-
sistent with the European approach to commingled
operations. Corresponding to each characteristic, the
manuscript also presented possible actions for pilot
application:
● Single authority/ownership of rail corridor/
dispatching control
● Uniform/single rule book authority
● Single communication network
● Consistent training/certification process – single
authority control
● Application of advanced technology – provide pos-
itive train stop or automatic train control capability
A recent study [2], sponsored by FRA, has reviewed
the technical feasibility of concurrent shared-track
operations in detail focusing on the 1999 FRA/FTA
Shared Track Policy Statement, which identifies six
factors to be considered in evaluating the operational
safety of light rail transit systems that would share track
with conventional railroad equipment. The research
team pointed out that five of these factors are primarily
operational in nature requiring appropriate organiza-
tion and training but no investment in technology. The
final factor, fail-safe train separation and collision pre-
vention, revolves around the deployment of appropri-
ate technology to prevent collisions between
trains. Relative to the primary technical consideration
of providing fail-safe train separation, the team has
identified a number of train control systems that ensure
trains sharing track do not collide. All of the systems
override the vehicle operator in the event of a signal
violation that could result in a collision with another
train.
Independent of the federal shared-track policy, the
research team identified and evaluated the four most
salient risks of concurrent shared-track operations.
The study team found that risk assessment is a valuable
tool that can contribute to assuring the safety of
commingled shared-track operations. Risk assessment
is any structured process of identifying the hazards pre-
sent in an operation, evaluating the level of risk associ-
ated with the hazard, assessing the acceptability of the
risk, and determining whether or not risk mitigation will
be required. Risk assessment may be applied at all stages
692 Light Rail Transit, Shared Infrastructural Issues
in the development and operation of a passenger rail
system, but the primary purpose for a shared-track
operation is to support the customary safety assurance
process for a new shared track operation through plan-
ning design and implementation stages.
In addition, the FRA study has also conducted
a market assessment to support federal investment in
mechanisms and methods that facilitate shared-track
operations. The market assessment indicates substan-
tial national applicability and considerable industry
interest in concurrent shared-track operations where
light transit cars share track with conventional railroad
trains. In the context of national applicability,
a nationwide review of the 88 urban areas with
populations greater than 500,000 inhabitants identified
165 corridors in 65 cities with potential for develop-
ment as shared track freight/transit rail services. Rela-
tive to industry interest, a brief scan identified 15 North
American corridors where shared track operations are
in current operation, design/construction, or active
planning.
The most recent TCRP Report 130 [3], produced
under the TCRP A-27 project, has focused on the pre-
sent and evolving state of the art in the shared-use
environment, specifically examining train control tech-
nology, operating rules and procedures, communica-
tions systems, vehicles, shared-use operations, and
governance and business models of existing and
planned systems.
The research team took a comprehensive view of
“Command and Control” as including the variety of
train control and communication systems, operational
rules and procedures that are particularly suitable or
adaptable to the shared-track environment. The
emphasis was on those features that compensate for
vehicle structural deficiencies by minimizing the risk of
a collision in such an integrated system.
Building on knowledge gained via previous studies,
the TCRP project presented the characteristics of
freight railroads, suggested business models for future
implementation, and included a business case and
practical guide for transit agencies or other stake
holders to be involved in the shared-use area. To carry
the TCRP A-27 to a new height, the study team devel-
oped a hypothetical case study and recommended
a demonstration program to explore the critical issues
and feasibility of sharing track between FRA
noncompliant LRT vehicles with lightly used freight
railroad operations.
There are various studies, trip reports, and research
summaries that document overseas practices in the
areas of commingled operations of freight and LRT.
The literature presented here may overlap on some
cases in Europe but the emphasis and time frame
may vary.
Geissenheimer [21] published a manuscript that
recognized the opportunities to operate efficient rail
service on sparsely used lines using a new generation of
self-propelled diesel multiple unit cars (DMUs) or
dual-power electric light rail vehicles (LRVs) combined
with changes in operation and organization. The focus
of this article related to vehicle technologies and their
applications, but also provided some pioneer study of
shared use cases in Duren and Karlsruhe, Germany;
Amsterdam, Holland; and briefly mentioned shared
use in Japan and Korea.
Another early introduction to overseas shared-use
practices was included in a short article in the Journal
of Transportation Engineering [22]. This article
highlighted two basic strategies to develop a local LRT
or streetcar network into a regional system at relatively
low cost. The first strategy is to take over a railroad line,
where most or all prior freight service has been or can
be abandoned. Examples mentioned in the USA
include the Riverside Line in Boston and LRT service
between downtown San Diego and the Mexican border.
The second strategy is to implement a shared track
agreement that provides LRT access to the railroad
with freight services. As documented in this manu-
script, in contrast to the requirement by the US
government regulations, for LRV to meet high buffing
strength standards to protect against damage in the
event of collision, there is no special buffing strength
requirement in Germany. The rationale is that LRV
braking rates, being superior to those of main line
trains, obviate the need for buffing strength.
The author presented a detailed network and vehicle
arrangements of Germany’s Karlsruhe system to dem-
onstrate the principles of shared-use applications.
A few other applications were also mentioned in the
article: Kassel, Saarbrucken, and Zwickau in Germany
and other cities in France and Switzerland.
Another detailed study of shared track operations
in Karlsruhe is presented in the FTA/FRA fact-finding

trip report [23]. In March 2000, a delegation of US
transit operators, equipment manufacturers, and con-
sultants led by FTA/FRA went to Karlsruhe to examine
shared track operations. After describing the nature
and importance of rail transportation in Germany,
the trip report documented how shared track operation
was achieved in Karlsruhe. Summarizing general
observations and specific case studies during the trip,
the delegation highlighted the following factors that
contributed to the success of shared track operation
in Karlsruhe
1. A government-owned national railway system
provides accessibility to all rail tracks.
2. Information, communications, and active train
control systems and equipment adaptable to
mixed usage are employed.
3. Government also requires the labor force to obtain
licensing certification to operate the different types
of trains running in Karlsruhe, including trains for
freight, intercity passengers, regional passengers,
and local trams.
4. Institutional factors, such as common operating
rules, dual dispatching facilities, and uniform safety
regulations, ensure the seamless integration of both
light rail and conventional rail operations.
As a member of the delegation of the fact-finding
trip organized by FTA/FRA in 2000, Phraner also vis-
ited additional cities, such as Saarbrucken, Duren, and
Koln. A supplemental report [24] was compiled to
provide an independent supplemental study of those
systems visited. Recognizing Karlsruhe as an instructive
case study but not a readily transferable model to North
America, the author defined Karlsruhe as representing
the most advanced example of shared track operations.
The Karlsruhe case study provided selective lessons and
precedents in institutional and regulatory innovations
that encourage shared track arrangements. Karlsruhe
practice demonstrated operating practice extremes by
dispatching tram and mixed LRT train traffic through
a peak volume point and confirmed that the German
train control technology and control center operated
safely.
Another important issue in the Karlsruhe practice is
the exemplification of strong political leadership and
commitment, necessary to accomplish a shared-use
project. Besides abundant field visit photographs of
Light Rail Transit, Shared Infrastructural Issues 693
each location, this report also included detailed net-
work maps, schedule, and operating arrangements for
Saarbrucken, Duren, and Kohn case studies. One of the
important contributions of this report was to identify
the potential issues to be encountered in the USA
when/if a particular model is to be transferred.
In addition, the report also devoted an entire chapter
to light-rail-derived DMU rail cars and discussed their
importance to shared track in North America.
Similar to previous fact-finding trip reports, the
TCRP Research Result Digest Number 47 [23]
documented a number of case studies of shared track
operations in Germany (Kohn, Durren, Kassel,
Karlsruhe, Saarbrucken, Rhine/Ruhr-Dortmund, and
Gelsenkirchen), France (Strasbourg), and Luxembourg
City. Different from the previous fact-finding trip
report is that this digest is organized by functional
areas, such as design, technology, and track-sharing
relationship to economic and urban development.
Each area is demonstrated via specific examples of
various cases in Europe. Its Appendix C provided
a summary table of German cities with track sharing
operations, which may be useful to identify the unique
characteristics and similarities of various shared track
operations.
Another recent description of shared track opera-
tions appeared in the Tramways & Urban Transit Mag-
azine [25]. In this article, the author described a visit to
a German city transformed since the end of communist
era. Nordhausen, a former part of East Germany, has
access to two meter-gauge tramlines. One runs to an
informal transfer center with two railway stations and
serves a large number of urban, suburban, and
interurban bus routes. The Deutsche Bahn Station,
a typical busy urban facility, has intercity and regional
trains arriving and departing at frequent intervals.
The Bahnhof Nord, the terminus of a privately oper-
ated but highly subsidized meter-gauge railway system,
runs through rural areas and forests and is used pri-
marily for passenger traffic, with diesel railcars and
steam-hauled trains attracting commuters,
schoolchildren, and tourists.
Similar to most US cities with growing suburban
and rising automobile ownership, Nordhausen wanted
to improve transit service and discourage motor traffic
from clogging its streets. Following the examples of
Karlsruhe, it was thought that a tram-train operation
694 Light Rail Transit, Shared Infrastructural Issues
providing through service might substantially reduce
the number of cars entering the city. Encouraged by
the no-transfer service and low-cost solutions,
Nordhausen simply built a track connection between
the two systems and a hybrid vehicle to run on busy
streets under typical 600 V tramway overhead and self-
propelled on traditional railway tracks. Service over the
connection began in April 2002 and an extended ser-
vice of 12 km beyond the privately operated station was
inaugurated in May 2004. As observed by the author
during his visit in 2005, the shared-use operation in
Nordhausen is “very elegant.” Using sophisticated lay-
over procedures, the shared track operation provides
sufficient service for commuters and makes significant
contributions to the local economy.
Current Operating Scenarios in the USA
It is important to have a general understanding of
the current LRT operations in order to explore the
issues of shared use of infrastructure with freight rail-
roads in the USA. A brief presentation of a national
survey of shared use between LRT and freight railroads
in North America is included below. The survey
results are categorized into four groups: physical char-
acteristics, operating privileges, insurance liability, and
regulation.
The difference between propulsion power, capacity,
and operating characteristics determined that LRT
operations have different issues from commuter rail
services, even though they may share certain legal,
administrative, and organization issues. According to
a recent survey [1], among the 34 LRT operations in
North America, about a quarter do not have any inter-
action with freight railroad and the rest all interact with
freight railroad to some extent, as shown in Fig. 5. The
majority of the LRTs surveyed, 61%, interact exclu-
sively with local freight; SEPTA light rail interacts
exclusively with mainline freight and another light
rail transit. It is interesting to note that, in the case of
Edmonton LRT, the interaction with the adjacent
freight main line has diminished due to changes in
freight operations since LRT was built in 1978. At that
time, the LRT route was located between main line
tracks. Now the alignment is a little used freight spur.
Platform access level is an important element in the
interaction between LRT and freight railroad. The
Types of Interaction Between LRT and
Freight Railroad
Shared Facilities
9%
Shared Track and
Time Seperated
Operations Shared Right of
36% Way
55%
Light Rail Transit, Shared Infrastructural Issues. Figure 5
Types of Interaction between LRT and Freight Railroad
Americans with Disabilities Act (ADA, 1990) requires
that transit systems provide level access for passengers.
Unlike a commuter railroad, whose platform access level
can be categorized as low-level platform and high-level
platform, LRT platform access level may be better
described as “low-floor” versus “high-floor” configu-
ration. In case of low-floor access, the LRT vehicle floor
is flush with station platforms that are 8–14 in. above
the track, depending on the system, so wheelchairs or
strollers can roll on to the vehicle directly. Low-floor
cars, which have been used in Europe for more than
a decade, were only first introduced to USA in 1997 on
Tri-Met MAX in Portland, OR.
Most early low-floor designs were intended for city
streetcar or tramway applications and have neither the
top speeds nor the ride quality suitable for US and
Canadian light rail operations. However, many Euro-
pean LRTs have extensive on-street operation and
recent new vehicle procurements have been predomi-
nantly for the low-floor type [26]. In the most recent
decade, partial low-floor LRVs capable of 55–65 mph
top speeds have been delivered to several US cities.
These vehicles are capable of providing the speeds and
levels of comfort needed to effectively serve regional
LRT lines of 10–20 miles in length.
When high-floor LRT vehicles are used, the floor is
higher than the station platform so wheel chairs have to
be aided by a lift to roll onto the vehicle. There are often
mini platforms situated at the end of the platform so
that they can be operated by the train driver when
required, which is the case in Sacramento, Dallas,
Baltimore, Denver, Salt Lake City. San Diego has eleva-
tors mounted on the car itself. High platforms are used

Types of Freight Railroad Interacting
with LRT
Both Mainline
and Local 8%
Mainline Only
8%
No Active Local Only
23% 61%
Light Rail Transit, Shared Infrastructural Issues. Figure 6
Types of Freight Railroad Interacting with LRT
exclusively in a number of systems including Los
Angeles, St. Louis, Edmonton and Calgary. They are
also used in combination with low level loading plat-
forms. Examples include Buffalo, Pittsburgh, and San
Francisco. A precise demonstration of both high- and
low-floor LRV may be observed in Tri-Met’s MAX in
Portland, OR. The agency practice coupling one high
floor with another low-floor vehicle, this ensures
adequate access for wheel chair users at all times and
all locations.
As documented in the study [1], the degree of
shared operation in light rail transit varies. Five sys-
tems, depicted in Fig. 6, share track and are temporally
separated from freight operations. They are MTA LRT
in Los Angeles, TRAX in Salt Lake City, the LRT system
in Baltimore, San Diego Trolley, and Newark Subway in
New Jersey. Light rail services usually operate during
the day, with freight operations from midnight until
the early morning hours. Since that survey was
conducted in 2003, a new kid on the block, South Jersey
Riverline LRT, has joined the group using the tempo-
rally separation arrangement. Six light rail transit sys-
tems share right-of-way with freight, with each mode
on separate tracks. These include New Orleans Street-
car; SEPTA light rail transit, Denver LRT, Portland
MAX, Sacramento LRT and Hudson Bergen LRT.
Only the Metro Blue, Green and Red LRT lines, in
MTA of Los Angeles, and Edmonton Transit have
some shared facilities with freight, but no joint usage.
In contrast to the mixed ownership of tracks used
for commuter rail operations, the ownership for LRT
Light Rail Transit, Shared Infrastructural Issues 695
tracks is distinctly defined. According to the same sur-
vey [1], about 40% of the shared tracks are owned by
freight railroad and 60% by the transit agencies. In
general, LRT either uses tracks abandoned by a freight
railroad, or the transit agency has acquired the right-of-
way, being abandoned, to preserve transit corridors.
Moreover, most of the light rail transits are
dispatched by transit agencies themselves, such as
Newark Subway, NJ River Line, Edmonton Light Rail
Transit, San Diego Trolley, SEPTA LRT, Baltimore light
rail transit, and Metro Blue, Green, Red lines in Los
Angeles. Two systems are dispatched by third parties,
which include TRAX in Salt Lake City and Hudson
Bergen LRT in New Jersey.
When compared to the commuter rail systems that
have different dispatching protocols during peak and
off-peak period, the majority of the LRTsystems imple-
ment the same dispatching priority timetable during
their operating period, which is largely caused by the
temporal separation.
None of the light rail transits surveyed has any
incentive plan for better coordination between freight
and transit operations. The obvious reason is that light
rail transit operators are less affected by freight opera-
tion, except during the time window that is allocated.
One of the questions included in the survey asked
what measures are provided to separate transit and
freight operations. The results indicate that the major-
ity of LRT systems do not have any physical barriers to
separate LRT and freight operation when they are in the
close proximity of each other. Among the ten valid
answers, seven do not have any separation measures
between transit and freight, two uses fences, and one
uses both fencing and intrusion detection. The last
example was dictated by the freight railroad due to
liability concerns.
The survey results revealed that generally not many
changes were made to the transportation system to
accommodate passenger operation for LRT. Most
changes were just a complete service time separation
schedule. For example, Edmonton Light Rail Transit,
where the LRT tracks were originally in the center of the
right-of-way with mainline tracks on both sides, gave
modification details to accommodate LRT operation:
“temporary relocation of CN track during construction
of a shared LRT/CN bridge. Also CN grade crossing
gates were connected to LRT signals system to allow
696 Light Rail Transit, Shared Infrastructural Issues
shared use of the gates. They extended the crossing ‘gate
call-on’ provided on CN line and tied it into the LRT
crossing ‘call on’ signal to allow gates called on by LRT
to stay down if CN train was approaching. There was
also a need for the modification of CN’s DC signal
system to ensure that there would not be interference
caused by LRT DC traction power return. A signalized
diamond crossing was installed on CN track to enable
LRT line to cross into their yard that was located across
the CN westbound track. All of the above were paid for
by the City of Edmonton.” Since the construction of the
LRT the main line freight service has largely been
relocated away from the shared right-of-way.
The settlement of insurance and liability issues
between the light rail transits and the freight railroads
is similar to that of commuter systems. The size of LRT
operation and the strict separation of operations
defined the insurance premium, liability, and deduct-
ible, which are generally smaller than for commuter rail
services. For example, the insurance and liability issues
for San Diego Trolley were handled through agreement
with Metropolitan Transportation Development Board
(MTDB) that both own the freight railroad and San
Diego Trolley. In the case of the Baltimore LRT, the
State MTA holds CSX harmless. The insurance require-
ment for Edmonton Light Rail Transit is as follows:
$5 million liability insurance covering persons using
LRT and CN passenger service; $1 million insurance
covers liability to cover third parties arising from utili-
zation by the City of the premises and all premiums are
paid by City of Edmonton. City waives its rights against
CN for damages arising from occupation/use of the
leased premises, except if caused by the negligence of CN.
No accidents were recorded as a result of freight and
LRT interaction in the survey [1]. This may reflect the
safety precaution of the separated operation. But at
the same time, we also have to admit that LRT systems
have only been in operation for a relatively short
period. The oldest among the several “new age” LRT
operations in the USA is the San Diego Trolley, which is
only 29 years old. Compared with the century old
commuter rail operations, the total accident number
is low or nonexistent.
Different from the commuter rail operation, most
LRT service providers or transit agencies have pur-
chased and now own the railroad lines. Then public
transit entities grant trackage rights to the contract
freight operators to access the shared-use tracks. For
example, Utah Transit Authority (UTA) bought the
railroad track from Salt Lake City to Sandy and beyond
from Union Pacific and contracted a short line to
provide the residual local freight switching service on
the LRT line. Similarly, New Jersey Transit purchased
the alignment from Conrail for the Riverline LRT in
South New Jersey. Conrail has a permanent operating
easement, which provides them trackage rights to use
the River LINE per the terms of the sale and operating
agreements. Many transit entities are in the process of
acquiring the shared-use asset; some properties still
inherited the historical arrangements. For example,
Edmonton Light Rail Transit and Newark Subway get
the corridor access through property lease; SEPTA light
rail gets their corridor access through a highway cross-
ing agreement between State and railroad, and San
Diego Trolley, Inc. via a contract through MTDB, of
which SDTI is a wholly owned subsidiary.
Demand for Commingled Operation
While North America pioneered commingled opera-
tion in the early twentieth century, the newest and most
innovative forms of joint use are found overseas. Tran-
sit professionals found that, in terms of commingled
operation between LRT and freight railroad, the com-
bination of US engineering and operations best prac-
tices is behind international trends. A few LRT service
providers argue that commingled operation is feasible,
if certain measures are taken to mitigate risks to toler-
able levels.
LRT is a relatively inexpensive transit mode to
build, compared with urban highway and heavy rail.
Yet, LRT is effective in attracting automobile users from
congested highways. For example, Baltimore’s LRT was
built in 1992, at the cost of $360 million, for the initial
22.5 miles. In year 2000 dollars, that’s approximately
$422 million, or $19 million per mile. The cost includes
right-of-way purchase, line construction, stations,
vehicle procurement, maintenance and storage facili-
ties, etc. According to APTA, Baltimore’s LRT carries
about 25,600 passenger trips per weekday, which is
about 2,500 trips in the highest peak hour, or about
2,400 trips in the peak direction, equivalent to almost
two freeway lanes. The LRT system has been planned to
accommodate future growth with three-car trains, and

with a maximum 2-min headways assumed, the system
could theoretically accommodate about 23,000 persons
per hour, the equivalent of about 20 freeway lanes [27].
LRTs also have the advantages of lower operating
cost. For example, the operating cost of LRT services in
Boston, which carries the heaviest LRT ridership in the
USA, is significantly less than that of bus services.
According to FTA’s National Transit Database [28],
the MBTA’s LRTsystem carried passengers at an average
cost of $1.25 per trip, and $0.51 per passenger mile,
compared with MBTA average bus costs of $2.04 per
trip and 0.71 per passenger mile.
LRT is the USA’s big winner in terms of ridership
growth among the major transit modes. Despite the
economic recession, Americans took more than
10 billion trips on public transportation in 2009, a 35%
increase since 1995 [4]. Among which, LRT ridership
remain the most stable, contributed by newly inaugu-
rated systems, in Phoenix, AR and Seattle, WA. In
addition, a number of existing LRT systems exhibited
large ridership increase, such as 12% in Baltimore, MD;
11% in Oceanside, CA; and 9% in Memphis TN.
LRT operations have been developing at an average
rate of nearly one new system per year over the past 3
decades. According to APTA [6], there are 27 modern
LRT systems (opened after 1981), and another 7 Vin-
tage Trolley systems in USA. The recent openings of
LRT include Portland’s MAX Yellow Line, streetcar
extensions in Memphis and New Orleans, and New
Jersey River Line, which is the first to use non-
electrified vehicles. Most extensions in San Diego, San
Francisco, San Jose, Sacramento, Denver, Dallas,
St. Louis, Philadelphia, Pittsburgh, Baltimore, Hud-
son-Bergen, and Newark subway are now open. New
start-ups in Phoenix, Memphis, Minneapolis, and
Charlotte are in operation also.
Most new light rail services operate through central
cities on streets from which the bulk of automobile
travel has been removed. However, outside of the center
cities, the new LRT services typically extend into the
suburbs on private right-of-way, with longer station
spacing. The general practice is that light rail agencies
own or control the track, operate transit service as
the host and lease or permit freight operations by
a contracting railroad in off-peak periods.
Though most LRTagencies own the right-of-way by
purchasing the abandoned or branch rail lines, the
Light Rail Transit, Shared Infrastructural Issues 697
demand for commingled operation between LRT and
freight railroad might increase when rail transit “new
starts” exhaust vacant right-of-way. It is possible that
the competition for track space will focus on the active
railroad lines, not only abandoned or underused lines.
On the other hand, rail freight is booming as well.
For the first time since World War II, some railroads
have capacity constraints on certain lines. According to
the Association of American Railroads [29], railroads
are experiencing a surge in traffic volumes for all com-
modities, including coal, and, for the first time, inter-
modal freight is the industry’s top revenue source. The
US railroads originated more than 330,000 carloads per
week in 2008. The transformation from once thought
to be stodgy and slow carrier to a good-service, high-
speed and modern carrier has made freight railroad an
indispensable link in providing “just-in-time” supply
chain.
The dilemma faced by both freight railroads and
rail transit operators is that rail segments, where pas-
senger service providers require additional capacities,
are often bottleneck areas within freight operations.
This fact can make it even more difficult to share tracks
for both freight and passengers.
Though commingled LRT operation with freight
railroad is not active in the USA, it is a common prac-
tice in Europe and Japan. The European approach to
mixed operations is significantly different from the US
approach. European operations maximize utilization
of the “track resource” through scheduled, on-time
freight and passenger operations [30].
Karlsruhe, Germany is acknowledged to provide
the most sophisticated and advanced examples of
shared-track practice between railroads and light rail.
As documented by the supplement of TCRP 52 [5]
Karlsruhe integrated its local streetcar and railroad
service on selected lines. LRVs and commuter trains
not only run jointly, but also provide joint service on an
integrated timetable, with freight and intercity trains
interspersed in the operating schedule. Karlsruhe dem-
onstrates operating practice extremes by dispatching
tram and mixed LRT and train traffic through peak
volume points. The observation made by TRB [5]
disclosed the highest level of operating discipline and
performance: 16 trains passing in an hour, nine of
which were scheduled northbound in the half-hour
interval between 12:30 and 13:00. A northbound
698 Light Rail Transit, Shared Infrastructural Issues
freight intercity train is followed 1 min later by a high-
speed Intercity Express (ICE) train, 10 min later by
a Euro International Express, 4 min later by a light
rail vehicle making a station stop, 3 min later by an
express freight, 4 min later by freight – all sharing the
same track [5].
Japan is another leader in commingled operation
between rail transit and freight railroad. Japanese
National Railways (JNR) split into six passenger car-
riers which are separate corporations but are marketed
as Japanese Railways (JR) and one freight carrier, JR
Freight in 1987. JR Freight and the six JRs operate on
the same track and catenary system, using the same
railway facilities. Unlike freight services in the USA,
freight in Japan is subordinate to passenger services
on the national railroad system. Freight operation is
consistently the tenant on the passenger railroads.
Although JR Freight does not pay track usage fees, it
is responsible for the incurred incremental costs of
operating freight trains on these tracks. An average
of 737 freight trains operates each day throughout
the country. Of these, 425 are container trains, while
the remainders are trains hauling bulky freight in
tankers and hoppers. About 90% of JR Freight’s rail
freight is transported at night [31]. JR Freight is
investing heavily to upgrade its fleet with faster and
powerful locomotives; developing new freight wagons
for high-speed trains; developing new containers for
easier loading and more capacity; and upgrading
freight station facilities. An advanced freight system
may also contribute to more harmonious operations
with the rest of the rail transit modes.
It remains to be seen whether FRA is willing to
adopt such practices even though Europe and Japan
has proven that commingled operation can provide
more line capacity. The RiverLine in Southern New
Jersey could be an example of efficient use in the
USA. NJ Transit’s initial plan, developed with Conrail
Shared Assets, called for a “limited mixed-use” opera-
tion with parallel running on adjacent freight-only
tracks, one freight train per day during off-peak
hours, and use of automatic trip-stops and derails.
The risk assessment indicated that the safety level
would meet FRA’s requirement. However, FRA rejected
this plan, claiming that freight traffic on the Trenton-
Camden corridor is expected to grow, making temporal
separation impractical. NJ TRANSIT then replaced the
original plan with a more rigid time-separated alterna-
tive, day and evening for passenger operations and late
night freight operations.
Challenges Faced by the Attempts
Though commingled operation could exhibit advan-
tages, such as lower capital investment for infrastruc-
ture, reduced operating cost, cheaper and quicker new
starts, etc., challenges are yet to be solved, such as
incompatible equipment, signals and communications,
and inspection and maintenance practices.
The FRA/FTA “Joint Statement” states, “Although
compatible in terms of track gage, these two forms of
rail service are incompatible in terms of equipment.
A collision between a light rail transit vehicle with
passengers aboard and heavy-duty freight or passenger
equipment would likely result in catastrophe.”
The main problem in the USA is the design of
different categories of rail vehicles: FRA requires rail-
road vehicles have certain crashworthiness or buff
strength, while these regulations do not apply to rail
transit and LRVs. Having different types of vehicles
with different weights, crashworthiness, and braking
characteristics commingling on the same track imposes
safety concerns. As stated in TCRP 52 [30], emphasis in
North America is focused on rail safety and collision
survival, versus European emphasis on collision avoid-
ance and impact attenuation. For example, railroad
buff strength for overseas railroads is in the 300,000–
400,000 lb range, while North American railroad
requirements are twice that, at 800,000 lb. Typical
light rail car buff strength generally is 200,000 lb or
less, and rarely, if ever, more than twice the empty
weight of the LRV such as CPUC General Order
143-B. Making rail cars stronger to sustain collision
forces creates a heavier car that is unable to stop in
shorter distances, thereby increasing costs and power
consumption, while sacrificing performance.
Related to the scarcity of North American main line
railroad electrification is the potential reintroduction
of Diesel Multiple Unit (DMU) rail car to North
America. DMUs are similar to LRVs in appearance
and performance, except that they can operate
independently of wayside traction power infrastruc-
ture. New Jersey Transit’s River Line is among the first
contemporary uses of the light DMU technology as

a surrogate for conventional electrified light rail. Anal-
ysis indicates that DMU can offer both cost savings and
operating advantages.
Operating trains that travel at different speeds
reduces capacity on a rail line. The optimum condi-
tion can be achieved by operating at about the same
speed by all types of trains. All other things being
equal, the longer the block length, the lower the capac-
ity. The block length is based on the distance it takes to
stop the train. Braking distance varies by the train
type. Signal design would have to conform to conven-
tional freight railroad requirements, but also accommo-
date design elements unique to LRV. For example,
freight trains require much greater braking distances
than do LRVs, so the design must consider the longer
freight braking distance when establishing the signal
block layout and establishing speed limits for the various
classes of trains to operate over a line. However, longer
signal blocks unnecessarily penalize LRVs by causing
them to run at reduced speeds under certain signal
aspects. This scenario could be compounded under cer-
tain topography, such as downhill or grade crossing
territories. Special operating rules must therefore be
developed to help reduce or eliminate such penalties.
Under the general guideline of the “Joint State-
ment of Agency Policy Concerning Shared Use of
General Railroad System by Conventional Railroads
and LRT systems” (Joint Policy Statement) issued in
2000, FRA does not consider the situation, where a rail
transit line shares a common right-of-way with
a general railroad without sharing track, involves
shared use of the general system by the rail transit
line, therefore a waiver is not necessary. Where a rail
transit line shares track with general railroad, but
under strict “temporal separation” as referred in the
Policy Statement, the FRA is likely to waive many of
its rules. However, the petitioner must demonstrate
that its system program and state safety oversight
program addresses all relevant safety issues.
For commingled operations on shared track, the
petitioner will face a “steep burden” of demonstrating
measures taken to reduce likelihood of collision
between light rail and conventional equipment,
because of concerns over major differences between
FRA-compliant cars and light rail vehicles (LRV). It
seems that, in this litigious society, chances are slim
for winning approval of such waivers from FRA.
Light Rail Transit, Shared Infrastructural Issues 699
In North America, liability is the single biggest
institutional obstacle to joint operations between rail
transit and freight railroad. In spite of many concerns
arising from joint operation, one of the primary fears is
a catastrophic event resulting in major loss of life and
property damage. Thus, an important condition for
a freight railroad to share its track with rail transit is
to be held harmless, regardless of fault.
According to General Accounting Office [32], in
1997, Congress limited the aggregate damages that
may be awarded to all passengers claims from
a particular rail accident to $200 million and permitted
providers of rail transportation to enter into indemni-
fication agreements. After reviewing the legislation, the
GAO concluded that the liability cap applies to com-
muter rail operations, which is certainly a good base for
commuter service providers to cap their liability insur-
ance to under $200 million per accident. It seems
unlikely that such a cap will apply to LRT as well,
since LRT is non-FRA compliant.
The level of risk can influence the level of liability.
As risk increases, the necessity for liability insurance
coverage increases too. Where passengers are involved,
risk increases dramatically and liability insurance can
become excessively onerous. No freight rail owner is
willing to risk the assets of its railroad, but looks for
return on investment to find joint operations attractive.
Higher risks for passenger operation create a dilemma
for transit operators.
Potential Solutions
It is understood that successful experience from over-
seas cannot be directly transferred to USA because of
cultural, regulation, intuition, and physical differences.
However, certain risk mitigation choices can be applied
to get USA prepared for future possible commingled
operation.
Advanced signal systems, positive train control, or
Intelligent Transportation System (ITS) may be applied
to improve the efficiency and communication of
rail transit. Railway signal systems, in general terms,
are made up of block signal system, interlocking sys-
tems, and control systems. Block signal systems are
designed to inform operators of the need to stop
when there is danger of rear end collision. Block lengths
are based on the distance it takes the heaviest train to
700 Light Rail Transit, Shared Infrastructural Issues
stop. To accommodate disparate operational trains on
the same tracks, block signal systems need to be
reconstructed to assure the appropriate block length
based on the train’s speed, braking ability, weight,
etc. Automatic cab signaling (ACS), displaying signal
aspect information on the operator’s control panel, can
be used to reduce the possibility of operators missing
signals. Further, positive train control systems, includ-
ing automatic train stop (ATS) and automatic
train control (ATC), can take control of the trains if
operator does not take appropriate actions upon
receiving the signals. The US government requires
that for operations above 79 mph, cab signaling or
train control systems are installed and used. In addi-
tion, all railroad lines carrying passengers must be
equipped with an acceptable system of Positive Train
Control (PTC) by 2015, as required by the recently
enacted Railroad Safety Improvement Act (RSIA,
2008).
Interlocking signals prevent trains from colliding
where railroad tracks meet. All interlockings for
commingled operations should be designed to ensure
that they do not delay trains or cause capacity prob-
lems. Centralized traffic control (CTC), commonly
used on busy railroads, allows trains to be routed
through complex track networks to optimize perfor-
mance of the overall railroad system. A CTC system
allows the dispatcher to safely direct overtaking trains
traveling in the same direction and trains meeting in
opposite directions. This enables the dispatcher to
place a train in a siding or on an additional mainline
track, to avoid delay [33].
As described earlier, its punctuality, superb service,
and the large crowds of people using it characterize
Japanese public transportation. The German rail sys-
tem is also credited with convenient, on time, superb
service. Both systems are the envy of the world. Such
success is achieved not only by innovative technologies,
but also by good communication, understanding and
attitudes among the parties sharing the facilities.
A sophisticated operational plan can be worked out
between willing parties. Communication must be
established and maintained between the operational
managers. Since passenger and freight operations
have different operating characteristics, employees
from a transit background may not be familiar with
freight rail operation and vice versa. Thus, it is
important for professionals from both parties to
reach a mutual understanding, not only the
dispatchers.
Insuring a high standard of employee qualifications
and performance is important in commingled opera-
tions. Cross training might be helpful to enhance
mutual understanding of employees in both opera-
tional environments. One of the transit agencies
suggested, “Get transit agencies to understand the
nature of freight rail business. Get freight railroads to
understand the requirements for public funding of
transit projects, lengthy time, many steps, public review
and comment, etc.”
According to a recent TCRP study [3], “Fail-safe”
means that any foreseeable failure will result in the
system reverting to a safe state. This approach works
well for mechanical and electromechanical devices, but
has proved very difficult to apply to the programmable
processor-based systems that have been coming into
use over the past 10–15 years. New safety assurance
methods and procedures have been developed for
application to safety-critical electronic systems and
software and related installation and maintenance
activities.
There are a number of train control systems that
ensure that trains sharing track do not collide [3].
These systems override the vehicle operator in the
event of a signal violation that could result in
a collision with another train. A list of various
approaches to provide fail-safe train separation in
approximate order of their common use is highlighted
here but detailed information on each should be
referenced in the TCRP Report No. 130 [3] and other
engineer publications:
● Automatic Train Stop (ATS)
● Train Protection and Warning System (TPWS)
● Indusi “Punktförmige Zugbeeinflussung” (PZB)
● Automatic Train Control (ATC)
● Communication Based Train Control (CBTC)
● Positive Train Control (PTC)
Three federal agencies, FRA, FTA, and Surface Trans-
portation Board (STB) govern different aspects of rail-
road operation. FRA focuses on safety; FTA controls
funding to transit projects; and STB is responsible for
economic regulation of railroads. However, none of

these agencies has responsibilities and authority to facil-
itate transit systems on freight track access issues, leaving
transit agencies who desire for a commingled operation
in a difficult situation when negotiate with railroad
companies. The difficulties create a discouraging effect
on the development and expansion of commuter rail
and light rail transits. Therefore, an affiliate agency
should be established to provide guidance or facilitate
the negotiation on track access. The Rail Transit Safety
Initiative recently submitted by Secretary LaHood [34]
has the potential to empower the safety oversight and
facilitate shared-use issues. However, detailed pro-
grams and procedures are still under development
and their impact on shared use remains to be seen.
FTA and FRA issued Joint Statement to “coordinate
the use of respective safety authorities with regard to
shared-track operations.” However, the waiver pro-
gram restricts the opportunity to commingled opera-
tion since it is a strict and tedious review process. FRA
officials are cautious to say “yes” to commingled oper-
ation. Though the technologies are sophisticated and
plenty experience has been gathered from overseas, no
one can guarantee 100% safety and no one can afford
the blame of permitting the commingled operation. It
is safe to assume that not only concerns over operation
safety, but also concerns over public reaction and polit-
ical impact are important in the FRA decision making.
The objectives of the business model are to establish
the core operational requirements essential in order to
satisfy oversight agency’s concerns for retention of an
equivalent level of safety for joint freight/LRT service
operations [35]. The business model addresses the spa-
tial, temporal, and institutional changes associated
with sharing track by freight and noncompliant pas-
senger vehicles without temporal separation. It will also
highlight strategies to deal with incremental risks and
liabilities associated with commingled operations.
It is a challenge to develop a business model of
commingled operations since there are no domestic
cases of “commingled” operations of freight and non-
FRA-compliant passenger services, and the number of
cases where freight and light transit equipment share
tracks on a temporal separation basis is limited. On the
other hand, greater opportunities may be created since
there are many domestic cases where public passenger
transportation and private freight operations share
track, and there are many proven international
Light Rail Transit, Shared Infrastructural Issues 701
experiences in commingled operations of LRT and
freight services [35].
As pointed out by Transportation Research Board
[29], the noncompliance, exemption, and buff strength
requirement are largely developed based on the Amer-
ican mentality, similar to larger SUVs on the road. That
is, if you drive a larger car or heavier train, your damage
would be less when an accident happens. On the other
hand, the European philosophy relies on accident
avoidance, that is, all parties try their best to avoid
collision and other accidents altogether. The philo-
sophical difference may serve as a fundamental reason
for Karlsruhe, Germany being safely and efficiently
operated for years while its American counterpart can-
not even pass the pilot stage.
Judging from the current extensive discussion of the
shared issues, newly commissioned studies by GAO
[32], FRA [2, 19], NJ Transit [1] and TCRP [3], it can
be safely predicted that the interest on shared track or
commingled operation has been reawakened. Along
with the increasingly widespread interest, a pilot pro-
gram could develop a realistic framework to explore the
potential of commingled operations.
Based on the obvious safety concerns, capacity expan-
sion and cost savings are also main factors driving
commingled operating practices. Before any commingled
operation may be even tested, a thorough cost-benefit
analysis coupled with a realistic operation analysis should
be conducted. A detailed risk assessment should also be
performed and a set of performance measures should
be developed to benchmark and mitigate the risks.
Future Directions
Interest in LRT as a viable urban transportation system
has been growing worldwide and new LRTs are
appearing on urban streets in many countries. Pressed
with a growing demand for extended LRT services, a few
LRT agencies seek the possibility of applying overseas
experience in USA. After examining existing practices in
the USA and overseas, it is hoped that this entry has
highlighted that commingled operation has many
advantages, including, but not limited to reduced con-
struction costs, increased track usage, and quicker new
starts. With strict train scheduling and on-time perfor-
mance control, the commingled operation proves to be
viable overseas. Japan and some European countries,
702 Light Rail Transit, Shared Infrastructural Issues

especially Germany pioneered in commingled operation PZB Indusi “Punktförmige

between LRT and freight railroad. However, many issues Zugbeeinflussung” System
need to be resolved before the USA can realize similar ROW Right-of-Way
implementations. RSIA Railroad Safety Improvement Act
Building on the current knowledge base and SDTI San Diego Trolley Inc.
momentum on transit demand under the large eco- STB Surface Transportation Board
nomic background, it is foreseeable that multiple fronts SWG Shared-Use Working Group
will be advanced along the lines of technology improve- TCRP Transit Cooperative Research Program
ment, regulation changes, and business model develop- TPWS Train Protection and Warning System
ment to promote shared use of railroad infrastructure TRB Transportation Research Board
between LRT and freight railroads. A logical next step UTA Utah Transit Authority
may be a pilot model to address all these issues in VTA Santa Clara Valley Transportation
a confined but realistic environment. Authority

Acronyms Bibliography
ACS Automatic Cab Signaling Primary Literature
ADA American with Disabilities Act
1. Liu RR (2004) Survey of transit/rail freight interaction final
APTA American Public Transportation report. Submitted to New Jersey Department of Transporta-
Association tion/Federal Highway Administration, submitted by New Jersey
ATC Automatic Train Control Institute of Technology, September 2004
ATS Automatic Train Stop 2. Alan B, Nelson D, Stangas P, Liu RR (2006) ITS technologies for
CBTC Communication Based Train Control integrated rail corridors. Submitted to Federal Railroad
Administration (FRA) US Department of Transportation,
CFR Code of Federal Regulation
Washington DC, December 2006
CN Canadian National 3. Transportation Research Board (2009) Shared use of railroad
CTC Centralized Traffic Control System infrastructure with non-compliant public transit rail vehicles:
DMU Diesel Multiple Unit a practitioner’s guide. In Transit cooperative research program
FHWA Federal Highway Administration report 130, TRB, National Research Council, 2009
4. America Public Transportation Association (APTA) (2010)
FRA Federal Railroad Administration
10.2 Billion trips taken on US public transportation in
FTA Federal Transit Administration 2009, http://www.apta.com/mediacenter/pressreleases/
GAO Government Accountability Office or 2010/Pages/100308_Ridership_Report.aspx. Accessed in
General Accounting Office, prior to March 2010
2004 5. Transportation Research Record (2001) Supplementing and
ICE Intercity Express Train updating TCRP report 52: joint operation of light rail transit
or diesel multiple unit vehicles with railroads. In Research
ITS Intelligent Transportation System
Result Digest, TRB, National Research Council, 2001
JNR Japanese National Railways 6. America Public Transportation Association (APTA) (2009) Pub-
JR Japanese Railways lic transportation fact book, 60th edn, American Public Trans-
LRT Light Rail Transit portation Association, April 2009
LRV Light Rail Vehicle 7. Boorse J, Tennyson E, Schumann J (2000) This is light rail
transit. Conference Proceedings of 8th national conferences
MBTA Massachusetts Bay Transportation
on light rail transit. Dallas, 2000
Authority 8. Grava S (2003) Urban transportation systems. McGraw-Hill,
MTA Maryland Transit Administration New York
MTDB Metropolitan Transportation 9. Thompson G (2003) Defining an alternative future: birth of
Development Board the light rail movement in north america, experience, eco-
NCTD North County Transit District nomics, and evaluation from starter lines to growing sys-
tems. Proceedings of the 9th national light rail transit
NJ TRANSIT New Jersey Transit
conference. Transportation Research Board, Portland,
PTC Positive Train Control System pp 25–36, 2003

10. US Department of Transportation (1996) Transportation
expressions, research and special projects administration,
Washington DC
11. Schumann J (1977) Historical of railroad and transit joint track
use, unpublished paper
12. Vigrass J (1995) Joint use of track by electric railways and
railroads: historic view. The proceeding of the 7th joint con-
ference on light rail transit, Vol 1,TRB, National Research Coun-
cil, Washington DC, pp 154–163, 1995
13. Phraner D, Roberts R, Stangas P, Korach K, Shortreed J,
Thompson G (1999) Joint operation of light rail transit or diesel
multiple unit vehicles with railroads, TCRP Report 52, TRB,
National Research Council, Washington DC, 1999
14. Federal Railroad Administration (2000) Shared use of the
tracks of the general railroad system by conventional railroads
and light rail transit systems. 2000
15. Schulte CF (2001) General system and rail transit common
corridor safety, Rail transit conference, Boston MA, 2001
16. Sheys KM, Spear T (2000) Safety oversight of shared track
operations: an update, the Proceeding of the 8th Joint con-
ference on light rail transit, Dallas, Texas, 2000
17. Phraner D (2000a) Light rail sharing track with other rail
modes: how far have we come at the millennium? The pro-
ceeding of the 8th joint conference on light rail transit, Dallas,
Texas, 2000
18. Christner P, Mauri P (2005) Sharing of track by transit
and freight railroads access, safety, and insurance issues.
Prepared for Walter Kulyk and Venkat Pindiprolu, Federal
Transit Administration, US Department of Transportation,
John A. Volpe National Transportation Systems Center,
United States Department of Transportation, Cambridge
MA, 2005
19. Ensco, ICF, KKO, and Mayville (2005) Safety of non-
compliant passenger rail equipment, Draft Report, Federal
Railroad Administration, Office of Research and Develop-
ment, 2005
20. Tereschuck P (2005) Draft of implementation strategy. Joint
LRT/Freight Operations. Unpublished report
21. Geissenheimer H (1997) New vehicle options for lower-cost rail
mobility: using diesel or dual-powered light rail cars as
a transfer of technology, Transportation Research Record, No.
1571, Transportation Research Board, Washington DC,
pp 125–133
22. Homburger W (1998) Practitioner’s forum. J Transp Eng
124(5):402–403
23. Phraner D (2002) German track sharing experience, TCRP
Research Result Digest 47. TRB, National Research Council,
Washington, DC
24. Phraner D (2000b) German shared track experience indepen-
dent supplemental technical study tour findings, TCRP Project
J06, Task 34. 2000
25. May J (2005) Nordhausen tram trains rule OK, Tramways &
Urban transit November 2005
Light Rail Transit, Shared Infrastructural Issues 703
26. Transportation Research Board (2003) Transit capacity and qual-
ity of service manual, 2nd edn. In Transit Cooperative Research
Program Report 100, TRB, National Research Council, 2003
27. Lightrail Now (2010) Rail transit systems in operation, http://
www.lightrailnow.org/. Accessed January 2010
28. Federal Transit Administration (2003) National Transit Data-
base 2003
29. Association of America Railroad (2010) Weekly traffic summary
report, http://www.aar.org/NewsAndEvents/WeeklyTraf-
ficReport.aspx. Accessed in January 2010
30. Transportation Research Board (1999) Joint operation of light
rail transit or diesel multiple unit vehicles with railroads. In
Transit Cooperative Research Program Report 52, TRB,
National Research Council, 1999
31. Iwasa K (2001) Rail freight in Japan – the situation today and
challenges for tomorrow. Jpn Railway Transport Rev 26:8–17
32. General Accounting Office (2004) Information and guidance
could help facilitate commuter and freight rail access negoti-
ations. GAO-04-240, 2004
33. Nash A (2003) Best practices in shared-use high-speed rail
systems. Mineta Transportation Institute, College of Business,
San Jose State University, 2003
34. Federal Transit Administration (2010) Rail transit safety initia-
tive, http://fta.dot.gov/11039_11052.htm. Access in January
2010
35. Liu RR, Nelson DO, Lu A (2009) Business model for shared
operations of freight and passenger services (09-2547).
J Transport Res Board (TRB) 2547:86–92, Transportation
Research Record (TRR), Washington DC
Books and Reviews
California Public Utilities Commission (1991) General order 143-B,
Safety rules and regulations governing light-rail transit, San
Francisco, CA, 1991 as amended
Choueiry E, Bullock D (2002) Light rail DMU on shared right
of way. Proceedings of commuter rail/rail transit conference,
June 9–13, 2002
Liu RR, Yang F, Chen M (2005) Understanding the shared operation
of commuter rail transit and freight railroads. J Transport Res
Forum 44(1):157–172
Liu RR, Yang F, Bing A (2007) State of the practice on commingled
operation of light rail transit and freight rail services. Proceed-
ings of the 86th Annual Conference of Transportation
Research Board. January 2007. Peer Reviewed
Liu RR, Yang F (2006) A national survey of shared use of railroad
infrastructure and LRT service. Proceedings of the 85th annual
conference of Transportation Research Board. January 2006.
Peer Reviewed
Sela E, Randolph R, Hickey T (2003) Shared use corridors: survey of
current practice and recommendations for the future. Pro-
ceedings of Transportation Research Board 9th national light
rail transit conference, November 16–18, 2003
704 Light Rail Transit, Systemic Viability
Light Rail Transit, Systemic Viability
GREGORY L. THOMPSON, JEFFREY R. BROWN
Department of Urban and Regional Planning, Florida
State University, Tallahassee, FL, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Historical Background
Conclusions
Future Directions
Abbreviations
Bibliography
Glossary
Average speed The speed that results by dividing the
round trip time of a transit vehicle to traverse
a transit route from beginning to end and back to
the beginning, including the layover time until the
beginning of the next trip, by the roundtrip length of
the route. Average speed also may be calculated by
dividing annual vehicle miles for a route by annual
vehicle hours for the route. Average speed is
slower than scheduled speed because average speed
includes layover time at each end of the route.
Bus rapid transit An application of buses in a way that
emulates rail transit. Bus rapid transit usually pro-
vides for a reserved right-of-way for buses that
includes wayside stations. Fare collection systems
allow passengers to board and alight buses at all
doors. Drivers are not involved in fare collection.
Buses usually are articulated and contain several
wide doors, all of which open when buses stop at
online stations. Rights-of-way are usually Type A or
B, and scheduled speeds of 20–25 mph are possible.
CBD Central business district. The CBD typically is
the largest business center in a metropolitan area
and often dates from the streetcar era. At one time,
CBDs contained almost all jobs, shops, and recrea-
tion in metropolitan areas, but since World War I
when autos became the principal form of urban
transportation, CBDs have been in relative decline
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
as most job, shopping, and recreation growth has
occurred in the suburbs.
Commuter rail A form of rail transit in which transit
vehicles use the same tracks (or have the capability of
doing so) as do long distance freight and passenger
trains. Commuter trains usually are comprised of
several large cars containing closely spaced seats and
two or three wide doors on each side of the car to
allow rapid boarding and alighting of passengers.
Commuter trains can be electrically or diesel powered
and may be either pulled by locomotives, or have
each car equipped with electric motors or diesel
engines under the floor. Conductors typically collect
fares from passengers while the trains are in motion.
Edge city One of several terms used to describe large
urban centers in metropolitan areas that rival cen-
tral business districts in magnitude and diversity of
employment. Edge cities typically cover a much
larger area than CBDs, however, with much of
the land devoted to parking lots and high-speed,
multilane roads. It typically is difficult or impossi-
ble to walk from one part of an edge city to
another part.
Express bus A form of bus transit in which buses run
nonstop over much of their routes. The most typ-
ical application is for express buses to run during
weekday rush hours between outlying affluent sub-
urbs and the metropolitan CBD. In the morning
express, buses typically circulate through neighbor-
hoods in which affluent commuters live, and when
they fill up the buses travel nonstop on a reserved
freeway lane or sometimes on a specially built
busway to the edge of the metropolitan CBD. The
express buses may use purpose-built ramps to gain
access to or to leave their freeway lanes. Express
buses then circulate through the CBD on city streets
to distribute their passengers. In the afternoon peak
period, express buses retrace their path to the sub-
urbs. Most express bus routes offer a handful of
one-way inbound trips in the morning and one-
way outbound trips in the afternoon. Passengers
pay their fares to the driver.
Grid route structure A multidestination transit
routing method, in which each north–south and
east–west arterial road in a metropolitan area has
its own bus, streetcar, light rail, or heavy rail route,
# 2012, DOI 10.1007/978-1-4419-0851-3

typically providing frequent service. To travel from
one part of a metropolitan area to another, passen-
gers typically walk to a nearby bus or streetcar line,
travel to a connecting route, and alight near their
final destination. Grid route structures afford pas-
sengers the ability to travel from any place to any
other place in a metropolitan area without out-of-
direction travel, but grid route structures require
most passengers to transfer. Grid route structures
typically attract much higher passenger volumes
than radial transit route structures and are more
productive, as well.
Heavy rail The word “heavy” refers to heavy infra-
structure that is associated with subway and elevated
rapid transit lines. Heavy rail lines are exclusively
Type A rights-of-way achieved by directing the paths
(or grades) of rail transit lines above or below paths
(or grades) of conflicting transportation. Thus,
heavy rail is also called, “grade-separated” transit.
Typically, heavy rail is operated by trains of up to ten
electrically powered cars, each car having two or
three wide doors on each side. Stations are designed
to be as long as the longest trains, an expensive
proposition when, as is typically the case, the entire
station is buried far below the surface or erected far
above ground. Passengers pay fares at turnstiles or
elaborate and expensive fare collection machines
before boarding trains.
Interurban A nearly extinct form of US rail transit in
which lightly built electric rail lines connected more
densely populated rural areas with nearby towns or
cities. Lines typically were 15–40 miles long, but
both shorter and longer lines were built. The lon-
gest through interurban run was from Detroit to
Cincinnati, a distance of over 300 miles. Typically,
interurbans entered city centers over the tracks of
local streetcar companies, but they operated
through the countryside either alongside state high-
ways or across county on their own rights-of-way.
Interurban cars stopped frequently and offered
scheduled speeds typically of about 20 mph.
Interurban cars were the size of smaller mainline
railroad coaches but typically each car was electri-
cally powered, using streetcar technology. Interur-
bans usually operated as one-car trains but could
operate in multicar trains controlled by
Light Rail Transit, Systemic Viability 705
a motorman in the lead car, much as rapid transit
trains. Interurban lines appeared in the USA during
the decade before the automobile appeared and
spread rapidly. They declined just as rapidly after
World War I and with a few exceptions disappeared
before the USA entered World War II.
Light rail “Light” refers to light infrastructure and
not to the weight of the rail or of the cars. Light
rail evolved from streetcars in northern European
cities in the three decades following World War II
and is characterized by short trains of high-
capacity rail cars operating though metropolitan
areas on mostly Type B rights-of-way. Each train
has multiple wide doors, and light rail systems
employ fare collection systems that obviate the
need for employees to be on trains to collect
fares. Passengers board and alight at all doors.
Light rail lines are intended to function like heavy
rail lines but at a much lower capital cost. In the
USA, they are most effective where they are con-
figured as long routes up to 30 miles long, along
which are situated major destinations. Scheduled
speeds typically are in the range of 19–30 mph in
contrast to typical local bus speeds of 10–15 mph.
Bus routes are reconfigured from CBD-radial ser-
vice to light rail stations at major destinations in
the suburbs, thereby creating multidestination
transit systems with improved productivity. Met-
ropolitan transit networks having such character-
istics are among the most successful in the USA
and Canada.
Mode A type of transportation technology, such as the
walk mode, the auto mode, the bus transit mode, or
the rail transit mode.
Multidestination transit route networks A transit
network so structured that users can reach many
destinations in a metropolitan area while avoiding
out-of-direction travel. Doing so usually requires
a transfer, however. A grid network is one type
of multidestination network, but not the only type.
Multidestination transit systems stand in contrast to
radial transit systems, where most transit routes radi-
ate outward from the one destination that is deemed
the most important to serve, usually the CBD.
PCC Streetcar A streetcar developed by the Presidents’
Conference Committee (a committee of the
706 Light Rail Transit, Systemic Viability
presidents of US and Canadian streetcar compa-
nies) between the late 1920s and mid-1930s. The
objective of the development was to create a quiet
and smoothly riding streetcar that could accelerate
and decelerate as rapidly as autos in stop-and-go
traffic while not jerking standing passengers off
their feet. The PCC car, particularly its control
system, is considered one of the marvels of North
American industrial engineering, but a streetcar
that was like a bus on rails competing with the
auto in its own environment was not what was
needed to save the transit industry.
Proof-of-payment (POP) Fare Collection System
Passengers are required to determine what their
fare for a particular trip should be and to then
purchase a ticket for that trip before they board
the vehicle. They may board any door. Nobody on
board monitors fares while passengers board, but
roving inspectors may ask for proof of payment at
any time. Passengers without valid proof are warned
or fined. Managers typically employ a sufficient
number of inspectors to inspect 1% or 2% of
passengers. The percentage of passengers without
valid proof of payment is the fraud rate, which is
generally lower than that of driver-administered
fare systems. Its chief advantage is allowing passen-
gers to board at all doors, important for high-
capacity lines but not important for low patronage
bus lines. Most passengers using such systems buy
monthly, weekly, or daily passes. Passengers buying
90 min passes may do so at simple vending
machines on train platforms.
Radial transit route systems Most transit routes radi-
ate outward from the one destination that is
deemed the most important to serve, usually the
CBD. Passengers are given a one-seat ride from
their homes to the CBD. Underlying a radial route
structure is the belief that passengers will not trans-
fer. Radial route systems stand in contrast to
multidestination route networks (of which the
grid network is an example). Most radial transit
systems in the USA are characterized by falling
demand and productivity. They typically have
higher operating costs per passenger mile than
multidestination networks.
Rapid transit The traditional name for what now is
known as heavy rail.
Scheduled speed The speed that results by dividing
the length of a route by the time that it takes for
a transit vehicle to travel from the beginning to the
end of the route. Scheduled speed is much slower
than the top speed at which transit vehicles operate,
because scheduled speed includes time spent at
stops, acceleration and deceleration time (which is
what transit vehicles are doing much of the time –
hence the design objective of the PCC car), and it
includes time lost in traffic congestion. Scheduled
speed is faster, however, than average speed, because
the later includes the layover time at each end of
the route.
Streetcar An electrically powered rail transit vehicle,
in North America traditionally about 40 ft long and
8 ft wide. Streetcars historically ran on tracks
imbedded in the middle of streets, and they com-
peted for street space with vehicular traffic. Thus,
they ran on right-of-way type C. Passengers board
and alight from streetcars in the centers of streets,
sometimes on raised concrete islands. In North
America, streetcars typically operated with two-
person crews, a motorman to drive the car and
a conductor to collect fares while the car was in
motion. PCC cars, which appeared in the mid-
1930s, were operated with foot pedals that left the
driver’s hands free to collect fares, enabling one-
person operation.
Traction Until the 1920s, streetcar and interurban
companies were known as traction companies.
One still hears the term, “electric traction.”
Transit-oriented development, or TOD A type of
development advocated by many planners but
not often seen that is comprised of dense, clusters
of various types of housing units, retail stores,
restaurants, bars, and offices clustered within
a quarter of a mile of a transit stop, preferably
a rail station. TODs also are characterized by
pleasant pedestrian paths linking all of their
parts with each other and with the transit station.
Studies of the behavior of residents of and visitors
to TODs indicate that most use automobiles but
that a significant minority do in fact use transit.
TODs built to date usually have large parking
structures as well as transit access.
Transit The generic name used today in the United
States for fixed-route, scheduled public

transportation services, whether they are bus or rail.
One may speak of transit agencies, rail transit, bus
transit, or just transit.
Trolley bus A bus powered by electricity delivered to
the vehicle by overhead wires. Upward-sprung dual
trolley poles (one for positive current and the other
for negative current) connect with dual parallel
overhead wires strung above the street. Trolley
buses may travel in the traffic lane under the
wires, or they may travel in lanes on either side of
the lane under the wire.
Vehicle miles traveled (VMT) An automobile or
light truck traveling 1 mile. VMT is the standard
unit for measuring the magnitude of auto use in a
metropolitan area.
Verkehrsverbund or federation of transit operators
A form of transit organization used in several Ger-
man metropolitan areas. The Verkehrsverbund is
characterized by separation of ownership of tran-
sit services in the metropolitan region among
many different transit operating companies or
agencies, and coordination between all of the
transit services in the region through a small,
centralized public marketing and planning body.
Passengers may navigate transit routes through-
out the metropolitan area without being aware of
boundaries between different transit operating
agencies. The idea behind this form of organiza-
tion is that smaller operators may control their
timeliness as well as their operating costs far bet-
ter than large operating companies, while
a central coordinating body insures coordination
among all operators through scheduling, fares,
and provision of information. The San Diego
MTDB created and ran this form of organization
from about 1979 to 2003 and experienced better
coordination between its various transit services
and lower costs per passenger mile than any of the
large public transit districts in the USA. It also
experienced among the highest rates of passenger
growth during this period.
Definition of the Subject
Light Rail Transit
Light rail transit is one of several forms of rail public
transportation designed to transport passengers within
Light Rail Transit, Systemic Viability 707
metropolitan areas, but there is no agreement on its
precise definition. The Transportation Research
Board of the National Academy of Sciences [1] and
the American Public Transportation Association [2]
define light rail to include streetcars and tramways.
Both organizations also define light rail to include
modern metropolitan rail systems characterized by
multicar trains operating on tracks that are segregated
from motor vehicle traffic except at cross streets.
Vuchic [3] defines light rail with respect to the nature
of metropolitan rail right-of-way that it uses. He cate-
gorizes metropolitan rail rights-of-way according to
three types. Type A denotes grade-separated rail
rights-of-way that have no conflicts with motor vehicle,
other rail, or bicycle and pedestrian traffic. Type A rail
rights-of-way characterize subway and elevated rail
rapid transit lines in the world’s cities. Type B denotes
rail rights-of-way where rail vehicles are segregated
longitudinally from motor vehicle traffic, bicycles,
and pedestrians but have conflicting traffic crossing
tracks at cross streets. Type C denotes rights-of-way
where the tracks are imbedded in streets shared by
motor vehicle and/or pedestrian traffic, and rail vehi-
cles have no preference over other vehicles with which
they compete for road space. Traditional streetcars and
trams, characterized by the operation of single rail cars
in mixed traffic with drivers collecting fares, are asso-
ciated with right-of-way Type C. Vuchic associates light
rail transit primarily with right-of-way type B, upon
which short multicar trains operate with fare systems
allowing passengers to board at all doors [4]. Vuchic’s
definition gives light rail much higher level of service,
measured in average speeds and passenger-carrying
capacity than streetcars and trams (typical scheduled
speeds of 10–14 mph), but lower speeds and passenger-
carrying capacity compared to subway and elevated rail
rapid transit lines (scheduled speeds of 20–45 mph),
whose long trains use Type A rights-of-way. According
to Vuchic’s definition, the cost of constructing light rail
often falls between the high cost of constructing sub-
ways and elevated lines and the much lower cost of
constructing streetcar lines.
Vuchic’s definition for light rail is closely aligned
with the historical rise of light rail transit in the 1950s,
1960s, and 1970s as a reform mode of metropolitan
transportation that addressed shortcomings of tradi-
tional streetcars and trams as they succumbed to local
708 Light Rail Transit, Systemic Viability
bus replacements in city after city throughout the
world [5, 6]. Political leaders in many cities wanted
a form of urban public transportation that provided
faster, higher capacity, and more productive service
than traditional streetcars or local buses but that cost
less than subways or elevated railways that only the
largest cities could afford. Light rail, and more recently
bus rapid transit, evolved to fill unmet demand in
many medium-sized metropolitan areas for a transit
mode offering intermediate capacity.
This entry follows Vuchic’s definition for light rail.
It defines light rail as the operation of short trains
operated with only one person. The trains have many
large doors, any of which passengers may board at
stops. Drivers are not involved in fare collection.
Light rail trains operate primarily on right-of-way
Type B, though they have the flexibility to operate
over rights-of-way Types A and C. Some segments of
a light rail system may exhibit all of the features of
a completely grade-separated metro line, while other
segments have lower-speed, at-grade operation. Selec-
tive investment based on need in specific areas is a key
characteristic. What is “light” about light rail is the
overall average level of capital investment per unit
length of a route; the term does not pertain in any
way to physical weight.
Generally, light rail trains operate at faster sched-
uled speeds (18–30 mph) than either streetcars or local
buses and provide higher passenger carrying capacity,
as well. Light rail systems also carry more passengers
per employee than buses or streetcars.
The intent of light rail construction is to attract
more passengers to transit by providing a quality of
transit service that approaches what metropolitan sub-
way or elevated rail rapid transit lines could achieve in
particular corridors but at a much lower overall capital
cost. In some metropolitan areas, the intent of investing
in light rail also is to improve the productivity of the
transit system by replacing numerous duplicative
bus routes running to the central business districts of
cities with one or two light rail lines to which passen-
gers transfer at outlying stations. In such cases, transit
operators restructure shorter bus routes to suburban
light rail transit stations: because buses meet trains at
the same time, they also connect with each other, and
therefore also make intra-suburban transit usage more
feasible. The intent is that fewer bus and train car miles
will be more aligned with dispersed travel demands in
today’s decentralized metropolitan regions, thus
attracting more passengers, resulting in more passen-
gers per employee and lower operating cost per passen-
ger. In recent years, some political leaders also have
intended light rail investments to attract new urban
development to the vicinity of light rail stations.
Viability
To measure viability of light rail transit is to measure
how well light rail investments achieve their objectives.
Following on Fielding’s treatment of transit perfor-
mance measurement [7], this entry evaluates how
light rail investments change overall transit perfor-
mance in the metropolitan areas where the light rail
investments are made compared to similar metropoli-
tan areas where light rail investments are not made.
In general, the entry examines trends in metropolitan
transit ridership per capita before and after light rail as
a measure of light rail’s effect on transit ridership in the
region. It examines trends in ridership per transit vehi-
cle mile over the entire region as a measure of produc-
tivity. It examines regional transit operating expense
per passenger or passenger mile as a measure of effi-
ciency. Where results of onboard passenger surveys of
passenger characteristics are available, this entry also
reports those. The entry also examines how light rail
has affected development in the areas that it serves by
referencing academic studies on that topic.
Introduction
Light rail transit evolved from streetcars in northern
European cities during the 1950s and 1960s. It was what
resulted as transit operators incrementally modified
their traditional streetcar systems to survive mass
motorization and rapidly rising wages. Light rail was
characterized by one driver (with the assistance of no
other labor) operating a short train comprised of large
articulated streetcars with numerous wide doors, fare
collection systems allowing passenger to enter and
leave all doors, and segregation of trains from automo-
bile traffic. Light rail was much more productive than
buses and proved to be a more successful adaption of
streetcars than the earlier American approach to saving
streetcars from auto and bus competition. The Amer-
ican approach was based on developing a single unit

streetcar about the size of a bus whose performance
mimicked that of automobiles, with which streetcars
competed for street space. Plagued by congestion from
passengers forced to board front doors to pay their
fares to the driver, as well as by auto congestion, the
American approach resulted in at best marginal pro-
ductivity superiority over buses. It failed to save street-
cars in most US and Canadian cities.
The European light rail idea, carried on the wings of
Canadian and US urban advocates and transit
reformers, was first adopted in a North American city
in 1978 when Edmonton opened a light rail line and
restructured buses around its stations, 27 years after it
converted its last streetcar line to bus operation.
Edmonton’s sister city, Calgary, opened its first light
rail line in 1981. The US transit industry resisted the
light rail idea, but reformers forced transit systems to
adopt light rail, first in San Diego in 1981, followed by
Portland, San Jose, and Sacramento in 1986–1987. The
US Urban Mass Transit Administration (UMTA) spon-
sored the opening of a demonstration light rail line in
Buffalo in 1984 with the full cooperation of the transit
system in that city. The next line opened in Los Angeles
in 1990, and 12 additional metropolitan areas added
light rail lines since then. Many of the metropolitan
areas that opened initial light rail lines since have
expanded their systems. Some of the North American
streetcar lines that survived into the 1970s also have
been rebuilt with some light rail characteristics.
Light rail advocates hoped that light rail invest-
ments would reverse the fortunes of declining North
American transit systems. Some hoped that the invest-
ments would do the same for declining downtowns and
inner city neighborhoods. Some actually hoped that
light rail combined with renewal of urban centers
could displace autos as the main form of urban trans-
portation. Actual results have been mixed. Some light
rail investments, such as in San Diego, Portland, and
St. Louis, have had highly positive impacts on the
performance of transit in their metropolitan areas.
Others have had little positive impact. Some have
been accompanied by greater population densities
and more walkable environments. None, however, has
reduced significantly the dominance of automobiles as
the primary mode of urban transport.
The purpose of this entry is to explain why results
are mixed, and in particular to understand what light
Light Rail Transit, Systemic Viability 709
rail innovations are most effective for better transit
system performance. It does this by first explaining
how streetcars arose and affected the growth of North
American cities before World War I. It then examines
reasons for streetcar decline, efforts to stem the decline,
and reasons why those efforts failed. It then explains
how light rail arose as a fix for streetcar defects, how
and why light rail was adopted in various North Amer-
ican cities. The analysis of light rail performance in
various US metropolitan regions precedes the entry’s
conclusions.
Historical Background
Light rail is not synonymous with streetcar (or tram-
way in English parlance); it is more accurate to see light
rail as a mode that evolved as a reaction to shortcom-
ings of streetcar technology and applications. This sec-
tion makes that evolution understandable. It provides
a history of the North American streetcar beginning
with a depiction of its technology and how it
influenced the form of American cities before World
War I. It then considers the streetcar’s increasingly
apparent shortcomings as auto use intensified, urban
form changed, and city bus technology improved. The
history considers the American and Canadian transit
industry’s attempt to modernize the streetcar with new
technology in the 1930s (the famous PCC car), and it
considers reasons for the failure of that effort, including
the American auto industry’s involvement in liquida-
tion of American streetcar systems. Finally, it considers
the appearance in the postwar period in western
Germany of a different, more successful approach
to modifying streetcar technology that is known as
light rail.
The American Streetcar at the Peak of Its Influence
and How It Shaped the City
From 1888 when Frank Sprague implemented the
world’s first successful streetcar system in Richmond,
Virginia through the 1920s, the electric streetcar sym-
bolized the American transit industry. In cities
throughout the country the press followed the expan-
sion and financial scandals of the traction industry,
“traction” being the term that the public then used
for streetcars. Land owners on the fringes of cities
clamored to have traction lines extended to their
710 Light Rail Transit, Systemic Viability
lots, thereby making the lots much more valuable. In as
little as 30 min and for a nickel fare (a fare the middle
class could afford), streetcars operating at average
speeds of 12 mph could connect lots in the countryside
with jobs and opportunity in the center. Such relation-
ships stimulated a huge demand for suburban living,
and to meet it, traction lines, sometimes owned by real
estate interests, expanded outward in all directions.
Population followed. For the first time in
their histories, the horribly overcrowded cities of
America began to empty outward in the form of new
streetcar suburbs, built on previously empty land
around the city edges. The center cities from which
the middle and working classes fled changed as well,
as department stores, specialized shops, corporate
offices, financial firms, hotels, theaters, concert halls,
and other less reputable types of entertainment rushed
in to fill the voids left by the departing middle and
working classes [8, 9].
By the 1910s, the streetcar created a new American
city characterized by specialized sections. Iconic was
the fashionable Central Business District, a vibrant
area with specialized subdistricts wherein were located
almost all jobs and activities that the middle class
wished to reach. Poorer groups remained in cheaper,
older housing surrounding the new center cities. Fac-
tories and warehouses also remained packed around
railroad yards. Further out, extending along the electric
car lines, were the new streetcar suburbs. These were
middle and working class bedroom districts, each
a new homogenous enclave catering to a specialized
income and ethnic group. Every day the middle and
working class populations surged into the center on
electric cars and surged outward in the evenings. Wives
and children filled the cars during the middle of the
days, and in the evenings and Sundays families used the
cars for recreation. For the middle and working classes,
the streetcar revolutionized the American city and the
way in which they experienced it [8–10].
The streetcar evolved to efficiently fulfill the mass
transit role called for it. By the 1910s, the modern
streetcar was about 40 ft long and 8 ft wide, riding on
two four-wheel trucks. Each axle was driven by its own
electric motor that was suspended from the axle. Air
pressure controlled the braking power. Cars typically
had top speeds of about 30 mph. A cabin (saloon) in
which passengers rode constituted about two-thirds of
the length of the vehicle. On cars operating in clement
climates, cabins often were divided into a weather-
protected section and an open-air part. The rest of
the length typically was comprised of large platforms
projecting from both the front and rear of the car. The
platforms, which were lower to the ground than the
center sections, contained large doors and vestibules
from which passengers could board and alight. The
driver, called a motorman, stood at the controls on
the front platform, and the conductor, who collected
fares, stood on the rear platform. The conductor com-
municated with the motorman with bells that he
(rarely she in 1910) could activate by an overhead
cord. Terms like, “platform employees,” or, “platform
hours,” still are used by some transit systems today to
denote transit operating personnel (usually bus
drivers) and the hours that they work serving the riding
public [8]. Figure 1 shows a state-of-the-art streetcar
from this period.
In operation, the streetcar was a crowd-eater.
Coasting into a stop thronged with waiting passengers,
the streetcar’s gaping rear door beckoned the crowd to
board. As the last passenger cleared the steps, the con-
ductor gave two quick yanks on the bell cord to the
motorman, who replied with two loud clangs on the
car’s traffic-warning brass gong affixed to the front.
The car lurched forward, and the conductor then col-
lected fares as the car swayed and trundled along to its
next stop, often interrupted by traffic congestion
and furious bursts of clangs from the motorman’s
gong. As passengers paid their fares, they proceeded
forward into the cabin to find a seat. Those not finding
a seat stood in the aisle holding onto leather straps
hung from the ceiling, or they stood on the front
platform, from which they ultimately alighted. Cars
from this period were not fast, but despite the huge
crowds they ingested and disgorged, they did not lose
time at passenger stops, because large numbers of pas-
sengers could board quickly – and fare collection
occurred while the car was under way. They truly
were mass transit vehicles [8].
The Decline of the American Streetcar Beginning
During World War I
Several factors began the streetcar’s long decline
around World War I. Most immediately, the cost of

Light Rail Transit, Systemic Viability. Figure 1
Traditional American streetcar. San Francisco Municipal Railway Car No. 1, dating from 1912, in a neighborhood it
helped to create. Author’s photo, taken in 1981
labor and supplies more than doubled during the war,
while streetcar companies often found it politically
difficult to raise the fare above the accustomed nickel,
sometimes written into corporate charters in the 1890s.
Profits turned to losses, and investment in tracks and
streetcars dwindled commensurately. Companies by
and large stopped extending lines to undeveloped
areas demanded by developers, and service became
less frequent and more overcrowded [11].
Beginning about 1910, automobile registrations
also climbed rapidly. Auto ownership initially rose
fastest in rural areas, but by World War I automobiles
were widely distributed in cities, as well. House builders
began shifting their products from clients using street-
cars to clients using autos. In cities throughout the
country, developers by World War I were offering
homes in areas not served by streetcars, though they
still were using their political muscle to attempt to force
streetcar companies to extend lines into new territory.
By 1925, however, housing developers no longer cared
whether streetcar companies extended lines into new
territory. Streetcars had become irrelevant to
them [11].
At the same time, motor trucks cut the tether
between retailing and railroad yards and led to the
Light Rail Transit, Systemic Viability 711
extraordinarily rapid decentralization of retailing.
Horse-drawn wagons were slow and expensive, and in
the pre-truck era, retailers could not afford to locate
their stores beyond a short wagon haul away from the
giant merchandise warehouses adjacent to the rail
yards. Almost all retailing was in the CBD. With the
adoption of trucks, however, retailers of all stripes
decentralized. The driving force was demand for free
parking from the increasingly large number of cus-
tomers who came downtown to shop in their autos.
Retailers quickly learned that the cheapest solution to
the parking problem was building major branch stores,
which could be easily supplied by truck from the cen-
tral city warehouses. By 1925, many established CBD-
located firms were opening new stores in the suburbs
that were grander and more posh than their original
CBD flagship stores. The new stores all were built
around auto access and free parking. The highly cen-
tralized, streetcar-oriented retailing regime character-
istic of the American city in 1910 had given way by 1925
to a decentralized regime characterized by rampant
suburban strip retailing and free parking. Already plan-
ners were talking about the decline of the CBD [10, 12].
The combined weight of these forces pushed down
streetcar patronage during the 1920s in all but the
712 Light Rail Transit, Systemic Viability

largest cities, and even there streetcar riders became
increasingly dominated by those going to and from
work. The smaller the city, the more rapid was the
loss of streetcar traffic [11]. Streetcar companies
began abandoning lightly used streetcar lines and
replacing other lightly used lines with buses.
Revolution in Streetcar Technology: The PCC
Most traction industry leaders thought that streetcars
would continue serving the more heavily traveled lines,
however, and toward the end of the 1920s, they formed
the Presidents’ Conference Committee to develop
a super streetcar to reequip the “trunk” lines.
The committee was comprised of presidents of various
streetcar companies and had at its disposal the
best electrical engineering talent from the larger oper-
ating companies as well as from the streetcar
manufacturing companies. The result was the Presi-
dents’ Conference Committee car, or PCC car, which
was unveiled in 1936, and which is considered today to
be one of the marvels of American industrial engineer-
ing. Figure 2 shows a train of two PCC cars coupled
together in Toronto. More than 5,000 PCC cars were
built between then and 1952 for US and Canadian
application, and even more were built for foreign use.
Despite the car’s success as a piece of industrial engi-
neering, however, it failed to save the North American
streetcar industry [8, 13].
In the PCC car, the traction industry sought and
obtained a streetcar that could keep up with autos in
stop and go traffic. The PCC with a full standing load of
passengers could accelerate smoothly and silently from
a stop as fast as an auto, and it did so without jerking,
an important consideration given that many of the
streetcar passengers were standees. It also could brake
to a smooth stop just as rapidly. Although its initial rate
of acceleration of about 3.5 mph per second fell off
rapidly as it gained speed, the PCC could reach about
45 mph on level ground, if given enough time. The
PCC also was quiet. It rode on trucks with rubber
inserts to reduce noise. Rubber pads even separated

Light Rail Transit, Systemic Viability. Figure 2

Two PCCs coupled together into a train in Toronto in 1974. There is a driver in the second car, as well, to collect fares and
transfers from boarding passengers, who are allowed to board only at the front doors of each car. Note passengers
queuing at front doors of both cars to pay fares. Also note sign on signal pole prohibiting motorists from turning, so as not
to impeded streetcars. Finally, note the illuminated white sign just to right of the streetcar, indicating a subway
entrance. This spot was (and remains today) an important transfer point between the north–south Yonge subway and
east–west Queen streetcar in Toronto’s gridded route network. Author’s photo

the steel wheel treads from the centers. The car was
streamlined, as well, and looked as modern as the latest
autos. Finally, the motorman could operate the car
entirely with foot pedals, leaving his or her hands free
to act as a fare cashier. The industry intended to operate
the PCC with a crew of one. Passengers would enter
one half of a double width front door, filing past the
driver to pay fares. They could exit from the other half
of the front door or from a double width center or rear
door [8, 13].
Improved Transit Bus Technology and Its
Replacement of Streetcars
Contributing to blunting the PCC’s promise for saving
American streetcars were major improvements to bus
technology. At the end of the 1920s, bus manufacturers
adopted monocoque construction that integrated body
and chassis construction into a high-capacity, strong,
and lightweight vehicle. Monocoque buses could carry
large passenger loads reliably and in comfort by the end
of the 1920s. In 1930, such buses began operating every
3 min through the new road tunnel connecting Detroit
and Windsor, demonstrating that buses could accom-
modate mass passenger movements. By 1940, the
Yellow Coach Division of General Motors had married
the monocoque bus with a lightweight, two-stroke
diesel engine connected to the rear wheels via
a hydraulic transmission. The result was a transit vehi-
cle that, according to technological historian Robert
Post could perform almost as well as the PCC car at
comparable operating cost without the need for tracks
or electrical overhead wires [8]. The bus had another
major advantage, as well. It did not get trapped behind
queues of left-turning autos and trucks at busy inter-
sections, increasingly the bane of streetcar operations.
The transit diesel hydraulic bus did have several
major disadvantages. Post cites choking diesel exhaust
as one. A second was the roar of accelerating diesel
engines [8]. Both exhaust and noise were particularly
objectionable on more heavily patronized lines with
frequent service, and they were severe nuisances in
areas of heavy pedestrian activity and confined spaces,
such as found in some downtowns. Although diesel
buses are far less polluting and much less noisy today,
pollution and noise emanating from them remain
issues and have led to the reduction of their use
Light Rail Transit, Systemic Viability 713
through downtowns in Denver and Minneapolis and
perhaps other cities [14]. Hilly terrain also intensifies
noise and pollution problems and slows heavily loaded
diesel buses to a crawl. Another problem was the fact
that bus drivers had to use their hands on the steering
wheel and thus could not complete dealing with pas-
sengers paying fares as the vehicle accelerated, as they
often did on PCC streetcars. Diesel buses also had
narrow doors, slowing boarding and alighting of
heavy passenger volumes.
By the mid-1930s, there was another bus technol-
ogy that performed about as well as PCC-equipped
streetcar lines but avoided air and noise pollution of
diesel buses. This was the electric trolley bus. Like the
PCC, trolley buses could accelerate rapidly with full
passenger loads, but unlike diesel buses trolley buses
could zoom quietly up and down steep hills with full
passenger loads. Trolley buses took their power from
dual overhead wires and could maneuver into lanes
right or left of the traffic lane over which the dual
wires were strung. Trolley buses did not require tracks,
thus saving a large investment, and while they did
require overhead electrical wires, they could use most
of the same electrical investment that already was in
place for streetcars [8].
With the availability of these two bus technologies,
the transit industry in America and Canada began the
full-scale replacement of streetcars on the eve of World
War II [8]. Most American transit systems still were
privately operated and, ignoring pollution issues
replaced streetcars mostly with diesel buses. There
were exceptions, particularly San Francisco and Seattle.
These two cities had hilly terrain and municipally oper-
ated transit and electric power systems and replaced
most streetcars (all in the case of Seattle) with trolley
bus systems. Canadian transit systems, most of which
were municipally rather than privately operated, were
more likely to turn to trolley buses. Most transit sys-
tems that installed trolley buses later replaced trolley
buses with diesel buses, however. Today, only two cities
in Canada and the USA continue to operate large trolley
bus systems: Vancouver, B.C. and San Francisco. Smaller
systems remain in Seattle, Dayton, Philadelphia, and
Boston.
By the end of the 1960s, vestigial streetcar services
remained in only a handful of US cities, and streetcar
patronage all but disappeared (Fig. 3). Most of the
714 Light Rail Transit, Systemic Viability
U.S. Transit Patronage 1930, 1950, 1970
20
Annual patronage (billions)
18
16
14
12
10
8
6
4 electric utility companies that owned transit systems to
2 divest themselves of the transit systems, thus ending
0
1930 1950 1970
Year Source: [15]
Bus and trolley bus passengers Streetcar passengers
Rapid transit passengers
Light Rail Transit, Systemic Viability. Figure 3
Between 1930 and 1970 streetcar passengers almost
disappeared. Local bus passengers declined dramatically
between 1950 and 1970. Only rapid transit passengers
(Type A right-of-way) maintained itself against the
onslaught of the automobile [15]
remaining street car services were operated with PCCs,
and most had elements of right-of-way Types B or A,
giving the streetcars a speed advantage over buses.
Toronto remained the sole city in Canada and the
USA that continued to operate a large and heavily
patronized streetcar systems mostly with right-of-way
Type C. The director of planning for the Toronto Tran-
sit Commission during the early 1970s, Robert Topp at
that time told one of the authors of this entry that the
TTC retained streetcars on heavily patronized routes
because in that environment PCCs cost less to operate
and provided faster service than buses. Philadelphia
also had several streetcar lines still running almost
entirely in mixed traffic, but the transit system there
converted those to bus operation during the 1980s
and 1990s. San Francisco operated one streetcar line
primarily in mixed traffic.
The General Motors Conspiracy Explanation for
Streetcar Decline
Whether streetcar loss was inevitable is hotly debated.
This history shows that the traction industry ceased
influencing the growth pattern of US cities between
roughly 1915 and 1925 as the auto became the primary
mode of urban transportation. The traction industry
became unprofitable during this period, and it began
losing passengers at an increasing rate as the 1920s
progressed. As a consequence, the industry had diffi-
culty financing track and car renewal. The Great
Depression accelerated patronage and financing diffi-
culties. Congress further compounded the financing
crisis during the mid-1930s by requiring those private
a source of cross-subsidy from utility customers to
transit users. This action deprived the transit industry
of yet another source of capital for renewing itself.
All of these developments, in the context of improved
bus designs, are enough to explain the near disappear-
ance of streetcars in Canada and the USA by 1970.
Despite the weight of such evidence, popular Amer-
ican culture views the elimination of streetcars as the
result of a conspiracy. The role of General Motors in
particular in financing conversions of transit services
from streetcars to diesel buses formed the basis of
Bradford Snell’s widely cited 1974 brief [16] arguing
that GM orchestrated the decline of rail transit in
America to increase the size of a stagnant auto market.
Snell was a staff member of the Senate Judiciary Com-
mittee. Research by some academics support the Snell
thesis [17, 18], and it has entered the popular culture in
outlets such as a popular Spielberg movie Who Framed
Roger Rabbit (1988) and a widely aired Public Broad-
casting System documentary Taken for a Ride (1996).
Informed rebuttals to the Snell thesis, such as Adler’s
[19] analysis showing that politics between suburban
and downtown real estate interests had a lot to do with
transit technology decisions, Bianco’s critique [20],
Post’s [8] and Black’s [21] analyses have done little to
dispel the GM conspiracy theory of streetcar
elimination.
It is a fact that General Motors and other automo-
tive interests financed a holding company in the 1930s,
called National City Lines, to purchase transit operat-
ing companies around the country. The purpose was to
scrap streetcar systems and replace them with buses
built by the General Motors subsidiary, Yellow Coach.
From then into the 1950s National City Lines
purchased streetcar and bus transit systems, and it
continued to operate many of its holdings into the
1960s. NCL eliminated streetcar and electric trolley
bus service on most but not all of the systems that it
purchased. It also outfitted its systems with new Yellow

Coach buses [8, 21]. National City Lines pared man-
agement of the companies that it took over to the bone
but provided staff management services for its operat-
ing properties. It also provided financing for new buses
and garages, as well. During this period, independent
transit companies, because of their unprofitability, had
great difficulty financing new streetcar purchases and
track and electrical system renewals. National City
Lines supporters point to these facts and state that
NCL injected desperately needed capital into the transit
industry, improved the efficiency of its management,
and kept it going in private hands for another couple of
decades [15].
In this debate, the authors’ side with Post [8] and
Black [21] who view, National City Lines as a symptom
rather than a cause of streetcar decline. Many private
streetcar companies in the USA were not taken over by
National City Lines, but they replaced their streetcars
with buses despite that fact. In some US cities, such as
Los Angeles and Pittsburgh, public agencies took over
private rail transit lines (including some from National
City Lines) and scrapped their newly acquired rail
systems in favor of new buses. It also is true that most
Canadian transit systems were municipally owned, but
save for Toronto, all converted from streetcar to bus or
electric trolley bus. Most cities in France and England
eliminated streetcars just before and after World War II,
and most cities in Spain did so in the 1950s and 1960s.
None of these conversions were the result of US auto-
mobile interests.
In the authors’ opinion, the US national interest
would have been served by modernizing parts of many
US streetcar systems at the end of the 1920s by segre-
gating them from auto traffic, but doing so would
have required municipalization and capital infusion
from higher levels of government. Neither municipal-
ization nor capital infusion from higher levels of
government were even remotely in the cards at that
time. Even if they had been, given Canada’s experience,
and experience with public authorities in Los Angeles
and Pittsburgh, it is uncertain whether streetcars
would have been modernized in the USA. It is safe
to say, however, that in the absence of such govern-
ment takeover and capital infusion, the American
streetcar industry could not have maintained itself in
private hands because of its unprofitability and inabil-
ity to raise capital. If National City Lines had not
Light Rail Transit, Systemic Viability 715
happened, some other combination of liquidators
probably would have.
The American and Canadian PCC streetcar operat-
ing in motor vehicle traffic almost disappeared because
it offered insufficient advantages over the diesel bus to
compensate for the expense of its infrastructure,
including track and electrical systems. As Post points
out, its strongest suit was quietness and absence of
fumes. Operationally, it was marginally faster than
buses in heavy patronage conditions, at least on roads
where left turns were banned, as in some intersections
in Toronto (Fig. 2). According to Toronto officials,
PCCs carried passengers at less operating cost than
buses under the heavy patronage conditions of that
city, but other studies showed that it carried passengers
at greater operating expense than diesel buses, perhaps
because patronage levels were lower in the cases studied
[21]. In any event, if there was a case to be made for the
PCC streetcar operating in mixed motor vehicle traffic,
it was not a strong enough one, and that primarily is
why it almost disappeared in North America.
The Light Rail Revolution
Light rail resulted by tweaking the streetcar concept to
reduce its drawbacks while enhancing its advantages.
The Canadian and American transit industry itself
initiated a few of the tweaks, but the industry initiatives
came too late in the decline of the Canadian and
American streetcar industry to change the trajectory
of decline. The progressive changes that did change
trajectory came from Europe and crossed the Atlantic
in response to the social protest movement in Canada
and the USA that began in the 1960s. Many in the
American transit industry, who came directly or indi-
rectly from the National City Lines mold, resisted the
European ideas but ultimately some were forced to
accept them by outside pressure.
In the USA, most of the few streetcar lines that
lasted into the 1970s possessed significant elements of
right-of-way Types B and A that gave them speed
advantages over buses. The speed advantage is why
they remained. Five streetcar lines remained in
San Francisco; four of them used long tunnels (one
2.5 miles long, the other 1 mile long) that took the
lines from the edge of the central business area to sub-
urbs that were cut off by a steep range of hills. Boston
716 Light Rail Transit, Systemic Viability
had a central streetcar subway that ran outward from
its business center, and several streetcar lines fanned
out from it like fingers on a hand. The fingers surfaced
on ramps, and then ran through inner suburbs in
mixed traffic or on reservations in the center of boule-
vards. Boston ran its PCC cars in one or two car trains,
with drivers in both cars collecting fares. In addition to
three or four remaining traditional streetcars that ran
almost entirely on narrow streets, Philadelphia still
operated remnants of three distinct streetcar systems.
It had a similar configuration of subway-surface street-
car lines to that of Boston. Philadelphia also continued
to operate three or four traditional long streetcar lines
running on narrow streets. Finally, it operated two
suburban streetcar lines that fanned out from the end
of a rapid transit line into the suburbs largely on rights-
of-way Type B. Pittsburgh still had one set of streetcar
lines that ran south from its business area, which the
lines exited through a long tunnel under a range of hills
to get to the southern suburbs. All that remained of
streetcars in New Jersey was a short streetcar subway in
Newark in which individual PCC cars shuttled back and
forth between several stations, crossing but not operat-
ing in city streets. In Cleveland two streetcar lines ran on
reservations in the centers of boulevards in the commu-
nity of Shaker Heights. The community’s developers in
1920 placed expensive homes so that residents had short,
pleasant walks to streetcar stops. The two streetcar lines
came together into one and then turned onto a railroad
corridor that the streetcar tracks followed with Type
A right-of-way to a terminal on the edge of the Cleve-
land CBD, 6 or 7 miles away. Short trains of PCC cars
operated on this commuter line at fast speeds.
Some transit advocates argued that rather than
representing the dying gasp of an old technology, the
surviving streetcar operations with their rights-of-way
Types B and A offered a model of a new mode of transit
needed in American cities. Bill Adams, an employee of
the Boston redevelopment authority and later the
Boston transit system wrote a thoughtful article in
1965 describing how the Boston streetcars could be
altered to improve their speed and productivity. He
was not aware of the light rail transformation then
happening in Germany, but what he advocated was
very much like what was happening there [22]. Boston
did make a much noticed improvement to its streetcar
system in 1959 by punching an additional ramp out of
its subway and connecting it to a commuter rail line
that just had been abandoned, and which then was
converted to streetcar operation. The Boston transit
system rebuilt the commuter tracks so they were suit-
able for streetcars, electrified them, and then it began
operating frequent service with two- and three-car
PCC trains from the outer terminal at Riverside to
central Boston via the streetcar subway [8]. An atten-
dant in each car collected fares. Patronage settled down
to about 25,000 per day, much greater than what the
commuter trains attracted, and Adams and others felt
the Riverside line could be a national model. This
model led to no emulation, however, except in 1964
for a short shuttle line in the suburbs of Chicago (on
right-of-way Type A, part of an abandoned interurban
line) called the Skokie Swift [8].
The impetus for the light rail revolution was the
social protest movement of the 1960s and the availabil-
ity of capital from higher levels of government [5, 6].
Continuously falling demand for transit, particularly
for bus service, and rising costs forced the US transit
industry into an unprofitable position. Between the late
1960s and early 1970s, almost all public transportation
in the USA was municipalized with federal and in some
cases, state financial assistance. The passage of federal
transit legislation in 1970 also made capital available
for building new rail systems [11, 23]. At the same time
citizen movements, such as those dedicated to preserv-
ing neighborhoods by stopping inner city freeways and
others to reducing air pollution, came together in some
metropolitan areas with the idea that better public
transportation systems could help with urban preser-
vation and revitalization [24]. Rail service was clearly
desirable, but there was little in the way of immediate
domestic examples that seemed relevant to the needs.
It was apparent from transit patronage trends (Fig. 3)
that bus and streetcar service was in decline, while rail
rapid transit was holding its own against the automo-
bile. Could surface rail systems be tweaked to perform
like rapid transit while costing much less to build? If the
answer was “yes,” there was hope that introduction of
such surface rail lines could reverse the declining for-
tunes of transit and perhaps contribute to renewal of
central cities. Thus was born the light rail movement.
Its model was not the American streetcar, even the PCC
car, but a new concept coming out of northern Euro-
pean cities.

Public transit in most German cities was centered
on streetcar technology before and during World
War II. As in the USA and Canada, German streetcars
ran on tracks located in the middle of streets. The
typical German streetcar was a much shorter vehicle
than its North American counterpart, however, and
German transit providers typically ran them in two-
and three-car trains. The first car was a motor car,
staffed with a motorman and a conductor. The motor
car pulled one or two small trailers, each with a con-
ductor, so the typical German streetcar train of three
diminutive cars had a staff of four. The passenger
(based on the capacity of the cars)-to-staff ratio was
45:1 [25].
When faced with rebuilding their war-ravaged cities
after World War II, German transit planners contem-
plated a future urban world dominated by the automo-
bile [26]. Should their streetcar systems be rebuilt?
Modern buses promised passenger-to-staff ratios of
60:1 for a regular bus and 75:1 for articulated buses,
and German transit systems probably would have made
the transition to buses at that time had they been able
to obtain buses. They were unable to obtain buses,
however, and so they put their streetcar systems back
together as best as they could. Rapidly rising wages,
however, forced German transit operators to seek ways
of improving streetcar productivity. In the mid-1950s,
they began obtaining larger vehicles, similar in size to
PCCs, and then they experimented with articulating
two of the large vehicles together into a long-bendable
vehicle. The experiments worked. Transit operators
improved productivity further by running large trailers
behind the articulated cars. They added third sections
to the articulated cars. They introduced proof-of-
payment (POP) fare systems. POP fare systems
required passengers who did not have weekly or
monthly passes to figure out what fares they owed
and purchase tickets from simple vending machines
or tobacco kiosks before boarding vehicles. No staff
would inspect their fare upon boarding, but roving
bands of inspectors could ask them to produce proof
of payment at any time during their journeys, and if
they failed to produce adequate proof, they could be
fined. This system added teams of roving inspectors to
the payroll, but it eventually eliminated all conductors
on all cars. Passenger-to-staff ratios rose to over 200
[25]. The increase in streetcar productivity was so large
Light Rail Transit, Systemic Viability 717
that many transport undertakings decided to retain
streetcars but adapt them to conditions where it was
expected that most of the public would use automo-
biles in the future [26]. A British publication Modern
Tramway reported regularly on the transition of the
streetcar systems in many northern European cities
into what later became known in North America as
light rail.
Figure 2 offers a vantage point for understanding
what the German’s were trying to do with their street-
car technology makeover, as evidenced from the pages
of Modern Tramway. Figure 2 shows a two-car PCC
train on the Queen Street line in central Toronto that
was given some priority over autos. Both the operation
of a second car as well as auto priority were significant
improvements over traditional North American street-
car practice and undoubtedly contributed to the higher
performance of the Toronto streetcars in handling high
passenger volumes compared to buses. Would it not be
possible to take such modifications further? Could
three-car trains be operated? Could drivers on the
second and third cars be eliminated to improve pro-
ductivity? Could the fare collection system be changed
so that passengers could board at all doors on the train,
without being impeded by an employee collecting fares
and transfer slips? Could motorists be kept off the
tracks? Could all left turn conflicts be avoided? Could
the service be speeded up by stretching out the distance
between stations, and could the stations be designed to
be safe from auto traffic and integrated with pedestrian
systems, with easy transfers between bus and rail?
Could the entire bus route structure in the region be
changed to dump the heaviest passenger loads onto
streetcar routes experiencing such modification, taking
advantage of their speed and high passenger carrying
capacity, leaving local diesel buses to handle lighter
loads? In short, rather than having the streetcar act
like a slow local bus on rails, could it act more like
a quiet fast subway train operating on the surface
(perhaps dipping into short subways in the heart of
the CBD), becoming the backbone to a regional transit
system? The answer to all of these questions was, and is,
“Yes,” and German transit thinkers (as well as transit
thinkers in Switzerland, the Netherlands, Belgium, and
Sweden) developed technology and operating proce-
dures for doing it. Getting transit systems in the USA
and Canada to adopt and implement such technology
718 Light Rail Transit, Systemic Viability
and procedures was what the light rail revolution was
all about.
In a major study that took place about 1959, Frank-
furt am Main in West Germany compared different
strategies for region-wide public transport improve-
ments. One was based on building a system of subway
lines fed by buses. The other was a system of light rail
trains running in short subways in the center of the city
and then using Type B rights-of-way everywhere else.
Frankfurt decided in 1960 to follow the second path
[27, 28]. Figures 4–7 offer several glimpses of the evo-
lution of the Frankfurt streetcar system to a light rail
network. Similar modifications were happening to
streetcar systems in cities throughout northern Europe
at this time [29].
The German concept of light rail entered North
America through different channels. Modern Tramways
was an important conduit, prompting at least two US
consultants to visit Germany to see for themselves and
then write about their observations in Traffic Quarterly.
In 1962, Dean Quinby, a consultant working on the
BART project in the Bay Area, captured the essence of
the idea, which was more than just separating streetcars
from autos. It was the idea of creating surface, or light
Light Rail Transit, Systemic Viability. Figure 4
Streetcar tracks being relocated from mixed traffic in center of road to side of road in right-of-way Type B. Frankfurt, 1966.
Author’s photo
infrastructure rapid transit that could be afforded by
midsized cities. Door design and efficient fare collec-
tion design were important parts of the concept [30].
Eight years later, Stuart Taylor captured the momen-
tum of the idea that was spreading across northern
Europe and again argued that the idea had great appli-
cability to the USA [31].
Other readers of Modern Tramways included Sena-
tor James R. Mills, president pro tempore of the
California State Senate, who brought light rail to
San Diego in 1981, the first application in the USA.
Mills’ most vehement opponent in introducing light rail
was San Diego Transit’s general manager, who also was
president of the American Public Transit Association.
San Diego’s MPO, the Comprehensive Planning Orga-
nization, who sought a Washington Metro-like regional
rapid transit system, also opposed light rail, but both
San Diego Transit and CPO lost the battle [32].
Readers of Modern Tramways also included Tom
Parkinson, a British electrical engineer, who introduced
the self service, barrier free fare (proof-of-payment or
POP) fare system, so critical to the success of light rail,
to North America. Parkinson installed POP on Seabus
in Vancouver (1978), then on the new light rail line in

Light Rail Transit, Systemic Viability. Figure 5
The result. Frankfurt right-of-way Type B, 1966. Author’s photo
Light Rail Transit, Systemic Viability. Figure 6
Light rail station being built in street right-of-way, Frankfurt 1968. Author’s photo
Edmonton (1978), and then on the San Diego Trolley
(1981). With the proof-of-payment fare system, pas-
sengers bought passes allowing from 90 min of riding
to a month of riding at vending machines on platforms
or at retail outlets. Passengers were required while on
Light Rail Transit, Systemic Viability 719
board trains to have in their possession a valid ticket or
pass, inspected at random by roving teams of fare
inspectors. This system eliminated the need for drivers
to inspect fares and allowed passengers to board
through all doors, making it possible for several
720 Light Rail Transit, Systemic Viability
Light Rail Transit, Systemic Viability. Figure 7
Delivery in Frankfurt of new light rail cars, four double doors to a side. Manufacturer: DueWag; Type: U2. 1968. A decade
to a decade and a half later, this car became the de facto standard light rail car in North America. Author’s photo
hundred passengers to board in a matter of seconds
[32]. Parkinson, with Walter Quintin of Bechtel, also
wrote the car specs for the first San Diego line. Only six
pages long, the spec was a stunning contrast to the
phone-book sized spec that the Urban Mass Transit
Administration issued for the flawed standard light
rail vehicle, which ultimately was ordered only by San
Francisco and Boston. The Parkinson/Quintin spec led
to San Diego selecting the DueWag U2 car (Fig. 7) [32].
The U2 was a rugged, reliable and easily maintained car
originally introduced in Frankfurt in 1968 and already
selected by Edmonton and Calgary for their light rail
lines. In addition to its use in San Diego, updated ver-
sions of the U2 car would be used later by Sacramento,
Pittsburgh, and other operators, and it became the de
facto standard light rail vehicle in North America [8].
The light rail idea also was brought into the USA by
academics and high government officials who lived in
Europe during this period and then moved to or back
to the USA. Such persons included Professors Vukan
Vuchic of the University of Pennsylvania, who
published a summary of European light rail practice
on behalf of the new Urban Mass Transit Administra-
tion. Another notable member of this initial group was
Professor John Bakker, a civil engineering professor at
the University of Alberta who worked closely with Don
MacDonald, the Superintendent of the Edmonton
Transit System to define a short rail system around
which a timed transfer bus system (based on ideas
from the Netherlands) could be structured. Edmonton
would be the first modern light rail adoption in North
America. The group included Bob Burco, who headed
the Oregon Department of Transportation when light
rail was adopted in Portland, and Ken Orski, who
as Deputy Administrator of UMTA wrote a policy
directing MPOs to consider light rail among alterna-
tives for transportation improvement in their long-
range transportation plans. Vuchic, Stuart Taylor, and
Orski worked together to establish what became known
as the Light Rail Subcommittee of the Transportation
Research Board’s Rail Transit Committee in 1974.
Shortly after its formation, the subcommittee branded
the German concept of limited tramways as light rail,
and held a national conference on light rail in June 1975
to popularize the idea among politicians, regional plan-
ning officials, and transit managers in cities in the USA
and Canada. Leadership of the American Public Transit
Association joined as a cosponsor of the conference at
the last minute, though many in leadership positions
then were opposed to the idea of light rail [5, 6].

A characteristic of German light rail that differed
from American streetcars was the German practice of
imbedding light rail lines into regional transit net-
works. The coordination of light rail and bus routes,
as it came to be practiced in North America, was done
to accomplish two objectives. One was to take advan-
tage of the high passenger-to-staff ratios of light rail
lines to improve overall transit productivity. The idea
here was to introduce light rail lines into corridors
served by many bus lines, particularly during rush
hours. Rather than operating all of the bus routes to
the CBD on arterial roads, where there were large
amounts of redundant bus service, the transit operator
would terminate bus routes at suburban light rail sta-
tions. High-capacity light rail trains then could take
passengers to the CBD, and perhaps to other important
destinations that bus passengers previously could not
reach. Staff and operating expense would be saved by
discontinuing the duplicative parts of the bus routes
running from the suburban stations to the CBD.
The other objective of bus and light rail service
coordination was the creation of multidestination tran-
sit networks that enabled those passengers without
access to autos to reach many metropolitan area desti-
nations in addition to the CBD. Suburban light rail
stations where many bus routes converged were excel-
lent places for passengers to transfer not only between
buses and trains, but also between trains and buses (so
that train passengers could reach nearby destinations),
and between one suburban bus line and another.
Ideas of bus/rail coordination to achieve the dual
objectives of greater productivity and multidestination
travel were new to transit systems in the USA. US
transit planners typically viewed the CBD as the only
important destination for transit to serve. They
believed that to attract as many passengers as possible
to the CBD, it was important not to force passengers to
transfer from buses to trains [39]. American (but not
necessarily Canadian) practice tended to view each
transit line as an independent entity that connected
suburban neighborhoods with the CBD.
The idea of multidestination transit came into
Canada and the USA from several directions. John
Bakker and Don MacDonald of Edmonton saw light
rail as the core of a restructured city-wide bus network
much as they saw happening in Europe. Brian Sullivan
of Edmonton, Calgary, and Vancouver also realized the
Light Rail Transit, Systemic Viability 721
importance of bus rail service integration for both
reducing costs and increasing patronage. Sullivan pas-
sionately argued for the network approach and
attempted to import the German idea of the Verkehrs-
verbund, or traffic federation, to North America. The
idea of the Verkehrsverbund, pioneered in Hamburg,
was that a central planning and marketing agency in an
urban region would design a regional transit network
to allow easy travel throughout the region by transfers
between lines and modes. The traffic union would
establish timetables, collect fare revenue, and provide
information to the public for all transit services in the
urban region. However, it would contract out the con-
struction and operation of different parts of the net-
work to many different entities. From the public’s
perspective, the entire regional network would seem
to be one coordinated entity. There would be only
one regional transit map, one source of timetables,
and one fare system. The public would be able to
transfer freely between all parts of the system, connec-
tions would be timed, and transfer stations would be
attractive and functional. However, different operators
would be able to keep close managerial control over the
independent parts of the network, insuring timely
operation and low operating costs. Sullivan, Bakker,
and MacDonald had considerable success in spreading
at least parts of these ideas either directly or
through coworkers (such as Tom Matoff and Greg
Thompson) to Vancouver, San Diego, Edmonton,
Calgary, Portland, Sacramento, and Denver. Under the
leadership of Tom Larwin, MTDB’s general manager
from 1978 to 2003, San Diego was the one region in
the USA and Canada that came closest to implementing
the Vehrkehrsverbund, where the concept was the orga-
nizing principle for regional transit in the southern part
of San Diego County from 1979 to 2003 [32]. Vancou-
ver also adopted many aspects of the traffic union
concept through British Columbia’s Bureau of Transit
Services. Sullivan, who was deputy director of the prov-
ince’s Bureau of Transit Services from 1973 to 1980, led
a team in preparing a transit plan for the region calling
for the region’s extensive local bus and trolley bus
system to be reorganized around a core of regional
light rail lines and regional buses, called Fast Buses.
The plan largely has been implemented, though the
region obtained the funding to build its rail lines as
entirely Type A rapid transit lines rather than as light
722 Light Rail Transit, Systemic Viability
rail. Other regions, such as Denver, also adopted
the concept of an integrated regional network but
attempt to manage the concept through large public
monopolies.
Also important to the transference of the light rail
idea from northern Europe to North America was the
mass travel of Canadian and US youth to Europe dur-
ing this period [24]. By and large, these travelers were
the offspring of middle class families who grew up in
the suburbs using automobiles. In Europe, however,
they rode between great urban centers on fast, com-
fortable, frequent trains on first-class Eurail passes
and used the modernizing tram systems in cities
throughout northern Europe. Many came back with
the realization that there was a viable urban alternative
to suburban, auto-based neighborhoods that they
experienced as kids. Such individuals, passionate
in their beliefs, entered transportation bureaucracies
in Edmonton, Vancouver, San Diego, Portland, San
Francisco, Sacramento, among other cities in the late
1960s and 1970s, and worked with other such individ-
uals in citizen activist groups to get the light rail idea
accepted by top decision makers [24].
Light Rail Adoption in North America
Table 1 summarizes the history of light rail adoption
in Canada and the USA as well as the streetcar systems
that survived to 1970 and then were modernized.
The table shows the year that a previously all-bus
urban area opened its first light rail line and the miles
of line that it originally opened. The table also indicates
growth, if any, from the original length, and the pas-
sengers boarding the light rail lines each weekday in
a recent year (generally 2006). Some systems cater to
passengers who travel only a short distance; others to
passengers who travel a long distance. Average distance
traveled for each city is shown, as is the speed of service.
The next to last column shows the current (usually
2006) traffic density on each mile of line. Density is
measured by the average number of passengers passing
each mile of line on a typical weekday. The primary
purpose of showing the relative traffic intensity of the
various systems is to offer a first cut to judging their
relative performance.
This section summarizes how the first group of
North American bus-only cities to adopt light rail
decided how to do so. There are seven such cities. The
section shows a great deal of commonalty of objectives
among five of the seven pioneer cities, and understand-
ing the commonality assists in evaluating the success of
the light rail movement in the section of this entry that
follows. This section stops at the first seven because
there is a clear gap between the seventh and eighth city
in terms of time, after which the rate of adoption
increased rapidly.
Edmonton was first. During the 1950s, 1960s, and
1970s, a city department, the Edmonton Transit System
operated the area’s bus and trolley bus system under the
direction of an electrical engineer, Don MacDonald.
MacDonald began his career with Edmonton
Transit by replacing the system’s streetcars with trolley
buses just after World War II, when he was made
Superintendant. During the 1960s MacDonald collab-
orated with a Dutch university transit engineering pro-
fessor, John Bakker, in forming strategy for system
change. MacDonald and Bakker developed a plan for
a short Type A rail transit line using railroad rights-of-
way modeled roughly after a similar line that opened in
Cleveland in1955. In contrast to Cleveland where the
rail investment resulted in no significant changes to
the city’s bus system, MacDonald and Bakker intended
to restructure the Edmonton bus and trolley bus net-
work around the rail line, eliminating duplication of
trunk bus routes running to the CBD, and making
suburban destinations more reachable. They could
not obtain funding or council approval for the project,
however. This changed about 1970 when two univer-
sity-affiliated activist groups organized to fight city
hall’s effort to bulldoze inner city neighborhoods,
nature corridors, and parks to make way for an inner
city freeway ring. Their 1971 book addressed to
Edmonton’s city council, “Immorality of the Motor
Car,” persuaded City Council to scrap the previously
secret freeway plan for the city. A year later, the activists
presented City Council with a second book, “Light
Rapid Transit, the (Immediate) Answer for Edmon-
ton.” This book, with a line drawing of the 1968 Frank-
furt U2 car on the cover and heavily footnoted with
references from Modern Tramway and Traffic Quarterly,
presented the light rail idea. The principal authors had
been acquainted with light rail on European trips.
“Light Rapid Transit . . . ,” then described how the
idea could be applied to Edmonton, including
Light Rail Transit, Systemic Viability. Table 1 Light rail adoption in Canada and the USA, 1978–2010
New light rail systems
Opening Recent one way Recent light rail average Distance traveled Average daily density of Scheduled
Opening one way miles, 2006 unless weekday boardings, 2006 by average passengers carried along speed,
Urban area year miles otherwise noted unless otherwise noted passenger miles each mile of route mph Notes
a
Edmonton 1978 4.3 15.3 74,000 4.0* 19,346 19.0
b
Calgary 1981 6.1 28.1 248,200 4.0* 35,331 18.0
San Diego 1981 15.9 51.7 100,111 6.2 12,006 19.0
Buffalo 1984 1.2 6.2 19,398 2.5 7,822 12.0
Portland- 1986 15.1 43.8 97,000 5.2 11,516 19.0
MAX
Sacramento 1987 18.3 37.8 50,088 5.2 6,890 21.0
c
San Jose 1987 6.6 41.6 34,837 5.1 4,271 16.7
Los Angeles 1990 20.7 54.7 136,000 7.2 17,901 21.0
Baltimore 1992 22.5 30 18,964 5.7 3,603 22.0
St. Louis 1994 15.5 46 44,297 7.2 6,933 27.0
d
Denver 1994 5.3 34.7 68,784 5.2 10,308 19.0
Dallas 1996 10.3 44 62,007 7.4 10,428 21.0
Salt Lake 1999 15 19 51,952 5.7 15,586 25.0
e
Jersey City- 2000 7.6 16.6 25,840 3.1 4,826 15.2
Hudson
Bergen
f
Minneapolis 2004 7.8 12 34,072 5.5 15,616 14.6
Camden- 2004 34.5 34.5 7,350 4.5 959 na
Trenton
River Line
Houston 2004 7.5 7.5 38,113 2.4 12,196 14.0
Light Rail Transit, Systemic Viability
723
 724

Light Rail Transit, Systemic Viability. Table 1 (Continued)

Light Rail Transit, Systemic Viability

New light rail systems

Opening Recent one way Recent light rail average Distance traveled Average daily density of Scheduled
Opening one way miles, 2006 unless weekday boardings, 2006 by average passengers carried along speed,
Urban area year miles otherwise noted unless otherwise noted passenger miles each mile of route mph Notes
g
Charlotte, 2007 9.6 9.6 16,000 4.0* 6,667 15.7
NC
h
Oceanside- 2008 22 22 6,560 10.4 3,101 22.0
Escondido
i
Phoenix 2009 19.8 19.8 37,000 6.0* 11,212 18.3
j
Seattle 2009 15.6 15.6 14,399 6.0* 5,538 na
Streetcar systems that survived to 1970 and were modernized subsequently
k
Toronto na 48.9 162,500 2.7* 8,972 9.0
Boston 26.9 26.9 245,236 2.7 24,615 14.0
Cleveland 10.5 12.7 13,379 5.8 6,110 19.0
New Orleans 6.5 12.9 24,000 2.0 3,721 9.0
2004
Newark 4.3 6.1 18,664 2.5 7,649 na
l
Philadelphia 22.3 22.3 80,994 2.7 9,806 11.0
subway-
surface
m
Philadelphia 11.8 11.8 7,700 3.6 2,349 16.0
Media/
Sharon Hills
 n
Philadelphia- 54.5 8.5 19,992 2.7* 6,350 8.3
surface
streetcars
Pittsburgh na 23.3 25,884 4.4 4,888 16.0
San 24.2 31.7 140,426 2.6 11,518 11.0
Francisco
o
Portland 2001 2.5 4.1 7,800 2.0* 3,805 10.0
streetcar
*
Average trip length not reported; authors estimated distance based on similar systems
a
Reflects April 2010 extension; speeds from [35]; other from: http://www.edmonton.ca/transportation/ets/lrt.aspx, downloaded on 29 June 2010
b
Average for 2006; speeds from [35]; other from: http://www.calgarytransit.com/html/technical_information.html, downloaded on 29 June 2010
c,d
2008 figure from source [36], transformed to weekday figure at rate of 300 days per year
e
2006 figures from source [33]
f
2008 figure from source [36], transformed to weekday figure at rate 300 days per year
g
Oct/Nov 2008 average. Taken from: http://www.norfolk.com/2008/12/charlotte-light-rail-ridership-down-november, downloaded on 29 June 2010
h
Figures are from 2008, taken from: http://www.ridecheckplusburst.com/development/Default.aspx?q=28&pdf=08_NS_WK_CompStatSysWide.pdf, downloaded on 29 June 2010
i
April 2010 weekday average from press report, found at: http://image.examiner.com/x-11410-Fresno-Transportation-Examiner~y2009m5d28-Valley-METRO-light-rail-reaching-new-
heights
j
November 2009 weekday average, http://seattletimes.nwsource.com/ABPub/2010/01/11/2010767928.pdf, downloaded on 29 June 2010
k
Newspaper story claiming annual streetcar ridership of 52 million in 2006, http://transit.toronto.on.ca/archives/data/200707150430.shtml, downloaded on 30 June 2010
l,m,n
From SEPTA Service Planning Department Annual Service Plan, June 2010, http://www.septa.org/reports/pdf/asp11-proposed.pdf
o
This is a new streetcar line but is included because its stats were mixed with those of Portland MAX, and authors wanted to preserve the stats after extracting them
Sources: References [33–36] unless otherwise noted.
Light Rail Transit, Systemic Viability
725
726 Light Rail Transit, Systemic Viability
diagrams showing how bus routes could be
restructured around suburban light rail stations
(Fig. 8). Oil royalties then flowing into provincial cof-
fers led to a provincial light rail construction fund
available to both Edmonton and Calgary, and Edmon-
ton City Council approved light rail development in
1973. Edmonton opened its first line with turnstiles but
within a year replaced them with the proof-of-payment
(POP) fare collection system after its successful debut
for Vancouver’s Seabus. The POP fare collection system
made it easier for passengers to transfer between buses
and trains, and station designs generally required short
walks between modes. Edmonton made the mistake of
putting the CBD part of its light rail line in a subway,
however, which greatly increased light rail construction
cost, thus slowing light rail expansion. Not until 2003
did the line finally break to the surface on its southern
face. Since then, it has expanded to the south rapidly,
and its patronage more than doubled since 2003 as
a consequence [5, 6].
Edmonton’s light rail system still lags the light rail
impact of Calgary, however. Under the direction of
William Kuyt, a Belgian-born traffic engineer who
already was familiar with light rail, Calgary put its
light rail line comingled with buses on a surface mall
through the CBD. With the same amount of money
that Edmonton received from the provincial govern-
ment, Calgary subsequently expanded more rapidly
and farther in several directions in the manner
described in, “Light Rapid Transit. . .”. The Calgary
light rail line directly serves most of the metropolitan
area’s major destinations, and Kuyt restructured its bus
system around its three lines in the manner projected
for Edmonton in Fig. 8.
San Diego was the first US urban area to open light
rail, and it followed the example of Edmonton (though
it avoided subway construction). San Diego’s light rail
was championed by the President Pro Tempore of the
California State Senate, James R. Mills of San Diego.
Opponents included the city-owned and federally
financed bus system and the federally sanctioned met-
ropolitan planning organization, called the Compre-
hensive Planning Organization. The bus system’s
general manager maintained that buses could do any-
thing a rail system could do better and at lower cost.
The MPO viewed light rail as interfering with its plans
for a Washington Metro-like system. Arguing that the
bus system already exhausted the region’s financial
resources for transit with unproductive bus service
and that the MPO’s plans were financially and politi-
cally infeasible, Senator Mills put forward the German
concept of light rail as a feasible alternative to both.
An avid reader of Modern Tramways, he persuaded the
San Diego County Department of Transportation to
study the feasibility of the German light rail concept in
San Diego. The department’s head, Rudy Massman
hired an engineer and a planner from the Edmonton
Transit System during a window when it looked like
political approval for the Edmonton system would not
happen. Massmann married the Edmonton transit
people with San Diego County roadway department
project planning and financing staff to prepare the
study. The study found that a regional network of
light rail trunk lines, around which bus service in the
county was restructured, was a viable transit concept
for the highly decentralized and auto-centric San Diego
region. On the basis of the report, the Senator carried
legislation that stripped the city bus system and the
MPO of much of their powers, created a new transit
planning, programming, and construction organiza-
tion (the San Diego Metropolitan Transit Development
Board, or MTDB), and gave it access to state and local
funding. The legislation creating MTDB envisioned
light rail becoming the backbone of a regional transit
system, and it gave MTDB the power to make that
happen. The first MTDB general manager, Robert Nel-
son sold the MTDB to business leaders in the region by
arguing that light rail was the only way to improve
transit productivity in the region by replacing a large
number of unproductive bus miles with productive rail
miles. Earlier as Deputy General Manager for the
Metropolitan Atlanta Transit Authority, Nelson made
similar arguments there. With such arguments, MTDB
gained local support for its plans, and when Mayor
Pete Wilson of San Diego swung on board, the project
moved forward. From the opening day, the San
Diego Trolley used the POP fare collection systems,
and as the LRT network expanded, bus routes have
been restructured around light rail stations. The
Verkehrsverbund form or organization, described ear-
lier, governed the region for over 20 years of light rail
expansion [14, 32].
As Edmonton’s, but unlike San Diego’s, Portland’s
light rail decision grew out of pro-neighborhood and
 Light Rail Transit, Systemic Viability 727

Light Rail Transit, Systemic Viability. Figure 8

Neighborhood activist vision for an integrated bus and light rail network for Edmonton in 1972. Taken from the Practicum
on Community Analysis, “Light rapid transit: the (immediate) answer for Edmonton”

anti-freeway organizations that coalesced in that city
during the early 1970s. The region used Interstate
Transfer Funds freed up by a deleted freeway to finance
part of the light rail line. As in San Diego, the concept of
light rail development in Portland was heavily
influenced by Edmonton and incorporated Canadian
ideas of light rail forming the armature of a res-
tructured regional bus system. The first document
advocating light rail in Portland was a state Public
Utilities Commission examination of the suitability of
using rail rights-of-way in the Portland region for rail
transit or busways. The author of the document pat-
terned it on, “Light Rapid Transit, the Immediate
Answer for Edmonton,” which he cited. Most of the
major players in the decision to adopt light rail in
Portland had been to Europe and seen the concept
728 Light Rail Transit, Systemic Viability
there. Some were readers of Modern Tramways. One of
them was the transit system’s planner, but the manage-
ment of the transit system was otherwise opposed to
light rail. The governor had to reorganize it, and
also change staff at the Oregon Department of Trans-
portation before light rail could move forward.
The reorganized transit system (TriMet) took the lead
in designing the integrated light rail/bus system.
It hired Don MacDonald from Edmonton to manage
Portland’s first light rail construction as well as to plan
for restructuring bus service around proposed light
rail trunk lines [24]. The Canadian approach was
strengthened in Portland by Tom Matoff, who brought
Portland’s first light rail line into operation and
restructured eastern city and suburban bus
lines around it in time for the first day of operation.
Matoff had worked with Brian Sullivan in Vancouver
on the Vancouver regional rail/bus integration concept
and its incorporation into the Greater Vancouver
Regional Plan. Sullivan brought Matoff to Vancouver,
and from Vancouver Matoff moved to San Francisco,
where he was director of planning for the San Francisco
Municipal Railway. In that capacity, Matoff initiated
and secured adoption of a revolutionary plan for
restructuring the San Francisco Municipal Railway’s
bus and trolley bus network around the new light rail
transit subway investment in that city. From there, he
moved to Portland, San Jose rail operations, and later
still to Sacramento.
As did Portland, Sacramento used Interstate Trans-
fer funds to finance part of the construction of its first
light rail lines, which also arose out of an anti-freeway,
neighborhood revitalization movement. Youthful
activists in that movement previously traveled widely
in Europe and read Modern Tramways. Despite the
project’s anti-freeway origins, the California Depart-
ment of Transportation, which had assisted in the San
Diego light rail project, contributed significantly to the
design of the Sacramento proposal. The bus transit
system, run by a regional district, originally was
opposed to light rail. However, Bob Nelson, who was
the San Diego Metropolitan Transit Development
Board’s first general manager, and as such led in the
design and adoption of San Diego’s first light rail line,
was appointed general manager of the Sacramento
system after leaving San Diego. Nelson was not directly
involved in light rail planning in Sacramento, but he
and John Schumann, who was (as director of a special
agency to plan and construct light rail), sold light rail to
the business community there as a means for improv-
ing transit productivity by eliminating bus route dupli-
cation. Nelson previously used this argument in
San Diego (and before that, he used it in Atlanta with
its heavy rail system) with success. After opening
Portland’s light rail system, and a year after
Sacramento’s first light rail line opened, Tom Matoff
became Sacramento Regional Transit’s general man-
ager and insured a coordinated light rail/bus system
there for its first several years. Unfortunately, more
recent extensions have not been supported by bus
restructuring, and the old bus people there have dis-
mantled much of the original coordinated rail/bus
scheduling from the earlier era [14].
The close linkages between development of five of
the pioneer light rail systems result in a commonality of
system design between them. In all five cases, light rail
was envisioned as the backbone of restructured
regional transit systems, and in four of the five cases
this idea has been kept as their light rail systems
expanded. Four of the five cases now experience high
passenger traffic density on their light rail lines, exceed-
ing 10,000 passenger miles per track mile of route.
Light rail in Sacramento does not achieve this intensity
of use, perhaps because that system’s commitment of
bus/rail coordination has waned.
Buffalo and San Jose followed different paths to
light rail development that resulted in less coordination
between their bus systems and light rail lines. Buffalo’s
light rail line was conceived by the region’s transit
operator originally as a heavy rail line using Type
A right-of-way. The transit authority requested federal
and state funding for their project but was turned
down. However, the Urban Mass Transit Administra-
tion was looking for a light rail demonstration project
that it could finance, and it chose Buffalo. To transform
the original proposal from heavy rail to light rail so that
it could obtain federal funding, the transit authority
redesigned the plans so that rail cars would run for
a mile along a transit mall, built on the right-of-way
of the downtown’s major shopping street. For the
remaining 5 miles from the edge of the downtown to
a university campus in the northeastern sector of the
city, the line ran in a subway, stopping at several under-
ground stations. The cost of the subway made it

impossible to extend the line any farther, though
a surface extension farther into the suburbs was in the
original plans. The line did open with the POP fare
systems, and there are unimpeded transfers between
bus and rail. However, it does not appear that there was
a significant restructuring of the bus network around the
light rail investment, even though the same agency in
the Buffalo area runs both the bus and light rail lines.
Local business support focused by a political leader,
Rod Diridon was the momentum behind light rail
development in the San Jose metropolitan area.
Diridon held several elected positions but was particu-
larly active as a Santa Clara County supervisor, where
he organized a county-wide transit system to develop
bus and rail service in the area. With support of the
business community leadership, Diridon obtained
state funding opportunities for transit made available
through the leadership of Senator James Mills of
San Diego, and he also obtained federal funding to
support light rail plans. The San Jose light rail initiative
differed from that in San Diego, however, by not fol-
lowing the transit planning objective of having light rail
forming the core of a regional bus system, which was
one of the principles around which the legislation
creating the MTDB was written. Both light rail and
buses now are operated by the same county transit
authority in the San Jose area, but the two systems are
not well integrated. In many locations, transfers
between modes require long walks in pedestrian-
hostile environments, and there never was a strategic
vision of integrating light rail and bus lines into
a regional system in which the heavy investment in
light rail would replace the trunk parts of bus routes
headed to major destinations. Matoff was brought in to
open the first light rail line in San Jose following his
successful opening of Portland’s first line, but unlike in
Portland he was not charged with establishing
a coordinated bus and light rail network in San Jose,
and there had been no ground work in that direction
before he got there. There is another problem, as well.
Because of a design flaw of the CBD mall that has
proven impervious to correction, the light rail system
is plagued with a 10-mph speed restriction through San
Jose’s CBD, which lengthens travel times between res-
idential districts in the southern part of the county and
employment concentrations, which tend to lie north of
the CBD [14].
Light Rail Transit, Systemic Viability 729
During this period, most of the streetcar lines that
survived to 1970s were modernized. All except New
Orleans and Toronto received fleets of light rail cars.
The Urban Mass Transit Administration wrote specifi-
cations for what it intended to be the standard light rail
vehicle for the United States, and it encouraged Boston
and San Francisco to be the first purchasers of the car.
Boeing-Vertol was selected to build it, but the car
proved to be complicated and unreliable. No other
orders followed. Other systems ordered the U2 car or
cars built by different builders but similar to the U2 car.
Toronto replaced its PCCs with new vehicles of similar
design built in Canada. New Orleans renewed its
old pre-PCC traditional streetcars and expanded the
fleet with replicas. Changes to the physical plants of the
streetcar systems varied widely from city to city. Pitts-
burgh rebuilt its South Hills lines to light rail standards
and placed the CBD part in a new streetcar subway.
Cleveland renewed its Shaker Heights lines, which
already were mostly on Type B right-of-way. San
Francisco placed the CBD portion of its streetcars
into a subway, which connects with the two long tun-
nels extending into the suburbs. The subway was a by-
product of regional rapid transit development. Even
with the streetcar subway running beneath Market
Street, the city’s transit system continues to operate
streetcars on tracks in the center of the street as part
of an historic streetcar line. In 2007, San Francisco’s
transit system opened a new light rail line to the south-
ern part of the city using right-of-way Type B. Boston
and Philadelphia renewed their already existing sub-
way-surface streetcar lines with Type B right-of-way.
Philadelphia made major streetcar cutbacks, however,
by eliminating all of its north–south Type C streetcar
routes. It since restored one of them using rebuilt
PCCs. Route 15 – Girard Avenue intersects both the
Broad Street Subway and Market-Frankford Elevated
rapid transit lines. Boston also abandoned some Type C
streetcar service. New Orleans renewed its traditional
St. Charles Street streetcar operation, and it expanded
its streetcar operation by converting its heaviest bus
route to streetcar operation. Before Hurricane Katrina
temporarily shut the system down in August 2005,
the New Orleans streetcar operation remained slow
and traditional.
How streetcars function as a part of the overall transit
system varies from city to city. Metropolitan Toronto’s
730 Light Rail Transit, Systemic Viability
transit system was one of the few to have been organized
as a grid network of north–south and east–west routes
from the 1920s. Streetcars serve some of the heavy east–
west routes in the grid, and they intersect with north–
south bus, subway, and in some cases, other streetcars at
every major north–south cross street. The transit system
provides many off-street transfer stations that allow pas-
senger to transfer between vehicles, using all doors, with-
out drivers supervising their fare collection [37].
For years, Boston’s streetcars have operated as radial
routes funneling into the CBD-destined streetcar sub-
way, which intersects with three rapid transit subway
lines in the CBD. The streetcar subway then continues
out the northern end of the CBD, where it serves
a major commuter train terminal before crossing the
Charles River and terminating at a streetcar/bus trans-
fer station at Lechmere. Very few bus routes enter
Boston’s CBD, so almost all bus passengers destined
for the CBD transfer to the streetcar subway or the
other three rapid transit lines.
During the early 1980s, under the direction of Muni
Planning Director Tom Matoff, the San Francisco
Municipal Railway extensively restructured bus and
trolley bus routes in the central part of the city to better
serve non-CBD destinations. The transit system
created new crosstown bus routes that intersected
with the light rail subway running to the downtown.
Parts of the bus routes previously running from the
stations to the CBD were discontinued, and passengers
heading to the CBD transferred to trains. The change
was intended to increase productivity and make more
destinations reachable by transit, thus increasing
patronage.
In Pittsburgh, Cleveland, and Philadelphia transit
service remains organized primarily as radial bus and
streetcar (and in the case of Cleveland and Philadel-
phia, rapid transit lines) operating to the CBD. In some
cases, passengers have to pay an additional fare to
transfer between routes, and most passengers do not
transfer. The subway-surface streetcar lines are not the
organizing principle of area-wide route restructuring
but function as just other transit routes heading down-
town. (However, the 15 – Girard is a crosstown line
functioning as part of the North Philadelphia grid
system.)
This entry generally does not consider previously
all-bus cities that built new streetcar lines during the
last two decades. There are several of these. Most of the
new streetcar lines are short urban circulator or tourist-
oriented loops in central city areas. Table 1 does show
one of these streetcars, however, partly because the line
is longer than others and functions as a transit line
connecting several destinations together. It also is
shown, because its statistics had to be disentangled
from Portland’s light rail statistics in the construction
of Table 1.
Analysis of Light Rail Transit Performance
Light rail advocates hoped that their efforts would
increase transit use, improve transit productivity, and
create North American lifestyles that could be indepen-
dent of automobiles. There are no comprehensive stud-
ies that evaluate the extent to which such objectives
have been reached, but there are academic studies that
touch on light rail performance. Some are critical of
light rail investments for costing a lot while achieving
very little, while other studies show that light rail
investments have had favorable impacts on transit sys-
tem performance if the investments were planned to
improve transit productivity. This section examines
light rail performance. It first considers academic crit-
icisms of light rail investments, followed by examina-
tions of how light rail investments have affected overall
transit performance in different metropolitan areas.
The primary conclusion from the various analyses
is that light rail investments that have been accompa-
nied by restructuring of bus systems have increased
transit riding per capita in their service areas, and
they have been more productive and more efficient in
this regard than alternative forms of transit investment.
Light rail investments accompanied by bus network
restructuring have been particularly effective in rapidly
growing parts of the country. Light rail investments not
accompanied by bus route restructuring have been less
effective. New streetcar investments that eschew light
rail productivity enhancements appear to be less effec-
tive, as well. Light rail investments have not, however,
substituted for auto traffic to a large extent.
Criticisms of Rail Investments
Light rail traffic densities shown in Table 1 suggest that
some light rail lines, such as in Salt Lake or Phoenix, are

carrying sizable traffic in sprawling, auto-oriented
regions where heavy rail transit volumes would not be
expected. Examining the patronage consequences of
new light rail transit investments throughout the
USA, Polzin [38] found that although light rail patron-
age initially was disappointing, it was growing consid-
erably as various systems expanded in scope. Light rail
critics counter, however, that traffic density figures are
not an indication of whether light rail investments
increased transit ridership or reduced transit costs.
They argue that without rail most of the same patron-
age would occur on buses at lower cost. Some critics
argue that patronage actually would be higher without
rail. Rail investments and operating expenses, they
argue, starves the bus system of funds, resulting in its
emaciation and loss of patronage. Without the rail
lines, the bus systems would be much healthier and
would have more patronage.
Arguments critical of light rail investments stem
from the 1966 work of Meyer, Kain, and Wohl [39],
which laid out an argument against investing in rail
transit. They argued that public transportation was effec-
tive in only one market, which was connecting suburbs
with CBDs. Unfortunately, CBDs had been in decline for
almost a half century, and census data suggested the
decline would continue into the foreseeable future.
Thus, transit patronage would continue to decline.
They also showed that for the suburb to CBD market,
under almost all conceivable residential densities in the
suburbs and employment densities in the CBD, express
buses on freeways were a better choice than rail transit.
Express buses could provide faster service with a one-seat
ride at lower cost per passenger than rail transit lines. The
latter were more expensive to build and operate, they
argued, and required passengers to transfer from local
buses to rail lines at suburban stations. Even worse, rail
transit lines sometimes required a second transfer from
rail to bus in the CBD. This type of argument has been
used against proposed rail transit investments in numer-
ous cities, and it also is used against restructuring of
bus services around rail transit lines. The transit system
opponents to light rail in San Diego, for example,
appealed to these arguments extensively.
Examining patronage and cost data from recent
heavy and light rail investments in the early 1990s,
Pickrell [40] argued that rail transit supporters
obtained political support for their projects by inflating
Light Rail Transit, Systemic Viability 731
patronage projections and underestimating cost
projections. Kain [41] offered a case study on how
such deception was used in Dallas. O’Toole [42] and
Richmond [43] also provided empirical evidence
showing that the actual costs of rail projects were
greater and patronage less than projected. They also
claimed that rail investments depressed transit patron-
age and increased transit costs compared to doing
nothing or using similar investments for express
buses. Richmond did report that there was one rail
investment that improved transit patronage and
reduced costs: San Diego.
In a forthcoming article, Giuliano argues that tran-
sit investments in the USA fail to achieve their objec-
tives. She argues that the investments are large. In 2006,
for example, all levels of government in the USA spent
$36 billion for transit capital investments and for sub-
sidizing two-thirds of transit operation expenses for
that year. Such large expenditures in that year and in
all prior years leading up to 2006 did little to expand
transit ridership or influence urban form. She further
argues that transit in the USA is useful primarily to the
poor, but that government expenditures in transit do
little to increase transit’s usefulness to the poor. She
includes expenditures on light rail development among
expenditures that fail to achieve objectives or help the
poor [44].
Comparison of Transit Performance in Urban Areas
With and Without Light Rail Investments
The criticisms of rail transit investments suggest that
metropolitan regions investing in rail transit will have
lower overall transit ridership per capita than compa-
rable regions that have bus-only transit systems. While
trains might carry substantial patronage, it is suggested
by critics that the overall train plus bus patronage
would be lower than the all-bus patronage in a region
that did not invest in rail. The criticisms also suggest
that the overall transit productivity, measured in pas-
senger miles per vehicle mile, would be lower in urban
areas with rail compared to similar regions with only
buses. Finally, rail critics imply that the rail plus bus
operating expense for moving each passenger 1 mile in
urban regions that invested in rail would be higher than
the bus-only operating expense for moving each pas-
senger 1 mile in urban regions that did not invest in rail.
732 Light Rail Transit, Systemic Viability
Brown and Thompson [45] tested these assertions
for US metropolitan areas that contained between
1 and 5 million people. There were 45 of such midsized
urban areas. Many light rail new starts were made
within this group of metropolitan areas, and other
metropolitan areas in this group considered light rail
but decided not to implement it, such as Kansas City,
Cincinnati, and Columbus. Brown and Thompson
divided the 45 metropolitan areas into four categories.
The first category contained bus-only metropolitan
areas where the transit systems organized bus routes to
carry passengers to many destinations. Six metropolitan
areas fell into this category (Table 2). The second cate-
gory contained metropolitan areas that operated both
bus and rail routes (light rail in most cases), and the bus
and rail routes were organized to reduce duplication
of bus routes and to carry passengers to many different
destinations. There were 10 metropolitan areas in
this category. The third category contained metropol-
itan areas that operated both bus and rail routes, but
most of the routes were radial in nature, focused on the
CBD. The fourth, and largest category (20 metropoli-
tan areas), contained metropolitan areas that operated
only buses, and the bus routes mostly focused on
the CBD.
Having established the categories, Brown and
Thompson then calculated median transit perfor-
mance in 1984 and 2004 for each of the four categories,
using three performance measures. One performance
measure was annual passenger miles per capita for the
region. This measure indicates the degree to which
the average inhabitant of the region uses transit.
Another performance measures was passenger miles
per vehicle mile, which indicates the average number
of passengers on board a transit vehicle each mile
that it operates. This is a measure of productivity.
The remaining performance measure is total transit
operating expense per passenger mile, expressed in
real dollars that remove inflation. This is a measure of
operating efficiency.
Tables 3–5 show the results. The multidestination
bus plus rail category outperforms the other three
categories by a wide margin. This is the category that
includes cities that invested in light rail and
reconfigured the bus route structure around new rail
lines (San Diego, Portland, Atlanta, Miami, St. Louis as
examples). Most of the rail transit investments were
light rail, though Atlanta’s and Miami’s rail investments
were rapid transit lines. This is the only category to
increase transit patronage per capita (Table 3). It is the
category to have the lowest reduction in transit pro-
ductivity, and three of the light rail/bus systems actually
increased their productivity (Portland, St. Louis, San
Diego). It also is the category to have the lowest
increase in operating costs. By 2004, operating costs
in this category were far lower than in the radial bus-
only category.
The multidestination bus and rail category
performed well despite poor performance of two of
its metropolitan areas: Sacramento and New Orleans.
As noted earlier, Sacramento partially abandoned coor-
dination between buses and trains that characterized
the system in its early years. New Orleans saw the worst
change in performance on all three performance indi-
cators. Slow traditional streetcar lines do not appear to
be good for transit performance.
Urban areas that invested in rail but operated most
of their rail and bus routes to the CBD did not do well.
Such urban areas include Pittsburgh, Cleveland, and
Buffalo. These metropolitan areas saw declines in most
of the performance indicators. Salt Lake and Memphis,
however, did well. Memphis is included in the rail
category because it invested in a short downtown cir-
culator streetcar. Carrying about 3,400 passengers per
day, the Memphis streetcar carries about 10% of transit
patronage in the Memphis metropolitan area [36]. Salt
Lake operates a long, fast light rail line with a short
branch to a university. The two light rail lines serve
several important destinations, while the bus system
was very sparsely developed before the light rail line
was built and changed little with its opening.
The worst-performing category for each perfor-
mance measure is the radial all-bus category, which is
the category that contains almost half of the transit
systems. In 2004, the median metropolitan area in
this category had only one-fifth the transit ridership
per capita as the median multidestination bus-rail sys-
tem (Table 3). Over the preceding 20 years, the median
decline in ridership per capita for the all-bus radial
category was 13%, while the median increase in rider-
ship per capita for the multidestination bus/rail cate-
gory was 9% (Table 3). Productivity for the median all-
bus radial category was about a third worse (Table 4),
while operating expenses per passenger mile were
 Light Rail Transit, Systemic Viability 733

Light Rail Transit, Systemic Viability. Table 2 Classification of 45 midsized US metropolitan areas into four categories
based on mode of transit and orientation of route structure

Multidestination MSAsa
Bus-only (in 2004)b % Non-CBD routes Bus-and-rail (in 2004)b % Non-CBD routes
Las Vegas 73.58% Atlanta 75.00%
Milwaukee 48.53% Dallas 61.08%
Norfolk 49.18% Denver 58.70%
Phoenix 61.36% Miami 67.61%
Rochester 45.00% New Orleans 50.50%
San Antonio 45.00% Portland 56.82%
Sancramento 69.05%
Saint Louis 54.55%
San Diego 81.87%
Seattle 53.88%
Radial MSAsa
Bus-only (in 2004)b % Non-CBD routes Bus-and-rail (in 2004)b % Non-CBD routes
Albany 9.52% Buffalo 34.92%
Austin 22.86% Cleveland 39.68%
Birmingham 5.41% Hartford 6.90%
Charlotte 27.16% Houston 38.32%
Cincinnati 7.14% Jacksonville 23.81%
Columbus 24.14% Memphis 20.90%
Dayton 23.53% Minneapolis-Saint Paul 34.80%
Grand Rapids 21.74% Pittsburgh 21.33%
Greensbroo 16.39% Salt Lake City 42.42%

Greenville 0.00%
Indianapolis 7.14%
Kansas City 40.82%
Louisville 29.63%
Nashville 0.00%
Oklahoma City 16.13%
Orlando 43.75%
Providence 29.31%

about a third more than the median for the bus/rail than real operating expenses per passenger mile for the
multidestination category (Table 5). Real operating median multidestination rail/bus system (Table 5).
expenses per passenger mile for the median all-bus Most if not all of the metropolitan areas in this category
radial category also increased almost four times faster invested very little in their bus systems for busways.
734 Light Rail Transit, Systemic Viability
Light Rail Transit, Systemic Viability. Table 2 (Continued)
Radial MSAsa
Bus-only (in 2004)b % Non-CBD routes
Raleigh 22.89%
Richmond 9.62%
Tampa 33.33%
a
We defined MSAs as multidestination or rail using the variable measuring percent of transit routes that do not serve the CBD. We
classified MSAs with 45% or more of the routes not serving the CBD as multidestination MSAs. We classified the remaining MSAs as radial
MSAs
b
We defined MSAs as bus-only or bus-and-rail based on the transit system in 2004. We placed two MSAs (Milwaukee and Tampa) with rail
transit routes totaling less than 4 miles in length in the bus-only group
Source: Brown and Thompson [45].
Over the 20 years of the study, they also tended not to
expand bus miles to provide more coverage or greater
frequency. In short, there is little evidence that because
they did not invest in rail transit that they improved
their bus systems instead.
Overall, the results of Brown and Thompson [45]
suggest that light rail investments around which bus
services are reconfigured into multidestination transit
systems increase their ridership more, have greater
ridership, have greater productivity, and cost less to
operate than all-bus systems and rail plus bus systems
that focus most of their routes on CBDs. These results
appear to occur, because travel markets are more com-
plex than the simple suburb to CBD model originally
postulated by Meyer, Kain, and Wohl in 1966. Residents
in metropolitan areas are traveling to multitudes of
different destinations in addition to CBDs. The
multidestination bus/rail systems appear to be better
suited to tapping the more complex travel markets than
the radial all-bus and bus/rail transit systems. Another
study confirmed this point. It shows that regional
employment reachable by bus and train combinations
in Atlanta is important for explaining rail ridership on
Atlanta’s rapid transit lines. Fluctuations of employ-
ment in Atlanta’s CBD are not important, primarily
because CBD employment is such a small part of
employment in the rest of the transit system’s service
area. Employment outside of the transit service area, or
which there is much in Atlanta, also affects transit
ridership, but negatively [46].
Surveys of passengers on board transit vehicles yield
additional insights into transit travel markets. Several
Bus-and-rail (in 2004)b % Non-CBD routes
of the multidestination bus/rail transit systems admin-
istered such surveys, and their results are similar. They
show that the bus parts of multidestination rail/bus
transit systems attract transit-dependent riders while
the rail parts attract both transit dependent and choice
riders. The large numbers of passengers who use com-
binations of buses and trains to complete their trips
have characteristics almost identical to the passengers
who use only buses [14]. In the context of the multides-
tination transit system attracting many more riders per
capita than the radial systems, these results suggest that
much of the added demand captured by the multides-
tination systems comes from the transit dependent. It
appears that the demand for transit by the transit
dependent population is highly elastic with respect to
directness of transit service between origin and desti-
nation. Another study of persons going to work on an
all-bus multidestination network in Broward County,
Florida, confirms this point. It reveals that almost all of
the workers are transit dependent and destined to
employment throughout the metropolitan area. Transit
travel time between origin and destination is the most
important variable for influencing their ridership, and
every 1% decrease in transit travel time between origin
and destination increases transit-dependent riding by
2.7% [47].
These results suggest that metropolitan areas that
invest in light rail systems achieve far greater transit
performance than metropolitan areas that do not invest
in light rail systems, if the metropolitan areas restruc-
ture their bus networks around light rail stations to
reduce bus miles and achieve multidestination travel
 Light Rail Transit, Systemic Viability 735

Light Rail Transit, Systemic Viability. Table 3 Ridership per capita in 1984, 2004 for transit systems in each category
Ridership (passenger miles per capita)
Multidestination bus- Percent change Radial bus-and-rail Percent change
and-rail MSAs 1984 2004 (1984–2004) MSAs 1984 2004 (1984–2004)
Atlanta 173.06 149.07 13.86% Buffalo 82.21 56.53 31.23%
Dallas 63.33 66.12 4.40% Cleveland 144.50 94.57 34.56%
Denver 131.74 149.01 13.11% Hartford 68.50 59.43 13.24%
Miami 125.14 163.80 30.90% Houston 97.41 99.28 1.92%
New Orleans 161.51 94.92 41.23% Jacksonville 55.01 48.28 12.23%
Portland 161.89 223.71 38.19% Memphis 40.87 55.85 36.66%
Sacramento 74.41 67.66 9.07% Minneapolis-Saint Paul 105.62 86.36 18.24%
Saint Louis 71.66 91.72 28.00% Pittsburgh 130.46 116.43 10.75%
San Diego 117.19 148.87 27.03% Salt Lake City 70.66 80.66 14.15%
Seattle 203.13 198.06 2.49% Median 82.21 80.66 12.23%

Median 128.44 148.94 8.76%

Multidestination bus- Percent change Percent change
only MSAs 1984 2004 (1984–2004) Radial bus-only MSAs 1984 2004 (1984–2004)
Las Vegas 22.67 107.05 372.17% Albany 53.40 45.90 14.05%
Milwaukee 143.49 101.99 28.93% Austin 60.16 80.02 33.02%
Norfolk 53.06 51.32 3.26% Birmingham 21.32 17.13 19.64%
Phoenix 40.16 53.49 33.20% Charlotte 22.23 36.68 65.02%
Rochester 53.67 39.13 27.10% Cincinnati 81.42 72.40 11.08%
San Antonio 101.14 82.84 18.09% Columbus 82.39 25.15 69.48%
Median 53.36 68.16 10.68% Dayton 85.99 41.61 51.61%
Grand Rapids 16.37 17.03 4.04%
Greensboro 11.67 9.50 18.65%
Greenville 4.46 4.94 10.69%
Indianapolis 44.54 22.82 48.76%
Kansas City 33.83 25.74 23.92%
Louisville 86.59 40.42 53.32%
Nashville 34.97 20.15 42.39%
Oklahoma City 12.52 16.75 33.80%
Orlando 27.24 68.62 151.96%
Providence 55.51 55.07 0.78%
Raleigh 15.96 27.64 73.13%
Richmond 69.86 29.39 57.93%
Tampa 38.64 38.82 0.48%
Median 36.80 28.51 12.56%
Source: Brown and Thompson [45].
736 Light Rail Transit, Systemic Viability

Light Rail Transit, Systemic Viability. Table 4 Productivity, measured as passenger miles per vehicle mile in 1984, 2004
for transit systems in each category

Service productivity (passenger miles per vehicle mile)

Multidestination bus- Percent change Radial bus-and-rail Percent change
and-rail MSAs 1984 2004 (1984–2004) MSAs 1984 2004 (1984–2004)
Atlanta 13.88 13.79 0.72% Buffalo 10.23 6.49 36.61%
Dallas 11.86 8.60 27.50% Cleveland 14.21 7.97 43.95%
Denver 9.17 7.47 18.51% Hartford 10.24 6.77 33.93%
Miami 11.38 10.34 9.13% Houston 9.81 9.56 2.62%
New Orleans 14.64 8.68 40.72% Jacksonville 7.56 5.64 25.44%
Portland 8.41 12.25 45.53% Memphis 6.67 8.18 22.61%
Sacramento 11.35 9.14 19.52% Minneapolis-Saint Paul 9.70 8.19 15.63%
Saint Louis 7.77 9.16 17.92% Pittsburgh 10.05 7.18 28.59%
San Diego 10.95 11.15 1.77% Salt Lake City 6.05 5.56 8.11%
Seattle 11.21 9.38 16.29% Median 9.81 7.18 25.44%

Median 11.28 9.27 12.71%

Multidestination bus- Percent change Percent change
only MSAs 1984 2004 (1984–2004) Radial bus-only MSAs 1984 2004 (1984–2004)
Las Vegas 10.90 11.23 3.05% Albany 9.15 6.97 23.83%
Milwaukee 9.29 7.19 22.64% Austin 7.24 6.98 3.55%
Norfolk 8.31 7.65 7.96% Birmingham 6.83 5.98 12.50%
Phoenix 8.88 6.29 29.18% Charlotte 9.46 6.90 27.02%
Rochester 8.97 6.59 26.45% Cincinnati 9.95 9.11 8.45%
San Antonio 8.59 8.01 6.74% Columbus 12.75 4.81 62.27%
Median 8.92 7.42 15.30% Dayton 12.96 5.56 57.12%
Grand Rapids 5.63 5.73 1.79%
Greensboro 8.57 4.92 42.63%
Greenville 5.24 7.00 33.73%
Indianapolis 10.66 6.31 40.76%
Kansas City 6.59 4.74 28.03%
Louisville 11.96 6.47 45.89%
Nashville 8.53 5.28 38.13%
Oklahoma City 4.80 5.11 6.46%
Orlando 7.22 9.37 29.93%
Providence 8.72 7.19 17.55%
Raleigh 6.18 4.55 26.29%
Richmond 11.57 5.96 48.50%
Tampa 8.26 5.99 27.52%
Median 8.55 5.98 26.65%
Source: Brown and Thompson [45].
 Light Rail Transit, Systemic Viability 737

Light Rail Transit, Systemic Viability. Table 5 Cost efficiency, measured as real annual operation expenses per passen-
ger mile in 1984, 2004 for transit systems in each category (all expenses expressed in real 2006 dollars)

Cost efficiency (real operating expense per passenger mile, 2006 dollars)
Multidestination bus- Percent change Radial bus-and-rail Percent change
and-rail MSAs 1984 2004 (1984–2004) MSAs 1984 2004 (1984–2004)
Atlanta $0.41 $0.57 40.08% Buffalo $0.66 $1.27 91.93%
Dallas $0.70 $0.71 1.60% Cleveland $0.54 $0.88 62.75%
Denver $0.70 $0.67 4.61% Hartford $0.56 $0.92 63.71%
Miami $0.62 $0.66 7.03% Houston $0.60 $0.55 8.57%
New Orleans $0.49 $0.98 97.42% Jacksonville $0.60 $0.87 45.16%
Portland $0.72 $0.60 16.08% Memphis $0.78 $0.65 15.70%
Sacramento $0.52 $0.87 66.07% Minneapolis-Saint Paul $0.66 $0.83 25.59%
Saint Louis $0.87 $0.65 25.92% Pittsburgh $0.76 $0.97 27.88%
San Diego $0.45 $0.52 15.96% Salt Lake City $0.60 $0.91 51.56%
Seattle $0.54 $0.82 51.67% Median $0.60 $0.88 45.16%

Median $0.58 $0.66 11.49%

Multidestination bus- Percent change Percent change
only MSAs 1984 2004 (1984–2004) Radial bus-only MSAs 1984 2004 (1984–2004)
Las Vegas $0.49 $0.46 5.40% Albany $0.52 $0.84 60.93%
Milwaukee $0.60 $0.87 43.97% Austin $0.99 $0.84 15.12%
Norfolk $0.62 $0.59 5.46% Birmingham $0.76 $0.69 8.33%
Phoenix $0.64 $0.94 46.67% Charlotte $0.52 $0.77 46.67%
Rochester $0.70 $1.05 49.73% Cincinnati $0.64 $0.59 8.33%
San Antonio $0.49 $0.57 16.61% Columbus $0.45 $1.45 225.23%
Median $0.61 $0.73 30.29% Dayton $0.41 $1.05 156.68%
Grand Rapids $0.83 $0.92 10.01%
Greensboro $0.47 $1.18 154.39%
Greenville $0.60 $0.59 2.41%
Indianapolis $0.50 $0.75 48.08%
Kansas City $1.09 $0.16 7.06%
Louisville $0.43 $0.84 97.51%
Nashville $0.50 $0.95 88.28%
Oklahoma City $0.97 $0.77 20.80%
Orlando $0.45 $0.50 12.40%
Providence $0.62 $0.91 46.10%
Raleigh $0.74 $0.89 20.14%
Richmond $0.47 $0.90 92.51%
Tampa $0.45 $0.82 83.94%
Median $0.52 $0.84 46.39%
Source: Brown and Thompson [45].
738 Light Rail Transit, Systemic Viability
capability. Productivity will improve, cost efficiency
will improve, and ridership per capita will increase.
Added ridership will be a combination of the transit
dependent and riders who have a choice for using an
auto. Metropolitan areas that invest in light rail but do
not do restructure their bus systems will not see such
improvements in system performance.
This is a suggestive conclusion and not a hard-and-
fast rule. There is the counter example of Salt Lake,
which invested in a fast light rail line that connected
a number of destinations together while not changing
its CBD-radial bus system very much. The bus system
was sparsely developed to begin with, and decision
makers did not want to be accused of cutting back the
already sparse CBD-focused bus system to pay for rail.
The result was that the rail system attracted new rider-
ship from previous auto users, and it also attracted about
half of the bus system’s ridership. The largely empty
buses became even emptier. The light rail line looked
like a winner, while the bus system looked like a loser
[14]. The Salt Lake result is consistent with the thrust of
a recent study of variables affecting rail transit ridership.
That study concludes that the most important variable
for increasing ridership is the magnitude of suburban
employment within walking distance of rail stations. Rail
lines that serve only employment in CBDs have low
ridership, because most employment is located outside
of CBDs. Rail lines with high ridership directly serve
employment in many locations. The study concludes
that multidestination rail networks serving large parts
of regional employment are essential for attracting rail
ridership but bus connections are irrelevant [48].
Consideration of Capital Investment
An unanswered question from the analysis of transit
performance in US midsized metropolitan areas is
whether the favorable light rail performance results
are worth the investment in light rail systems. If annu-
alized capital expenses per passenger mile were added
to operating expenses per passenger mile, would
light rail appear so attractive? Brown and Thompson
did not include capital investments for either buses
or light in their analyses, because there is no source
from which to obtain capital investments in US
or Canadian transit systems. In the absence of such
a data source, one could attempt to estimate capital
investment in a metropolitan area’s transit system in
a given year, but doing so is a huge task and rarely has
been attempted [49].
One study that did estimate comparative capital
costs between light rail and buses was Kain and Liu’s
1999 study comparing transit development in Houston
and San Diego [49]. Both metropolitan regions aggres-
sively development transit over a period of years,
Houston with express buses on busways and high occu-
pancy vehicle (HOV) lanes on freeways, and San Diego
with light rail lines, around which the regional bus
system was restructured. Kain and Liu found that
both metropolitan regions achieved comparable gains
in ridership with comparable investment in transit
fixed facilities and rolling stock, measured per passen-
ger mile. In coming to this conclusion, Kain and Liu
prorated only a portion of the cost of HOV facilities to
transit passengers, because motorists also used the
facilities. These results suggest that metropolitan
regions that invested heavily in express bus on freeway
or busway facilities, such as Houston, Minneapolis-St.
Paul, Seattle, and Pittsburgh, may incur capital
expenses per passenger mile as high or higher (as was
the case in Houston compared to San Diego) as met-
ropolitan areas investing in light rail systems, such as
San Diego, Denver, Dallas, St. Louis, or Portland.
One may conclude, then, that light rail capital
investment, if it is accompanied by a bus network
restructuring, buys a much better-performing transit
system than no capital investment over and above what
is required to buy buses and bus maintenance facilities.
The multidestination transit network resulting from
the light rail investment carries many more passengers,
and the increment of additional passengers comes from
both the ranks of auto users and of the transit depen-
dent, who now use transit much more frequently than
when they were forced to use a system focused on
downtown. Presumably, the transit dependent who
now use transit much more, and the former auto
users experience greater welfare for having made the
switch to transit. Are these results worth the capital
investment? A benefit-cost analysis is required to
answer that question, and at the moment the authors
are aware of none to which to turn.
Suppose that it is decided to spend a certain
amount of capital in a metropolitan region that
would buy either a core light rail system, around

which buses could be restructured into a multides-
tination network, or an express bus network, using
busways and HOV lanes on freeways, all focused on
the CBD. Many light rail critics claim that money
would be much better spent on the latter. How does
a light rail investment in fact stack up to an express bus
investment? The next section addresses that question
by turning to three actual cases.
Comparison of Light Rail and Express Bus
Performance
Brown and Thompson [51] examined express bus and
light rail performance in three metropolitan areas that
operate both modes: San Diego, Minneapolis-St. Paul,
and Pittsburgh. In San Diego, light rail has been the
center of transit development in the southern part of
San Diego County since 1978, but express buses and
commuter rail are the enhanced mode in the northern
part of the county. Light rail and express buses function
quite differently in San Diego County. In the light rail
corridors, light rail stations are the focus of transit
activity. Local bus routes focus on light rail stations,
and passengers transfer not only between buses and
trains, but between buses and buses (Fig. 9). Two of
the stations shown in Fig. 9, 12th and Imperial and Old
Town, also are transfer points between light rail lines,
and the Old Town station also accommodates transfers
with commuter trains.
The express bus corridors in the San Diego area
have much less interaction with the local transit net-
work. Buses operate long distances from northern
points to the San Diego CBD. Some of the express
buses operate nonstop, but some of them leave the
freeway at intermediate locations to serve important
destinations and transfer stations with light rail and
other bus routes. There is a large time penalty in doing
so, because of heavy traffic congestion around the free-
way on and off ramps, but the destinations are impor-
tant enough to warrant the loss of time. Nonetheless,
there is much less emphasis on express bus transferring
with local bus lines than in the light rail corridors, and
because of the time penalty associated with express bus
stations, there are far fewer express bus stations than
light rail stations.
Figure 9 shows the patronage results in the San
Diego area. The area of each of the blue circles is
Light Rail Transit, Systemic Viability 739
proportional to transit passengers getting on and
off vehicles. The figure shows that most passenger
activity in the San Diego region is associated with
the light rail stations. Express bus stations are consid-
erably less important, even though they serve areas of
heavy population and employment. Express buses in
each of the two corridors to the north carry about
5,000 daily passengers in each corridor. The light rail
corridor running from Old Town through the CBD to
the border (the Blue Line) carries about 65,000 passen-
gers per day. (The express bus that operated in the
CBD to border part of this corridor until the light rail
line opened in 1981 carried about 900 passengers per
day.) The light rail line running from the CBD to El
Cajon to the east (the Orange Line) carries
about 25,000 passengers per day. Not shown in the
figure is the eastern half of the newest light rail line in
San Diego, which runs from Old Town east to El Cajon
(the Green Line, operating through the west–east
Mission Valley). Even though the Green Line does not
serve the CBD, it does serve many other important
destinations, including two regional malls, suburban
office buildings, apartment and condominium com-
plexes, complexes of big box stores, San Diego State
University (SDSU in Fig. 9), and a sports stadium. It is
carrying about 22,000 passengers per day [52]. By far
the most heavily patronized transit stop in the region is
not in the CBD but is the Old Town Station, where the
Green Line intersects with the Blue Line running to
the CBD and the south, numerous bus lines running to
the northeast, north, northwest, west, and southwest,
and a commuter rail line running to the north. A large
park and ride lot also is located at the Old Town station.
Before the Green Line was completed to El Cajon,
the Old Town Station accommodated about 20,000
boarding and alighting passengers per day in 2005;
that number increased to 32,000 daily boarding
and alighting passengers in 2006 after the Green
Line opened in its entirety. The second most heavily
patronized station in the region is the 12th and
Imperial Station, which is the transfer point between
the south (Blue Line) and east (Orange Line) light
rail corridors. The 12th and Imperial station (called
the Mills Center) handled about 22,000 boardings
and alightings per day in 2006, and most of the pas-
sengers were transferring between the two light rail
routes [14].
740 Light Rail Transit, Systemic Viability
Light Rail Transit, Systemic Viability. Figure 9
Transit lines and daily boardings and alightings at transit stops, San Diego 2005, a year before the full Mission Valley line
opened [51]
The top panel of Table 6 also shows the
much greater productivity of light rail compared to
express buses in the San Diego area. Express
buses account for 12% of the revenue hours of transit
service in the region but carry 7% of the transit
passengers. The three light rail lines accounts for 7%
of the revenue hours of transit service in the region
but carry 33% of the transit passengers. These
numbers illustrate what is meant by light rail being
the backbone of transit service, around which the
rest of the transit system is organized. Tables 3–5
show that transit in San Diego is among the best
performing of the transit systems in midsized US met-
ropolitan areas.
 Light Rail Transit, Systemic Viability 741

Light Rail Transit, Systemic Viability. Table 6 Express bus and light rail comparisons in three metropolitan areas.
2006 data

Trips per revenue Trips per revenue

Transit service Passenger trips Revenue hours Revenue miles houra milea
San Diego
Express bus 21,606 717 13,839 23.80 1.10
Percent of system total 7% 12% 14%
Light rail transit 106,087 437 9,552 242.65 11.11
Percent of system total 33% 7% 10%
Minneapolis-Saint Paul
Express bus 31,097 742 16,871 39.66 1.77
Percent of system total 13% 15% 22%
Light rail transit 26,407 142 2,818 186.36 9.37
Percent of system total 11% 3% 4%
Pittsburgh
Bus routes using 43,582 1,878 33,915 18.62 1.12
busway
Percent of system total 19% 22% 28%
Light rail transit 26,038 340 4,863 76.54 5.35
Percent of system total 11% 4% 4%
a
Performance for express bus routes refers to the median-ranked route
Source: Brown and Thompson [51].

The San Diego light rail and bus network is orga-
nized so that passengers could travel to many destina-
tions in addition to those in the CBD. Most bus routes
and one of the three light rail routes do not go to the
CBD, and one could infer that most of their passengers
are not going to the CBD, either. Even transit routes
directly serving the CBD have most of their passengers
going to other destinations, as shown in Table 7.
Constructing Table 7 was possible, because the San
Diego MPO keeps detailed records of passenger
boardings and alightings for all stops for all transit
routes at different times of the day. During the morning
rush period, 65% of the riders of the two light rail lines
directly serving the CBD did not get off in the CBD.
They got off at other stations. Eighty-five percent of
express bus riders, 68% of commuter train riders,
and 77% of local bus riders got off at destinations
other than the CBD. These figures indicated that
a large market does exist for non-CBD transit service.
The Canadian approach to transit network design,
structuring bus routes around light rail lines serving
numerous destinations, thus appears to be an approach
to providing transit service that would yield much
higher ridership and greater productivity than CBD-
focused approaches, such as in the next two examples.
Minneapolis-St. Paul offers a different perspective,
in that the region operates both light rail and express
buses, but light rail is not the favored mode in that area.
Since the 1970s, the area’s transit provider, Metro Tran-
sit focused on connecting suburban neighborhoods to
the CBD with express buses operating on freeways.
Metro Transit has done an excellent job expanding the
express bus system, and it is considered one of the more
progressive transit systems in the USA for its efforts in
this regard. More than 90 of the region’s 200 plus
bus routes are express bus routes, and 85% of them
742 Light Rail Transit, Systemic Viability

Light Rail Transit, Systemic Viability. Table 7 Weekday morning peak period transit alighting for transit routes serving
the San Diego CBD directly. FY 2006 data

Type of service
Trolley (LRT) CBD-serving express bus Coaster (commuter rail) CBD-serving local bus
Location Alightings Percent Alightings Percent Alightings Percent Alightings Percent
Inside CBD 6,687 34% 400 15% 670 32% 2,517 23%
Outside CBD 13,000 66% 2,349 85% 1,447 68% 8,254 77%
Total 19,687 100% 2,749 100% 2,117 100% 10,771 100%
Source: Brown and Thompson [14].

Anoka

Ramsey
Washington

Hennepin

W E
Carver
S
Dakota

Scott Legend
Express bus

0 5 10 20 miles LRT stations

LRT line
Bus routes

Light Rail Transit, Systemic Viability. Figure 10

Transit routes in Minneapolis-St. Paul [51]

focus on the Minneapolis or St. Paul CBDs. Figure 10 facilitate the express bus service. Express buses make
shows in heavy black lines the area’s CBD-oriented use of 15 miles of busways, 38 miles of high-occupancy
express buses and light gray lines other bus routes. vehicle lanes, and 268 miles of bus-only freeway shoul-
There also is a heavy investment in infrastructure to der lanes [51]. Metro Transit did not open a light

rail line until 2004. The agency had opposed light rail
transit but acceded to outside political pressures to
build a line. To its credit it built a state-of-the art line,
around whose stations it restructured local bus lines
operating into the light rail corridor. The line serves not
only the Minneapolis CBD, but also two stations at the
international airport and a terminal station at the Mall
of America, perhaps the largest suburban mall in the
USA. The Minneapolis light rail line thus is multides-
tination in character. The light rail line and its stations
also are shown in Fig. 10 [14, 51].
Express buses in the Minneapolis-St. Paul region
are among the most productive in the nation [14].
Despite this, the express buses are not nearly as pro-
ductive as the region’s light rail line, as shown in the
middle panel of Table 6. The express bus system
accounts for 15% of the region’s hours of transit service
and carries 13% of the region’s transit passengers. The
one light rail line accounts for 3% of the region’s hours
of service but carries 11% of the region’s transit pas-
sengers. The percentage may be higher today. The more
recent daily light rail patronage figure for Minneapolis
of 34,000 daily passengers in 2008, shown earlier in
Table 1, indicates light rail passenger growth.
The trend of patronage for the overall transit sys-
tem, however, has been downward, as shown earlier in
Tables 3–5. The downward trend could be caused by
decentralization. The transit system is exceptionally
well run, and the region is progressive and growing.
Service reliability and regional decline are not causes of
transit decline. The region continues to decentralize,
however, and the CBDs are becoming relatively if not
absolutely less important as time passes. Metro Transit
recognizes this point and is building more intermediate
stations at emerging destinations along its express bus
routes to tap new, non-CBD travel demands. Overall,
however, the transit system in the Minneapolis-St. Paul
region remains CBD-centric.
Pittsburgh also operates both light rail and express
buses in a transit system where about 85% of total
regional transit service connects suburban neighbor-
hoods with the CBD. The transit system, the Port
Authority of Allegheny County (PAT) invested in
three busways and a light rail line. The busways are
Type A grade-separated facilities that run to the south,
east, and west from the CBD. Express buses coming
from the suburbs funnel into the busways and head to
Light Rail Transit, Systemic Viability 743
the CBD. The busways cost several hundred million
dollars to build, and each one was designed to serve
up to 50,000 passengers per day. Actual usage is much
lower, varying between 9,000 and 13,000 daily passen-
gers, depending on the busway. The light rail line is
a several hundred million dollar rebuilding of the South
Hills streetcar corridor, including a subway in the CBD.
Its primary purpose is to connect residential neighbor-
hoods in the south of the city to the CBD. The line does
serve a second destination, however, which is South
Hills Village, a suburban shopping center. Unlike stations
on the light rail line in Minneapolis-St. Paul, Pittsburgh
light rail line stations are not focal points for local bus
routes. Bus lines tend to parallel the light rail line into the
CBD. The south busway also parallels part of the light
rail line and uses the light rail Mt. Washington Tunnel to
gain access to the CBD. Transferring between buses and
light rail trains is difficult at suburban stations. Not only
do few bus routes intersect with the rail route, but trans-
ferring passengers must pay a second fare on the vehicle
to which they transfer [14, 51].
As noted earlier in Tables 3–5, the Pittsburgh transit
system has been in decline. Its CBD focus may be the
cause. Although population in the overall region has
not been growing, it has been decentralizing. The CBD
also has been losing employment relative to the rest of
the region [14]. The CBD is the only market that the
transit system has identified as being important to
serve; thus, as the CBD declines, so does the transit
system. Because of its isolation from the rest of the transit
system, light rail in Pittsburgh is not as well used or
productive as light rail is in many other US cities. Table 1
presented earlier shows its passenger density is among
the lowest of light rail lines in the USA. Despite that, it
still is more productive than the Pittsburgh express bus
system, as shown in the bottom panel of Table 6.
Express buses account for 22% of daily transit service
hours in Pittsburgh and carry 19% of daily transit
passengers. Light rail accounts for 4% of daily service
hours and carries 11% of daily transit passengers.
Light Rail Traffic Density Related to Strategic
Development of Light Rail Systems and to Land Use
Table 1 presents average light rail passenger densities.
This section discusses traffic densities in the context of
light rail development strategies and land uses in
744 Light Rail Transit, Systemic Viability
several of the urban areas listed in the table, coming to
conclusions about the relative effectiveness of those
strategies in the context of different land uses. Traffic
densities, for example, suggest that of the pioneer light
rail systems, Edmonton and Calgary are highly success-
ful, and that the systems in San Diego, and Portland are
less successful, but on the other hand, they are relatively
more successful light rail initiatives than those in San
Jose, Sacramento, and Buffalo. The transit systems in
Edmonton, Calgary, San Diego, and Portland all devel-
oped with similar visions of light rail serving as the core
of multidestination light rail/bus networks. Differences
in transit development strategy do not explain differ-
ences in traffic densities between the Canadian and US
light rail systems, but differences in land uses may.
Edmonton and Calgary are more compact metropoli-
tan areas than San Diego and Portland, with fewer
freeways and more apartment living. Distances
between places where people begin and end trips are
shorter. Such land use differences may account for the
greater light rail traffic densities in the Canadian met-
ropolitan areas compared to their US counterparts
employing similar principles of transit development.
Jones [11], on the other hand, argues that the higher
transit ridership seen in Canadian cities compared to
transit use of similar cities in the USA is due largely to
lower personal income in Canada.
Higher light rail passenger densities in San Diego
and Portland compared to those in San Jose, Buffalo,
and Sacramento likely arise as a consequence of the role
of the light rail lines in San Diego and Portland as the
cores of their multidestination transit networks. Such
strategic vision has been less well articulated in San Jose
and Buffalo, and it has not been consistently embraced
in Sacramento.
The traffic densities of Edmonton, Calgary, San
Diego, and Portland light rail lines also equal or exceed
the traffic densities of most of the older streetcar oper-
ations that have been rebuilt. Boston is the only older
streetcar operation whose traffic density exceeds that of
most of the modern light rail adoptions, and that is
because many passengers use the Boston streetcar sub-
way as a shuttle to their final destination after alighting
from rapid transit lines, bus lines, and commuter
trains. Hardly any buses serve the Boston CBD, and
bus passengers destined to the CBD transfer to light rail
or rapid transit lines. The light rail traffic densities of
Edmonton, Calgary, San Diego, and Portland light rail
lines are much greater to slightly greater than streetcar
traffic densities in Toronto and San Francisco, two
cities where the streetcars function as important parts
of bus/streetcar/rapid transit integrated route net-
works. Toronto and San Francisco both are densely
populated cities, and one would expect that their
streetcar traffic densities are greater than they are rela-
tive to the lower-density, more sprawling metropolitan
areas of Edmonton, Calgary, San Diego, and Portland
because of their central city locations. However, street-
cars in San Francisco (even in the subway) and in
Toronto generally are much slower than the light rail
lines, and slow speed may account for the lower-than-
expected streetcar traffic densities in Toronto and San
Francisco. Similarly, the subway-surface light rail lines
in Philadelphia are both relatively slow and are not part
of integrated bus/light rail networks.
Light rail traffic densities in Edmonton, Calgary,
Portland, and San Diego greatly exceed the light rail
traffic densities of Cleveland and Pittsburgh, which are
other suburban light rail lines, but where there is
no route coordination with buses. Cleveland and
Pittsburgh operate CBD-radial transit systems. The
light rail traffic densities in Edmonton, Calgary,
Portland, and San Diego also greatly exceed the traffic
densities of the New Orleans and Portland streetcars.
Similar patterns are exhibited by more recent light
rail adoptions. Those that are characterized by bus
systems being restructured around light rail lines
(Dallas, Denver, Minneapolis, St. Louis) or where the
light rail lines have been imbedded into bus grid
networks (Los Angeles in the cases of the Blue and
Green Lines but not the Gold Line) generally have
higher traffic densities than those where bus/light rail
integration does not exist (Baltimore, Pittsburgh,
Hudson-Bergen).
The importance of the place of light rail in
a multidestination bus network for high traffic density
is illustrated by the Green Line in Los Angeles. As the
Green Line in San Diego, Los Angeles’ Green Line does
not serve the CBD. It is an east–west crosstown line that
runs in the median of huge freeway about 8 miles south
of the CBD, and pedestrian access between its stations
and surrounding land uses very difficult. The line does
not serve directly any major trip generators and is
dubbed locally, “The line from nowhere to nowhere.”

Yet, it carries between 30,000 and 40,000 passengers per
day. It operates entirely on a Type A right-of-way, is
very fast, and at every station it intersects with a north–
south arterial road, each one of which carries a long,
frequent, heavily patronized bus line offering free
transfers to and from the rail line. Transfer movements
are heavy. The Green Line also connects with the
north–south Blue Line, the first light rail line in the
region. The Blue Line carries 75,000–85,000 passengers
per day (the most heavily traveled single light rail line
in the USA) because it is fast and because almost every
station serves a major east–west arterial road, each of
which hosts a long, frequent, heavily patronized bus
line with free transfers to and from trains [53].
There are two light rail applications with lower
traffic densities that restructured bus services around
the light rail investments. Neither one disproves the
proposition that multidestination transit systems
achieved by restructuring bus networks around light
rail investments improves transit performance. One of
these is in St. Louis. St. Louis operates a fast light rail
spine through the center of the metropolitan area serv-
ing the CBD and several other important destinations.
Within the metropolitan area, the transit operator
restructured bus routes around light rail stations, pri-
marily to improve transit productivity rather than to
create a multidestination network. The line, which
originally terminated across the Mississippi River in
East St. Louis, later was extended for about 20 miles
into Illinois farmland to achieve equity of light rail
investment on the Illinois side of the river. The exten-
sion dilutes light rail traffic density [33, 54]. Even so,
the St. Louis line’s patronage density is much higher
than that of Pittsburgh and Baltimore. Light rail lines
in Baltimore and Pittsburgh (as well as in Cleveland)
compete with radial bus lines running to the CBD.
The fact that Cleveland’s light rail traffic density is
considerably higher than Pittsburgh’s and Baltimore’s
when its service philosophy of serving the CBD is much
the same may be related to both speed and land use.
Cleveland’s light rail line was built in 1920 to serve
a series of neighborhoods that were designed concur-
rently, with the rail line in mind. The neighborhood
design objective was the creation of short and attractive
walks from all houses in the Shaker Heights develop-
ment to streetcar stops. This factor may set it apart
from light rail line in Baltimore, which is even faster
Light Rail Transit, Systemic Viability 745
but where favorable urban design around stations does
not exist. The Pittsburgh line is somewhat slower than
the Cleveland line, and its access to residents may not
be as good, either.
The Oceanside-Escondido line is the remaining
light rail line, around which bus services are structured,
where traffic density is lower. Land uses may be the
explanation. The line serves a sprawling, low density
and hilly part of San Diego County almost 40 miles
north of the CBD, offering service every 30 min with
diesel rail cars. The area that it serves is more sparsely
settled than the other areas considered, and the fre-
quency of service is much less, as well. These attributes
likely result in its low patronage density, even though
there are high transfer movements between buses and
trains at three transit centers along the line. The three
bus transfer stations are the most heavily patronized
stations on the line.
The examination of traffic densities in the context
of transit service approach and metropolitan context
suggests several points. First, it strengthens earlier
conclusions that light rail being imbedded in a multi-
destination transit network is important for achie-
ving high traffic density, which appears to be related
to improved overall regional transit performance.
The analysis also suggests that faster light rail lines
that are backbones of multidestination transit networks
perform much better than slower lines in a similar
context. The message on the effect of urban form is
mixed. Light rail experiences higher passenger densities
in Calgary and Edmonton than in San Diego and
Portland, even though all four metropolitan areas
embraced similar transit development strategies, pos-
sibly because the Canadian metropolitan areas are
more compact than their US counterparts. On the
other hand, lower incomes in the Canadian cities may
explain the difference. Urban form may influence
higher-than-expected patronage for Cleveland’s light
rail line. On the other hand, high densities in Toronto
and San Francisco do not translate into streetcar traffic
densities that are higher than traffic densities for many
of the new light rail lines serving much lower density,
more sprawling metropolitan areas. The overall con-
clusion is that where light rail has been employed to
take advantage of the high productivity characteristics
that were bred into its technology, it has done extraor-
dinarily well for improving transit performance. Where
746 Light Rail Transit, Systemic Viability
policy makers have not made use of those characteris-
tics, light rail has had a less favorable effect on transit
performance.
Effects of Light Rail Investments on Land Use
Patterns
As discussed earlier, growing automobile ownership
weakened and all but eliminated the power of streetcar
investments to affect metropolitan growth patterns
around World War I. Giuliano [55] summarizes theo-
retical and empirical literature on why automobiles had
that effect. Her argument is that transportation invest-
ments alter land use patterns when the investments
make parcels of land relatively more accessible to jobs
and employment than other parcels. More accessible
parcels have higher rents and will be developed at
higher intensities. When the streetcar era began, exten-
sion of streetcar lines into green fields around cities
conveyed huge accessibility increases for those parcels.
Streetcars thus had a large impact on the pattern of
development. The automobile, however, conveyed high
levels of accessibility to all parcels in metropolitan
areas. After auto ownership in cites was widespread
and roads around cities were paved, parcels of land
served by streetcars no longer had an accessibility
advantage compared to other parcels. All parcels in
the metropolitan area had roughly equal accessibility,
and so development spread out at low density. The star-
shaped city morphed into the blob. Giuliano argues
that the condition of relatively uniform accessibility
continues today. Parcels proximate to new light rail
transit stations are unlikely to have much of an acces-
sibility advantage compared to other parcels and are
unlikely to develop much differently if left to the whims
of the land market. She then marshals abundant empir-
ical evidence to show that rail transit investments dur-
ing the post World War II period have had little if any
effect on development patterns in US cities. Some other
researchers, such as Bernick and Cevero [56] or Knaap,
Ding, and Hopkins [57], examined whether property
located in proximity to rail stations has fetched higher
prices in comparison to similar properties not near rail
stations and found that rail transit proximity does
bring higher rents, but the effect is marginal. Others
have surveyed residents of new transit-oriented devel-
opments built adjacent to light rail stations in
Washington County west of Portland and Pasadena,
California to determine the extent of light rail use.
Both studies found that around 10% of residents use
light rail stations, which is low but still higher than
typical suburban residential areas [58, 59]. The effect
appears to be the greatest for inner city parcels near
stations on large, comprehensive rapid transit systems
operating on right-of-way Type A, of which the Wash-
ington Metro is an example [60].
Portland’s light rail line often is cited as the best
example of light rail investment affecting land uses. The
first line extended 16 miles from the CBD through
medium to low density blue collar and middle income
suburbs to the east and opened in 1986. Its opening
was accompanied by planning and zoning policies
that allowed higher density development around sta-
tions. Little redevelopment happened through the first
15 years of the lines’ existence, except in one area called
the Lloyd Center. Located across the Willamette River
from the Portland CBD, Lloyd’s Center was the original
suburban mall in the Portland region. By 1986, it was
starting to become run-down. After light rail arrived,
commercial and government redevelopment started,
and over time the Lloyd Center has become an eastside
extension of the Portland CBD. It is included in the free
fare CBD zone (formerly multimodal, but now limited
to LRT and streetcar lines). The extension of light rail
16 miles to the west in 1999 was accompanied by
stronger planning and zoning regulations that distin-
guished between transit-friendly and transit-hostile
land uses (authors’ terms). Only transit-friendly land
uses were permitted near stations. Metro, Portland’s
regional government, also employed a developer who
knew construction techniques for medium density res-
idential complexes (in the range of 20–60 dwelling
units per acre). His job was to teach these techniques
to local developers, whose expertise was in low density
residential or steel-framed high rise building. These
changes resulted in some higher density residential
redevelopment around stations on the original line
and higher density residential development in green
field sites along the new western line. Perhaps the
biggest decision affecting land use, however, was to
take light rail to where growth was happening rather
than to naively think that light rail investments, wher-
ever they were made, would attract development.
So, the transit district extended the west line into

a region that already was growing very rapidly, both
with new employment and new residences. Planners
purposefully chose an alignment that offered many
green field sites near stations. The net result of all of
these actions is the construction of some higher density
development connected with pedestrian paths and
sidewalks to rail transit stations.
Another slightly different approach to achieving
coordination between land use and light rail is to mod-
ify auto-oriented commercial development so that
light rail can serve it. This also is an example of taking
light rail to where growth is happening, rather than
trying to entice growth to come to light rail. The City of
San Diego and the San Diego MTDB followed this
approach with the building of the Green Line through
the Mission Valley. Mission Valley, running in an east–
west direction about 5 miles north of the San Diego
CBD, began attracting intense retailing and commer-
cial development in the 1950s that largely replaced the
CBD as the retailing hub of the region and rivaled the
CBD as a commercial center. By the time MTDB
designed the Green Line in the early 1990s, Mission
Valley contained an east–west interstate freeway, an
east–west high speed six lane arterial road, and cross-
ings of four north–south freeways built to interstate
standards. The valley also contained a regional mall,
a mall of big box stores, other scattered big box stores,
scattered high rise office building, smaller malls and
strip malls, and a sports stadium. San Diego State
University rose on the south wall of the valley toward
its east end. Another regional mall was located just east
of that and already was served by the Orange Line
running from the CBD. There also were apartment
complexes built on the floor of the valley and on its
side. Running through the middle of all of this was the
San Diego River, little more than an open storm drain
that was dry most of the year but sometimes flooded.
Light rail in the Mission Valley was part of a city-led
remodeling of the valley. The city rebuilt the river into
an attractive, natural-looking wetlands lined on both
sides with swales. The light rail line was pieced together
from sections of swales, from bits of access street rights-
of-way, and from truck alleys serving the backside of
mall buildings. The light rail line also used elevated
structures liberally. To access San Diego State University,
the Green Line dove into the side of the valley in a short
subway, with a station directly under the university’s
Light Rail Transit, Systemic Viability 747
bus transfer hub. The city built a pedestrian path along
the river and connected it laterally to existing develop-
ments. MTDB built pedestrian connections from its
light rail stations to the river path. The city also
commissioned a nationally known architect, Peter
Calthorpe, to design a new apartment complex centered
on a light rail station in one of the valley’s remaining
green field sites. The development is called Rio Vista and
is what is known as a Transit-Oriented Development,
or TOD. The Green Line, which opened in late 2005,
runs from the Old Town station on the west where it
connects with numerous bus routes and two rail routes.
It connects to another bus transit center at Fashion
Valley mall adjacent to one of the freeways coming
from the north. Express buses from the north call at
this station, as do several local bus routes. It connects
with another bus station at San Diego State University,
still another bus station as the Grossmont Mall (where
it joins the Orange Line from the CBD), and it termi-
nates at still another bus station in El Cajon. The valley
still is auto oriented, and most people who go there
use autos. The difference is that transit now works
there, as well (Fig. 11). The Green Line now is carrying
about 22,000 passengers per day to and from valley
destinations [52].
Conclusions
Light rail arose during the 1950s and 1960s as a new
transit mode in northern European cities that provided
higher quality, higher capacity, substantially more pro-
ductive, and faster transit service than traditional street-
cars and local buses. Light rail generally was cheaper
to build than traditional subway and elevated railway
systems, and as a consequence, it could be afforded by
many midsized metropolitan areas that could not afford
to build rapid transit lines. These qualities attracted the
attention of advocates of transit and urban renewal in
Canadian and US cities, leading to light rail’s first
North American adoption in Edmonton in 1978.
Since then, 21 Canadian and US metropolitan areas
whose transit systems operated only buses at the begin-
ning of the 1970s opened one or more light rail lines in
addition to buses.
The scope of light rail implementation in Canada
and the USA also may be responsible for a renewal of
interest in traditional streetcar investment. Eight
748 Light Rail Transit, Systemic Viability
Light Rail Transit, Systemic Viability. Figure 11
The first section of San Diego’s Mission Valley Line. Looking east in 2002 along the valley from a high rise building at the
Hazard Station. Note train of U2 cars. Author’s photo
Canadian and US metropolitan areas still operated
streetcars in 1970 along with large bus systems. Six of
these areas subsequently rebuilt their streetcar systems
with at least some light rail characteristics, generally
leaving out those characteristics that led to productiv-
ity improvement. One city with streetcars renewed and
expanded its traditional streetcar operation as
a traditional streetcar operation (New Orleans), using
a mix of historic cars and replicas of historic cars with
more modern electromechanical gear. (The eighth city,
El Paso abandoned its remaining streetcar line in the
early 1970s.) In the last several years, a few other cities
built new streetcar lines primarily as tourist-oriented
circulator lines in downtown areas. Some of the new
lines use restored streetcars, some use replicas of his-
toric cars, and some use modern European-designed
articulated streetcars.
Results are mixed. Some of the light rail investments
have had strongly positive results. They substantially
improved regional transit performance in metropolitan
areas such as San Diego, Portland, St. Louis, and Dallas
by increasing per capita ridership, improving transit
productivity, and stabilizing or decreasing real operat-
ing costs per passenger mile. In these metropolitan areas
light rail investments have all or most of the following
characteristics: light rail lines are significantly faster
than local buses, they are long, and their outlying sta-
tions are loci of bus route restructuring. Some of the
stations in each of these metropolitan areas also have
large park and ride lots. In all of these metropolitan
areas, light rail lines serve sectors of their respective
regions that were experiencing robust growth in both
population and employment before the onset of the
Great Recession. Zoning and development codes
favored denser development around light rail stations.
As a consequence, some of the stations in several of
these metropolitan areas have been surrounded by
dense new development whose residents and employees
can and do walk to light rail trains. In each of these
metropolitan areas, light rail lines account for between
10% and 20% of the transit service in their respective
metropolitan areas but carry 30–50% of the daily pas-
senger miles of transit traffic. These systems clearly are
reaping the productivity capability of light rail invest-
ments. It also appears that the larger the share of met-
ropolitan area employment directly reached by light rail
lines, the more favorable the impact of the light rail
investments on all aspects of transit performance.
Light rail lines with such favorable design charac-
teristics have limitations, however. They are powerless
to attract economic growth in depressed areas, for
example. They also have not substituted for auto travel

on a wide scale. The best performing systems (in San
Diego and Portland) have achieved patronage per
capita increases in the range of 30–50% over a 20 year
period, impressive when compared to the bus patron-
age magnitudes and trends of most midsized US cities
but not impressive when compared to the magnitude of
auto use in US metropolitan areas. Autos and light
trucks generally account for more than 90% of metro-
politan travel, and even a 50% increase in per capita
transit use makes little inroads into auto use [61].
Many light rail and streetcar investments result in
less favorable results. Results are less favorable where
light rail investments cover only limited parts of their
metropolitan areas or are placed in depressed parts of
metropolitan areas. Light rail and streetcar investments
also are less effective where the resulting service is as
slow (or slower) than local bus services, where the
investments result in the operation of one-car trains
(buses on rails), or where the investments result in staff
being placed in each car of multicar trains to collect
fares. They also are less effective where there has been
no coordination of bus and light rail service. If some or
all of these qualities are characteristics of the light rail
investments, the investments may result in higher
real operating expenses per passenger mile, lower
productivity, and reduced patronage per capita. (The
investments may possess compensating attractive char-
acteristics, such as tourist appeal, or adding cache to
a neighborhood.)
Future Directions
Continued light rail investments in US and Canadian
cities are consistent with the idea of sustainability if
benefits from the investments outweigh the costs of the
investments. There are two types of light rail invest-
ments to be considered: those that result in improved
regional transit performance, and those that do not.
Light rail investments that enhance regional transit
productivity have the following characteristics: trains
have lots of capacity; trains have lots of wide doors;
multicar trains are operated by one person; proof of
payment fare systems are used with free transfers to and
from buses and between buses; rail routes are long
and fast (short routes offer no time advantage over
buses); part of the speed is achieved by trains wasting
little time at intermediate stops boarding and
Light Rail Transit, Systemic Viability 749
deboarding large passenger loads, because passengers
use all of the wide doors and do not pay fare-collecting
personnel; rail stations serve important destinations in
addition to CBDs; bus networks are restructured
around light rail stations with a substantial part of
bus mileage replaced by train mileage; restructured
bus routes not only take passengers from residential
areas to trains, but train passengers to important des-
tinations located close to but not directly served by
stations; rail lines are located to serve growing parts
of metropolitan areas, and development in such areas is
retrofitted so that passengers can walk between rail/bus
stations and adjoining development. In short, light rail
works well where planners place it within a region in
such a way to make use of its high capacity, productiv-
ity, and speed characteristics.
Such light rail investments will enhance regional
transit performance. The outstanding and unanswered
question is whether they will do so sufficiently to make
the light rail investment worthwhile. While the authors
do not have quantified benefits from such light rail
investments, they can enumerate the categories of ben-
efits and leave it to the reader to judge their signifi-
cance. To begin with, benefits include consumer
surpluses to transit dependent persons who choose to
use restructured transit service to reach more quickly
destinations throughout metropolitan areas. Benefits
also include consumer surpluses to previous auto
drivers who find it more convenient to switch to light
rail services. Benefits also include fuel savings and pol-
lution reductions resulting from fewer auto trips. There
are two categories of fewer auto trips. One category
includes previous auto trips made by those who now
leave cars at homes or in suburban lots. The other
category includes previously chauffeured auto trips of
those without cars (perhaps friends and relatives were
the chauffeurs). Such trips often involve considerable
driving by the chauffeur to pick up the rider at a remote
location or to return to the chauffeur’s place of resi-
dence or work once the rider is dropped off. The
elimination of such trips could reduce considerable
regional auto and light truck VMT. Another significant
saving is the value of time saved by previous chauffeurs.
Still other benefits include lower real operating expense
per passenger mile carried on the transit systems.
Other benefits include option value to environmentally
conscious citizens who value having a multidestination
750 Light Rail Transit, Systemic Viability
transit network that they might wish to use some day.
The aggregate of such benefits, in the authors’ opinion,
could outweigh resources required to build regional
light rail systems that will enhance transit system
performance.
The other category of light rail investments does
not enhance regional transit performance. Light rail
lines belonging to the second category typically are
designed as just another slow bus route to the CBD,
having no transfer connections with bus lines, no free
transfers, displacing no bus miles, and either operating
only one-car trains, or having drivers collecting fares
on each car of multicar trains. Such lines have little
justification. They will not perform well, and neither
will the overall transit system. They probably are not
consistent with sustainability.
In summary, a midsized metropolitan area into
which light rail is introduced to take advantage of its
high capacity, productivity, and speed characteristics
can achieve substantial patronage growth per capita,
improve productivity, and stabilize real costs for mov-
ing passengers within a metropolitan area. Stations can
be nodes for development if they are located where
development activity of the private market is active.
Planning intervention in such areas can create attrac-
tive development that is accessible by transit, although
autos will remain the principle form of access for some
years into the future. Accounting for as few as 10% of
regional transit service, such light rail lines can carry
30–50% of regional transit passenger miles that con-
tinue to grow year by year. Over time auto-less, middle-
class subcultures could evolve around such systems.
The benefits of such systems might outweigh their
construction costs. If they do, this category of light
rail investment is consistent with the concept of
sustainability.
Abbreviations
APTA American Public Transportation
Association, the lobbying organization for
transit agencies in the USA
CBD Central business district
FRA Federal Railroad Administration, a division
of the US Department of Transportation
FTA Federal Transit Administration, a division of
the US Department of Transportation
MTDB The San Diego Metropolitan Transit
Development Board, which planned and
built light rail in the San Diego region but
which organized a Verkerhsverbund
arrangement (of which it was the centralized
coordinating body) to operate all transit
services in its jurisdiction
PCC Presidents’ Conference Committee, which
conducted basic research to develop a
radical new streetcar design between the late
1920s and mid-1930s
POP Proof-of-payment fare collection system.
Also called self-service, barrier-free fare
collection system
TOD Transit-oriented development
TRB Transportation Research Board, a division
of the National Academy of Sciences
devoted to expanding knowledge about
various transportation systems of all modes
TriMet The tri-county transit district that builds
and operates light rail lines and bus services
in Portland, Oregon
UMTA The Urban Mass Transit Administration,
which is what the FTA was called before 1991
USDOT The US Department of Transportation,
headed by the Secretary of Transportation, a
member of the President’s cabinet
VMT Vehicle miles traveled (generally by autos
and light trucks) during a certain time
period over a particular road or within a
given area, such as a metropolitan area. One
can speak of, for example, daily VMT in a
metropolitan area, such as Portland, or
VMT per capita
Bibliography
Primary Literature
1. Boorse J (2001) This is light rail transit. Transportation
Research Circular Issue Number: E-C033. Transportation
Research Board, Washington, DC. ISSN: 0097-8515
2. APTA (1994) Glossary of transit terminology. American Public
Transit Association, New York
3. Vuchic V (1981) Urban public transportation systems and
technology. Prentice Hall, Englewood Cliffs, pp 62–63
4. Vuchic V (2005) Urban transit: operations, planning, and eco-
nomics. Wiley, Hoboken, pp 380–383

5. Thompson G (2005) The birth of the light rail movement in
North America and its results. Transp Serv Telecomun J
8:142–164
6. Thompson G (2003) Defining an alternative future: birth of the
light rail movement in North America. Transportation
Research E-Circular Number E-C058, ISSN 0097-8515. Trans-
portation Research Board, Washington, DC, pp 25–36. http://
gulliver.trb.org/publications/circulars/ec058/ec058.pdf
7. Fielding G (1987) Managing public transit strategically:
a comprehensive approach to strengthening service and
monitoring performance. Jossey-Bass, San Franciso
8. Post R (2007) Urban mass transit: the life story of a technology.
Greenwood, Philadelphia
9. Warner S (1968) The private city: Philadelphia in three
periods of its growth. University of Pennsylvania Press,
Philadelphia
10. Longstreth R (1997) City center to regional mall: architecture,
the automobile, and retailing in Los Angeles, 1920–1950. MIT
Press, Cambridge, MA
11. Jones D (2008) Mass motorization + mass transit: an American
history and policy analysis. Indiana University Press,
Bloomington
12. Longstreth R (1999) The drive-in, the supermarket, and the
transformation of commercial space in Los Angeles, 1914–
1941. MIT Press, Cambridge, MA
13. Wyse W (1968) The PCC car: a marvel of control engineering of
the 1930s, part I: background and general principles. Modern
Tramways, September, pp 296–303
14. Brown J, Thompson G (2009) The influence of service planning
decisions on rail transit success or failure. MTI Report 08-04.
Mineta Transportation Institute, San Jose
15. Saltzman A (1992) Public transportation in the 20th century.
In: Gray G (ed) Public transportation, 2nd edn. Prentice Hall,
Englewood Cliffs, pp 24–45
16. Snell B (1974) American ground transport: a proposal for
restructuring the automobile, truck, bus and rail industries.
Government Printing Office, Washington, DC
17. St Clair D (1986) The motorization of American cities. Praeger,
New York
18. Yago G (1984) The decline of transit: urban transportation in
German and U.S. cities, 1900–1970. Cambridge University
Press, Cambridge
19. Adler S (1991) The transformation of the Pacific Electric Rail-
way: Bradford Snell, Roger Rabbit, and the politics of transpor-
tation in Los Angeles. Urban Aff Q 27:51–86
20. Bianco M (1998) Kennedy, 60 minutes, and Roger Rabbit:
understanding conspiracy-theory explanations of the decline
of urban mass transit. Discussion paper 98-11. Center for
Urban Studies, Portland State University, Portland
21. Black A (1995) Urban mass transit planning. McGraw-Hill, New
York, pp 31–39
22. Adams W (1965) The central subway of the Massachusetts Bay
Transportation Authority. Traffic Q 19:443–457
23. Smerk G (1991) The federal role in urban mass transportation.
Indiana University Press, Bloomington
Light Rail Transit, Systemic Viability 751
24. Thompson G (2007) Taming the neighborhood revolution:
planners, power brokers, and the birth of neotraditionalism
in Portland, Oregon. J Plan Hist 6:214–247
25. Waltking D (1965) The articulated tramcar in West Germany. In:
Joyce J (ed) Modern Tramway review. Ian Allen, London, pp 72–88
26. Schmucki B (2003) Cities as traffic machines: urban transport
planning in East and West Germany. In: Divall C, Bond W (eds)
Suburbanizing the masses: public transport and urban devel-
opment in historical perspective. Ashgate, Aldershot,
pp 149–170
27. Clark R (1978) Light rail vehicles for Edmonton Transit. Presen-
tation at American Public Transit Association rapid transit
conference, Chicago
28. Wyse W (1968) Germany goes underground, part 1: Frankfurt’s
Feiertag. Modern Tramway, December, pp 404–409
29. Walker P (1968) The tram subway (a rose by any other name).
Modern Tramway, May, pp 162–171
30. Quinby D (1962) Major urban corridor facilities: a new concept.
Traffic Q 16:242–259
31. Taylor S (1970) The rapid tramway: a feasible solution to the
urban transportation problem. Traffic Q 24:513–529
32. Thompson G (2010) Technological change in transit: how the
U.S. light rail movement won its first victory in San Diego.
Working paper. Department of Urban and Regional Planning,
Florida State University, Tallahassee
33. Demrey L (2008) U.S. urban rail transit lines opened from 1980.
http://www.publictransit.us/ptlibrary/Railopenings_Feb2008.
pdf. Accessed 29 June 2010
34. Schumann J, Francis L (2003) Status of North American light
rail transit systems: year 2003 update. Transportation Research
E-Circular Number E-C058. Transportation Research Board,
Washington, DC, pp 3–24. ISSN 0097-8515. http://gulliver.trb.
org/publications/circulars/ec058/ec058.pdf
35. Schumann J (2007) Status of North American light rail transit
systems: year 2006 update. Transportation Research E-Circular
Number E-C112. Transportation Research Board, Washington,
DC, pp 3–20. ISSN 0097-8515. http://onlinepubs.trb.org/
onlinepubs/circulars/ec112.pdf
36. Florida Department of Transportation (2010) Florida transit
information system. http://www.ftis.org/INTDAS/tableflat.
aspx. Accessed 2 July 2010
37. Thompson G (1977) Planning considerations for alternative
transit route structures. J Am Inst Plan 43:158–168
38. Polzin S, Page O (2003) Ridership trends of new start rail pro-
jects. Report 350-11. National Center for Transit Ridership
39. Meyer J, Kain J, Wohl M (1965) The urban transportation
problem. Harvard University Press, Cambridge, MA
40. Pickrell D (1992) A desire named streetcar: fantasy and fact in
rail transit planning. J Am Plan Assoc 58:158–176
41. Kain J (1990) Deception in Dallas: strategic misrepresentation
in rail transit promotion and evaluation. J Am Plan Assoc
56:184–196
42. O’Toole R (2005) Rail disasters 2005: the impact of rail transit
on transit ridership: an update to great rail disasters. Reason
Foundation, Los Angeles
752 Light Rail Transit, Systemic Viability
43. Richmond J (2005) Transport of delight: the mythical concep-
tion of rail transit in Los Angeles. University of Akron Press,
Akron
44. Giuliano G (2010, forthcoming) Transportation policy: public
transit, settlement patterns, and equity in the U.S. chapter 28.
In: Brooks N, Donaghy K, Knaap G, (eds) Oxford University
Press handbook of urban economics and planning. Oxford
University Press, Oxford
45. Brown J, Thompson G (2008) Service orientation, bus-rail ser-
vice integration, and transit performance: an examination of
45 U.S. metropolitan areas. Transp Res Rec J Transp Res Board
Natl Acad Sci 2042:82–89
46. Brown J, Thompson G (2008) The relationship between transit
ridership and urban decentralization: insights from Atlanta.
Urban Stud 45:1119–1139
47. Brown J, Thompson G, Bhattacharya T (2010) Evaluation of
land use and transportation strategies to increase suburban
transit ridership in the short term. Florida Department of
Transportation, Public Transit Office, Tallahassee
48. Wood J, Zimmerman M, Poticha S, Center for Transit-Oriented
Development (2009) Destinations matter: building transit suc-
cess. Report FTA CA-26-1007. Federal Transit Administration,
U.S. Department of Transportation, Washington, DC
49. Kain J, Liu Z (1999) Secrets of success: assessing the large
increases in transit ridership achieved by Houston and San
Diego transit providers. Transp Res A 33:601–624
50. Thompson G, Matoff T (2003) Keeping up with the Joneses:
planning for transit in decentralizing regions. J Am Plan Assoc
69:296–312
51. Brown J, Thompson G (2009) Express bus versus rail transit:
how the marriage of mode and mission affects transit perfor-
mance. Transp Res Rec 2110:45–54
52. Bates T, Jablonski P (2007) San Diego, California, Trolley’s new
green line: early success for distinctive service. Transp Res Rec
J Transp Res Board Natl Acad Sci 2006:40–49
53. Mieger D, Chu C (2007) Los Angeles, California metro green
line: why are people riding the line to nowhere? Transp Res
Rec J Transp Res Board Natl Acad Sci 2006:50–59
54. Tennyson E (2005) Analysis of metro-link performance in Saint
Louis, Missouri. Transp Res Rec 1930:63–78
55. Giuliano G (2004) Land use impacts of transportation invest-
ments, highway and transit. In: Hanson S, Giuliano G (eds) The
geography of urban transportation, 3rd edn. Guilford Press,
New York, pp 237–273
56. Bernick M, Cervero R (1997) Transit villages in the 21st century.
McGraw-Hill, New York
57. Knaap G, Ding C, Hopkins L (2001) Do plans matter? The
effects of light rail plans on land values in station areas.
J Plan Educ Res 21:32–39
58. Bae CC-H (2002) Orenco station, Portland, Oregon: a success-
ful transit oriented development experiment? Transp Q 56(3):
9–18
59. Lund H, Willson R (2005) The Pasadena Gold Line: develop-
ment strategies, location decisions, and travel characteristics
along a new rail line in the Los Angeles region. Mineta Trans-
portation Institute, San Jose State University
60. Schrag Z (2006) Great society subway: a history of
the Washington metro. Johns Hopkins University Press,
Baltimore
61. Polzin S, Chu X (2010) A closer look at public transportation
mode share trends. Research and Innovative Technology
Administration, Bureau of Transportation Statistics. http://
www.bts.gov/publications/journal_of_transportation_and_
statistics/volume_08_number_03/html/paper_03/. Accessed
8 July 2010
Books and Reviews
Cervero R (1998) The transit metropolis: a global inquiry. Island
Press, Washington, DC
Committee for the Study on the Relationships Among
Development Patterns, Vehicle Miles Traveled, and Energy
Consumption (2009) Driving and the built environment:
the effects of compact development on motorized travel,
energy use, and CO2 emissions – special report 298. Transpor-
tation Research Board, National Research Council,
Washington, DC
Downs A (2004) Still stuck in traffic: coping with peak-hour traffic
congestion. Brookings, Washington, DC
Mees P (2010) Transport for suburbia: beyond the automobile age.
EarthScan, London
Pucher J, Lefevre C (1996) The urban transport crisis in Europe and
North America. MacMillan, London

MAGLEV Technology Development
JAMES R. POWELL1, GORDON DANBY2
1
Maglev-2000 Corporation, Shoreham, NY, USA
2
Maglev-2000 Corporation, Wading River, NY, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
The Physics and Engineering of Maglev
Maglev Applications
Future Directions
Bibliography
Glossary
Gap The physical clearance between a Maglev vehicle
and the guideway it travels along.
Guideway The structure that holds Maglev vehicles as
they travel. Guideways can be elevated above the
local terrain, or on-grade, depending on the design,
need for isolation from the local surroundings, and
other factors.
Maglev (magnetic levitation) Transport of vehicles
that are magnetically levitated above a guideway
and magnetically propelled along it.
Magnetic Induction The generation of induced
current in a conducting loop by the movement of
a magnet past it.
Magnets, electromagnetic The generation of
a magnetic field by a current in a loop of conductor,
usually wound around an iron core to increase the
strength of the magnetic field.
Magnets, permanent A physical object that is the
source of a magnetic field, without the need for an
electric current to produce the field. The movement
of orbiting electrons in the atoms of the permanent
magnet is locked in place to create the magnetic field.
Magnets, superconducting The generation of power-
ful magnetic fields from a loop or loops of zero
electrical resistance superconducting wire, usually
wound as an air core magnet without iron.
Superconductors Materials that when cooled below
a transition temperature completely lose electrical
resistance. They can carry very large currents at very
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
MAGLEV Technology Development 753
high current densities, hundreds of thousands
of amps per square centimeter of cross section,
indefinitely without any decay in their current and
without any voltage being applied to them.
Definition of the Subject and Its Importance
Maglev (magnetic levitation) is the first new mode of
Earth-based transport since the invention of the
airplane. Maglev vehicles are magnetically levitated
and propelled along a guideway, with no mechanical
contact and friction, their speed limited only by
atmospheric drag. In evacuated tunnels, Maglev
vehicles can potentially reach orbital speeds of 8 km/s
(18,000 mph). Traveling in the atmosphere, Maglev
vehicles have reached 580 km/h (361 mph).
Maglev vehicles have no engines, do not burn oil or
other fuels, do not emit pollutants and greenhouse
gases, and are quiet, with no rail or engine noise.
They are propelled by the interaction between AC cur-
rents in the guideway and the magnets on the vehicle.
The speed and position of Maglev vehicles on the
guideway is determined by the frequency of the AC
current that propels them. The distance between
individual Maglev vehicles on the guideway can be
designed to always remain constant, regardless of
what conditions the different vehicles encounter, that
is, up or down grades, curved guideways or straight,
head or tail winds, etc. The individual vehicles are not
controlled by on-board engineers. Instead a central
traffic control center views the entire guideway in
real time to detect potential hazards, controlling the
location and speed of all vehicles by means of the
frequency of the AC propulsion current fed to the
energized block of guideway that each individual vehi-
cle travels on.
Maglev vehicles can travel as individual single units
on the guideway, or during high levels of traffic, can be
coupled together to form multi-vehicle consists. Each
vehicle is individually propelled by the AC current
applied to the energized block of guideway the vehicle,
or the multi-vehicle, consist is traveling on.
There are two types of Maglev: electromagnetic and
electrodynamic. In electromagnetic Maglev the vehicles
are equipped with conventional electromagnets in
which a DC current winding energizes an iron structure
# 2012, DOI 10.1007/978-1-4419-0851-3
754 MAGLEV Technology Development
to create a strong magnetic field. The vehicle’s electro-
magnets are positioned underneath two iron rails
attached to the guideway. The magnetic force between
the vehicle’s electromagnets and the guideways iron
rails located above them levitate the vehicle.
Electromagnetic Maglev has three major limita-
tions. First, its magnetic suspension is inherently
unstable. As the vehicle electromagnets move closer to
the iron rails above them, the attractive force automat-
ically increases. To prevent a crash, the DC current in
the vehicle’s electromagnets must be reduced by a servo
control system that operates on a very fast time scale,
that is, thousandths of a second. The servo control
system thus continuously adjusts current to maintain
a safe gap between the vehicle and the guideway.
The second limitation is that to keep the DC magnet
power required to levitate the vehicle at an acceptable
level, the gap between the vehicle and guideway must
be very small, on the order of 1 cm (3/8 in.).
This requires a very precise and high cost guideway.
The third limitation is that the lifting capability of the
Maglev vehicle is small, because of the heavy magnet
and power system weights.
The second type, electrodynamic Maglev, uses
permanent magnets on the vehicle, not electromagnets.
Instead of having iron rails on the guideway, electro-
dynamic Maglev uses ordinary aluminum loops that
are located beneath the vehicle’s permanent magnets,
not above them. As the Maglev vehicle moves along the
guideway, its magnets induce currents in the aluminum
loops. These induced currents magnetically interact
with the vehicle’s permanent magnets, automatically
levitating it. The suspension is automatically and
inherently stable – the closer the vehicle gets to
the aluminum loops beneath it on the guideway, the
stronger the levitating force. With appropriate design,
Maglev vehicles using electrodynamic forces can be
strongly stable in the vertical, horizontal, pitch, yaw,
and roll directions, but able to move freely in the
longitudinal direction along the guideway.
Electrodynamic Maglev can use either conventional
permanent magnets or superconducting magnets. The
conventional permanent magnets are much weaker in
magnetic strength than superconducting magnets. As
a consequence, Maglev vehicles that use permanent
magnets have very small gaps between the vehicle and
the guideway, on the order of 1 cm (3/8 in.) and are
very limited in their capability to lift heavy weights.
In contrast, Maglev vehicles that use super-
conducting magnets have much larger gaps between
the vehicle and the guideway, on the order of 10 cm
(4 in.) or more, and have much greater weight-lifting
capability. Moreover, they are inherently very strong,
stable, and can be designed to resist hurricane force
winds and strong earthquakes without contacting
the guideway.
The technology for superconducting magnets is
commercially well developed. There are thousands of
superconducting MRI medical units all around the
world, and superconducting magnets are widely used
in R&D laboratories. The Large Hadron Collider
(LHC) at the CERN Laboratory in Switzerland, for
example, has 54 km (31 miles) of superconducting
magnets, all of which must function perfectly for the
LHC to operate. The same length of magnets on
Maglev vehicles would provide half of the highway
and airplane passenger miles in the USA, plus half of
the intercity truck long distance miles.
Japan is now operating a first-generation passenger
Maglev system in Yamanashi Prefecture that has carried
well over 80,000 passengers, with accumulated running
distances of hundreds of thousands of kilometers.
Japan plans to extend their system to become a
500 km (300 miles) route between Tokyo and Osaka
that will carry 100,000 passengers daily, with a trip time
of 1 h.
Germany has developed a first-generation electro-
magnetic Maglev passenger system called Transrapid.
A commercial 31 km Transrapid system now operates
between the center of Shanghai, China, and its airport.
An advanced second-generation superconducting
Maglev system, termed Maglev-2000, is being devel-
oped in the USA. The basic components – quadrupole
magnets, aluminum loop guideway panels, guideway
beam and vehicle – have been fabricated and success-
fully tested, but as yet funds for operational testing at
high speed on a guideway have not been available.
The second-generation Maglev-2000 system will
have important new and unique capabilities beyond
those already achieved by the present Japanese and
German first-generation Maglev systems that will
make it a major mode of sustainable transport in the

rest of the twenty-first century. These new capabilities
include:
● Transport of autos and trucks on Maglev vehicles,
enabling electric vehicles to make long trips without
the need to consume gasoline
● Much lower cost guideway system that can be
privately financed and will not require government
subsidies
● High speed electronic skip-stop switching of
vehicles to off-line stations for unloading/loading
operations enabling high average speeds with many
closely spaced stations
● Ability to adapt Maglev-levitated travel to existing
railroad tracks at very low cost, enabling Maglev
travel inside urban and suburban areas without
needing to build new infrastructure
● Ability to greatly reduce highway deaths and
injuries by transporting people, autos, and trucks
on Maglev rather than on highways
● Ability to use superconducting Maglev for other
important applications, including low cost storage
of electrical energy from variable wind and solar
farms; the long distance, low cost transport of
fresh water to drought areas; and very low cost for
launching payloads into orbit, particularly
space solar power satellites that can beam
massive amounts of clean power down to Earth
to meet power needs in an environmentally accept-
able manner
The implementation of Maglev as a major
mode of transport will enable much better sustainable
living standards for the world’s population, without
environmentally damaging the planet by the
massive use of fossil fuels.
Introduction
History and Impact of New Modes of Transportation
Over the last 200 years, the development of revolution-
ary new modes of transport have transformed human
society, greatly raising the standard of living and
enabling people to do things that were previously
undreamed of.
Maglev, magnetic levitation, is such a revolutionary
new mode of transport. It will dramatically change how
MAGLEV Technology Development 755
humanity transports people and goods, launch pay-
loads into space, store large amounts of electrical
energy generated by wind and solar sources, and deliver
billions of gallons of water over long distances to
drought stricken regions. Moreover, it will be sustain-
able for the long term, and not require the use of
destructive fossil fuels. In 1800, the only modes
of travel were walking, horses, wagons, and sailing
ships. In the USA, the per capita annual GDP was
only $1,200 (in constant 2007 dollars) and the annual
per capita travel distance was just 1,500 miles [1].
Today, the annual per capita GDP is $50,000
(in constant 2007 dollars) and the per capita annual
travel distance is 15,000 miles [1].
What happened? A series of revolutionary new
modes of travel that made it much faster, better, and
cheaper to transport people and goods. During the
Constitutional Convention, it took many days to travel
from New England to Philadelphia, and to cross the
ocean, many weeks. Even in the late 1800s, with steam
boats and railroads, it took Phileas Fogg 80 days to go
around the world.
The eight major transport revolutions initiated by
the USA were (Fig. 1):
1. The Erie Canal (1814)
2. Steamboats on the Hudson and Mississippi (1807
and 1811)
3. The Transcontinental Railroad (1869)
4. The Panama Canal (1914)
5. Henry Ford and the Model T (1908)
6. The Wright Brothers and Kitty Hawk (1903)
7. The Interstate Highway System (1956)
8. The Moon Landing (July 20, 1969)
Maglev will be an equally transforming new
mode of transport. On Earth, it will move people and
goods without being powered by fossil fuels that emit
massive amounts of greenhouse gases and pollutants,
in vehicles that are very safe, efficient, convenient,
quiet, and comfortable. Maglev will also store large
amounts of electric energy at very high efficiency
and low cost, transport very large amounts of water
for long distances over difficult terrain, and launch
payloads into space at very low cost, including space
satellites that can beam large amounts of power to the
Earth.
756 MAGLEV Technology Development
Erie Canal (1)
The Transcontinental Railroad (3)
Henry Ford and the Model T (5)
The Interstate Highway System (7)
MAGLEV Technology Development. Figure 1
The eight American transport revolutions
Why Fossil Fueled Transport Is Not Sustainable
Today’s Transport Systems – autos, trucks, airplanes,
most trains, and ships – are critically important to
present society and its standard of living. All are
Steam boats Hudson and Mississippi (2)
The Panama Canal (4)
Wright Bothers at Kitty Hawk (6)
The Moon landing (8)
critically dependent on oil. Without oil, humanity
would be back in the 1700s, with only horses and
wagons and sailing ships.
Since Edwin Drake started pumping oil in Western
Pennsylvania in 1859, the world has consumed
 MAGLEV Technology Development 757

approximately one trillion barrels of oil, most of it for
transport. How much is a trillion barrels? Visualize
a lake of oil that is one square mile in cross section
containing a trillion barrels of oil. The oil lake would be
40 miles deep.
Another trillion barrels of oil is still in the ground,
but it will not sustain the world for very long. World oil
production has plateaued at about 30 billion barrels per
year and will soon start to steadily decline. More sig-
nificant than the arguments over when the “peak oil”
crisis will occur is the “oil competition” crisis. As
countries like China and India rapidly industrialize,
competition with the developed countries for ever
scarcer oil will drive up the price of oil. Today, highway
traffic in China is increasing 18% annually, and more
cars are sold per year in China than in the USA. Today
(2010) oil costs about $80 per barrel; in a few years,
analysts predict $200 per barrel.
For most of the rest of the twenty-first century, the
present world transport systems will not be sustainable,
using conventional oil for fuel.
Can society sustain existing autos, trucks,
airplanes, trains, and ships with some alternative fuel
as the oil runs out? Biofuels? Hydrogen? Synthetic fuels
derived from coal, oil shale, tar sands, methane
hydrates from ocean beds? No. None of these will be
able to maintain society’s present transport
systems powered by the internal combustion engine.
There are a variety of reasons why these alternative
fuels are not practical on the scale needed to support
a world population of nine billion people in 2050
(Fig. 2).
Biofuels? Not practical for the amounts needed.
The acreage required is prohibitively large. In a world
where there is not enough arable land to adequately
feed today’s population of 6.6 billion people, which is
projected to be 9 billion by 2050, biofuels can supply
only a tiny fraction of future transport needs. Not only
is the population growing, but the fertility of the
world’s arable land is declining, due to over use, ero-
sion, increasing acidification due to excessive use of
fertilizers, and increasing droughts. Humanity will be

THE ELECTRIC TRANSPORT ROAD

TRANSPORT 2010:
LOTS OF CHEAP FUEL

HIGH ELECTRIC
SPEED CARS &
OR
RAIL MAGLEV
TOO SMALL STABLE ENVIRONMENT
FOR TRANSPORT AND HIGH LIVING
NEEDS STANDARDS

THE INTERNAL COMBUSTION ENGINE ROAD

BIOFUELS HYDROGEN SYNFUELS

NOT PRACTICAL- NOT PRACTICAL- GLOBAL WARMING

ONLY TINY SAFETY AND AND
AMOUNTS ENERGY PROBLEMS MASS EXTINCTION

MAGLEV Technology Development. Figure 2

The decision tree for future transport
758 MAGLEV Technology Development
lucky to avoid massive famines in the years ahead, even
if biofuels are not produced.
Hydrogen fuel would require 1,000 new nuclear
reactors, each of 1,000 MW electrical capacity – ten
times more than now operated in the USA – to make
enough hydrogen fuel to equal the gasoline and diesel
fuel currently consumed in USA’s transport systems.
Hydrogen cars raise a serious security issue. Over
one million cars are stolen in the USA per year. After
stealing a hydrogen car, a terrorist need only attach
a small bomb with an explosive-driven penetrator to
the hydrogen tank, and then park it in a public parking
garage, or shopping mall, or on a busy city street. The
penetrator bomb could be then remotely detonated by
cell phone or an attached timer, releasing the hydrogen
gas to cause a massive explosion. Result? Many deaths
with no way to prevent them. A terrorist would need no
special skills. The penetrator bombs could probably be
bought in the black market.
While more expensive than oil, synthetic gasoline
and diesel fuel can be manufactured from coal, tar
sands, oil shale, and methane at an acceptable cost.
The reserves are enormous and could sustain oil-fueled
transport for hundreds of years.
The problem with synfuels is not the sustainability
of fuel supply, but sustainability of Earth’s environment
due to the release of carbon dioxide to the atmosphere
from the manufacture of the synfuels, and the release of
carbon dioxide from the tailpipes of the vehicles that
burn the synfuels. Today, the US transport sector emits
two billion tons per year of carbon dioxide into the
atmosphere [2], roughly one-third of total US
emissions. The carbon dioxide emissions from
the manufacture of the synfuels needed to replace
the oil now consumed by US transport would add
another two billion tons per year to the atmosphere,
for a total transport emission of four billion tons
annually.
Would leaders call for a reduction of 80% in carbon
dioxide emissions from today’s levels, to be achieved by
2050? This corresponds to a total world emission of five
billion tons annually in 2050. Four billion tons of just
US transport emissions at today’s level of transport
would be four-fifth of the world’s total.
With synfuels for transport, there is no way to
reduce carbon dioxide emissions below today’s level,
even if all other present sources – coal-fired power
plants, industry and home heating, etc. – were to go
to zero. In fact, as the developing countries industrial-
ize, their transport needs will greatly increase, making
the world’s total emissions from transport alone in
2050 much greater than today’s total world emissions
of 25 billion tons per year [3].
There is no practical way to capture the carbon
dioxide emissions from autos, trucks, airplanes,
trains, and ships. Inevitably, they will enter the
atmosphere.
The climate effects of global warming are
very bad – higher temperatures, melting glaciers and
ice sheets, increased droughts and food shortages, ris-
ing sea levels and flooding, species shifts and extinc-
tion, etc. Even worse, at some point, high levels of
carbon dioxide emissions risk environmental catastro-
phe that cannot be stopped, even if all emissions went
to zero.
As global warming from the release of greenhouse
gases continues, the ocean is becoming more acidic,
threatening sea life. Natural carbon dioxide emissions
from decaying organic matter in the permafrost
regions, plus thermal decomposition of unstable meth-
ane hydrates in the sea bed, may cause an irreversible
runaway warming of the Earth.
Humanity should not risk global disaster by
continuing to emit massive amounts of carbon
dioxide from its transport systems. A transition to
electrically powered transport must soon start, with
the electricity coming from nonpolluting energy
sources (Fig. 2).
Maglev and Electric Transport: A Necessity for
Long-Term Sustainability
There are three options: electric cars and trucks, electric
rail, and Maglev. Electric cars are desirable and
practical, but have limitations. The Chevrolet Volt, for
example, which will soon be commercially available,
has a maximum range of about 40 miles before it must
be recharged or shifted to an auxiliary gasoline engine
to extend its range [4]. Longer trips almost certainly
will use the gasoline engine because nobody will want
to stop every 40 miles or so and wait several hours for
a recharge.
There are other limitations. In colder regions,
batteries should not be charged or discharged unless

they are warmed, as this may degrade them. Heating
the car from the battery will reduce the car’s range.
In hot regions, running the air conditioner from the
battery will also reduce range. The battery pack will be
very heavy, hundreds of pounds, increasing the weight
of the car.
In combination with Maglev, electric cars and
trucks can provide practical transport, for both short
and long trips. The electric cars can operate locally for
trips of a few miles to the nearest Maglev station, there
to drive onto a Maglev vehicle that will transport the
drivers and their automobiles to any other station in
the continental USA at speeds of 300 mph. At the
station nearest the destination, the car would be
driven off the Maglev vehicle, a few miles by highway
to the final destination.
During the trip on the Maglev vehicle, the electric
cars would be fully recharged, using on-board power,
and ready to travel once driven off. The cost to drivers
traveling on a high speed Maglev trip would be less than
if driving by highway with a gasoline-powered engine.
Maglev would provide the same kind of service for
long distance trucks. The average haul distance for
intercity trucks, which currently transport 1½ trillion
ton miles of freight per year at an annual cost of over
$300 billion, is 500 miles. There is no way that electric
trucks can make that haul distance if traveling by
highway.
With a National Maglev Network, almost all travel
could be electrically powered, with long distance travel
for autos, trucks, and passengers by Maglev, while
electrically powered autos and trucks handled short
local trips. There would be little domestic air travel,
with most flights being across the oceans and to remote
areas not served by Maglev.
Electrically powered conventional rail will still
transport bulk freight – grain, cement, etc. – but, the
role of high speed rail (HSR) will be limited. This is
because HSR only carries passengers, is inherently
expensive, and must be subsidized by the government,
even in Europe and Japan. Even in countries like France
which is much smaller than the USA, with considerably
high population density, for example, HSR plays only
a small role in transport [1]. Per capita, on average
the French travel only 400 miles per year on the TGV,
while driving 7,600 miles per year and only taking a 1¼
trip per year on the TGV.
MAGLEV Technology Development 759
The Physics and Engineering of Maglev
The Physics of Maglev
It has been well known for well over a century, since the
beginning of the electrical age, that electromagnetic
forces can be very strong, even stronger than
gravity. This principle makes magnetic levitation
possible. Moving charged particles, which usually
means electrons, create magnetic fields. Matter con-
tains atomic electrons perpetually whirling around in
orbits. Most materials are largely nonmagnetic because
the moving electrons cancel each others’ magnetic
fields, even when an external magnetic field is applied.
A few materials are ferromagnetic. This material
responds to external magnetic fields by aligning the
electron orbits to produce very strong induced mag-
netic fields within the ferromagnet that extend beyond
it into the surrounding air. When the external magnetic
field is removed, the ferromagnetic object essentially
returns to the unmagnetized state. It is fortunate that
iron, a common and cheap structural material, is
strongly ferromagnetic. This enabled the creation of
the generators and transformers that made AC electric
systems very practical.
It was quickly realized, however, that stable suspen-
sion in air is not possible with ferromagnetic materials
using static, that is, non-time-varying, external
excitation. This is known as Earnshaw’s Theorem.
The unstable force causes magnetized iron to clamp
itself to another piece of iron with a force which gets
stronger as it gets closer. This is clearly not suitable for
levitation.
Permanent magnets are familiar to people and can
be used for levitation. These magnets are made
of unusual materials which resist easy magnetization
or demagnetization. When exposed to very strong
external magnetic fields during manufacture, a large
fraction of the electron orbits line up parallel and retain
strong magnetization alignment after the charging
magnetic field is removed.
Two such permanent magnets will attract each
other when the magnets’ electron orbits are parallel.
However, if one magnet is turned over, the magnets will
strongly repel each together. As the magnets get closer,
the repulsive force becomes stronger. At some distance,
the repelling force is sufficiently strong to equal that
of gravity.
760 MAGLEV Technology Development
This then leads to suspension at a height where the
repulsion force equals that of gravity. However, without
additional constraint, this type of suspension is
not practical for levitation. The suspended magnet
is unstable at right angles to the direction of gravity.
A familiar example is two donut-shaped permanent
magnets centered on a stick. With the magnetic
polarity opposing, the lower magnet will suspend the
upper magnet. However, if the lower magnet is moved
up on the stick to the point where the upper magnet
leaves the top of the stick, the latter will be violently
thrown off to the side.
Figure 3 shows how stable levitation can be
produced. This is a familiar Maglev model construction
for school science projects. The side mounted magnets
and the vertically mounted magnets in combination
create forces which are highly stable in both directions.
For a real Maglev transport system, low cost familiar
ferrite magnets can only levitate a very small weight.
Moreover, to build many miles of permanent magnet
guideway would require an enormous volume of very
expensive material to be purchased, installed, and
maintained. There is a class of Rare Earth permanent
magnets which produce forces at least ten times greater
than ferrite magnets. However, this material which
weighs as much as iron costs hundreds of times more
per pound than iron. In summary, for a robust Maglev
system with substantial suspension clearances that can
levitate heavy weights, using a permanent magnet
approach is not practical.
Opposing
Vehicle
Magnet
Pairs
Guideway
MAGLEV Technology Development. Figure 3
Magnetic suspension based on the repulsive force
between permanent magnets on a vehicle
Let us return to the ferromagnetic circuit which is
statically inherently unstable. Figure 4 illustrates that if
the applied magnetizing current in a coil is electrically
varied over time, the unstable force can be controlled so
that a practical suspension is possible. Sensors contin-
ually monitor the air gap and signal the power
supply to adjust the circuit current in a suitable time-
dependent manner so that the air gap is maintained at
a constant value. This is the system used by Transrapid
[5]. As was the case for permanent magnets shown in
Fig. 3, both horizontal and vertical magnets are
required for stability in both directions. On each side
of the guideway, magnets attached to the vehicle are
attracted upward to the bottom of the iron rails on the
guideway, providing levitation. Magnets of each side
are attracted sideways to the edge of the guideway, as
illustrated in Fig. 5, providing lateral guidance.
The operating gap between the vehicle magnets and
the iron rails on the guideway is typically very small
with the electromagnet iron rail system, on the order
of 1 cm (3/8 in.).
Another type of magnetic suspension results from
applying alternating current (AC) to a current-carrying
coil located above a sheet of electrically conductive
normal metal, such as copper or aluminum. If the
frequency is sufficiently high, currents are induced in
the metal sheet that keeps the AC magnetic field from
penetrating the sheet. These induced currents are
opposite in direction to the currents in the AC coil
above the sheet. The magnetic interaction between the
opposing currents produces a repulsive force similar to
that of opposing permanent magnets which can levitate
Fe Gap
Sensors
Air Gap
Fe
Current
Control
Power
Supply
MAGLEV Technology Development. Figure 4
Magnetic suspension based on servo control of the
attractive force between an electromagnet and
a ferromagnetic sheet
 MAGLEV Technology Development 761

Current
Loop (ac)

d
Aluminum
Sheet
Images

MAGLEV Technology Development. Figure 6

Magnetic suspension based on the repulsive force
between a coil carrying alternating current (AC) and
a conducting sheet

MAGLEV Technology Development. Figure 5
Cross section view of Transrapid Maglev vehicle showing
electromagnetic suspension
the AC coil (Fig. 6). As early as 1912, Batchelet [6]
obtained a patent on such a system for vehicle
transport. While not practical for transport because of
its very inefficient use of energy, it does work from the
physics viewpoint.
The concept of induction brings us to electrody-
namic Maglev. Note that induction occurs because of
time-varying applied magnetic fields. The usual
example is an applied AC current, for example, as
seen in transformers, which make our electric utility
power distribution practical. In the case of Maglev,
static (DC) magnets on a vehicle moving along the
guideway can also apply a time-varying, that is,
AC field to normal conductors that they pass on the
guideway.
Powell and Danby shared a house from 1960 to
1962. This was about the time superconducting wires
capable of producing strong superconducting magnets
were first becoming available. Dr. Powell had studied
the use of superconductors positioned on a guideway
and the vehicle for Maglev. Then Dr. Danby provided
Conducting
+ W0 – Sheet
Y0
– – – – – – – – – – – +++++++++++
Y0
– W0 +
Distributed Induced
Currents In
Image SC Loop
Guideway Sheet
MAGLEV Technology Development. Figure 7
Superconducting (SC) inductive suspension conducting
sheet guideway
key electrodynamic insights into using induction from
superconducting magnets on a vehicle to generate
currents in normal conductor loops on the guideway.
Together Danby and Powell proceeded to develop and
patent superconducting electrodynamic Maglev for
which the Benjamin Franklin Medal for Engineering
in April 2000 was awarded [7].
These innovations are described on the following
figures. Figure 7 shows a superconducting coil
moving above a conductive sheet. This configuration,
while possible, is less desirable, but is readily visualized.
Figure 8 shows the conductive sheet replaced by a
bundle of wire forming a shorted coil. Figures 9–13
illustrate a variety of the many configurations for use of
what the authors’ term “null flux” loop geometries.
762 MAGLEV Technology Development
These null flux configurations, as discovered by the
authors, have proved to be a very desirable approach
for superconducting electrodynamic Maglev. The use
of strong superconducting magnets enables the vehicle
to levitate very heavy loads while at the same time
keeping electrical I2R losses in the normal guideway
conductors at a minimum level [7, 8].
In Fig. 9 the superconducting (SC) magnet loop,
which is attached to the vehicle, is shown halfway
between the guideway loops. The induction in the top
and bottom guideway loops is equal and opposite,
since these loops are added in series opposing. As
a result, no current flows, no force is exerted, and no
power is consumed. As the vehicle magnet drops down
because of the vehicle weight, a current and force
proportional to displacement occurs. The vehicle is
Actual SC Loop
(Typ) + –
W0
Guideway
Y0
– +
Induced Current In
Simple Multiturn
Guideway Loop
MAGLEV Technology Development. Figure 8
Superconducting (SC) inductive suspension simple loop
guideway
+ –
Y0
+ W0
Y0
– +
Surface Of
Guideway
MAGLEV Technology Development. Figure 9
Series opposed loops
suspended with a restoring force as in a spring. In
Fig. 10, two loops on the guideway are mounted side
by side and connected in series opposition. If the
SC loop on the vehicle is centered, there is no net
induction or current or force. This gives a strong linear
restoring force sideways as the vehicle moves sideways
off center. Figure 11 is identical to Fig. 10, but oriented
at 90 . Now the force is vertical instead of horizontal.
Figure 12 shows another example. When centered there
is no induction or sideways force due to the guideway
loop. Figure 13 shows canceling of the force between
the two sides when the vehicle is centered. In all the
above cases, strong restoring forces are developed
proportional to the displacement from the ideal posi-
tion of the vehicle, that is, centered. The figure of
8 configuration is used in the very successful Japanese
Maglev [9].
Why is the null flux principle so powerful? For the
simple loop of Fig. 8, the geometry and the total
ampere turns in the superconducting magnet
completely determine the induced current in the
shorted guideway loop. With the null flux configura-
tion, the number of ampere turns in the vehicle SC
magnets can be much greater than the guideway
ampere turns and still have a very strong suspension.
This reduces material requirements for the guideway
coils and results in low I2R losses in the guideway.
Stronger SC magnets on the vehicles are well within
the state of the art. These stronger magnets in turn
result in less power needed for the linear synchronous
Top Half of Multiturn
Null Flux Loop
Circuit
Structural Arm
–
Connecting
SC Loop To Vehicle
Bottom Half of Multiturn
Null Flux Loop
Circuit
 MAGLEV Technology Development 763

Gravity SC Loop Throughout the history of technological evolution,

ideas that were known in principal had to await new
+ – technical advances to make them truly practical. That
is the case with Maglev. Early visionaries of Maglev
1 1 2 2 over the last 100 years include Bachelet [6],
Goddard [10], and Kemper [11]. Lightweight and
Horizontal Surface Of Guideway
strong superconducting magnets provided a great step
#1 and #2 loops wound in opposite forward for electrodynamic Maglev. The materials that
directions and coupled to form made this possible appeared about 1960. This in com-
a complete circuit bination with the authors’ null flux loop guideway
MAGLEV Technology Development. Figure 10 made Maglev eminently practical. Similarly, the
Horizontal figure of 8 development of modern power supplies, controls, and
sensors permits fast-acting electronic control of the
iron magnets of the Transrapid system which has
Gravity been demonstrated in Germany and commercially
1
implemented in China.
From the Maglev point of view, a superconducting
+ winding behaves like a very light and strong permanent
Vertical magnet which is located in a very low temperature
1
Side container, or cryostat. Superconductivity is now in
Of Guideway routine use, especially in the MRI scanners found in
2 hospitals and doctors facilities. Figure 14 shows the
–
history of superconducting development. At the very
low temperatures of liquid helium, 4.2 K, Kammerlinge
SC Loop Ohnes [12] discovered that various pure metals had no
2 resistance to current flow. In turn, the flowing electron
MAGLEV Technology Development. Figure 11
current generates magnetic fields without the I2R losses
Vertical figure of 8
experienced in normal conductors. While the discovery
was of great interest to physicists, it was almost 50 years
before superconductors were developed which could
SC Loop carry large currents in the presence of high magnetic
fields as necessary for magnets. The most common
+ – superconducting wire for magnet builders for the last
several decades has been Niobium-Titanium (NbTi)
followed by Niobium 3-Tin (Nb3-Sn).
Figure 15 shows the behavior of NbTi alloy.
The lower the operating temperature, the greater the
Guideway superconducting current and magnetic field that can be
Loop sustained, that is, the more powerful the magnet capa-
MAGLEV Technology Development. Figure 12 bility. Figure 16 shows the behavior of these two most
Central orthogonal loop (horizontal or vertical orientation) common superconductors at the temperature of boil-
ing helium. Note the much greater current densities
and resulting magnetic fields that can be sustained
motor discussed later. Moreover, because of the very compared to conventional iron electromagnets.
efficient null flux levitation configuration, most of the Figure 17 shows NbTi wire drawn down to a circular
propulsion power required for Maglev vehicles at high or an approximately rectangular form. The small dots
speeds is due to air drag, not electrical loss. are the NbTi superconductor. The surrounding
764 MAGLEV Technology Development

SC SC
Loop Loop
+ +
1 Center 2
Vehicle Section Vehicle
Vertical
Side Arm of Arm
Of Guideway Vehicle
– –
1 2

#1 and #2 loops wound in opposite

directions and coupled to form
a complete circuit

MAGLEV Technology Development. Figure 13

Cross-coupled lateral loops

140 Superconductivity: TiBaCaCuO

DC Resistance = 0 TiSrCaCuO
120
BiSrCaCuO
Transition Temperature

YBaCuO
100
Space Temperature
80
Liquid Nitrogen
60

40 LaSrCuO

NbN NbGe
20 Liquid
Hg
Helium
0
1900 1920 1940 1960 1980 2000 Year

High Field/Current
Meissner BCS
Discovery Superconductivity
Effect Theory

MAGLEV Technology Development. Figure 14

The time evolution of the superconducting transition temperature

material is usually copper. This cross section starts out
as a short cylinder about 10 in. in diameter. It is then
drawn down to a small wire by pulling through orifices
of diminishing sizes. The superconducting wire and
cryostat technology required for Maglev has existed in
the USA for several decades. For example, Fig. 18a
shows a large superconducting magnet system which
was operated for over a decade as a central part of
a high energy physics facility. It was designed and
built by Gordon Danby and his group and
commissioned in 1973. Figure 18b shows a large
superconducting magnet system used as a high energy
 MAGLEV Technology Development 765

Current Density (kA mm4)

3

) 1
(°K
tur
e 2
era 2
mp 4 Fie
Te ld
4 6 (te
sla
)
6 8
8 10
10 12
14

16

MAGLEV Technology Development. Figure 15

The critical surface for niobium-titanium superconductor

particle storage ring for a fundamental physics study of
the anomalous magnetic moment of the muon,
a fundamental point-like particle.
The largest and longest particle accelerator that uses
superconducting magnets is the Large Hadron Collider
(LHC) now operating at the CERN Laboratory in
Switzerland [13]. The LHC has two rings of many
thousands of superconducting magnets that accelerate
and guide high energy particles for experiments.
The total circumferential length of the magnets is
54 km (31 miles). The same length of superconducting
magnets on Maglev vehicles could transport half of the
4 trillion passenger miles per year traveled by car,
airplane, and rail in the USA, plus half of the 1.5 trillion
ton miles per year of freight transported by intercity
highway trucks.
The LHC superconducting magnets are much
more technically complex and challenging than
superconducting Maglev magnets. To achieve maxi-
mum field capability, LHC superconducting magnets
operate with subcooled liquid helium at 1.7 K, rather
than the normal boiling point of 4.2 K, where Maglev
magnets operate. The refrigeration requirements at
1.7 K are much more difficult than at 4.2 K. Moreover,
766 MAGLEV Technology Development

operation of the other magnets on the vehicle. Several

1010
of the multiple independent magnets on the vehicle
could fail without compromising safety. If a magnet
were to fail, the vehicle would be taken off-line at the
Current Density (Am–2)

Nb3Sn next station for repairs. The probability of multiple

109
NbTi
108 Conventional
Iron-Cored
Electromagnets
107
0 2 4 6 8 10 12 14 16 18 20 22
Magnetic Field (tesla)
MAGLEV Technology Development. Figure 16
Critical current densities of niobium-titanium and niobium-
tin at a constant temperature of 4.2 K
a b
MAGLEV Technology Development. Figure 17
Cross sections of two typical filamentary composites of
NbTi in a copper matrix
the LHC superconductors must operate at higher mag-
netic field strengths and with much greater spatial
precision than Maglev magnets again making them
much more challenging. Finally, the many thousands
of LHC magnets must all work perfectly for the LHC
facility to operate. In Maglev, the vehicles carry multi-
ple independently operating magnets with very high
system redundancy and robustness. If one magnet were
to fail, a very unlikely event, it would not affect the
failures during this time would be many orders of
magnitude smaller than the probability of multiple
engine failures on commercial jet aircraft, which in
turn are extremely small.
Returning to the cost of the magnets for
superconducting Maglev, a good measure is the cost
of the magnets for the LHC. The total capital cost of the
much more complex and difficult LHC magnets was
several billion dollars. The equivalent length of mag-
nets on Maglev vehicles would carry half of the annual
US passenger and intercity highway freight traffic,
which presently costs on the order of $1 trillion per
year [14]. The amortized cost of the Maglev magnets is
trivial in comparison. Also, the annual cost of the
40,000 deaths and 500,000 serious injuries per year on
US highways is estimated at $500 billion annually [1].
The cost of superconducting Maglev magnets is again
trivial by comparison, plus the vast amount of human
suffering that would be avoided by traveling on much
safer Maglev systems.
The newer high temperature superconductors
(HTS) shown in Fig. 14 are extremely promising.
HTS conductors operate at much higher temperatures
than the 4.2 K for NbTi, typically in the range of
30–40 K, with the capability of operating at 77 K, liquid
nitrogen temperatures, depending on the desired
current capacity. Presently, the most promising HTS
conductor is yttrium barium copper oxide (YBCO).
Deposited as a very thin film, 1–2 m in thickness, on
a thin tape substrate, YBCO conductors can achieve
engineering current densities of 50,000 A/cm2 or more,
depending on operating temperature. Engineering
current density is the overall current density of the
conductor tape, including the superconductor film,
which has a current density of millions of amperes
per square centimeter, plus the substrate tape that it is
deposited on.
HTS conductors are attractive for Maglev, because
their refrigeration systems are simpler and have lower
energy requirements. The present Japanese Maglev
system uses NbTi superconductor at liquid helium
temperatures without problems, and the refrigeration
 MAGLEV Technology Development 767

MAGLEV Technology Development. Figure 18

(a) Large superconducting magnet systems for a high energy physics facility. (b) Large superconducting magnet system
used as a high energy particle storage ring for study of the magnetic moment of muon particles
768 MAGLEV Technology Development
requirements are modest and very acceptable. As
HTS conductors come into wider use and larger
production, superconducting Maglev will probably
shift to their use.
A further question could be raised by people
unfamiliar with superconducting technology. Is it too
complex for everyday transportation systems? One area
where superconductivity is intensively used is MRI
medical scanners. The technology of superconducting
coils and cryogenics has evolved considerably in
scanners over the last 20 years, and thousands of
superconducting MRI magnets are now in use all over
the world. Figure 19 shows an open MRI, for which
Gordon Danby was the leader of the magnet design
effort. The important thing for Maglev is that the
commercially made superconducting coils in the MRI
scanner as yet require elaborate refrigeration systems.
Two small cryocoolers approximately a foot in diame-
ter and an arm’s length cool the coils. Once a year the
cryocoolers are taken off and replaced with another set.
The first set is sent back to the manufacturer for
MAGLEV Technology Development. Figure 19
Magnetic Resonage Imaging (MRI) machine for medical diagnostics, using superconducting magnets
reconditioning. No cryogenic fluid has to be added
during this year of continuous trouble-free operation.
This example shows that superconductivity can be very
reliable, just like the refrigerators in people’s kitchens.
It is definitely ready for efficient, reliable transportation
with extremely low maintenance.
Engineering Maglev
If Maglev is to become a major mode of transport in the
twenty-first century, as important as road, rail, air, and
water have been in the twentieth century, it must be
safe, reliable, cost effective, environmentally benign,
operate as a nationwide network, not a small number
of isolated routes, be readily accessible, and integrate
easily with other transport modes. To fulfill Maglev’s
promise, the proposed Maglev systems must be
engineered with respect to the following six develop-
ment goals.
The first development goal is to have a substantial
physical clearance between the Maglev vehicle and the

guideway. Large clearances are desirable for safe oper-
ation, and allow Maglev vehicles to operate in all types
of weather, including strong winds, snow, and ice,
that are not possible with conventional transport.
Superconducting Maglev enables clearances of 10 cm
(4 in.) or more between a vehicle and the guideway.
Electromagnetic Maglev is constrained to much
smaller clearances, on the order of 1 cm (3/8 in.).
Also, large clearances reduce guideway cost because
construction tolerances are less demanding.
Building guideways to a tolerance of 10% of the
vehicle/guideway clearance is much cheaper when
the clearance is 4 in., compared to guideways
where the clearance is a fraction of an inch.
The second development goal is to have a very
stable and safe magnetic suspension. Electromagnetic
Maglev systems, such as Transrapid, achieve stability by
very fast, for example, thousandths of a second, servo
control of the current in the electromagnets on the
vehicle. If the gap between the vehicle electromagnets
and the iron rails becomes smaller than desired, the
magnet current is reduced. If the gap becomes too
large, the magnet current is increased. In electrody-
namic superconducting Maglev, the vehicle
superconducting magnets and the normal conductor
guideway loops are arranged so that the moving vehicle
will automatically be magnetically suspended at an
equilibrium position above the guideway. Any external
force on the vehicle – for example, a cross wind or the
centripetal force on a curve – that acts to displace the
vehicle from its equilibrium position is immediately
countered by an inherent, passively generated opposing
magnetic force. This opposing magnetic force always
acts to push the vehicle back toward its equilibrium
position. The magnitude of the displacement from
equilibrium is proportional to the strength of the exter-
nal force. Electrodynamic superconducting Maglev can
be designed so that no conceivable external force, even
if it were twice the weight of the vehicle, would produce
a displacement great enough to make the vehicle con-
tact the guideway. In effect, superconducting Maglev
creates a five-dimensional “magnetic well” in which the
vehicle is trapped so strongly that any conceivable
external force cannot make it climb out of the “well.”
The five dimensions correspond to vertical (“heave”),
lateral (“sway”), pitch, yaw, and roll movements of the
vehicle. The vehicle is not constrained in the sixth, or
MAGLEV Technology Development 769
longitudinal, dimension, so that it can move freely
along the guideway, with only a small magnetic drag
force acting on it.
Electromagnetic Maglev systems, that is,
Transrapid, maintain their levitation at all speeds,
including zero speed at stations and if the propulsion
power to the guideway is cut off, subject to the electri-
cal capacity of the batteries on-board the vehicle.
Superconducting Maglev vehicles can automati-
cally and inherently remain stably suspended as long
as they move faster than a transition speed of approx-
imately 20 mph. Below this speed, resistive IR losses in
the guideway loops reduce the induced currents in the
guideway loops to the point where the magnetic force is
insufficient to levitate the vehicle.
Superconducting Maglev vehicles can remain levi-
tated at zero speed and speeds below 20 mph on the
short sections, for example, 50 m in length, of guideway
at stations where the vehicle is decelerating into
a stop at the station or accelerating out of the station.
To remain levitated in those special sections, one can
power the aluminum guideway loops with DC current
or use high temperature superconducting (HTS)
guideway loops in place of aluminum loops. The HTS
guideway loops will only require a small fraction of the
HTS conductor used on the vehicle, and their refriger-
ation power will be small. Maglev vehicles are
magnetically propelled by a set of alternating current
propulsion windings in the guideway. In the unlikely
event that the electrical power grid that energizes the
guideway propulsion windings was to fail, vehicles
would simply coast to a safe stop at a special location
on the guideway or at the next station where the vehicle
remains levitated.
The third development goal is to have the vehicles
operate with very low magnetic drag. Low magnetic
drag reduces both the energy consumed by the mag-
netic propulsion system, as well as its capital cost.
Because the energy consumed by the superconducting
magnets on the vehicle is negligible – the electrical
resistance of superconducting windings is zero – low
magnetic drag can be achieved by having the magni-
tude of the currents induced in the guideway loops be
much smaller than the magnitude of the currents in the
superconducting magnets on the vehicle. The magnetic
levitation force on the vehicle is proportional to the
product of vehicle magnet current and guideway loop
770 MAGLEV Technology Development
current. Maximizing vehicle current relative to guide-
way loop current minimizes magnetic drag since it
minimizes the I2R losses in the guideway loops, the
only contributor to magnetic drag. By using a unique
configuration of guideway loops and vehicle magnets,
termed the null flux suspension, to be described later,
the guideway currents can be made very small, resulting
in magnetic drag forces that are negligible compared to
air drag.
Following publication of Powell and Danby’s
pioneering 1966 paper [15] on superconducting elec-
trodynamic Maglev, Kolm and Thornton proposed
a different Maglev system, called the Magneplane, that
also used superconducting magnets on a vehicle to
provide levitation and propulsion. Instead of using
null flux aluminum guideway loops, the Magneplane
[16] used a curved solid aluminum sheet as the guide-
way underneath the Maglev vehicle. The magnetic
fields from the superconducting magnets on the vehicle
induced currents in the aluminum sheet guideway. The
magnetic interaction between the induced currents in
the aluminum sheet guideway and those in the
superconducting magnets on the vehicle generated
the levitation force.
However, in this configuration, the magnitude of
the induced currents in the sheet guideway equaled the
magnitude of the currents in the vehicle magnets,
resulting in large I2R losses in the aluminum sheets
and strong magnetic drag in the vehicle. Passenger
aircraft typically have a lift to drag ratio on the order
of 20/1, resulting in high energy consumption per
passenger mile. The magnetic lift/drag ratio for the
Magneplane is comparable to that for airplanes, again
resulting in high energy consumption.
In comparison, by using the null flux guideway
configuration [7, 15], the induced currents in the
guideway loops are much smaller than those in the
vehicle magnets, resulting in much less magnetic drag
due to I2R losses in the in the guideway conductors,
and a much larger magnetic lift/drag ratio. Also,
the amount of aluminum used in the null flux config-
uration is much less than in the Magneplane
configuration.
The fourth development goal is low capital and
operating costs. Most – approximately 80% – of the
projected total life cycle cost of a Maglev system is
associated with the construction cost of the guideway.
A low cost for the guideway appears feasible through
the use of prefabricated beams and piers which are
mass produced in factories and shipped to the route
site where they can be easily assembled into the finished
guideway. Not only will the prefabricated, factory-
produced beams and piers be substantially lower in
cost than if they were fabricated at the construction
site, but their quality control and tolerances will be
better. The prefabricated guideway loops can be
attached to the sides of the beams before they are
shipped to the route site, which further reduces cost.
The finished prefabricated beams and piers can be
shipped by truck, or transported along completed por-
tions of the guideway to where the new sections are
being assembled.
The fifth developmental goal is to achieve high
revenues from the Maglev system, so as to avoid the
need for government subsidies for construction and
operation of Maglev routes. Today, as government
budgets are tightening and deficits and debt are grow-
ing, there is great resistance to having government
finance new transportation projects. In the USA,
intercity and commuter rail systems all require large
subsidies for operation, with fare revenues only
accounting for 30–40% of total operating cost. Simi-
larly, the high speed rail (HSR) passenger transport
systems in other countries all require major govern-
mental subsidization. For Maglev to be widely
implemented, it should require little or no subsidiza-
tion, and as a goal, be attractive to private investment,
without the need for any government funding.
The sixth developmental goal is high speed,
nonmechanical switching to off-line stations for
unloading/loading operations. Requiring high speed
vehicles to operate with online stations greatly limits
their convenience and number of stations. Decelerating
into and accelerating out of stations greatly reduces the
average travel speed unless the stations are far apart, for
example, spaced every 50–100 miles or so. Long dis-
tances between online stations reduce accessibility and
ridership, while short distances negate the benefits of
high speed systems. The ability to switch at high speeds
to closely spaced off-line stations, when scheduled, and
to bypass them at high speeds when not scheduled,
enables a high speed transport system to maintain
high average speeds while at the same time having
many closely spaced stations with easy accessibility.

Superconducting Electrodynamic Maglev
Enabling Inventions
• Magnetic levitation using superconducting
magnets and inductive normal guideway
• Null flux guideway
• Linear synchronous motor (LSM)
• Quadrupole vehicle magnets
• High speed electronic switches
MAGLEV Technology Development. Figure 20
Superconducting electrodynamic Maglev-enabling
inventions
The nonmechanical contact nature enables a unique
new capability not possible with rail systems. Suitably
designed, Maglev vehicles can electronically switch at
high speeds without the need for mechanically moving
long sections of the guideway. Not only does electronic
switching avoid the cumbersome mechanical move-
ment, it enables switching at much higher speeds than
if the vehicle had to slow down to navigate the mechan-
ical switch section.
Figure 20 lists the superconducting electrodynamic
Maglev inventions pioneered by the authors that
address the six development goals outlined above.
The first invention is the use of superconducting
magnets on the vehicle to induce currents in normal
metal – for example, aluminum – guideway loops to
levitate and automatically stabilize high speed vehicles
[7, 15]. This combination efficiently uses the very
strong magnetic fields generated by a small number of
superconducting magnets on the vehicle that interact
with simple, low cost, low drag normal aluminum
guideway loops to levitate the moving vehicles. The
combination is ideal, since using normal electrical con-
ductors on the vehicle would require an impractical
amount of on-board power for the large clearances
involved, and superconducting loops on the guideway
would be far too expensive.
The second invention, illustrated in Fig. 21, is the
null flux guideway. The principle of the null flux guide-
way is simple. The guideway loops are arranged so that
they have symmetry positions relative to the vehicle
magnets. When a vehicle magnet is at the symmetry
position, the net magnetic flux through the given
guideway loop or loop circuit is zero. (The null flux
condition can be designed to apply to either a simple
MAGLEV Technology Development 771
loop or a circuit of two or more connected loops.)
When the net flux through the null flux loop or circuit
is zero, the resultant induced current in the loop or
circuit is also zero. If the vehicle magnet moves away
from its symmetry position, the net flux through the
loop or circuit becomes nonzero, and a current flows
that acts to push the vehicle magnet back toward its
symmetry position. A wide variety of loops and loop
circuits can be used in the null flux guideway. A very
useful configuration is the figure of 8 null flux loop
circuit, as illustrated in the upper right corner of
Fig. 21. The left-hand loop of the figure of 8 circuit is
wound in a clockwise direction, while the right hand
loop is wound counterclockwise. When the dipole loop
vehicle magnet is centered on the figure of 8 loop
circuit, the net flux through the circuit is zero. If the
dipole loop moves to the left, a net current and force
develop to push it back toward the right. Conversely, if
it moves to the right, an opposite current and force
push it back to the left.
The lower left corner of Fig. 21 shows the guideway
and vehicle loops for the narrow beam guideway. The
figure of 8 loops, on the sides of the beam, levitate the
vehicle. The vehicle magnets move slightly below the
symmetry position, generating a current and force that
push them upward. The figure of 8 null flux loops,
besides levitating the vehicle, also stabilize it against
vertical and roll displacement from external forces. The
vehicle is stable against pitch displacements by having
independent figure of 8 loops along the beam. If the
vehicle tends to pitch downward due to an external
force, the pitching motion is automatically countered
by magnetic forces. A second null flux loop circuit on
the narrow beam provides lateral and yaw stability. It
consists of two dipole loops, one on each side of the
beam, electrically connected to form a null flux circuit.
When the vehicle is laterally centered on the beam, the
net flux through the null flux circuit is zero. When the
vehicle is laterally or yaw displaced from its centered
position, a net flux and current develop in the
dipole null flux circuit that pushes the vehicle back to
the center.
The lower right corner of Fig. 21 shows a fabricated
guideway loop assembly for one side of the narrow
beam. Besides the dipole and figure of 8 loops, there
is also an LSM propulsion loop. The loop assembly
is encapsulated in a thin panel of polymer concrete,
772 MAGLEV Technology Development
Null Flux Guideway
• Vehicle Magnet Automatically Centered
On Null Flux Loop By Induced Currents
• Null Flux Guideway Loops Levitate
Vehicle and Passively Stabilize It,
Vertically and Horizontally
• Null Flux Currents Are Small, Resulting
In Low Magnetic Drag
Figure Of 8 Null Flux Loops
Dipole Null Flux Loops
Levitation
Levitation
Force
Force
Narrow Beam
Option
Vehicle Vehicle
Magnet Magnet
MAGLEV Technology Development. Figure 21
Null flux guideway
a very strong, tough form of concrete, and attached to
the sides of the narrow beam and shipped to the con-
struction site. The I2R losses in the guideway loops, and
the corresponding magnetic drag, are much smaller in
the null flux guideway than in a guideway that
uses a simple conducting sheet or a set of simple
dipole, non-null flux loops. Moreover, the magnetic
“stiffness” – that is the magnitude of the magnetic
restoring force per inch of displacement – is much
greater for the null flux guideway.
The third invention is the linear synchronous
motor (LSM) [7] illustrated in Fig. 22. Prior to the
LSM, a method proposed for the propulsion of high
speed ground transport vehicles included conventional
steel wheel on rail traction motors, the linear induction
motor (LIM), and jet engines. These methods required
either on-board engines or collecting power from
a high voltage line along the guideway using a sliding
contact. In the linear synchronous motor, the
CL
Vehicle
Magnet
Net Magnetic
Force
Null Flux
Figure of 8 Loop
Net
Magnetic
Vehicle Magnet
Force
Null Flux
Figure of 8 Loop
superconducting magnets on the vehicle magnetically
interact with a traveling wave of AC current in a set of
aluminum conductor windings on the guideway, as
illustrated in Fig. 22, producing a longitudinal mag-
netic force that propels the vehicle along the guideway.
The vehicle remains locked in phase with the traveling
AC current wave, with its speed determined by the
frequency of the LSM current wave, in accordance
with the relation n = fl, where l is the pitch of the
alternating polarity magnets on the vehicle (l = twice
the length of one magnet), f is the frequency of the AC
wave, and n is the vehicle speed. Vehicle speed is con-
stant, even if it is subjected to head or tail winds or it
climbs or descends grades, as long as sufficient phase
margin is maintained by the LSM current wave. As
a result, the distance between vehicles on the Maglev-
2000 guideway is fixed, even if the external forces on the
individual vehicles vary. This is extremely important
since it eliminates any possibility of vehicle collisions.
 MAGLEV Technology Development 773

Linear Synchronous Motor (LSM)

LSM Top View

• LSM Magnetically Propels Vehicle Energized LSM Winding in Guideway

Without Physical Contact

• Vehicle Does Not Neet On-Board

Propulsion Power
Upwards Vehicle Magnetic Propulsion
• LSM Propels Vehicle At Constant Magnetic Field Magnet Force On Vehicle
Speed Regardless of Head or Tail From LSM Winding
Winds, Up or Down Grades, etc. LSM End View
• Distance Between Vehicles Always Magnetic Vehicle
The Same Field Magnet
+ −
• LSM ls Highlty Efficient (>80%) Propulsion Force Is
Into Screen
Crossover Current In
LSM Winding

Energized LSM Block Operation

Constant Separation Distance

Non-Energized LSM Block

Energized LSM Block
Propulsion Force
Vehicle 10 km 10 km 10 km

Head Tail Up Down

Wind Wind Grade Grade
Power Line Non-Energized
–100 meters LSM Block

MAGLEV Technology Development. Figure 22

Linear synchronous motor (LSM)

To minimize I2R losses in the guideway LSM windings,
the entire guideway is not energized. Instead, only the
section on which the vehicle is traveling is switched on
from the DC power line that runs along the guideway,
using solid state switches to control local frequency and
power flow. The switched-on section length is typically
100–200 m, resulting in high motor efficiency. Approx-
imately 90% of the input electrical power to the
switched section is delivered as propulsive power to
the Maglev vehicle. As the vehicle leaves the old
switched-on section, the electrical power input is trans-
ferred to the new section ahead, as illustrated in the
drawing in the lower left corner of Fig. 22. By control-
ling the frequency of the AC power fed to the guideway,
the central traffic control facility can speed up or slow
down the vehicles if conditions warrant – again,
a major advantage for safe operation.
The fourth invention is the Maglev-2000 quadru-
pole superconducting magnet. It differs from the
dipole magnet used in the Japanese Maglev system,
and achieves new capabilities. Dipole magnets employ
two lines of current that run in opposite directions,
forming a simple rectangular loop. Quadrupole mag-
nets employ four lines of current, at the corners of
a square, and alternating in direction as shown in
Fig. 23. Fringe magnetic fields from quadrupole mag-
nets are much smaller than from dipole magnets, as
illustrated in the lower left corner of Fig. 23. A 600 KA
dipole magnet has a fringe field of 15 gauss at 6 m,
30 times the Earth ambient field, while the quadrupole
magnet fringe field is only 0.07 gauss, one-seventh of
Earth ambient, at the same distance. The much smaller
fringe fields make it simple to keep the magnetic field
strength inside the vehicle passenger compartment and
774 MAGLEV Technology Development
MAGLEV Technology Development. Figure 23
Quadrupole magnet
at stations at Earth ambient level. Quadrupole magnets
enable Maglev vehicles to travel on and smoothly
transition between both narrow beam and planar
guideways. Most of the time, Maglev-2000 vehicles
travel on the low cost narrow beam guideway. Near
stations, the vehicles transition to the planar guideway
where the vehicle can switch from the main guideway
to off-line stations to stop and carry out loading/
unloading operations. The switch section has two
lines of overlapping planar guideway loops, with one
line going on as the main high speed route, and the
second line directing the vehicle to an off-line station
away from the main route. The switching process uses
electronic solid state switches to control which line of
guideway loops the vehicle will follow. Mechanical
movement of the guideway is not required. Depending
on schedule, the vehicle may not switch off the guide-
way, but continue at high speed on the main line to
a station further down the line.
The capability to switch at high speed to off-line
stations for loading and unloading using an electronic,
nonmechanical switching process – the fifth fundamen-
tal invention – is very important for countries like the
USA, where the population density is relatively low and
spread out, not concentrated at a few high density
points. High speed switching enables Maglev vehicles
to employ “skip-stop” schedules to maintain high aver-
age speed, as illustrated in Fig. 24. If vehicles had to stop
at or slow down for every station along the route, the
average speed would be considerably less than 300 mph,
unless stations were a hundred miles apart or more. With
“skip-stop” scheduling, stations can be spaced more

MAGLEV Technology Development. Figure 24
High speed electronic switch
closely to more efficiently serve the spread-out popula-
tion. Using “skip-stop” travel, a passenger could board
a Maglev vehicle at his/her local station and skip all other
stations in the area, traveling directly to his/her final
destination in some distant city.
Operational First-Generation Maglev Systems
Following the publication of Powell and Danby’s 1966
paper on superconducting Maglev [8], a number of
countries started Maglev development programs.
Japan, and also Germany, visited Powell and Danby to
learn about their inventions. Japan chose to focus their
Maglev development program on superconducting
Maglev because of its large clearance of 10 cm or
more – a very important factor in Japan because of
the strong earthquakes there – and its strong inherent
stability, a very important safety feature.
Germany’s Maglev development program focused
on both superconducting electrodynamic Maglev and
MAGLEV Technology Development 775
electromagnetic Maglev systems, and tested both
approaches. They then selected electromagnetic Mag-
lev as a system that could be commercialized relatively
quickly.
Japan and Germany have both achieved technically
successful first-generation Maglev systems that have
transported many thousands of passengers safely and
comfortably at speeds on the order of 300 mph. The
Japanese Maglev system presently operates on Japan
Railways’ 20 km demonstration track in Yamanashi
Prefecture [17]. Japan Railways plans to have the
Yamanashi track become part of a 500 km (300 mile)
Maglev route to be constructed between Tokyo and
Osaka that will carry 100,000 passengers daily with
a trip time of 1 h [18].
The German electromagnetic Maglev system, called
Transrapid, was extensively tested on a 20 km
demonstration track at Emsland, Germany [5]. The
Transrapid Maglev system now operating in Shanghai,
776 MAGLEV Technology Development

China [19], has the distinction of being the first com-
mercial Maglev system. The 30 km route runs between
the center of Shanghai and its Pudong airport.
Both the Japanese and German first-generation
Maglev systems are limited to transporting passengers
and lightweight, high-value FedEx-type freight. They
cannot transport heavy trucks, autos, and freight con-
tainers. They cannot electronically switch from the
main guideway to a secondary guideway, but require
mechanically moving long sections of the guideway.
Guideway construction cost is very high, on the order
of $60 million per two-way mile or more [20].
More detailed descriptions of the development and
current status of the Japanese and German first-
generation passenger Maglev system is given below.
Table 1 [21] lists the significant milestones in the
development of the Japanese first-generation Maglev
system from its start in 1970 through 2004, the last
available update from Japan Railways’ Railway Technical
Research Institute (RTRI). By 2004, the Yamanashi Mag-
lev Demonstration Line had carried over 80,000 passen-
gers [21] with total accumulated running distances of
over 400,000 km (240,000 miles). Since then, the num-
ber of passengers and running miles has grown consid-
erably, but data is not available.
The 20 km Yamanashi line located north of Mt. Fuji
was dedicated in April 1997. Before then, R&D tests
were carried out at the Miyazaki Test site. The first test
run of a levitated Maglev vehicle was in 1977.
The unmanned ML-500 vehicle (Fig. 25) ran on an
inverted T-shaped guideway. In 1977, it attained
a speed of 517 km/h [21].
Japan Railways then transitioned to a U-shaped
guideway, which is the configuration for the Yamanashi
ML-500, 500R

MAGLEV Technology Development. Table 1 Milestones

in the development of the first-generation Japan Railways’
superconducting Maglev system [21]
Date Milestone
1970 Study of electrodynamic levitation system
(EDS) started
1977 Miyazaki Maglev Test Center opened – test run
of unmanned ML-500 vehicle on inverted
T-shaped guideway started
1980 Test run of MLU 001 vehicle on U-shaped
guideway started at Miyazaki
1982 Manned 2 MLU 001 vehicle set tests started
1986 3 MLU 001 vehicle set achieved 352 km/h
ML-500
1987 400 km/h achieved by manned MLU 002
vehicle set
1997 First running test of MLX 01 vehicle on
Yamanashi Demonstration Line Manned 5 MLX
01 vehicle set attained 552 km/h
1999 Test of 2 MLX 01 vehicle sets passing each
other in opposite directions at 1,003 km/h
2003 Manned 3 MLX 01 vehicle set attained
581 km/h (361 mph) 13500
2004 Cumulative number of passengers carried at ML-500 (side veiw, dimension in mm)
Yamanashi exceeds 80,000; cumulative
distance traveled exceeds 400,000 km MAGLEV Technology Development. Figure 25
ML-500, 500R
 MAGLEV Technology Development 777

demonstration route, and the planned line between MLU002

Tokyo and Osaka. The first test run of the MLU 001
(Fig. 26) vehicle [21] on the U-shaped guideway was
carried out in 1980 at the Miyazaki Test site. The
U-shaped guideway permitted a more efficient config-
uration for carrying passengers. Testing of the MLU001
vehicle carrying passengers started in 1982, with a
two-car train set. In 1987, the two-car manned vehicle
set attained 400 km/h. Also in 1987, test runs of a new
Maglev vehicle, MLU002 (Fig. 27), began [21]. It was MLU002
longer than the MLU001 – 22 m compared to 20 m for
the two-car MLUJ001 vehicle set. MLU002 could carry

3,700
44 passengers and weighed 17 t. In 1989, it attained
a speed of 394 km/h. 22,000

In 1991, MLU002 was damaged by a fire started in MLU002 (side view, dimensions in mm)

its braking system caused by excessive heating. MAGLEV Technology Development. Figure 27
MLU002 was then replaced by MLU002N (Fig. 28) MLU002
which was essentially the same length as MLU001 and
slightly heavier, 19 t compared to 17 t [21]. Test runs of
MLU002 started in 1993. In 1995, MLU002 attained
a speed of 411 km/h with passengers.
By 1996, the Yamanashi Test Center opened, and
MLU002N
was officially dedicated in April 1997. A few months
MLU002N
later, in December, the new MLX01 vehicle (Fig. 29)
attained 531 km/h carrying passengers. In 1999, a five-
vehicle MLX01 set reached 552 km/h with passengers
[21] and two-vehicle sets passed each other going in

MLU001

MLU002N
3,000

MLU001 (3-car train)

3,700

10103 8200 10100 22,000

MLU001 (side view) MLU002N (side view, dimensions in mm)

MAGLEV Technology Development. Figure 26 MAGLEV Technology Development. Figure 28

MLU001 MLU002N
778 MAGLEV Technology Development
MLX01
MLX01 leading car, aero-wedge
MLX01 leading car, double cusp
MAGLEV Technology Development. Figure 29
MLX01
MLX01-901
New leading car, MLX01-901
MAGLEV Technology Development. Figure 30
MLX01-901
opposite directions with a relative speed of 1003 km/h
(600 mph). Tests on a new vehicle set, MLX01-901
(Fig. 30) started in 2002. A three-vehicle MLX-01 set
attained the world speed record of 581 km/h (361 mph)
on December 2, 2003, carrying passengers [21].
The Japanese Maglev system uses a U-shaped
guideway, with the aluminum levitation, stability, and
Propulsion coil
Panel
Side-wall
Levitation and
guidance coil
3,000
Hole for
bolt-setting
Wheel support pass
MAGLEV Technology Development. Figure 31
Panel method
propulsion located on the beams at the sides of the
guideway (Fig. 31). The concrete panels with their
attached aluminum loops are manufactured in
a factory and then transported to the construction
site [21] where they are bolted to the concrete side
beams of the massive field constructed U-shaped
guideway. Figure 32 shows a photo of the completed
guideway.
Figure 33 shows a photo of the Yamanashi Maglev
Demonstration line, with two Maglev vehicles on
a bridge that crosses a local highway. In the distance,
the JR Maglev guideway enters a tunnel through the
mountain. Much of the proposed 500 km (300 miles)
Maglev route will be in deep tunnels through the
mountains of Central Japan. Three routes are under
consideration [22] as illustrated in Fig. 34. Depending
on the route, as much as 60% of the Tokyo to Osaka
line will be in deep tunnels.
The principal constraints for the first-generation
Japanese Maglev system are its high construction cost
and its limitation to passenger-only transport. The
high field construction cost results from the massive
U-shaped guideway, plus the high cost of tunneling for
much of the Tokyo to Osaka route. The limitation to
passenger-only transport can be a revenue problem in
low population countries, like the USA. In the USA,
for example, the transport outlay for intercity trucks
is much greater than for intercity air and train
passengers – the USA spends over $300 billion per
year for intercity truck transport, compared to only
60 billion for air passengers and three billion for
intercity rail passengers.
 MAGLEV Technology Development 779

MAGLEV Technology Development. Figure 32

Guideway of the Yamanashi Maglev test line

MAGLEV Technology Development. Figure 33

Vehicles crossing bridge over highway on Yamanashi Maglev test line

Turning now to the German electromagnetic Maglev system [5] which is levitated by conventional
Transrapid Maglev system, it is the basis for the first electromagnets on the vehicle that are attracted upward
high speed commercial Maglev system in the world. to two iron rails on the guideway. The physical gap
Transrapid is a first-generation electromagnetic (EMS) between the vehicle’s electromagnets is small,
780 MAGLEV Technology Development
Possible Maglev Routes
Route B (Backed
by Nagano Pref.) Suwa
Chino
Kiso-fukushima
Ina
Route A
lida
Nagoya Route C (Preferred
choice of JR Tokai)
Tokaido Shinkansen
MAGLEV Technology Development. Figure 34
Possible Maglev routes
approximately 1 cm (3/8 in.), necessitating very precise
and accurate guideway construction. EMS suspensions
are inherently unstable because as the gap between the
vehicle’s electromagnets and the iron rail decreases,
the levitating force on the vehicle increases, acting to
pull the vehicle into contact with the rail. The
Transrapid system rapidly servo controls the currents in
the electromagnets on the vehicle to keep the gap
between the vehicle and the guideway constant. If
the vehicle moves toward the iron rails above the electro-
magnets, the servo control system quickly decreases the
current in their windings, reducing the levitating force. If
the vehicle moves down and away from the iron rails, the
current in the electromagnets is quickly raised, increas-
ing the levitating force. The servo control system
operates on a time scale of thousandths of a second.
The 30 km long Shanghai Maglev Line runs from
the outskirts of Central Shanghai to Pudong Airport.
Figure 35 shows a photo of the Transrapid vehicle
leaving Pudong Airport, while Fig. 36 shows a photo
of the Longyang Road Station at the outskirts of Shang-
hai. The Transrapid Maglev Line does not service the
Yamanashi Maglev Test Line
Tokyo
Kofu
Southern Alps
center of Shanghai. Instead, there is a 20 min subway
ride from the center of Shanghai to the Longyang Road
Maglev Station [19].
Maglev service on the Shanghai route started on
January 1, 2004. Depending on time of day, trip time
for the 30 km run is 7.20 min for peak travel hours
and 8.10 min for nonpeak hours. The corresponding
average speeds are 251 km/h (156 mph) for the faster
trip and 224 km/h (139 mph) for the slower trip.
Maximum speeds during the trip are 431 km/h
(268 mph) and 301 km/h (197 mph). The record
speed on the route was 501 km/h (311 mph),
achieved on November 12, 2003. The interval
between trains is 15 min, except in the evening, when
it is 20 min [19].
The 30 km (19 miles) Shanghai Transrapid line
cost $1.8 billion (US dollars) and was built in 2½
years. The construction cost was $60 million per
kilometer of route ($95 million per mile) [19]. The
proposed 200 km (124 miles) extension to Hangzhou
was approved in March 2010, with construction
planned to start in 2010 [19]. However, given the

MAGLEV Technology Development. Figure 35
Maglev train coming out of the Pudong International Airport in Shanghai
MAGLEV Technology Development. Figure 36
Longyang road Maglev station
high cost per kilometer and China’s plans to build
thousands of kilometers of high speed rail, the pro-
posed Maglev extension may be postponed or even
canceled.
The development of the Transrapid Maglev System
has proceeded through a number of vehicles, as illus-
trated in Fig. 37, eight during the years 1976–2008.
MAGLEV Technology Development 781
A number of EMS vehicles were tested in Germany in
prior years, before 1976, by various companies before
settling the final Transrapid system. These included
the MBB “Prinzipfahrzeng,” the MBB Komet, and the
Krauss-Maffei Transrapid TR-02 and TR04 vehicles.
In 1974, MBB and Krauss-Maffei joined together to
form Transrapid.
782 MAGLEV Technology Development

Transrapid Milestones
1934–1977: From the idea to the system decision
1978–1991: From the test facility to technical readiness for application
1992–1999: The first application in Germany is planned
2000–today: Alternative routes in Germany and abroad

TR09 (2008)

Shanghai (2003)
HMB2 (1976)

TR08 (1999)

TR05 (1979)

TR07 (1989)

TR06 (1984)

MAGLEV Technology Development. Figure 37

Transrapid milestones. 1934–1977, from the idea to the system decision; 1978–1991, from the test facility to technical
readiness for application; 1992–1999, the first application in Germany is planned; 2000–present, alternative routes in
Germany and abroad

The speed records achieved by the various 1987 TR06 406 (manned)
Transrapid vehicles and their predecessors are given
1988 TR06 412 (manned)
below [23]:
1989 TR07 436 (manned)
1993 TR07 450 (manned)
Year Vehicle Speed, km/h
2003 TR08 501 (manned)
1971 Rinzipfahrzeg 90
1972 TR02 164 For extended testing of the Transrapid vehicles,
1973 TR04 250 (manned) Germany constructed a 31.5 km test track at Emsland
in Germany. The single track had turning loops at
1975 Komet 401
both ends, permitting continuous running tests.

Construction of the Emsland Facility began in 1980
and was completed in 1984. It was the site of the 2006
accident when a moving Transrapid vehicle collided
with a stopped maintenance vehicle that was working
on the guideway [5].
The Transrapid vehicles through TR08 obtained
power for levitation and electronics below 80 km/h by
physical contact with the track. The new TR09 vehicle
(Fig. 38) needs no physical contact with the track at any
speed, but receives power by inductive transmission
with the track’s propulsion system. If the track’s
propulsion power system fails, TR09 uses on-board
batteries for back power to maintain levitation [5].
Application of Transrapid in Germany has been
prevented by two factors: (1) the existing network of
ICE high speed rail lines, and (2) high construction cost.
The proposed Hamburg to Berlin line was canceled
because of cost, as was the Munich to its airport route.
The 40 km (25 mile) route between Munich Central
Station and Munich Airport was originally estimated
to cost 1.85 € billion, but later raised to over 3 € billion
($4.7 billion) or almost $200 million per mile. High
tunnel and civil engineering costs were significant
factors in the 2008 decision to cancel the project [5].
Transrapid has been intensively lobbied in the USA
to build Maglev routes between major metropolitan
MAGLEV Technology Development. Figure 38
Transrapid 09 at the Emsland test facility in Germany
MAGLEV Technology Development 783
areas, for example, Las Vegas to Anaheim, California,
Baltimore to Washington, DC, Pittsburgh to its airport,
etc. So far, however, no projects have gone forward.
The Development of the Second-Generation
Maglev-2000 System
The first-generation Maglev systems, while technically
successful, have two factors that limit their implemen-
tation, particularly in the USA.
First, the guideway construction cost is very
high, $60 million or more per mile. Second, the
first-generation Maglev systems only carry passengers.
While useful in densely populated Europe and
Japan, passenger-only systems are less useful in lower
population density large countries like the USA.
US transport outlays for intercity trucks over $300
billion annually; for intercity air passengers, $60 billion
per year; and for intercity rail passengers, only $3
billion per year [24]. Maglev or high speed rail passen-
gers only systems in the USA will require major gov-
ernment financing for construction and large subsidies
for operation and maintenance. Because of USA’s large
debt, major government financing is unlikely. Maglev
routes will need private investment. To achieve this,
they must be profitable, with a short payback time on
invested capital, less than a decade.
784 MAGLEV Technology Development

The second-generation Maglev-2000 system operating on a planar guideway (Fig. 23), the bottom
addresses these two factors. First, the projected faces of the quadrupoles magnetically interact
guideway construction cost is about $25 million per with aluminum loops located on the guideway beneath
mile, a factor of 2 or more lower than first-generation the vehicle.
systems. Low cost prefabricated monorails are used for The ability to operate on a planar guideway as well
the elevated guideway. Figure 39 shows an artist’s view as monorail reduces construction cost. When operating
of a Maglev-2000 passenger vehicle on the monorail in densely populated urban and suburban areas,
guideway. Maglev-2000 vehicles do not need a new, very expen-
The prefabricated monorail beams would be mass sive guideway with its accompanying disruptions to
produced in factories, with their guideway loop panels, existing infrastructure. Instead, the vehicles can transi-
sensors, electronic equipment, etc., attached to them at tion to, and operate on, existing RR tracks to which
the factory. The beams and piers would then be aluminum loop guideway panels have been attached on
transported by truck or rail to the construction site, the cross-ties (Fig. 40). Conventional trains can con-
where they would be quickly erected on pre-poured tinue to use the RR tracks, given appropriate schedul-
concrete footings or pilings, using conventional cranes. ing. The cost of attaching guideway panels to enable
Low guideway cost would be achieved by the use of levitated Maglev-2000 operation is very small, only
conventional box beams for the monorail, which
minimizes the materials required, and prefabrication,
which minimizes expensive field construction.
Disruptions to local infrastructure would also be
minimized, which would reduce local opposition to
guideway construction.
The superconducting quadrupoles on the vehicles
(Fig. 23) have four magnetic poles that alternate in
polarity around their circumference. When on the
monorail guideway, the vertical sides of the quadru-
poles interact with the aluminum loops attached to the
adjacent vertical sides of the monorail guideway. When

TIE
RAITAST

MAGLEV Technology Development. Figure 40

MAGLEV Technology Development. Figure 39 Drawing of levitated Maglev-2000 vehicle traveling on
Artist’s drawing of Maglev-2000 vehicle on monorail conventional RR tracks to which aluminum loop panels
guideway have been attached to the cross-ties

about $4 million per mile, compared to the much
higher cost of building a new elevated guideway.
Turning to the second factor affecting Maglev
implementation, revenues and net profits,
Maglev-2000 can carry heavy trucks as well as passen-
ger vehicles on its dual-use guideway. In contrast to the
superconducting dipole loops used in the Japanese
first-generation Maglev system, the powerful
superconducting Maglev-2000 quadrupoles can be
located along the length of a Maglev vehicle, increasing
its lifting capability without producing magnetic fields
inside the vehicle that significantly exceed the natural
Earth ambient value. This is a result of the considerably
lower value of the magnetic fringe fields from
a quadrupole, compared to a dipole system.
Figure 41 shows a drawing of a passenger and truck
carrier Maglev-2000 vehicles on the dual-use Maglev-
2000 monorail guideway. The gross revenue from
transporting 3,000 trucks daily on a Maglev-2000
route (one-fifth of the daily truck traffic on a typical
Interstate Highway) is equivalent to 180,000 passengers
per day, assuming a typical US outlay of 30 cents a ton
Maglev-2000 System Can
Handle Both Freight
and Passengers
Passenger
Compartment
Equipment Equipment
Compartment Compartment
MAGLEV Technology Development. Figure 41
Maglev-2000 passenger and truck carrier vehicles on dual-
use guideway
MAGLEV Technology Development 785
mile for trucks, and 10 cents a passenger mile. The
truck revenues alone would pay back the Maglev-2000
guideway in less than 5 years, attractive for private
investment.
The same Maglev-2000 guideway could also trans-
port personal autos together with their passengers, on
long trips at lower cost than by highway. It could also
transport high-value freight.
Using Maglev, a trucker could pick up a load, drive
a few miles to the nearest Maglev station, travel across
the USA in a few hours rather than a few days, and drive
off the Maglev vehicle to deliver the load to its destina-
tion a few miles away. Wear and tear on the truck would
be much less and it could make five deliveries in the
time it would take to go by highway.
The ability of Maglev-2000 vehicles to travel on
planar guideways also enables high speed electronic
switching to off-line stations. Maglev-2000 vehicles
can bypass high speed stations they are not scheduled
to stop at, enabling stations to be closely spaced for
convenient access. This increases revenue potential
compared to only a few stations in a metropolitan area.
Figure 42 shows a drawing of the Maglev-2000
aluminum wire loop guideway panel. It has three sets
of multi-turn aluminum loops: (1) a sequence of four
short figure of 8 loops; (2) a sequence of four short
dipole loops; and (3) one long dipole loop.
When the panels are attached on the vertical sides of
the monorail guideway beam, the figure of 8 loops
provide levitation and vertical stability. The dipole
loops on each side of the beam are connected together
to make a null flux circuit that maintains the vehicle in
a centered position on the beam – when centered no
current flows in the aluminum null flux circuit. When
an external force (wind, curve, etc.) acts to push the
vehicle away from its centered position, a magnetic
force develops that opposes the external force. The
long dipole loop is part of the linear synchronous
motor (LSM) propulsion system, in which the loops
on a sequence of panels are connected in series to form
an energized block along which the Maglev vehicles
travel. The energized block is typically on the order of
100 m in length; as the vehicle leaves an energized
block, its AC propulsion current is switched into the
next block that the vehicle is entering.
For the planar guideway, the panels are laid flat on
the planar surface beneath the line of quadrupoles on
786 MAGLEV Technology Development

FIGB 20.56X36.19

LSM25.56X64.56

DPL.17.19X19.19

3/4TRUX1.5C⬘BORE 1
INCH DEEP 6PLCS ALUM SLEEVE 1⬘ODX3/4IDX2⬘LG.

MAGLEV Technology Development. Figure 42

Drawing of aluminum loop guideway panel providing vertical life and stability, lateral stability, and linear synchronous
propulsion

the moving vehicle. The figure of 8 loops now provide
lateral stability, generating magnetic restoring forces if
an external force acts to displace the vehicle from its
centered position on the guideway. The dipole loops act
individually, with inductive currents that levitate over-
head. The LSM loops function as they do on the mono-
rail guideway.
The planar guideway panel configuration can levi-
tate and propel Maglev vehicles along existing RR
tracks, with the panels attached to the cross-ties of the
RR tracks.
The planar guideway also enables high
speed switching. At a switch section, there are two
lines of overlapping guideway loops, which can be
either electronically open-circuited or close-circuited,
depending on the desired switching action. Line
A of guideway loops runs straight ahead on the main
line, while the second line (Line B) of loops
diverges laterally at a rate acceptable to passengers.
If the loops in Line A are close-circuited and those in
Line B are open-circuited, the vehicle travels
straight ahead on the main route. If Line A is open
and Line B is closed, the vehicle diverges laterally
from the main route onto a secondary guideway
leading to the off-line station. The high speed
vehicle then decelerates on the secondary guideway
that leads into the off-line station. When the vehicle
leaves the station to rejoin the main line, it accelerates
on an out-bound secondary guideway to a second
switch section where the high speed vehicle reenters
the main line.
The Maglev-2000 components discussed above
have been fabricated and tested at full scale in order
to determine their performance and validate their
projected costs.
Figure 43 shows one of the two wound
superconducting loops used for the Maglev-2000
quadrupole. The loop has 600 turns of NbTi

MAGLEV Technology Development. Figure 43
NbTi superconductor loop for Maglev-2000 quadrupole
superconducting wire, supplied by Supercon, Inc. of
Shrewsbury, MA [25]. At the design current of 1000
A in the NbTi wire, the Maglev-2000 quadrupole has
a total of 600,000 A turns in each of its two
superconducting (SC) loops. The SC winding is
porous, with small gaps between the NbTi wires to
allow liquid helium flow to maintain their temperature
at 4.2 K, and to stabilize them against flux jumps and
micro movements [25].
Figure 44 shows the SC loop enclosed in its stainless
steel jacket. Liquid helium flows into the jacket at one
end and exits at the end diagonally across from the
entrance, providing continuous helium flow through
the SC winding. Before insertion of the SC loop into
the jacket, it is wrapped with a thin sheet of high purity
aluminum (5000 residual resistance ratio) to shield the
NbTi superconductor from external magnetic field
fluctuations. After closing the jacket, a second layer of
high purity aluminum is wrapped around it for addi-
tional shielding.
Figure 45 shows a CAD-CAM drawing of the
complete Maglev-2000 cryostat that holds two
superconducting quadrupoles. The magnetic polarity
of the front SC quadrupole is opposite to that of the
rear quadrupole. This allows levitation at lower speed
MAGLEV Technology Development 787
than if the two quadrupoles had the same polarity, due
to less L/R decay of the currents induced in the alumi-
num guideway loops. The two SC loops are supported
by a graphite-epoxy composite structure that resists the
magnetic forces – due both to the forces in a loop from
its self current, and to the forces between the two
loops – that act on them.
Figure 46 shows the SC loops, support structure,
and cooling currents for the Maglev-2000 quadrupole
being assembled in Maglev-2000’s facility on Long
Island. The SC loops have a 10 K thermal shield,
which is cooled by helium exiting from the jacket
holding the SC loop. The SC quadrupole structure is
then enclosed by an outer layer of multilayer insulation
(MLI) consisting of multiple alternating layers of glass
fiber and aluminum foil. A second thermal shield
encloses the SC quad, and maintained at 70 K by
the helium outflow from the 10 K primary thermal
shield.
Figure 47 shows the completed SC quadrupole
enclosed in its vacuum cryostat, while Fig. 48 shows
testing of the quadrupole magnetic levitation and pro-
pulsion forces using DC current in the aluminum loop
guideway assembly beneath the quadrupole as a stand-
in for the induced currents. The quadrupole was
788 MAGLEV Technology Development

MAGLEV Technology Development. Figure 44

NbTi superconducting loop enclosed in stainless steel jacket

MAGLEV Technology Development. Figure 45

CAD-CAM drawing of Maglev-2000 superconducting quadrupole
 MAGLEV Technology Development 789

MAGLEV Technology Development. Figure 46

Assembly of Maglev-2000 superconducting quadrupole

MAGLEV Technology Development. Figure 47

Completed Maglev-2000 quadrupole enclosed in its cryostat

successfully tested to its full design current of 600,000 A displacement from the centered position, and longitu-
turns. The magnetic forces between the quadrupole dinal position in the direction of movement along
and the guideway loop assembly were measured as the guideway. The measured forces agreed with
a function of vertical separation and lateral three-dimensional computer analyses.
790 MAGLEV Technology Development

MAGLEV Technology Development. Figure 48

Testing of magnetic forces on Maglev-2000 quadrupole using DC current in aluminum loop panel

In the time following the Maglev-2000 quadrupole

tests, high temperature superconductors have become
much more capable, and are commercially produced in
substantial amounts. Using YBCO high temperature
superconductor wire, it appears very possible to fabri-
cate Maglev-2000 quadrupoles that would be much
simpler in construction, with much easier refrigeration
requirements. The YBCO superconductor would oper-
ate at 40 K with a much simpler on-board cryocooler
than for NbTi superconductor at 4.2 K. One
Maglev-2000 quadrupole requires 3,600 kA turns of
superconductor. At $10 per kiloamp meter, which
appears achievable with large-scale production of
high temperature superconductor, the superconductor
MAGLEV Technology Development. Figure 49
for it would cost $36,000. A passenger vehicle with
Wound dipole loop for guideway panel using nylon-coated
eight quadrupoles would then have a superconductor
aluminum conductor
cost of $288,000, while a truck carrying a vehicle with
16 quadrupoles would then have a superconductor cost composed of a set of four figure of 8 loops, a set of
of $576,000, both of which are very reasonable for four dipole loops, and one long LSM propulsion loop.
a projected total cost of $5 million per vehicle. Future Figure 49 shows a wound dipole loop, to be used in the
tests of Maglev-2000 quadrupoles will probably involve panel. The aluminum conductor has a 10 mil layer of
high temperature superconductors rather than NbTi nylon using a dip process to coat the conductor. The
superconductor with liquid helium coolant. nylon insulation withstood 10 kV tests without break-
As described previously, the guideway loop panels down. Figure 50 shows a completed guideway loop
contain three sets of wound aluminum loops, panel with all of its loops.

MAGLEV Technology Development. Figure 50
Completed guideway panel with figure of 8 dipole, and LSM propulsion loops
MAGLEV Technology Development. Figure 51
Guideway loop panel enclosed in polymer-concrete matrix
The completed panel is then enclosed in a polymer-
concrete structure for handling and weather protection
(Fig. 51). Polymer concrete – a mixture of aggregate,
cement, and plastic monomer – can be cast into virtually
any form as a slurry. When the monomer polymerizes
(the rate of polymerization is controlled by the amount
of added promoter), the resulting concrete-like structure
is much stronger – a factor of 4 or greater – than
MAGLEV Technology Development 791
MAGLEV Technology Development. Figure 52
Polymer-concrete panel with enclosed aluminum loop
exposed for 2 years to outdoor environment with multiple
freeze-thaw cycles
ordinary concrete, and not affected by freeze-thaw
cycles, salt, etc. Figure 52 shows a completed polymer-
concrete panel left outside of the Long Island facility for
2 years. It was subjected to a wide range of weather
conditions and multiple freeze-thaw cycles over the
2 year period, without any degradation.
792 MAGLEV Technology Development
After being fabricated at the Maglev factory,
the guideway panels would be attached to the sides of
the monorail or the surface of planar guideway
beams to be shipped to a construction site for an
elevated guideway, or transported to existing RR track-
age that was to be modified for use by Maglev-2000
vehicles.
Figure 53 shows the basic design for the monorail
guideway beam. It is a hollow box beam made with
reinforced concrete. The beam length is 22 m and
weight is 34,000 kg. It uses post-tension construction,
which allows the tensioning cables in the base of the
beam to be retightened if some stretching were to
occur. The beam is tensioned to have a 0.5 cm upward
camber at the midpoint of the beam when it is
not carrying a Maglev vehicle. When the Maglev vehi-
cle is on the beam, the beam flattens out to a straight
line condition, with no vertical dip or camber along
its length.
Figure 54 shows a photo of the fabricated beam
after transport by highway truck from the manufactur-
ing site in New Jersey to Maglev-2000’s facility in
Florida. No problems in the 800 mile transport by
highway were encountered. The first beam cost
$45,000 with a projected large-scale production cost
of $25,000 per beam. Since 1999, construction costs
have increased. At $50,000 in today’s dollars, 140
beams for a two-way monorail guideway would cost
$7 million per mile.
Figure 55 shows a CAD-CAM drawing of the
aluminum chassis that was fabricated for a 20 m long
Maglev-2000 test vehicle, designed to carry 60 passen-
gers in urban and suburban service. Figure 56 shows
the fuselage for the test vehicle.
Maglev Applications
High Speed National Maglev Networks
As in other developed nations, USA’s three main trans-
port systems – motor vehicles (autos, buses, and trucks),
airplanes, and conventional rail – operate as national
networks. From any given area in the USA, passengers
and freight can drive, fly, or go by bus or train to any
other area in the USA. To reach a particular location, it
may be necessary to transition from one system to
another; for example, one can fly from one airport to
another airport, with a short drive to one’s final desti-
nation. However, such local transitions are usually easily
accommodated.
For Maglev to be an important mode of trans-
port in the twenty-first century, it must also
function as a national network. A few isolated Maglev
routes, while helpful to local travelers, will provide
only minor benefits. They will not substantially
reduce oil consumption and greenhouse gas emis-
sions, and not significantly increase domestic jobs
and exports.
The potential 25,000 mile National Maglev Net-
work (Fig. 57) would interconnect virtually all major
metropolitan areas in the USA. The intercity Maglev
routes, following the vision of the late Senator Daniel
Patrick Moynihan, would primarily be on the rights of
way of the Interstate Highway System [26]. Had his
$750 million 1990 Senate passed legislation for a US
Maglev R&D program not been killed in the House of
Representatives, USA would be well on its way to its
National Maglev Network.
One of the unique features of the second-
generation Maglev-2000 system is the ability of
Maglev-2000 vehicles for levitated travel along existing
RR tracks, to which thin panels holding aluminum
loops have been attached to the cross-ties.
Conventional trains can continue to use the RR
tracks, given appropriate scheduling. This capability to
use existing RR tracks enables Maglev-2000 vehicles to
travel in densely populated urban and suburban areas,
without needing to construct very expensive new infra-
structure, with its inevitable disruptions to the existing
infrastructure and the local population. Combined with
high speed elevated Maglev guideways between metro-
politan areas, this would result in fast convenient travel
by Maglev, both within a given metropolitan area, and
from one metropolitan area to another.
In addition to the attractive transport features and
economic benefits of the National Maglev Network
benefits and compared to present transport systems,
its ability to effectively and cheaply transport passen-
gers and freight without the need for oil will become
extremely important in the following decades as world
oil runs out.
 3′ 7′′

10 1/2′′ 1′ 10′′
6′′ R. ALL 1" PLATE (NOTE 7) NOTES:
4 CORNERS 12′′ 1/2′′ PROJ.
CONC. INSERTS SEE DRAWING NUMBER 2000–2 FOR NOTES.
(SEE NOTE 8)

3′′
2 1/2′′ (TYP)

3′′
ANCHORS FOR (TYP)
BOLTING 6′′ WEB (TYP)
MAGNETS
(NOTE 6)
2 × 2–#4 SS STIRRUPS
O16′′(MIDDLE 40′)STAGGER O 8′′
4′ 6′′ O12′′(BOTH ENDS)STAGGER O 6′′
TWO SETS OF CLOSED TIES(TYP)
1.5′′ MIN. COVER (TYP)
SEE NOTE 10 (TYP)

40 1/2′′ MAGNET POCKET

16–#5 SS BARS
REV. DATE DESCRIPTION

1 12/12/98 CLOSED STIRRUP LOOPS

2′′

6 1/2

LEDGE

7′′ HIGH
BOTTOM
2
FLANGE TYP MAGLEV-2000 NARROW BEAM
2 × 10 = 20. 0.6′′ DIA STRAIGHT STRANDS (NOTE 2) 72′′ – 2 1/8′′ LONG

DESIGNED BY M. REICH, Ph. D. 7/13/98

CHECKED BY A. CHIDAMBARAM, P. E. 7/13/98

MAGLEV GUIDEWAY CROSS–SECTION DRAWN BY W. GROSSMAN, P. E. 7/13/98

SCALE: 1′′ = 1′′− 0′′

APPROX. ESTIMATED QUANTITIES: DRAWING NUMBER

CONCRETE: 21.12 CY; SS REBARS: 2,310 lb.
SHEET 1 OF 6 2000–1
STRANDS: 1070 lb; WT OF BEAM: 86.72 KIPS

MAGLEV Technology Development. Figure 53

Design for 22 m long monorail guideway beam
MAGLEV Technology Development
793
794 MAGLEV Technology Development

MAGLEV Technology Development. Figure 54

Photo of 22 m long monorail guideway beam delivered to Maglev-2000 facility in Florida from construction site in New
Jersey

MAGLEV Technology Development. Figure 55

CAD-CAM drawing of aluminum chassis for 60 ft long Maglev-2000 vehicle

The desired capabilities for a US National Maglev use guideways with high energy efficiency in all
Network include: weather conditions
3. Able to be privately financed without the need for
1. Low guideway cost
government funding and subsidization for con-
2. Able to transport passengers, personal autos,
struction and operation
highway trucks, and freight containers on dual-
 MAGLEV Technology Development 795

MAGLEV Technology Development. Figure 56

Photo of fuselage for 60 ft long Maglev-2000 vehicle

MAGLEV Technology Development. Figure 57

25,000 mile national Maglev-2000 network
796 MAGLEV Technology Development

4. Rapid installation of guideways with minimal those for passenger air ($60 billion annually) and
disruptions and modifications to existing dwarf passenger rail (only $3 billion per year) (Fig. 58).
infrastructure Intercity highway trucks haul about the same
5. High speed electronic switching to off-line stations, amount of freight, 1.5 trillion ton miles per year as
with service to multiple convenient stations in met- conventional rail, but charge ten times as much per
ropolitan areas ton mile transported. They are preferred for high-value
6. Earth ambient magnetic field levels in passenger freight transport over conventional rail because of their
cabins much faster delivery and their convenient service,
direct from origin to destination. Shipping by rail
Construction costs for the Japanese and German
takes much longer and is much less convenient. The
first-generation Maglev systems are high, for example,
average haul distance for highway trucks is 500 miles.
$50 million or more per two-way mile [27]. To be
Truckers would need far fewer trucks to deliver the
widely implemented in the USA, construction cost
same volume of goods if they went by Maglev at 300
should be substantially lower (Capability #1).
mph rather than by highway at 60 mph, and at less cost
The National Maglev Network would not just
per ton mile delivered (Fig. 59).
carry passengers, but also intercity highway trucks,
Similarly, drivers could take their personal
personal autos, and freight containers (Capability #2).
autos with them on Maglev vehicles configured to
US transport outlays [24] for intercity highway
carry 15–20 autos and their passengers. Travel would
trucks (over $300 billion) are much greater than

1400
Personal Auto

1200 2001
Intencity Truck Freight

2025
Annual Outlay, Billions of 2001 Dollars

1000
Local Bus & Transit Passengers

800
Local Truck Freight

Intercity Air Passengers

600
Domestic Air Freight

RR Commuter

400
Rail Freight

Intercity Rail

Intercity Bus

200

0
1 2 3 4 5 6 7 8 9 10
2001 838 309 158 66 37 31 28 4 3 2
2025 1340 620 320 152 37 50 98 7 5 3
Mode of Transport

MAGLEV Technology Development. Figure 58

Annual outlays current and future, for US transport modes (billions of dollars) (US census Bureau [24])

Passenger
Compartment
Equipment Equipment
Compartment Compartment
MAGLEV Technology Development. Figure 59
Maglev-2000 system can handle both freight and
passengers
be much faster and cheaper than by highway, counting
fuel, lodging, and auto maintenance costs.
Private financing is essential for a National Maglev
Network (Capability #3) because of the very high gov-
ernment debt levels. Passenger-only Maglev systems
will not attract private financing. Figure 60 shows the
payback time for a Maglev route as a function of the
number of trucks transported daily, for the operating
parameters summarized in Table 2 assuming zero pas-
senger revenues (curve A). At 3,000 trucks daily (20%
of typical highway truck traffic on an interstate) the
payback time is less than 5 years. At 6,000 trucks daily
(40% of truck traffic) payback time is less than
2.5 years. Without trucks, payback time is many
decades for realistic passenger ridership levels of
10,000–20,000 passengers daily (curve B).
Rapidly installing guideways with minimal disrup-
tion and modification of existing infrastructure
(Capability #4) minimizes the cost and local
opposition to the construction of a Maglev route.
This is especially important in densely populated
urban and suburban areas. Whenever possible, existing
infrastructure including existing rail trackage should be
used to avoid disruptions to local inhabitants.
High speed electronic switching (Capability #5) is
very desirable. Not being able to switch off the main
MAGLEV Technology Development 797
line, or having to mechanically move a long cumber-
some guideway switch section significantly reduces
average speed. For example, HSR systems, though
capable of maximum speeds of 200 mph, have
considerably slower average speeds when station stops
and their slow acceleration and deceleration rates are
taken into account [28]. Electronic switching to
off-line stations will enable Maglev vehicles to bypass
at high speed stations they are not scheduled to stop at,
maintaining high average speeds. This will allow many
closely spaced stations within a metropolitan area, for
convenient access.
Earth ambient magnetic field levels in passenger
compartments (Capability #6) are not only desir-
able, but a necessity. The US public strongly resists
new technologies that appear to deviate from their
normal environment.
There is no evidence that DC magnetic fields of
a few gauss have any effect on the body (there is
a limit of 5 gauss for people with pacemakers) and
people experience much greater fields – at the kilogauss
level during MRIs – without problems. However, to
avoid possible controversy and opposition to Maglev,
it is desirable to keep the magnetic field inside passen-
ger cabins at Earth ambient level.
The guideway panels can also be mounted on the
cross-ties of existing RR tracks (Fig. 40), enabling lev-
itated travel of Maglev-2000 vehicles along existing RR
tracks in a planar guideway mode. The panels do not
interfere with the operation of conventional trains,
which could continue to use the tracks for bulk freight
transport, given appropriate scheduling, probably at
nighttime. The ability of Maglev-2000 vehicles to travel
as individual units would enable much more frequent
and convenient passenger service, rather as long trains
of many RR cars. Also, because the Maglev vehicle loads
are distributed along the vehicle and not concentrated
at wheels, its local track loading is much less than
conventional trains, resulting in much longer track
life and reduced maintenance.
The capital cost of the monorail guideway has been
examined in detail, based on fabrication experience
and costs of magnets, panels, and the guideway beam.
In 2000 dollars, the projected cost was $11.4 million per
two-way mile for “greenfields” construction (no land
acquisition or infrastructure modification costs) [29]. In
2010 dollars, the corresponding construction cost would
798 MAGLEV Technology Development

10,000 100,000
Basis
9000 90,000
25 M$/2 way mile
8000 Passengers 80,000
Gross Revenue 10¢/PM
7000 Net Revenue 7.4¢/PM 70,000

Daily Passenger Traffic

Daily Truck Traffic
6000 Trucks 60,000
Gross Revenue 25¢/ton mile
5000 Net Revenue 17.7¢/ton mile 50,000

4000 40,000

3000 30,000

2000 20,000
Cu
(Pass rve B
Curve A enger
1000 s only 10,000
)
(Trucks only)
0
10 20 30 40 50 60 70 80 90 100
Payback Time (years)

MAGLEV Technology Development. Figure 60

Payback time for Maglev-2000 guideway as a function of truck and passenger traffic

MAGLEV Technology Development. Table 2 Vehicle MAGLEV Technology Development. Table 3 National
O&M costs network and Maglev-2000 system parameters
Revenues and Passengers Trucks (cents/ton Network 25,000 miles, 300 mph max
costs (cents/pm) mile) Passenger 100 passengers
Gross rev 10 25 vehicles

Energy cost 1.2 4.0 Truck carriers Two types (one and two trucks)

Am&M cost 0.9 2.8 Auto carriers 15 autos + passengers

Personnel cost 0.5 0.5 Travel Either as single units or multiunit

consists, depending on traffic
Net Rev. 7.4 17.7
Quadrupole 600,000 A turns 18 in. wide, hi temp
$5 million vehicle cost; 10 year amortization; 5%/year maintenance magnets superconductor, 8 magnets for
100 passenger or 30 t capacity; 80% load factor; 12 h op/day passenger vehicle, 16 magnets on truck
250 mph average speed; 3 MW propulsion power for passenger and auto carriers
vehicles
4 MW for trucks; 6 cents/KWH Magnetic 10 cm gap between vehicle and
suspension guideway, 0.3 g/cm automatic magnetic
restoring force

be $25 million per two-way mile. The cost to adapt

existing RR tracks for Maglev-2000 operation would be
much less, on the order of $4 million per two-way mile,
since an elevated guideway would not be required.
Table 3 summaries the nominal operating parame-
ters for the Maglev-2000 system. A key element of the
Maglev-2000 system that results in low construction
cost is to mass produce its prefabricated components
in large factories. The components, beams with
attached panels, piers, controls, etc., can then all be
shipped to the construction site and quickly erected at
low cost on pre-poured concrete footings using
 MAGLEV Technology Development 799

conventional cranes. The components can also be read- MAGLEV Technology Development. Table 4 Vehicle
ily exported to other countries. A container ship, for flow and congestion along the I-5 corridor
example, can carry 20 miles of prefabricated two-way
Traffic flow on corridor
guideway.
2007 2035
The proposed 2000 mile West Coast Maglev Net-
work (Fig. 57) along the I-5 corridor and its branches Avg. Max Avg. Max
would connect the metropolitan areas in California, Vehicles/day 71,000 300,000 150,000 600,000
Nevada, Oregon, and Washington State, plus the Trucks/day 10,000 35,000 22,000 70,000
Vancouver, British Columbia, area into a high speed a
Urban segments, 65% 95%
Maglev system that would quietly transport many
% congestion
thousands of passengers, highway trucks, and personal
Rural segments, 31% 85%
autos 6 in. off the rails at speeds of 300 mph and one-
% congestion
fifth the cost of airline travel. Forty-two million persons
a
in California, Nevada, Oregon, and Washington, 85% 550 miles of 1,381 mile total length are urban segments
of the total population of 49 million in the four states,
would live within 15 miles of their local Maglev
station, from which they could reach any other station Table 4 gives data on the vehicle flow and conges-
in the Maglev Network within a few hours. Another tion along the I-5 corridor.
1.3 million persons in the Vancouver area would be Based on the data provided in Table 4 [30], a refer-
served by the network, making the total population ence case for the average flow on the West Coast Maglev
within 15 miles of a Maglev station equal to 43.3 Network was assumed, with
million people. Total government funding is limited
● 5,000 trucks per day (2 per Maglev vehicle)
to $600 million over 5 years for up front demonstration
● 30,000 passengers per day (100 per Maglev vehicle)
and certification activities. After that, freight capability
● 20,000 personal autos carried on Maglev per day
enables building the entire network with private
(10 per Maglev vehicle)
financing.
Typical trip times on the Maglev Network, com- The above traffic flows on Maglev are a significant
pared to going by highway would be: fraction of the 2035 traffic flow, but could be increased
substantially. For the reference case, travels by Maglev
San Diego to Seattle 4 h 30 min vs 25 h 15 min results in the major benefits given below:
San Francisco to Los Angeles 1 h 45 min vs 9 h 40 min ● $21 billion annual reduction in cost of transport
Portland to San Francisco 2 h 30 min vs 12 h 45 min ● 53 billion KWH annual reduction in energy for
Los Angeles to Las Vegas 1 h vs 5 h 30 min transport
● 27 million tons annual reduction in CO2 emissions

In addition to much shorter trip times by Maglev,
the cost of travel by Maglev would be significantly less
for passengers, highway trucks, and personal autos as
compared to existing transport modes:
Passengers – 3 cents per passenger mile (PM) on
Maglev, compared to 40 cents per PM for driving by
auto, 15 cents per PM by air, and 50 cents per PM
by high speed rail.
Highway trucks – 10 cents per ton mile by Maglev
compared to 30 cents per ton mile by highway.
Personal autos – 30 cents per mile by Maglev
transport compared to 40 cents per mile by highway.
The West Coast Maglev Network can be in full oper-
ation by 2019, given an aggressive program with ade-
quate funding. Construction ground breaking would
start in 2015, following a 5 year program to demonstrate
and certify the various types of Maglev vehicles – pas-
senger, auto with passengers carrier, and highway truck
carrier on a test guideway at commercial operating con-
ditions. Planning activities, environmental permits, and
startup of guideway production plants would also be
carried out in parallel with the testing of commercial-
type vehicles. The network would be constructed over
a 4 year period by simultaneously carrying out
800 MAGLEV Technology Development
construction of nine segments of the 2,000 mile network.
Each segment would run between major metropolitan
areas on the network, for example, San Diego to Las
Angeles would be one segment, Anaheim to Las Vegas
another segment, and so on. Each segment of the Net-
work would have one or more guideway beam and pier
manufacturing plants. The projected construction cost is
on the order of $60 billion, based on an average unit cost
of $25 million per two-way mile.
Maglev Energy Storage
Electrical power systems typically experience large
swings in power demand, with low demand during
the night and high demand during the daytime. Even
in the daytime, power demand varies widely, with
major peaks during the morning and late afternoon.
Existing technologies for electrical power storage are
generally too expensive and difficult, for example,
batteries, flywheels, superconducting energy storage
(SMES), or too limited in siting, for example, pumped
hydro. Today, the great majority of peak power demand
is supplied by fossil fueled peaking power plants, for
example, gas turbine, or by purchase from distant
power grids.
A new approach for the storage and rapid delivery
of electric power, based on Maglev, is proposed [31].
This new method, termed MAglev Power Storage
(MAPS) uses Maglev technology similar to that
employed for the transport of passengers and freight
to lift masses a kilometer or more in altitude. In doing
so, electrical energy is stored as gravitational potential
energy. Raising a 100 t mass 2 km in altitude, for
example, stores 0.5 Megawatt Hour (MWH) of energy.
Raising 2,000 such masses would store 1,000 MWH of
energy, which would be returned to the power grid at
appropriate times by simply transporting the masses
down to a lower altitude. During periods of electrical
storage, Maglev vehicles would transport masses up to
a higher altitude storage facility, with the vehicles’
magnetic propulsion system operating in the motor
mode. During periods of electrical power delivery, the
masses would be transported down to lower altitude,
with the propulsion system operating in the generator
mode.
The total energy delivered is determined by the
number of storage masses transported up and down,
while the power delivered is determined by the rate at
which they are transported. The MAPS system is very
flexible and can rapidly alter its power level by changing
the rate at which masses are moved up or down the
guideway. MAPS power level, for example, could go
from zero to 100% of full capability, or from 100% to
zero, in less than 2 min. Peak power capabilities
of 1,000 MW(e) can be generated using MAPS.
Because of its rapid response capability, in addition to
providing peak power, MAPS system could also be used
for low cost spinning reserve. Presently, spinning
reserve is provided by keeping expensive generator
units hot and ready to generate large blocks of power .
While the basic Maglev technology for energy stor-
age is similar to that for passenger and freight trans-
port, there are differences, that is, steeper grades for the
MAPS system (up to 45 vs a maximum of 10 for
passenger/freight transport), and considerably heavier
vehicles (100 t vs a maximum of 50 t for freight
transport). The heavier loaded and much shorter vehi-
cles (6 m for MAPS vs 30 m for passenger/freight
vehicles) use a magnet configuration which maximizes
propulsion force relative to I2R losses in the propulsive
windings. To efficiently levitate the short heavy vehicle
with minimum I2R losses, iron plates in the narrow
beam guideway provide most of the levitation force,
with null flux loops for stability and oscillation
damping. The storage masses are simple reinforced
large concrete blocks on the flat upper surface of the
MAPS vehicle. Typical dimensions for a 100 t storage
mass are 3.5 m wide, 2.5 m high, and 4.5 m long. The
blocks can be rapidly moved onto or off from the
vehicle using a roller bar surface and stored on
pads at the guideways high and low altitude points.
A 100 m  600 m long (6 acres) storage facility could
handle and deliver 2,000 storage masses, equal to
1,000 MWH of electrical energy at an elevation change
of 2 km (6,000 ft). Overall energy efficiency, output
electrical energy/input electrical energy, would be well
above 90% – much higher than any other electrical
storage technology. For a mass unload time of
50 s and a mass load time also of 50 s (both steps are
necessary during a vehicle round trip, whether MAPS is
operating in the power storage or power delivery
mode) a given MAPS vehicle could make 20 round
trips per hour. This is quite conservative, since the
50 s intervals could probably be reduced to about

30 s, considering the simplicity of the transfer process.
The total number of round trips per hour would
then increase.
A very attractive feature is the completely
nonpolluting environmentally benign nature of
MAPS, and its ability to store power at very low cost
from renewable wind and solar energy sources. MAPS
facilities could be built in a short time, for example,
3 years or less, to meet increasing peak power demands.
The total land use for a 2,000 MWH facility, including
a dual 3 km long guideway, is only about 20 acres. Most
MAPS facilities would be sited in remote area and
would not impact the environment. At the design
speed of 130 mph, the vehicles would be virtually silent.
In locations with flat terrain, MAPS guideways
could be installed underground in inclined tunnels or
vertical shafts. Many mining tunnels and shafts operate
at depths of 1 km (3,300 ft) or more. MAPS would
transport its concrete blocks between a storage yard
on the surface, and underground storage cavities
positioned along a guideway in an underground hori-
zontal tunnel. There are many mines, both coal and
hard rock, in the USA that are no longer in use. Such
mines would be attractive sites for MAPS facilities in
flat regions of the USA.
Maglev Water Transport
The availability of ample clean water is a major concern
in many regions, and is expected to be even worse in the
decades ahead – in fact, many experts believe it to be
the number one problem facing the world. A large
fraction of the world’s population, most of them
poor, already lives in a state of water scarcity. World
population is expected to grow to nine billion
(mid-range projection) by 2050. This growth,
along with increasing industrialization, urbanization,
and irrigation, will put even more stress on water
resources.
The advanced second-generation Maglev-2000
system described previously is designed to transport
passengers and freight at very low cost. With design
modifications to the vehicles and guideway, the
second-generation Maglev-2000, termed the Water
Train [32] system, can transport large amounts of
water, that is, on the order of one billion gallons daily
(1,000 megagallons/day), for hundreds of miles.
MAGLEV Technology Development 801
The Water Train vehicles transport a higher load per
unit length of vehicle than for passenger and truck
transport. The weight of loaded Maglev passenger/
freight vehicles is nominally 50 t for a 30 m length.
A water transport vehicle will probably weigh 200–300 t
for the same length. This much heavier loading requires
that the guideway be on-grade, or if it utilizes discrete
supports, that they be relatively close together, for
example, every 20 ft or so. The Water Train uses the
“iron lift” configuration proposed by Danby and
Powell [33], in which superconducting magnets on
the vehicle are attracted upward to laminated steel
plates attached to the guideway. The suspension is
designed so there is a vertical equilibrium point at
which the vertical lift force from the steel plates equals
the weight of the vehicle.
The “iron lift” configuration is vertically stable,
since as the vehicle moves upward, the lift force
becomes smaller, and as it moves downward, the lift
force becomes larger. The iron plates, however, make
the “iron lift” configuration horizontally unstable. This
horizontal instability is countered by the null flux sta-
bility loops on the guideway. The stable restoring force
from the null flux loops is greater than the unstable
force from the attraction to the steel plates, making the
suspension horizontally stable.
An important feature of the Water Train is the
flexible “balloon” that carries the water on each vehicle.
The “skin” of the balloon is a relatively thick, for
example, 1 in., reinforced flexible composite material.
When pressurized and filled with water, the container
forms a streamlined cylinder that runs the full length of
the vehicle. After delivery of the water load and depres-
surization, the flexible skin is depressurized and col-
lapsed minimizing the frontal area and air drag on the
vehicle. The air drag resistance for a long multi-vehicle
train will be much smaller than for a single vehicle.
The equivalent drag per vehicle of the multiunit train is
only 24% of a single vehicle.
The three principal cost components for the Water
Train are propulsion energy and amortization of the
vehicles and guideway. Energy cost scales as V2, and
vehicle amortization scales as 1/V. The faster the vehicle
travels, the more water it can carry in a given period of
time, reducing its amortization cost. Guideway cost is
independent of vehicle speed, and depends only on the
total amount of water carried. For a 1,000 megagallon
802 MAGLEV Technology Development

per day delivery rate – the rate for a large city – Future Directions
amortization of the 300 mile long guideway only costs
Maglev is still at an early stage of development, both in
about 16 cents per 1,000 gal. Adding vehicle amortiza-
its technology and in its various applications. In the
tion and propulsion energy costs, the total cost for
technology area, Maglev will evolve to use high tem-
a 300 mile delivery is only about $1.50/1,000 gal. If
perature superconductors (HTS) that will enable oper-
the Water Train delivers water from a higher altitude
ating temperatures in the range of 20–40 K, with much
to a lower one, part or all of the energy cost would be
lower refrigeration power requirements, that is, W(e)
offset by gravitational energy input from the decrease
of refrigeration electric power per watt(th) of thermal
in altitude. A 600 m drop from the origin to its desti-
input through the cryostat insulation, than those for
nation would enable the transport of water at zero
the present liquid helium cooled superconductors that
energy cost. Over the 500 km delivery distance, this is
operate at 4.2 K. HTS superconducting Maglev mag-
an average grade of about 1 m/km – less than many
nets will require only one-tenth as much refrigeration
natural river systems. A greater drop in altitude would
power input as liquid helium cooled magnets.
enable the sale of surplus electric power generated by
The lower refrigeration power and greater simplic-
the moving vehicles to the electric grid. Accordingly,
ity of HTS magnets will enable Maglev vehicles to be
the Water Train can also generate large amounts of
smaller, carry fewer passengers, and serve a wider range
hydroelectric power at sites where there is substantial
of transport applications. Maglev vehicles can be
water flow and rapid drop in elevation. As an alterna-
adapted for levitated travel in existing subways and
tive to large dams, such an electric generation system
commuter rail lines, providing faster and more fre-
would have significant advantages, including a much
quent service in urban and suburban areas.
lower capital cost and much smaller environmental
Maglev platforms will be developed to provide
impact.
faster, easier, and safer movement of heavy objects in
Elevation changes in terrain traversed by the Water
industrial plants, and movement of ores dug from open
Train are easily accommodated by small changes in
pits and underground mines at lower cost. Maglev
train speed. Even for changes of 200 m (660 ft), the
vehicles will provide fast, convenient, and nonpolluting
velocity changes are modest and acceptable, without
movement of heavy freight containers to and
the need for propulsive power. The Water Train would
from seaports, in place of the thousands of greenhouse
simply speed up slightly as it coasted downhill, or slow
gas–emitting trucks that now provide the service.
down slightly as it coasted uphill, depending on the
local terrain.
Finally, in contrast to long distance pipelines, which Bibliography
require many pumping stations along their length, the
1. O’Toole R (2009) “Gridlock: why we are stuck in traffic and
Water Train, because of its high kinetic energy and low
what to do about it.” CATO Institute
air drag, can coast for hundreds of miles without addi- 2. http://www.eia.doe.gov/bookshelf/brochures/greenhouse/
tional propulsion. After the Water Train attains an Chapter1.htm
initial velocity of 100 m/s (225 mph), and coasts for 3. http://en.wikipedia.org/wiki/List_of_Countries_by_canbon
a distance of 100 miles, the Water Train velocity has dioxide_emissions
4. http://en.wikipedia.org/wiki/chevrolet_volt
dropped to 80% of the initial value, with the average
5. http://en.wikipedia.org/wiki/Transrapid
velocity over the 100 mile segment at 90% of its initial
6. (1914) The Bachelet levitated railway. Nature 93(2324):273
speed. While the Water Train could coast the entire 300 7. Danby G, Powell J (2000) Magnetic levitation: a new mode of
miles without propulsive input, it probably would be transport for the 21st century, lecture given at the award for
desirable to reaccelerate at intervals of 100–150 miles in the Franklin Medal for Engineering, Franklin Institute, 26 Apr
order to keep the average velocity close to the optimum 2000
8. Powell J, Danby G (1966) High speed transport by magneti-
for minimizing cost. At its destination, the Water Train
cally suspended trains. Paper 66-WA/RR-5, presented at 1966
would electromagnetically de-accelerate down to zero Winter ASME Meeting, New York City
velocity, and its kinetic energy fed back as electric 9. www.rtri.or.JP/rd/maglev/html/english/maglev_frame.E.html
power into the grid. 10. http://en.wikipedia.org/wiki/vactrain

11. http://magnetbahnforum.de/index.php?_mf_steering_
committee
12. http://en.wikipedia.org/wiki/Heike_Kamerlingh_Onnes
13. http://en.wikipedia.org/wiki/Large_Hadron_Collider
14. Census Bureau (2006) US statistical abstracts, Table 1046,
transportation outlays by type of transport: 1970 to 2001.
Census Bureau, Washington, DC
15. Powell J, Danby G (1967) High speed transport by magneti-
cally suspended trains. Paper 66-WA/RR-5, presented at 1966
Winter ASME Meeting, New York City; Mech Eng 89:30–35
16. Kolm H et al (1975) The magneplane system. Cryogenics
15:377–384
17. http://en.wikipedia.org/wiki/Maglev_(transport)
18. http://www.upf.org/bering-strait. . ./101. . ./479-Japanese-
maglev-technology
19. http://en.wikipedia.org/wiki/Shanghai_Maglev_Train
20. News.smh.com.au  Technology
21. http://www.rtri.or.jp/rd/maglev/html/english/maglev-frame_E.
html
22. http://www.digitalworldtokjo.com. . ./500kph_maglev_flying_
trains_get_green_light_to_burn_up_japan
23. http://magnetbahnforum.de/index.php?speed-records
24. US Census Bureau (2006) Statistical abstracts of the United
States 2006, tables 1046 and 1047. US Department of Com-
merce, Washington, DC
MAGLEV Technology Development 803
25. Powell J et al (2008) Fabrication and testing of full scale
components for the 2nd generation Maglev-2000 system. In:
Maglev 2008 Conference, San Diego
26. Powell J, Danby G (eds) (1989) Benefits of magnetically
levitated high speed transportation for the United
States, report to United States Senate Committee on environ-
ment and public works, Grumman corporation. Bethpage,
New York
27. Maglev I (2005) The Pennsylvania project: FRA high speed
Maglev deployment program – project description. Transrapid
International and Maglev, Pittsburgh
28. http://www.o-keating.com/hsr/tgv/htm
29. Powell J (ed) (1999) Cost projections for the Maglev-2000
system. Report DPMT-20, Maglev 2000 of Florida
30. http://www.thwa.dot.gov/pressroom/dot0795.htm
31. Powell J (2000) Electrical power storage and delivery using
Maglev: the Maglev power storage system (MAPS). Report
DPMT-14, DPMT Corp., 1 Nov 2000
32. Powell J, Danby G The water train: long distance delivery of
water by Maglev. Report DPMT-3
33. Danby GT, Jackson JW, Powell JR (1974) Force calculations for
hybrid (ferro-null flux) low drag systems. IEEE Trans Magnetic
Mag O:443
804 Mass Transit Science and Technology, Introduction
Mass Transit Science and
Technology, Introduction
GARY L. BROSCH
Center for Urban Transportation Research,
University of South Florida, Tampa, FL, USA
Mass transit, or public transportation, has and will
continue to play a key function in the shaping of
communities. Public transportation plays an enor-
mous role in the economic and social well-being of
individuals and societies. Some have said the study
and understanding of public transportation is not
“rocket science.” Indeed, it is much more difficult, as
it involves not just science, but also public policy and
individual choices.
Mass Transit Science and Technology, Introduction
of the encyclopedia will examine a wide range of public
transportation alternatives from conventional buses
and steel wheeled trains to state-of-the-art BRT (bus
rapid transit) systems and magnetically levitated trains.
It will examine specially designated high-occupancy
vehicle lanes (HOT lanes) and the role of bicycle trans-
port. Also examined will be public policy decisions
concerning a wide range of issues from parking policy
to land use decisions related to transit-oriented
development.
Bus Rapid Transit (BRT) combines technological
advances and public policy changes to provide
a transit system that is faster than conventional buses
and often less expensive and more flexible than light
rail systems. Development of BRT systems is occurring
worldwide as examined in ▶ Bus Rapid Transit: World-
wide History of Development, Key Systems and Policy
Issues. Technological advances such as automated
guidance and signal priority are of use only if institu-
tional issues are addressed. Public policy decisions that
play a crucial role in the implementation of such sys-
tems are addressed in ▶ Bus Rapid Transit, Institu-
tional Issues Related to Implementation.
As Bus Rapid Transit systems have developed, so
has the controversy and debate over choice between
Bus Rapid Transit and conventional light rail (LRT)
systems. Which is better in the long run, which is more
costly, which will attract more riders, which is better for
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
economic development have all been the questions in
this ongoing debate. Two articles, such as ▶ Bus Rapid
Versus Light Rail Transit: Service Quality, Economic,
Environmental and Planning Aspects and ▶ Bus Rapid
Transit and Light Rail Transit Systems: State of Discus-
sion, provide useful insight into the issues involved.
Included in these discussions is a review of the trends of
the last 50 years in urban transportation, the differences
between developed and developing countries, the char-
acteristics of light rail systems and bus rapid transit
systems, and direct comparisons of the alternative modes.
The use of public transportation is determined by
a wide array of direct factors including its cost, conve-
nience, travel time, and perceived safety. Ridership also
is affected by external factors such as the cost of
alternatives, congestion on roadways, and economic
conditions. Public transportation is both affected by
land use patterns and affects land use. Suburban versus
urban developments clearly affect transit ridership
potential. Similarly, transit centers clearly promote
greater density. Balancing public policy goals for land
use development and for use of mass transit can be
facilitated through the use of transit-oriented develop-
ment (TOD). ▶ Bus Versus Rail Implications for
Transit-Oriented Development examines the differences
and opportunities for transit-oriented development when
comparing light rail systems and bus rapid transit sys-
tems. Urban transport systems clearly affect economic
productivity and economic development. It is impor-
tant to understand the differing implications of bus,
bus rapid transit and rail systems on commercial and
residential development surrounding transit stations.
Recent technological advances have enabled policy
makers the opportunity to consider managing highway
express lanes with high occupancy toll (HOT) lanes.
Managing congestion and effective throughput of major
expressways is a challenge. The use of high-occupancy
vehicle lanes is gaining increased acceptance as is the
increasing use of toll roads to finance new construction.
Finding methods and technologies to combine these two
offers the potential to increase the effectiveness of major
roadways. Two articles, ▶ HOT Lanes/Value Pricing:
Planning and Evaluation of MultiClass Service and
▶ High-Occupancy Vehicle and Toll Lanes provide an
understanding of the technological advances and pol-
icy issues surrounding the potential of HOT lanes.
# 2012, DOI 10.1007/978-1-4419-0851-3

Congestion pricing alternatives, equity issues, and
future developments are included.
Light Rail Transit (LRT) is a staple of mass transit
systems in both developed and developing countries.
Its applications range from streetcars and tramways to
modern automated systems. In spite of significant
history and experience with light rail systems, under-
standing the conditions under which new or expanded
systems are appropriate remains a challenge. ▶ Light
Rail Transit, Systemic Viability provides important his-
torical background on the development of LRT, a look
at over 25 new systems developed in the last 30 years, as
well as an analysis of the use of transit in cities with and
without LRT.
As new light rail systems have been implemented
and existing systems expanded, there have been
increasing concerns related to their operation and
interaction with pedestrians in urban areas. Light rail
is most effective when pedestrian interface is easiest.
Yet, this easy access leads directly to greater safety
concerns. Technological and policy innovations such
as regulatory and warning devices, delineation marking
and positive control devices offer potential to reduce
pedestrian injuries. ▶ Light Rail Transit in the US and
Abroad, Examination of History and Innovations
offers a comprehensive examination of the latest inno-
vations in pedestrian/LRT safety.
Traveling on a cushion of air in high-speed trains
is a dream that technological innovation has brought
to reality in the form of magnetically levitated trains,
both high speed and low speed. ▶ MAGLEV Technol-
ogy Development provides the reader with a compre-
hensive review of magnetically levitated vehicles. The
physics, engineering, and costs of magnetic levitation
are covered as well as the history and state of the art of
its development.
High-Speed Rail systems developed in the last
50 years continue to be expanded and implemented
in countries all over the world playing a significant
role changing public transport opportunities. Eco-
nomic changes and technological advances have led
to the increasing popularity of high-speed trains, par-
ticularly in Europe and Asia. ▶ High Speed Rail, Tech-
nology Development of examines technological
developments of high-speed rail as well as the resultant
challenges of implementation.
Mass Transit Science and Technology, Introduction 805
Light rail systems can be effective in moving larger
numbers of people from a limited number of origins
and destinations. Whereas, moving smaller numbers of
people from greater numbers of origins to greater num-
bers of destinations calls for a smaller-scale system.
Personal Rapid Transit (PRT) systems attempt to pro-
vide such a smaller-scale system. Sometimes referred to
as automated people movers, these systems offer an
intriguing concept of quickly moving small groups of
people from one place to another. ▶ Personal Rapid
Transit and Its Development looks at the history of
PRT in several countries and the current state of PRT
development. Also provided is a discussion of the sus-
tainability/energy issues and planning/architectural
issues related to the future of personal rapid transit
systems.
At the other end of the speed spectrum from high-
speed rail is the integration of bicycle transport with
mass transportation. Often overlooked, bicycles are an
important element of a comprehensive transit system.
Finding techniques to successfully integrate the use of
bicycles with mass transit bus and rail systems can
significantly increase the overall effectiveness of transit
systems. A series of case studies and best practices is
found in ▶ Bicycle Integration with Public Transport.
Transit-oriented development, sometimes referred
to as “smart growth” seeks to integrate public trans-
portation investments and land use near transit
stations to improve the sustainability and effectiveness
of both. ▶ Transit-Oriented Development and Land
Use provides a comprehensive look at such develop-
ments in countries ranging from Sweden, to Columbia
and to the USA. Successful examples of the transpor-
tation and environmental benefits of transit-oriented
development are presented. Also presented are the
challenges, the financing, and the future of transit-
oriented development.
▶ Advanced Public Transport Systems, Simulation-
based Evaluation offers the state of the art in computer
simulation of transit operations. The use of these sim-
ulation techniques is valuable in areas such as fleet
management, traveler information systems, electronic
fare payments, and transportation demand manage-
ment systems. Examples are provided of both
microscopic transit simulation and mesoscopic transit
simulation.
806 Night Vision Pedestrian Warning in Intelligent Vehicles

Night Vision Pedestrian Warning in CFAR Constant False Alarm Rate – An adaptive algo-
rithm used especially in radar systems to detect
Intelligent Vehicles targets in noisy measurements.
MATTHIAS SERFLING, OTTO LÖHLEIN CMOS Complementary Metal Oxide Semiconductor –
Machine Perception Systems and Reliability, A technology for constructing integrated circuits.
Environment Perception Daimler AG, Ulm, Germany CTRV Constant Turn Rate and Velocity – A motion
model which assumes an object moves in sample
time T with constant velocity and constant turn rate.
Disparity A term in multiple view geometry in com-
Article Outline puter vision which describes the distance of two
points in two different images by identifying the
Glossary
same point in the scene observed from different
Definition of the Subject and Its Importance
positions.
Introduction
Digital map The knowledge database used in automo-
Discussion of Fusion Concepts
tive navigation systems which provide information
System Overview
about the road network.
Sensor Configuration
Doppler An effect that describes the change in fre-
Boosted Feature-Level Sensor Fusion
quency of a wave for an observer moving relative
Global Positioning and Road Course Estimation
to the source of the wave.
Threat Analysis
EU European Union
Experiments
Epipolar line A line describing the positions in
Conclusion
a sensor coordinate system, which correspond to
Future Directions
a single position in a second sensor coordinate
Acknowledgments
system. Typical term used in stereo camera systems.
Bibliography
False alarm rate A failure measure in detection sys-
tems describing the amount or the rate of detec-
Glossary tions which do not correspond to a proper object.
ACC Adaptive Cruise Control – A cruise control FIR (far-infrared) A device analyzing the infrared
system which controls the vehicle speed in order electromagnetic radiation with a wavelength
to keep a set distance to the vehicle in front. between 0.7 and 300 mm.
AdaBoost Adaptive Boosting – A supervised machine Flat world assumption A simple model in the field of
learning algorithm which combines several weak computer vision which assumes that all objects
learners (very simple classifiers) to construct perceived by a sensor (e.g., a camera) are positioned
a strong learner classifier. in a flat world.
Baseline The baseline of a stereo camera system GPS (global positioning system) A space-based
defines the distance of the corresponding camera global navigation satellite system that provides reli-
coordinate systems and influences the accuracy of able location and time information of an object on
the depth analysis. or near the Earth.
Bayes classifier A simple probabilistic classifier based Haar wavelets A basic wavelet method to decompose
on applying Bayes’ theorem from Bayesian a signal into independent signal parts.
statistics. INS Inertial Navigation System – A system that
CAN Controller Area Network – A standard vehicle uses motion sensors and rotational sensors
bus system which is designed to allow microcon- installed on a platform to continuously calculate via
trollers and devices to communicate with each dead reckoning the position, orientation, and velocity
other within a vehicle. of a moving object without any external references.

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3

MCL Monte Carlo Localization – A method in robot-
ics and sensors to determine the position of a robot
given a map of its environment based on Markov
localization.
NIR Near-Infrared – A device analyzing the electro-
magnetic spectrum from approximately 800 nm to
2,500 nm.
NIRWARN Near-Infrared Warning
Particle Filter A probabilistic method in the field of
computer vision to track objects using a Monte
Carlo approach.
PReVENT Preventive and Active Safety Application,
European founded project.
ProFusion ProFusion Subproject of the PReVENT
project.
RADAR Radio Detection and Ranging – An object-
detection system that uses an electromagnetic wave
to identify the range, direction, (and velocity) of
a target.
ROI Region of interest – A small region selected for
further processing.
ROC Receiver operating characteristic – A graphical
plot to show the fraction of true positives vs. the
fraction of false positives for a binary classifier
system as its discrimination threshold is varied.
Stereo camera system A type of camera system
which observes the environment with two lenses
in order to simulate human binocular vision, and
thus reconstructs the depth information of the
environment.
Two-class problem A typical classification problem
with an object class and a non-object class.
UTM Universal Transverse Mercator – A grid-based
two-dimensional Cartesian coordinate system,
which specifies locations on the surface of the Earth.
VGA Video Graphics Array – A standard VGA camera
has a resolution of 640
480 pixels.
Definition of the Subject and Its Importance
Today’s second generation of automotive night vision
systems uses near-infrared (NIR) or far-infrared (FIR)
cameras combined with integrated pedestrian detec-
tion systems. These systems direct the driver’s attention
by showing all detected pedestrians on a display inside
Night Vision Pedestrian Warning in Intelligent Vehicles 807
the car. This enables the driver to analyze an imminent
situation earlier and react appropriately to prevent
a potential hazardous situation, but the driver has to
decide on his own whether a detected pedestrian is
a potential risk or not. Thus, the next generation of
night vision systems needs to integrate an additional
warning component which signals to the driver only
the relevant objects, that is, all objects on the road or
near the road. Such a warning system needs to detect
objects more reliably and sooner and has to
provide information about the object’s position. More-
over, the system has to know the road course in front of
the vehicle to do a threat analysis of the detected objects
by their position. According to [1] and [2], the
brake reaction time under real-life conditions is about
1.5–2 s. Incorporating also the braking time to stand-
still at a common speed of 100 kph on rural roads, the
driver needs to be warned at least 6 s in advance.
In this context, a reliable and robust pedestrian
detection system as well as a road course estimation
system which covers both factors at a distance up to
120 m is essential for the development of the next
generation of night vision warning system.
As shown in a German funded project NIRWARN
[3], single NIR sensor systems can provide pedestrian
detection reliably up to 90 m. For the enlargement of
the detection range to 120 m, the illumination of the
road and the camera resolution and sensitivity are
limiting factors. Neglecting the cost factor of an auto-
motive FIR camera, which can achieve these ranges, the
problem of providing information about the object’s
position still remains with any kind of single camera
sensor system. In fact, a stereo camera system can
provide position information, but a large baseline
(i.e., at least 0.8 m using a VGA imaging sensor with
lenses of focal length 12 mm) is required to realize
detection ranges of 120 m. A large baseline is difficult
to integrate into a passenger car.
The known approaches for lane or road recognition
systems perform well under daylight conditions on
motorways [4–9], but these systems mostly fail or
have a very limited prediction horizon on rural roads
during the night due to high curvature road types,
missing lane markings, heavily changing lighting con-
ditions, and varying road surfaces.
808 Night Vision Pedestrian Warning in Intelligent Vehicles
This contribution addresses both problems of
a reliable and robust pedestrian detection system with
enlarged detection range, and also provides informa-
tion about the object’s position and a reliable road
course estimation system covering distances up to
120 m in order to carry out a threat analysis and to
signal to the driver only the relevant objects. To achieve
the full functionality, the presented approach combines
the sensor information provided by a NIR camera, an
imaging radar, and a digital map.
Unlike other mid- and high-level radar camera
fusion approaches [10–15], the proposed system real-
izes an early feature-level fusion to detect pedestrians at
night within 120 m in rural and suburban environ-
ment. The presented low-level fusion system shows
better results than a camera-only system, outperforms
a fusion system on higher levels, and manages the
higher computational load of processing raw data.
In order to overcome the drawback of the visibility
constraints of an image sensor, a digital map in com-
bination with a new highly sensitive imaging radar is
used to provide a reliable prediction of the road course.
Unlike an image sensor, the used imaging radar sensor
is capable of clearly detecting the road boundary inde-
pendent of lane markings and for the most part inde-
pendent of weather and lighting conditions at least up
to 60 m. Reflections from trees, posts, traffic signs,
parked cars, and guardrails enlarge the detection
range of the road boundary up to 100 m.
Introduction
Many approaches for pedestrian recognition under
daylight conditions for vision systems have been devel-
oped, especially for shorter detection ranges. Single-
sensor pedestrian recognition at night based on image
processing is widely discussed in [16–19]. The pre-
ferred detection method applied to near-infrared
(NIR) images [16, 17] and far-infrared (FIR) images
[18, 19] is the cascaded classification using Haar wave-
lets. Note that [19] uses a pixel classification prestage to
reduce the input space and uses in addition to the
cascaded classifier, a second classifier based on object
shape matching. A particle filter fuses the detection
results from both classifiers.
On the other hand, multi-sensor pedestrian recog-
nition has been presented in [20] and [21]. A two-step
fusion of a radar sensor and monocular vision sensor is
applied in [20]. The first step generates a list of poten-
tial targets using the radar sensor. The second step
utilizes the images from a vision sensor to verify the
targets, applying flexible shape models trained from
manually extracted pedestrians. Thus the false alarm
rate can be reduced analyzing the image in small
regions of interest (ROI), but the detection rate is
strictly limited by the radar sensor.
A three-level early fusion system for road user
detection including cars and pedestrians is presented
in [21]. The fusion process combines raw sensor data
provided by a laser scanner, a long and a short range
radar, a gray-scale camera, and a FIR camera at an early
stage of the data processing chain and performs a joint
data interpretation with respect to a common model
basis. The first step generates hypotheses for each sen-
sor and merges overlapping hypotheses of different
sensors in one hypothesis and assigns the involved
sensor measurements. In the second fusion level, dif-
ferent sensor features (e.g., velocity, object dimensions,
and FIR image edges) are used to classify the hypoth-
eses. Finally, the third fusion level takes into account
the assigned measurements (e.g., object position) of
each hypothesis to track the object.
This contribution presents a robust and reliable
system for the recognition of pedestrian at night. In
contrast to the fusion techniques above, the developed
detection system fuses raw sensor data using a boosted
classifier. In order to reduce additional costs for new
sensors, fusion approach is applied to a scanning ACC
radar sensor and an NIR camera sensor, commonly
available in modern cars.
The next section gives a general overview of fusion
concepts. Moreover, the advantages of feature-level
fusion compared with a high-level fusion approach are
discussed based on a simple experiment. The developed
application of a pedestrian warning system is presented
in sections “System Overview” to “Threat Analysis.”
First, a system overview gives a short description of all
components. Section “Sensor Configuration” describes
the sensor configuration and the calibration algorithm
to associate measurements from different sensors.
Section “Boosted Feature-Level Sensor Fusion” pre-
sents the developed feature-level fusion approach to
detect pedestrians reliably and robustly with an
enlarged detection range. The developed road course

estimation subsystem is described in section “Global
Positioning and Road Course Estimation.” The last
subsystem, discussed in section “Threat Analysis,”
combines the two subsystems to realize a threat analysis
of detected pedestrians depending on their position on
the road. Section “Experiments” analyzes the perfor-
mance and the results of these subsystems. Finally,
section “Conclusion” provides a conclusion and
section “Future Directions” discusses the impact on
the development of upcoming fusion systems.
Discussion of Fusion Concepts
The special requirements for an automotive pedestrian
recognition system can be achieved by means of two
strategies. First, enhance the sensing performance of the
sensors used in terms of higher spatial resolution and
sensitivity and an improved recognition technology.
Second, exploit synergistically the electromagnetic spec-
trum by fusion of different physical sensor technologies.
Sensor fusion allows exploiting the best of all sen-
sors to overcome the shortcomings of one individual
sensor. This provides for an enriched knowledge about
the environment in comparison to only one sensor or
sensor technology.
The EU projects ProFusion 1 and ProFusion 2 [22]
investigated five multiple-sensor data fusion approaches
for safety applications, namely, early fusion (also named
feature-level fusion), track-level fusion, multilevel
fusion, feedback fusion, and grid-based fusion.
Early fusion techniques combine data provided by
multiple sensors at an early stage of the data processing
chain and perform a joint data interpretation on aggre-
gated data with respect to a common model basis. Two
different models for early fusion are possible: pixel-level
fusion, only applicable for fusion of images with the
same mapping rule, and feature-level fusion, also adapt-
able for arbitrary sensor combinations. Early fusion
techniques demand extensive data processing, which
directly requires more complex hardware, and are less
flexible compared to high-level fusion methodologies.
The standard approach of sensor fusion in the
automotive industry is high-level, or the so-called
object-level fusion or decision-level fusion. Each sensor
provides high-level track information aligned in space
and on a temporal basis, before they are fused by
a track-level fusion module. From an economical
Night Vision Pedestrian Warning in Intelligent Vehicles 809
standpoint [23], there are several advantages. At an
object level, it is easy to modify the constellation on
the sensor setup, for example, the incorporation of
innovative new sensor types, and it is flexible in
means that the interface between sensor suppliers can
be defined at the object level.
However, one potential drawback is the applica-
tion-specific definition of objects, leading to an extra
effort in adaption. Another disadvantage is a potential
loss in raw data or feature information during the
object formation process. Subsection “Motivation for
Feature-Level Fusion” shows a simple example com-
paring feature-level fusion and decision-level fusion.
Approaches like multilevel fusion and feedback
fusion were investigated in the ProFusion projects
[22] and are heavily dependent on the respective appli-
cations, incorporating advanced situation refinement
techniques.
Grid-based fusion techniques require distance
information (e.g., using stereo systems for imaging
sensors) for all sensors, attended with additional
costs. Nevertheless, this contribution shows in which
way it is possible to generate a grid-based representa-
tion fusing an imaging sensor with a ranging sensor by
feature-level fusion.
The elected fusion approach is the feature level
fusion. It uses features extracted from each sensor’s
data building a composite feature vector to classify
the objects. The major advantage of feature-level fusion
is that all information, including the sensors, is com-
bined in an optimal manner. Thus, detecting pedes-
trians can incorporate the information from both
sensors at the same time. Even if neither of the two
sensors is able to detect any single pedestrian on its
own, the fusion approach at the feature level could still
be capable of doing the task. Although the advantages
are well known, feature-level fusion is rarely used in
automotive applications. As mentioned above, this fact
can be explained by unavailable low-level sensor data in
automotive applications and the higher costs for the
transmission with the private CAN or FlexRay link
which is generally necessary.
Motivation for Feature-Level Fusion
The following example demonstrates the advantage of
feature-level fusion over decision-level fusion. A
810 Night Vision Pedestrian Warning in Intelligent Vehicles

10 20 5 10 20 5

Feature 2

Feature 3

Feature 4
Feature 2

Feature 3

Feature 4
5 10 5 10
0 0
0 0 0 0
−5 −10 −5 −5 −10 −5
−10 −5 0 5 −10 −5 0 5 −10 −5 0 5 −10 −5 0 5 −10 −5 0 5 −10 −5 0 5
Feature 1 Feature 1 Feature 1 Feature 1 Feature 1 Feature 1
20 5 20 5

Feature 3

Feature 4
Feature 3

Feature 4
10 10
0 0
0 0

−10 −5 −10 −5
−5 0 −5 0 −5 0 5 10 −5 0 5 10
5 10 5 10
Feature 2 Feature 2 Feature 2 Feature 2
5
5

Feature 4
Feature 4
0
0
−5
−5 −10 0 10 20
−10 0 10 20 Feature 3
Feature 3

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 1

Feature space of a Gaussian distribution with dimension four for a two-class problem. The features are depicted by pairs.
For both classes, the three sigma standard deviation line is illustrated. The left side presents the distribution with
a correlation factor 0. The right side shows it for the maximum correlation

two-class problem is assumed with a four-dimensional 0.05

Feature level fusion
feature space extracted from the measurements of two
Decision level fusion
sensors. The first sensor measures feature 1 and 2. The 0.04
second sensor observes feature 3 and 4. Two classifiers
have been adapted to compare a feature-level fusion
0.03
Error rate

approach with a decision-level fusion approach.

The feature-level approach combines all features in
0.02
one Bayes classifier. The decision-level approach uses
two Bayes’ classifiers to separate the two classes into
their two sensor subspaces of the feature space. The 0.01
inferences of the two classifiers build a new feature
vector, which is then classified by a trained following 0
0 0.2 0.4 0.6 0.8 1
classifier. Scaled correlation factor
For the experiment, the correlation factor between
the features of different sensors is enforced simulta- Night Vision Pedestrian Warning in Intelligent Vehicles.
neously, whereas the correlation factor for features of Figure 2
one sensor is held constant. This is done as long as the Comparison of error rate between feature- and
correlation matrix persists semidefinitely. Figure 1 decision-level fusion: The performance enhancement of
shows the feature space of a Gaussian distribution. the feature level fusion depends on the correlation factor
The features are depicted by pairs. For both classes between the features of the two sensors
the three-sigma bound is illustrated. The left side pre-
sents the distribution with a correlation factor zero maximum arising correlation factor is scaled to 1. Fea-
between the features of different sensors. The right ture-level fusion clearly outperforms decision-level
side shows the feature space for the maximum simu- fusion. The performance enhancement of the feature-
lated correlation. level fusion depends strongly on the correlation factor
In Fig. 2 the achieved error rates as a function of between the features of the two sensors. The risk of
the correlation factor are illustrated, whereby the applying feature level fusion in place of decision-level

fusion is the increased feature space for the classifier.
With the expansion of the feature space, the risk of
over-adaption in the training stage increases.
System Overview
Figure 3 shows an overview of the structure and the
relations between the information sources and the
implemented subsystems necessary to construct
a pedestrian warning application. The application is
composed of three subsystems.
The first part combines the information provided
by the camera sensor and the radar sensor to detect
pedestrians reliably and robustly based on the
presented feature level fusion approach.
The second part estimates the road course using the
information provided by a digital map (i.e., a fixed road
shape in front of the vehicle), the radar observations,
the inertial vehicle measurements, and the initial posi-
tion of a GPS receiver. The goal is to fit the road shape
into the radar measurements, which is achieved by
testing different global vehicle positions in a local area
around the last estimated vehicle position. The GPS
receiver provides the first position estimate to initialize
the algorithm.
The third part combines the results of the previous
subsystems to analyze the relevance of each detected
pedestrian depending on their position on the road.
This threat analysis gives the opportunity to direct
the driver’s attention to a situation of imminent
danger.
Camera
Pedestrian detection
Radar
INS
Road course
Digital estimation
map
Initial
GPS
position
Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 3
System overview: The application is composed of three subsystems, the pedestrian detection, the road course estimation,
and threat analysis
Night Vision Pedestrian Warning in Intelligent Vehicles 811
Sensor Configuration
The presented pedestrian recognition system fuses data
from a camera sensor and a radar sensor. Figure 4a
shows the sensor mounting positions in the research
vehicle: the camera behind the windshield and the
radar sensor in front of the radiator grill.
The vision component corresponds to the
Mercedes-Benz Night View Assist, consisting of an NIR
CMOS camera sensor and infrared headlamps provid-
ing roughly the same range of visibility as high beam
headlamps – but without impairing the vision of the
oncoming traffic. The camera is configured with
12 mm optics and supplies 12 bit gray-scale VGA
images.
The advanced radar sensor ARS300 is a new
mechanically scanning 77 GHz FMCW radar sensor
developed by the Continental AG. Figure 5 gives an
impression of the radar measurement raw data which
consists of the intensity iðr; ’Þ of the reflected signal in
dB and the relative speed vrel(r, ’) of reflecting objects
in kph with range fr 2 j0m  r  199mg and azi-
muthal angle f’ 2 j  8  ’  8 g. In this config-
uration each cell (r, ’) has the dimension of 1 m
1 .
The resolution in relative speed is specified with
2.76 kph.
Sensor Alignment
In order to associate measurements from multiple sen-
sors, the alignment of each sensor to the vehicle system
Threat analysis
812 Night Vision Pedestrian Warning in Intelligent Vehicles

a b

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 4

(a) Research vehicle with sensor mounting positions and (b) 3rd generation ACC advanced radar sensor ARS300 from the
Continental AG

60 m

40 m

20 m

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 5

Radar measurements: The left image depicts the experienced scene from the camera. The image in the middle shows the
radar intensity (value) and relative speed (hue). The three moving pedestrians can be recognized by the green hue
component at about 41 m, 75 m, and 108 m. The right image shows all radar intensity values (blue corresponds to low and
red to higher intensity)

(DIN70000) needs to be accurately determined. The
radar system is calibrated using the alignment proce-
dure developed by the Continental AG. The centered
radar beam is aligned in azimuth and elevation to
the longitudinal vehicle axis. However, to calibrate
the camera system, the fundamental matrix
F ¼ Fintr Fextr containing the intrinsic parameters and
the extrinsic parameters needs to be estimated. While
the intrinsic camera parameters (e.g., focal length,
principal point, skew and distortion coefficients)
are estimated using a standard camera calibration tech-
nique [24–26], the extrinsic parameters (mounting
position and Euler angles) are assessed by solving
a nonlinear least-square data fitting problem of the
calibration target for N positions:
N 
X   2
^extr ¼ arg min u
F Fintr Fextr xðri ; ’i Þ  ;
Fextr i¼1
v i
where xðri ; ’i Þand (u, v)T determine the position of the
calibration target in the vehicle coordinate system and
in the image plane, respectively. An optimal calibration
target in the radar plane is a trihedral corner reflector
[27], reflecting the emitted electromagnetic wave
 Night Vision Pedestrian Warning in Intelligent Vehicles 813

35

30

25

20

x[m]
15

10

0
5 0 –5
a b y[m]

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 6

Calibration result: (a) Composed image of M target positions with M  N and the projection of N target positions into the
image. (b) N radar measurements of the trihedral corner reflector in the vehicle coordinate system

directly backward to the radar sensor. A 2D Gaussian

Set of camera Set of radar
function is fitted to the intensity data i(r,’) to improve hypothesis hypothesis
the measured discrete position of the target in the radar
frame with Dr = 1 m and D’ = 1 . The tip (u, v)T of Model of camera

each corner position is used as a reference point in the Hypotheses association

image plane. Figure 6 shows the result of the Model of radar
calibration.

Feature Feature
Camera Radar
Boosted Feature-Level Sensor Fusion extraction extraction

Fusion Architecture Feature level fusion

In order to increase robustness of the pedestrian detec-

tion, the system combines the information from com-
plementary sensors (i.e., a camera and a radar sensor)
on a feature level using the supervised training algo-
rithm AdaBoost [28]. The sensor fusion architecture
consists of three layers (see Fig. 7):
1. Firstly, hypotheses are generated within the search
space. Hypotheses originating from the different
sensors are associated based on model informa-
tion (camera and radar), and eventually on detec-
tion results from the previous detection stage
(coarse-to-fine search, see section “Multi-Sensor
Hypotheses”).
Classifier
Night Vision Pedestrian Warning in Intelligent Vehicles.
Figure 7
Sensor fusion architecture
2. In the second processing step, the features are cal-
culated and organized in a combined set for both
sensors.
3. The classification itself is done in the third
processing step based on the calculated features.
814 Night Vision Pedestrian Warning in Intelligent Vehicles
The layers two and three are optimized simulta-
neously. Feature level fusion is achieved in the following
way: Features are obtained from each sensor stream
simultaneously. These features are held in one set. Dur-
ing the training stage, the source of the individual
features and their type do not matter. AdaBoost selects
incrementally those features having the lowest weighted
error on the training set, until the algorithm achieves
a predefined correct and false detection rate. This tech-
nique assures an optimal selection and weighting of the
features from both sensors for the classification task.
All features are defined within search windows
belonging to the different sensors. The tuple of search
windows (one for each sensor input) describes
a hypothesis presented to the classifier. The latter
assumes that both search windows identify the same
localization of the scene. This requires an accurate
mapping between the camera plane and the radar
plane. However, combining multiple sensors also
increases the input space and, consequently, increases
the computational effort due to the ambiguous map-
ping between the camera plane and the radar plane.
The radar measurement iðr; ’Þ defines an ROI in
the camera image (see Fig. 8), which is much bigger
than the projected object at the measured position
ðr; ’Þ. Moreover the object height cannot be measured
by the radar system. Thus multiple search windows in
Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 8
Ambiguous mapping between the camera and the radar plane: A radar measurement iðr; ’Þ defines a much bigger ROI
(green) in the image plane than the corresponding pedestrian object (dashed blue). Thus the measurement ðr; ’Þ belongs
to multiple search windows in the corresponding image area
the image are possible to be associated to a single radar
measurement iðr; ’Þ. Section “Multi-Sensor Hypothe-
ses” describes the association rule in detail.
In order to realize a real-time detection system, the
presented fusion system incorporates cascaded classi-
fiers introduced by Viola and Jones [29], [30]. Such
a classifier is capable of separating regions in the input
space into two classes, namely, the object and back-
ground class using a chain of classifier stages (see
Fig. 9). With an increasing number of stages, the com-
plexity of these classifiers rises by the number of fea-
tures used. Each classifier stage is trained using the
AdaBoost algorithm to choose the most significant
features from both sensors. Depending on the discrim-
inative power of the features from the camera sensor
and the radar sensor, the resulting cascade can perform
the classification task very efficiently. Subsection
“Radar Features” describes the derived radar features
in detail.
The AdaBoost Approach to Machine Learning
Boosted classifiers and the related AdaBoost algorithm
introduced by Freund and Schapire [28] are very well
described in [31–33]. The basic idea is that a set of weak
classifiers build a strong classifier. The decision of the
strong classifier is a weighted voting of the decisions of
 Night Vision Pedestrian Warning in Intelligent Vehicles 815

Classifier N
Classifier 1

Classifier 2
Hypotheses Detection

Rejected hypotheses

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 9

Cascaded classifier: Each stage discards hypotheses and passes promising samples onto the successive and more
complex stage

Require:
(1) (2) (M )
N samples (xj ,xj ,...,xj ,yj ) for M sensors with
j = 1,...,N and yj ∈{–1,+1}
τ : minum detection rate
T : maximum number of selectable weak classifiers
Ensure:
The final strong learner decision H(x) is composed of the most discriminative weak
learners ht(k) and achieves the required detection rate.

1: initialize wj = 1
N
2: for feature t = 1, . . . , T do
3: determine weak learner ht(k) that minimizes the weighted error et on the
training data
4: choose a1, the importance of ht(k)
5: update sample weights and normalize so that w remain a distribution:
(k) (k)
wj = 1 wj exp −αtyt (x1 )
nt
with nt as a normalizing factor
6: end for
7: adjust q of the final strong learner to realize the minimum detection rate t :
T
(k) (k)
H(x ) = sgn ∑ αt ht (x ) − θ
t=1

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 10

Pseudo-code description of the multi-sensor AdaBoost algorithm

the weak classifiers. The weak classifiers are selected
subsequently in the trainings phase. On each round,
the weights of incorrectly classified examples increase,
the weights of correctly classified examples decrease, so
that the next weak classifier focuses more on incorrectly
classified samples. The importance of the weak classi-
fier depends on its error on the weighted sample set.
A pseudo-code description of the multi-sensor algo-
rithm regarding a two-class classification scenario is
given in Fig. 10.
816 Night Vision Pedestrian Warning in Intelligent Vehicles

The algorithm requires a multi-sensor train- ● Subset S1 consists of filters describing the average

ing set of  
N samples 
with value of the reflected object intensity
ð1Þ ð2Þ ðMÞ ð1Þ ð2Þ ðMÞ
x1 ;x1 ;.. .;x1 ;y1 ;. ..; xN ; xN ;. ..;xN ;yN as ● Subset S2 consists of complex signal filters with
ðkÞ
input where xj represents the input of the kth sensor a structure similar to the radar CFAR (constant
and yj 2 f1;þ1g the corresponding manually false alarm rate) filters [27]
extracted label (i.e., 1 for a background and +1 for
Contrary to the image plane, a pedestrian cannot be
an object sample). For all iterations t ¼ 1; ...; T
distinguished clearly in the radar plane because the
AdaBoost calls a weak learning algorithm to train the
radar signal of a typical pedestrian is similar to many
base classifier with importance at 2  on a weighted
other objects, for example, posts, street lamps, and
version of the training set by minimizing the weighted
trees. In other words, these filters are not capable of
error et :
unambiguously classifying a pedestrian. However, they
 ðkÞ 
XN h ðx k Þ  yj  can be used to reject background samples very effi-
 t;i j 
et ¼ min wi;j   ciently. The manually labeled pedestrian in the radar
i;k  2 
j¼1
image at position ðr; ’Þ defines the center of the search
  window with a fixed size 5 m
5 . Figure 11 shows the
1 1  et
at ¼ ln : expanded field of view by mirroring the corresponding
2 et
beam directions at the left and the right border. This
Thus each round AdaBoost chooses the best weak allows for apply the filters to all beam directions ’.
ðkÞ
learner hi from the set of specified features of all
sensors (e.g., an image feature or a radar feature). The 1. Average object intensity filter: The reflected signal
final strong learner decision H is a weighted majority intensity iðr; ’Þ can be used to very efficiently sep-
vote of the chosen T weak learners: arate between object and background. This can also
8 be seen in Fig. 5 (in the radar intensity measure-
< PT
ðkÞ ments). Within the search window, a rectangular
þ1 : at ht ðx ðkÞ Þ  0
HðxÞ ¼ filter is used to calculate the mean signal intensity.
: t¼1
1 : else: Thus subset S1 consists of different rectangular fil-
Applying this algorithm to each stage of a cascaded ters varying in position and size within the search
classifier, the decision in the first layers with less weak window.
learners can be dominated by one of the sensors with 2. Complex signal filter: The complex signal filter com-
the most discriminative power of their features. bines the reflected signal intensity iðr; ’Þ and the

Image Features
Lateral position y [m]

20
In the image plane, an object (e.g., pedestrian) is
10
represented by a rectangular search window, which is
0
defined by its upper left and lower right corner with
a fixed width-to-height ratio (e.g., 0.5). The feature set −10
used in the image plane consists of Haar-wavelet filters −20
[34, 35], which vary in size and position within the 0 20 40 60 80 100 120
rectangular search window. These features have been Distance x [m]
successfully applied in many pedestrian recognition Night Vision Pedestrian Warning in Intelligent Vehicles.
systems [16, 17, 35], and are able to compute efficiently
Figure 11
using integral images.
Definition of the search window in the radar plane
projected onto the Cartesian vehicle plane: Each position
Radar Features
ðr; ’Þ indicates the centre of the search window with
In the radar plane, the feature set is composed of two a fixed size of 5 m
5 . The field of view is expanded in ’ to
different filter subsets: apply the filters to all beam directions

relative object velocity vrel ðr; ’Þ to detect object
clusters of similar velocity. With regard to the the-
ory of radar systems, the measured object intensity
and relative velocity represent the highest peak in
the Doppler spectrum at the measured
positionðr; ’Þ. Hence the complex number zðr; ’Þ
describes the radar measurement ði; vrel Þ at position
ðr; ’Þ equally:
zðr; ’Þ ¼ Az expjpo ; z 2 
Az ¼ iðr; ’Þ
o ¼ f ðvrel ðr; ’ÞÞ þ vego ; fo 2 j jo  1:0jg ;
where Az corresponds to the measured intensity, vego
defines the driving speed, and o describes the mea-
sured absolute velocity of measured objects after the
normalization.
A complex signal filter is composed of three rectan-
gular zones (see Fig. 12): a background zone A, a blind
zone B, and a test zone C with C  B  A. This corre-
sponds to the CFAR filter structure [27] extended
to two dimensions. Arranging the three zones in all
positions and sizes within a search window of the size
5 m
5 results in 44,100 different filters. The real-
valued filter value is defined by:
 
 
jAnBj X X 
 zðr; ’Þ zðr; ’Þ 
 jC j  of a minimum size and a maximum size pedestrian can
 8ðr;’Þ2C 8ðr;’Þ2AnB 
which is equivalent to the norm of normalized differ-
ences between the integrated complex values zðr; ’Þ
over C and A\B. Figure 13 shows the filter calculation
and the resulting filter value which describes a local sim-
ilarity measure of velocity differences. These filters sup-
press noisy measurements indicated by a small complex
modulus and a random complex argument in zðr; ’Þ.
Night Vision Pedestrian Warning in Intelligent Vehicles.
Figure 12
A complex signal filter is defined by three zones:
a background zone A, a blind zone B and a test zone C with
CBA
Night Vision Pedestrian Warning in Intelligent Vehicles 817
However, the defined filters cannot distinguish between
moving pedestrians and small stationary objects with
a similar intensity pattern (e.g., posts): both result in
high filter responses. In contrast, low filter responses
are caused by bigger objects with intensity patterns
different from pedestrian patterns (e.g., cars, houses).
In order to reduce the computational complexity, the
algorithm decomposes the complex number z into their
components Re{z} and Im{z} and uses integral images
to calculate the filter value components independently.
Multi-Sensor Hypotheses
The presented cascaded classifier requires a tuple of
search windows identifying the same scene in the
image plane and in the radar plane. Search windows
are generated in a range from 20 to 120 m, which
corresponds to pedestrian sizes in the image plane
from 120 to 21 px. Due to the ambiguous mapping
between both sensors (see Fig. 8), this results in multi-
ple search windows in the image plane associated to
one search window in the radar plane. In order to
efficiently perform on the increased input space, the
classifier scans the image plane restricted to the ROI
using the relaxed flat world assumption presented in
[19] that compensates the pitch of the vehicle.
Using the camera model, the search window height
be calculated at each distance r of the radar sensor.
These heights represent the limits of scaled versions of
the minimum search window height at distance r. The
combination of the restricted image plane and the set
of search window heights define the resulting grid of
search windows in the image plane associated to
the search windows with distance r in the radar plane.
The resulting amount of hypotheses to be classified
remains unmanageable for real-time applications.
With a coarse-to-fine search, generating a set of search
window grids in the image plane with different position
sampling increments the problem can be solved. First,
the detector classifies the coarse grid and continues the
classification in the fine grid for positive classified search
windows.
Global Positioning and Road Course Estimation
The previous sections presented the feature level sensor
fusion approach to detect pedestrians at night using an
818 Night Vision Pedestrian Warning in Intelligent Vehicles

15 250

200
10

150
Imaginary part

Imaginary part
805.76
5

100

0
50

−5
0

−10 −50
−20 0 20 40 60 −500 0 500 1000
Real part Real part

15 20

15
10

10
Imaginary part

Imaginary part

5
148.04
0
0

−5
−5

−10 −10
−20 0 20 40 60 −50 0 50 100 150
Real part Real part

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 13

Complex signal filter response: Given a search window, the corresponding radar measurements ði; vrel Þ are assigned to the
filter zones (left figures): background (red), blind (the measurements which are not taken into account) and test (green).
The upper graphics show the optimal filter and the lower graphics show a suboptimal filter in the complex plane related
to the given radar measurements. The filter response (right graphic) is shown by the complex number in blue with the
corresponding real-valued filter result as text

NIR camera and an imaging radar. In order to realize
a warning system which directs the driver’s attention
primarily to the relevant objects, the information about
the pedestrian’s position is combined with a new road
course estimation approach. Instead of estimating the
shape of the road, this contribution estimates the posi-
tion on a digital map and extracts the preceding road
course directly from the map. In contrast to loosely and
tightly coupled INS/GPS solutions, the main concept of
this contribution is based on the method to localize
a robot in a known defined environment, also known as
Monte Carlo Localization (MCL) [36, 37].
MCL algorithm uses a particle filter approach [38].
Each particle describes a position and an orientation of

the robot in a restricted environment which has been
mapped using the sensor measurements of the robot
(e.g., ultrasonic, radar, laser, and camera). At the begin-
ning, the position of the robot is unknown. Thus each
position in the restricted environment is possible, lead-
ing to a uniform distribution of the particles on the
map. In order to localize the robot on the map, the
algorithm verifies each position based on the local
sensor measurements. After a few iterations, the parti-
cles will converge to the most likely position and thus
the robot is localized.
A significant difference to a robot is that the envi-
ronment of a passenger car corresponds to an
unrestricted network of roads. Thus, the position cannot
be uniformly distributed in the initialization cycle. This
problem can be solved by means of a low-cost GPS
sensor which restricts the area of possible initial posi-
tions to the maximum position uncertainty of 25 m.
The position and the orientation of the ego vehicle are
described in UTM (Universal Transverse Mercator)
coordinates [39] which project meridian segments into
a rectangular plane with negligible distortion error. In
order to verify a position hypothesis, the observations of
the imaging radar are compared with the extracted road
course of an electronic horizon component (e.g., [40]).
Note that the electronic horizon component also
performs a map matching using the GPS position
and a digital map. But the calculated position does
not regard the width of the road, and the orientation
of the vehicle cannot be estimated precisely.
The resulting state vector of a position hypothesis is
defined by
 
W ðiÞ
qðiÞ
k
¼ ekW ; nW
k ; Fk ; bk
W
:
The scalar components ekW ; nW
k ; Fk ; bk of the par-
W W
ticle number (i) denote the position in UTM (east and
north coordinate), the orientation of the vehicle, and
the width of the road for iteration k.
The main steps of the positioning algorithm using
a particle filter approach can be outlined as follows:
1. Sample from the probability density pðqk1 jy 0:k1 Þ
of the previous iteration generates a new set of
ðiÞ
particles. Each particle qk describes a vehicle coor-
dinate system in a global UTM coordinate system.
2. Apply a deterministic and a stochastic drift to pre-
dict the hypotheses to the most likely position of
Night Vision Pedestrian Warning in Intelligent Vehicles 819
the current iteration. The deterministic portion
corresponds to the egomotion of the ego vehicle
during an iteration step which is estimated by an
extended Kalman filter (EKF). The stochastic por-
tion models the process uncertainties by an unbi-
ased Gaussian and yields to a random distribution
of equally sampled particles.
3. Calculate the likelihood for each particle based on
the measurements of the imaging radar sensor. This
step is based on an intensity model of the road,
which expects that the road itself is free of objects
and narrowed by high intensities from stationary
objects. Hence a preprocessing step of the
radar measurements is required to compensate
the high intensities caused by moving objects on
the road.
4. The final step is to generate the probability density
pðqk jy 0:k Þ based on the weighted particle set.
The following subsections discuss the egomotion
estimation, the radar preprocessing, and the likelihood
function in detail.
Egomotion Estimation
In order to measure the egomotion of the vehicle, an
EKF approach is used which considers the inertial
sensor measurements providing information about
_
the turn rate Cand the wheel rates dfl,fr,rl,rr of all four
wheels. As discussed in [41], a Constant Turn Rate and
Velocity (CTRV) motion model is totally sufficient to
describe the common movements of a vehicle. The
state space and the continuous differential equations
are defined by
_ T
x ¼ x; y; v; C; C
with
x_ ¼ vðtÞ cos CðtÞ;
y_ ¼ vðtÞ sin CðtÞ;
v_ ¼ 0;
C€ ¼ 0:
The scalar components x, y, v, C and C _ denote
the longitudinal and the lateral (to the left) movement,
the longitudinal velocity, the rotation angle, and the
turn rate. This curvilinear motion model involves
a nonlinear state transition as well as a nonlinear mea-
surement model using a four-wheel vehicle model.
820 Night Vision Pedestrian Warning in Intelligent Vehicles

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 14

Egomotion accuracy analysis: The figure shows the estimated course of a closed-loop scenario of length 4.5 km in
UTM coordinates. The resulting estimation error between the starting and the end position (represented by the arrow in
driving direction) is about 2.4 m

The discretized formulation based on the sample time motion (occurs mostly in sharp turns and when the
T can be derived by solving vehicle accelerates). The accuracy of the egomotion
estimation algorithm is shown in Fig. 14.
ZtþT
The required incremental movement between con-
xkþ1 ¼ xk þ xðtÞdt:
secutive radar measurements is provided by a reset
t
operation after each cycle of the positioning algorithm.
This yields to For this purpose, the scalar components x, y, and C are
vk set to zero and the corresponding entries in the covari-
xkþ1 ¼ xk þ ð sinðCk Þ þ sinðwk T þ Ck ÞÞ; ance matrix are set to their initial values.
wk
vk
ykþ1 ¼ yk þ ðcosðCk Þ  cosðwk T þ Ck ÞÞ;
wk
vkþ1 ¼ xk ; Radar Preprocessing
Ckþ1 ¼ Ck þ wk T ; Moving obstacles in the field of view of the radar sensor
wkþ1 ¼ wk ðthe discrete turn rateÞ; have a huge impact on the evaluation of position
when jwk j ! 0 then use hypotheses because the algorithm assumes a road
model with low intensities on the road and high inten-
xkþ1 ¼ xk þ vk T cosðCk Þ;
sities on the road border. Due to the high radar cross
ykþ1 ¼ yk þ vk T sinðCk Þ : section of vehicles, a moving vehicle on the road vio-
In order to handle the asynchronous arriving turn lates this assumption. In order to diminish this impact,
rate and wheel rate measurements, the EKF implemen- a preprocessing step is applied replacing the intensity
tation corresponds to the sequential measurement measurements by their corresponding noise level value
update formulation presented in [42], which is further (measured by the radar sensor for each cell (r; ’)) when
proved to be efficient and numerically stable. The filter the measurement represents a moving object. The
consistency is maintained by testing each measurement moving object decision is based on the measured radial
innovation against its predicted innovation uncertainty relative velocity vDR (Doppler), its direction ’R and the
in a three-sigma bound which covers 99.7% of the estimated velocity of the ego vehicle vego .
expected measurements. A failed test will reset the filter  
Moving object , vDR cosð’R Þ  vego   gv ;
to the initial state covariance, which happens when the
CTRV process model does not represent the vehicle’s where gv denotes the velocity threshold.
 Night Vision Pedestrian Warning in Intelligent Vehicles 821

Evaluation of Position Hypotheses The inflection point a denotes the barrier to evalu-
ate a hypothesis as a positive road course candidate.
Each hypothesis or particle defines a position and an
The nonlinear characteristic of gð Þ improves the con-
orientation of the vehicle in UTM coordinates. Given
vergence behavior. Furthermore, the factor b models
that position and the road course from the digital map,
the ratio of the estimated road course in the radar’s
an individual course of the preceding road segment can
field of view.
be described in vehicle coordinates. This provides the
Figure 15 shows three iteration steps of the position-
possibility to establish a relationship between the digital
ing algorithm. The expected road course after the 5th
map data and the radar observations of the environ-
iteration is shown in the Cartesian representation of the
ment. The estimation of the road width is used to define
radar measurements. Both polylines are partly outside of
two polylines describing the boundary of the road. The
the field of view which yields to the ratio b < 1:0.
next step transforms these two polylines for each particle
into the bilinear interpolated Cartesian representation of
the preprocessed polar radar intensity measurements Threat Analysis
(see Fig. 15). Thus, a goodness of fit measure can be
After the detailed description of the pedestrian detec-
computed for each particle. This evaluation is based on
tion system and the road course estimation system, the
the mean intensity value between the polylines U1 and
final task is to fuse both information sources to realize
the mean intensity value below the polylines
a pedestrian warning system. A range-dependent object
U2 according to the logistic function
warning is straightforward using the position informa-
tion provided by the detection system. The more chal-
gðU2  U1 Þ ¼ b ð1 þ expððU2  U1  aÞÞÞ1 : lenging task, discussed here, is to warn the driver about

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 15

Results of the position estimation process: The small maps show the particle distribution at initialization of the algorithm
after the 2nd and after the 5th iteration. The sensor observations of the camera and the radar sensor as well as the
estimated road course after the 5th iteration step is shown above
822 Night Vision Pedestrian Warning in Intelligent Vehicles
1 Figure 17 shows the result of the realized threat analy-
left left right right
sidewalk lane lane sidewalk
0.5
0 Pedestrian Detection System
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1
Night Vision Pedestrian Warning in Intelligent Vehicles.
Figure 16
The desired warning scheme on a two-lane road: The most
threatening object position is in the centre of the right
lane. The width of the road is normalized to one
objects depending on their lateral position to the road
which requires a threat analysis based on both infor-
mation sources. Figure 16 shows the desired warning
scheme with low ratings on the sidewalk and the
highest rating in the center of the right lane. Moreover,
a pedestrian on the right sidewalk poses a higher threat
to the driver than a pedestrian on the left sidewalk,
under the assumption of driving in the right lane.
An advantage of the presented system architecture
is the shared usage of the radar data in the detection
system and in the road course estimation system. This
gives the opportunity to measure directly the relevance
of an object. The detection system describes each object
position in polar radar coordinates ðri ; ’j Þ and in
image coordinates ðu; v; w; hÞ. The road course estima-
tion system delivers the polylines of the road boundary
in Cartesian radar coordinates which yields to a lateral
resolution problem when using ðri ; ’j Þ directly. Hence
the image position is used to define a more precise
object position in ðri ; ’j Þ. The calculation is based on
a linear interpolation between the calibrated left and
right image columns of the object beam ’j .
This refined position defined by ðri ; ’
j Þ is then by the pedestrian label in the image plane. While
transformed into the Cartesian view to measure the
distance d to the left polyline of the estimated road
course. The final warning level of a detected object
corresponds to
!
ðd
 0:75Þ2 Þ option 3 achieves the most promising results. Using
ðd Þ ¼ exp  : this assistance the labeler can use the initial conjecture
2s2
The distance d
is normalized by the estimated road in order to achieve good distance estimates of labeled
width and s2 chosen to 0.1 defines the warning focus.
sis. The warning level changes from red to green for
a pedestrian walking from the center of right lane to the
left sidewalk.
Experiments
The following subsections discuss the performance of
the feature-level fusion approach used in the pedestrian
detection system. This requires methods for ground
truth generation, a set of recorded sequences for train-
ing and testing and an evaluation criterion.
Defining a Ground Truth Each supervised classifi-
cation task requires a consistent ground truth for
training and evaluating a classifier. With regard to a
multi-sensor configuration, consistency is much more
important because an object label in one sensor plane
has to identify the same object in the other sensor
plane. Figure 17 shows a typical scene of pedestrians
in a suburban environment. Labeling the pedestrians in
the image is rather simple and easy to learn. However,
labeling the same pedestrian in the radar image is much
more complicated and requires an expert or some kind
of label assistance. Different label assistance options are
possible:
1. Monocular distance estimation assuming a fixed
pedestrian size of the label in the image plane
2. CFAR detector applied to the projected epipolar
line in the radar plane given by an assumed mini-
mum and maximum pedestrian size
3. Stereo distance estimation analyzing the pixel dis-
parity within the label in the image plane
Note that they all utilize the information provided
option 1 fails due to poor label accuracy, especially in
a high detection range, and option 2 fails due to the
fixed structure of the CFAR filter mismatching the
pedestrian with other objects on the epipolar line,
and correct the position if necessary. This method
implies also a stereo camera setup with a large baseline
pedestrians up to 120 m.
 Night Vision Pedestrian Warning in Intelligent Vehicles 823

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 17

Result of the threat analysis: The left image shows a walking pedestrian from the right lane centre at iteration k to the left
sidewalk at iteration k + N. While the pedestrian is walking to the left sidewalk the warning level of the pedestrian is
decreasing from red to green. The figures on the right show the corresponding pedestrian position as well as the
estimated road course in the Cartesian radar measurement view

Training and Evaluation Settings To demonstrate
the performance of the presented detection system,
two pedestrian detectors were trained: an image-based
reference detector using only the Haar-wavelet filters in
the image plane and a fusion detector using addition-
ally the defined radar features.
The training set consists of 36,562 images taken
from 122 pedestrian sequences and 71 non-pedestrian
sequences recorded nightly in the period from July
2007 to June 2008. The collection of images contains
17,833 pedestrian samples and represents a good mix-
ture of pedestrians in rural and suburban environments
within 120 m. Both detectors use 30,000 background
samples selected randomly in each stage of the cascade
from the image set.
The test set contains 7,887 pedestrian samples in
8,449 images. In order to compare the performance of
both detectors equitably, an evaluation algorithm ana-
lyzes the detections in the image plane restricted by the
field of view of the radar sensor. All detections are
separated in true and false positives using a coverage
measure cov(A,B) between the rectangular detection
A and the pedestrian label B:
A\B
covðA; BÞ ¼ : system using only the information provided by the
A[B
The threshold gmin defines the minimum required
coverage to assign detections to the set of true positives.
Night Vision Pedestrian Warning in Intelligent Vehicles.
Table 1 Detector characteristics
Stages Image features Radar features
Fusion 28 1061 407
Reference 30 1621 –
Detection Results The trained fusion detector signif-
icantly outperforms the reference detector while
requiring less stages and thus less features (see Table 1).
Furthermore, AdaBoost selects up to 38% of the fea-
tures from the set of radar features which emphasizes
their discriminative power. These radar features afford
a fast and cost-effective reduction of hypotheses in
early stages of the cascaded fusion detector. Figure 18
compares the computational cost of both classifiers in
environments of different complexity. Figure 19 shows
the ROC curves comparing the performance of the
fusion detector and the image based reference detector
for gmin ¼ f0:5; 0:6g. For the fusion detector, the
detection rate is higher and the false alarm rate is
lower than the performance values of the reference
NIR camera. Furthermore, system performance
increases with the required minimum coverage gmin .
This behavior indicates the detection quality in spatial
824 Night Vision Pedestrian Warning in Intelligent Vehicles

Reference Suburban road (lights, buildings, pedestrians)

Fusion

Reference Rural road without pedestrians

Fusion

Reference Forest bounded road with pedestrians

Fusion

Reference Country lane with pedestrians

Fusion

0 1 2 3 4

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 18

Computational cost analysis in the early stages of the cascade up to layer 12: The image based reference detector and the
fusion detector classify the same set of hypotheses. In stage one both detectors analyze about 2 million hypotheses using
five weak learners (referenced as cost 100%). Due to the combination of radar features and image features, the fusion
detector can reject more hypotheses than the image-based reference detector. Thus, the number of weak learner
calculations can be reduced more efficiently. The fusion detector also shows similar results in highly structured
environments, such as suburban areas with buildings and heavily changing light conditions

0.9

0.8

0.7
Detection rate

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
False alarms per image

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 19

Performance comparison between the image based reference detector (solid) and the fusion detector (dash-dot) for the
coverage 0.5 (blue, upper two graphs) and 0.6 (cyan, lower two graphs)
 Night Vision Pedestrian Warning in Intelligent Vehicles 825

resolution: While the fusion detector uses the informa-
tion provided by the radar sensor to suppress invalid
hypotheses and effectively achieves a grid based fusion
approach, the reference classifier cannot distinguish
between pedestrian hypotheses at different distances.
Hence, the latter produces detection results incapable
of localizing the pedestrian in space. Figure 20
shows the precise pedestrian localization of the fusion
system compared to the uncertainty of the reference
detector.

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 20

Detection examples in a suburban environment (camera image with green detections left and radar data with cyan
detections right): (a) Typical result of the image based reference detector. The uncertainty in spatial localization is very
high using only image features for classification. Figure (b) shows the increased performance of the fusion detector
incorporating features provided by the radar system too. The result is a precise localization of the pedestrians at 100 m
((a) Image-based reference detector: Typical detection of a pedestrian at 100 m with very imprecise spatial localization
(70–120 m). (b) Fusion detector: Accurate detection of a pedestrian at 100 m)
826 Night Vision Pedestrian Warning in Intelligent Vehicles

Road Course Estimation Moreover, the map-based position estimation

In order to evaluate the estimated road course, or rather
the estimated vehicle position, a ground truth is required.
The best, but unfortunately the most expensive, equip-
ment is clearly a RTK GPS measurement unit. It is able to
measure the vehicle position with a very high accuracy
within a range of a few centimeters. For this examination,
such a device was not available.
Nevertheless, the estimated position can also be
analyzed considering the position uncertainty in
a driving scenario. The position uncertainty calculation
transforms each position hypothesis into the vehicle
coordinate system of the expected position value of
the distribution and determines the standard deviation
for longitudinal and lateral position as well as for the
orientation angle. Figure 21 shows the lateral and
the longitudinal deviation for three scenarios. Whereas
the lateral position deviation remains low and nearly
constant, the longitudinal position deviation fluctuates
more strongly. This behavior is associated with the
current road course. A course with frequent curves
provides more precise information for longitudinal
localization than a straight road, which yields to ambig-
uous position hypotheses because the digital map fits
anywhere along the road.
can be compared to an approach using only the vehicle
data that is the result of the egomotion estimation
without resetting the position regularly. The
better performance of the presented map matching
algorithm against the egomotion approach is shown
in Fig. 22.
Note that the collection of investigated scenarios
has been checked manually to diminish the impact of
errors or outdated information from the digital map
which cannot be corrected by the system itself. Correct
information of the digital map is a necessary prerequi-
site condition of the realized road course estimation
system.
Threat Analysis of Pedestrians The most interesting
part is the evaluation of the warning performance
answering the question: How many pedestrian detec-
tions have been classified correctly according to their
relevance for the driver? This has been investigated on
a ground truth of 1,945 labeled pedestrians in a range
from 20 to 120 m with manually assigned discrete
warning levels. Figure 23 shows the different warning
levels. Manual labeling of the warning levels in the
camera image is partially very complicated, mostly for
detections in the far field of curved roads.

14
Longitudinal
12 Laterale

10
Sigma [m]

0
0 100 200 300 400 500 600 700 800
Time [s]

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 21

Standard deviation of the estimated longitudinal and the lateral position for three driving scenarios
 Night Vision Pedestrian Warning in Intelligent Vehicles 827

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 22

Position trajectory with and without localization on the map: The course using only the vehicle data in an egomotion filter
(slash-dotted) deviates clearly from the course of the map-based positioning algorithm (dashed)

1 1. The calculated warning level ðd

Þ is in the speci-
fied range defined by the ground truth label.
2. The coverage between the ground truth label and
the detection is at least 80%.
0.5
The presented pedestrian warning system achieves
84% correct warnings. The remaining 16% mostly refer
to a position error of the presented road course estima-
0
−1 −0.75 −0.5 −0.25 0 0.25 0.5 0.75 1 tion system.

Night Vision Pedestrian Warning in Intelligent Vehicles.

Figure 23
The defined discrete warning levels
In order to reduce the resulting ground truth errors,
the warning levels overlap between adjacent levels by 6%.
This corresponds to a tolerance band of 0.4 m for a typical
road width of 6.5 m. The evaluation algorithm classifies
a detection considered to be correctly warned if at least
one detection exists satisfying the following conditions.
Conclusion
The described novel pedestrian warning system has
been implemented in a test vehicle equipped with an
NIR camera sensor, an imaging radar system, and
a digital map.
In order to achieve a robust, reliable, and real-time
capable detection system, the information provided by
the camera sensor and the radar sensor are fused on
feature level using a cascade of boosted classifiers. That
is, both sensors define a set of features. For the
828 Night Vision Pedestrian Warning in Intelligent Vehicles

Night Vision Pedestrian Warning in Intelligent Vehicles. Figure 24

Some results of the pedestrian warning system: the left camera image with highlighted pedestrians depending on their
calculated threat. The next image corresponds to the interpolated Cartesian view of the radar raw data which is the input
data to estimate the road course. The right image shows the radar raw data including the detected pedestrians

classification task, the relevant feature are selected and
parameterized using the supervised training algorithm
AdaBoost in multiple stages of increased complexity
(number of features). The feature set is composed of
Haar-wavelet filters in the image plane and average
object intensity as well as complex signal filters in the
radar plane. The newly derived radar filters are charac-
terized by their efficiency to separate between objects
with a local velocity similarity and background. Exper-
iments have shown that the resulting fusion detector
dramatically outperforms a single sensor reference sys-
tem. Unlike the image-based reference system, the
presented multi-sensor fusion approach can reject
hypotheses in earlier stages of the cascade and gives
a precise spatial localization of the resulting object
hypotheses. In this way, the fusion system also achieves
a grid-based fusion with the digital map and the radar
image.
In order to signal to the driver only the relevant
objects, a new road course estimation concept has been
implemented. The new concept maps the information
provided by a digital map to the observations of the
radar sensor, similar to the Monte Carlo localization
approach in the field of autonomous robotics. Thus
this component is more likely a global positioning
approach which is apart from the digital map module,
independent of inaccurate and error prone GPS mea-
surements. Finally, the implemented threat analysis
algorithm combines the pedestrian detection system
and the road course estimation system. The evaluation

results show a good warning performance. About 84%
of detected pedestrians triggered a correct warning.
Figure 24 gives an impression of the pedestrian warning
system in a rural environment.
Future Directions
Future work will incorporate object classes other than
pedestrians and more sensors to improve the environ-
mental perception of vehicles, and will enhance the
proposed approach.
The feature-level fusion for sensors of different
physical principle is a possible solution to meet the
requirements of safety application where high-level
fusion is not sufficient. The described method is appli-
cable not only for automotive applications, but also for
example in the areas of monitoring, inspection, sur-
veillance, and medicine.
Digital maps are currently the exclusive technology
to provide information of the course of road for far
distances over 100 m independent of weather condi-
tions, which is necessary to realize a warning system.
Today, neither a camera nor a radar sensor can provide
this information. This shows that the future develop-
ment of driver assistance systems should focus more on
digital map technologies.
Fusion of local maps with digital maps will offer
a lot of applications where a high-precision localization
is necessary, and so far no low-cost solution is available.
Acknowledgments
This research was supported by PROPEDES
(Predictive Pedestrian Protection at Night), BMBF
FKZ 13N9750-13N9754, Germany.
Bibliography
Primary Literature
1. Green M (2000) ‘How long does it take to stop?’ Methodolog-
ical analysis of driver perception times. Transp Hum Factors
2(3):195–216
2. Zhang Y, Antonsson E, Grote K (2006) A new threat assessment
measure for collision avoidance systems. In: International con-
ference on intelligent transportation systems (ITSC), 2006 IEEE,
Toronto, pp 968–975.
3. Beutnagel-Buchner U, Arnon M, Sparbert J, Moisel J, Rode M,
Ritter W, Oberländer M, Löhlein O, Serfling M, Strobel M, Graf HG,
Night Vision Pedestrian Warning in Intelligent Vehicles 829
Valipour H, Müller F, Schweiger R (2007) N I R W A R N
Gesamtschlussbericht, BMBF, Forschungsbericht 01M3157A-D
4. Franke U, Loose H, Knöppel C (2007) Lane recognition on
country roads. In: Intelligent vehicles symposium (IV), 2007
IEEE, Istanbul, pp 99–104
5. Jung CR, Kelber CR (2005) Lane following and lane departure
using a linear-parabolic model. Image Vis Comput 23(13):
1192–1202
6. Apostoloff N, Zelinsky A (2003) Robust vision based lane
tracking using multiple cues and particle filtering. In: Intelli-
gent vehicles symposium (IV), 2003 IEEE, Columbus,
pp 558–563
7. Southall J, Taylor C (2001) Stochastic road shape estimation.
In: International conference on computer vision (ICCV), 2001
IEEE, vol 1. Vancouver, pp 205–212
8. Franke U (1992) Real-time 3D road modeling for autonomous
vehicle guidance. In: Johansen P, Olsen S (eds) Theory and
applications of image analysis: selected papers from the 7th
Scandinavian conference on image analysis, Aalborg, Den-
mark, vol 2, Machine perception artificial intelligence. World
Scientific Publishing Company, Singapore, pp 277–284
9. Dickmanns ED, Zapp A (1986) A curvature-based scheme
for improved road vehicle guidance by computer vision. In:
Conference on mobile robots, 1986 SPIE, vol 727. Cambridge,
pp 161–168
10. Bertozzi M, Bombini L, Cerri P, Medici P, Antonello P, Miglietta
M (2008) Obstacle detection and classification fusing radar
and vision. In: Intelligent vehicles symposium (IV), 2008 IEEE,
Eindhoven, Netherlands, pp 608–613
11. Alessandretti G, Broggi A, Cerri P (2007) Vehicle and guard rail
detection using radar and vision data fusion. IEEE Trans Intell
Transp Syst 8(1):95–105
12. Haselhoff A, Kummert A, Schneider G (2007) Radar-vision
fusion with an application to car-following using an improved
AdaBoost detection algorithm. In: Intelligent transportation
systems conference (ITSC), 2007 IEEE, Seattle, pp 854–858
13. Dehesa M, Allezard N, Furca L, Kaiser A, Kirchner A, Meinecke
MM, Vinciguerra R (2005) SAVE-U: high and low level data
fusion algorithms, Public Deliverable 19, 2005. Commission
of the European Communities, Brussels
14. Enzweiler M, Gavrila DM (2009) Monocular pedestrian detec-
tion: survey and experiments. IEEE Trans Pattern Anal Mach
Intell 31(12):2179–2195
15. Milch S, Behrens M (2001) Pedestrian detection with radar and
computer vision. In: Symposium on progress in automobile
lighting, Darmstadt, Germany, pp 657–664
16. Schweiger R, Hamer H, Löhlein O (2007) Determining
posterior probabilities on the basis of cascaded classifiers as
used in pedestrian detection systems. In: Proceedings of
IEEE intelligent vehicles symposium, Istanbul, Turkey,
pp 1284–1289
17. Kallenbach I, Schweiger R, Palm G, Löhlein O (2006) Multi-class
object detection in vision systems using a hierarchy of cas-
caded classifiers. In: Proceedings of IEEE intelligent vehicles
symposium, Tokyo, pp 383–387
830 Night Vision Pedestrian Warning in Intelligent Vehicles
18. Miyamoto R, Sugano H, Nakamura Y (2007) Pedestrian recog-
nition suitable for night vision systems. Int J Comput Sci Netw
Secur 7:1–8
19. Maehlisch M, Oberländer M, Löhlein O, Gavrila D, Ritter
W (2005) A multiple detector approach to low-resolution FIR
pedestrian recognition. In: Proceedings of IEEE intelligent
vehicles symposium, Las Vegas, pp 325–330
20. Milch S, Behrens M (2001) Pedestrian detection with radar and
computer vision. In: Proceedings of the symposium on
progress in automobile lighting, Darmstadt, Germany,
pp 657–664
21. Lindl R, Walchshusl L (2006) Three-level early fusion for road
use detection. PReVENT Fusion Forum J 1(1):19–24
22. ProFusion (2004) Subproject of the PReVENT-project (Preven-
tive and Active Safety Applications), founded by European
Commission
23. Dickmann J et al (2009) Sensorfusion as key technology for
future driver assistance systems. In: Workshop Sensor- und
Datenfusion – Architekturen und Algorithmen, Berlin, 20 Nov
2009
24. Zhang Z (1999) Flexible camera calibration by viewing a plane
from unknown orientations. In: Proceedings of the
international conference on computer vision, Kerkyra, Corfu,
pp 666–673
25. Bouguet J (1999) Visual methods for three-dimensional modeling.
Ph.D. dissertation, California Institute of Technology, Pasadena, CA
26. Tsai R (1987) A versatile camera calibration technique for high-
accuracy 3D machine vision metrology using off-the-shelf TV
cameras and lenses. IEEE J Robot Automation 3(4):323–344
27. Skolnik MI (1990) Radar handbook, 2nd edn. McGraw-Hill,
New York
28. Freund Y, Schapire RE (1997) A decision-theoretic generaliza-
tion of on-line learning and an application to boosting.
J Comput Syst Sci 55:119–139
29. Viola P, Jones M (2002) Robust real-time object detection. Int
J Comput Vision 57(2):137–154
30. Viola P, Jones M (2001) Rapid object detection using a boosted
cascade of simple features. In: Proceedings of IEEE computer
society conference on computer vision and pattern recogni-
tion, Kauai, HI, vol 1, pp 511–518
31. Freund Y, Schapire RE (1996) Experiments with a new boosting
algorithm. In: Proceedings of the international conference on
machine learning, Bari, Italy, pp 148–156
32. Freund Y, Schapire RE (1999) A short introduction to boosting.
J Jpn Soc Artif Intell 14(5):771–780
33. Schapire RE (2002) The boosting approach to machine learn-
ing: an overview. In: MSRI workshop on nonlinear estimation
and classification, Berkeley, CA
34. Papageorgiou C, Oren M, Poggio T (1998) A general frame-
work for object detection. In: Proceedings of international
conference on computer vision, Bombay, India, pp 555–562
35. Oren M, Papageorgiou C, Sinha P, Osuna Z, Poggio T (1997)
Pedestrian detection using wavelet templates. In: Proceedings
of IEEE computer society conference on computer vision and
pattern recognition, San Juan, Puerto Rico, pp 193–199
36. Dellaert F, Fox D, Burgard W, Thrun S (1999) Monte Carlo
localization for mobile robots. In: International conference
on robotics and automation (ICRA) 1999, Detroit, vol 2. IEEE,
Washington, DC, pp 1322–1328
37. Fox D, Burgard Z, Dellaert F, Thrun S (1999) Monte Carlo localiza-
tion: efficient position estimation for mobile robots. In: National
conference on artificial intelligence (AAAI), Orlando, pp 343–349
38. Isard I, Blake A (1998) Condensation – conditional density prop-
agation for visual tracking. Int J Comput Vis (IJCV) 29(1):5–28
39. Snyder JP (1987) Map projections – a working manual. US
Government Printing Office, Washington, DC
40. Ress C, Balzer D, Bracht A, Durekovic S, Löwenau J (2008),
ADASIS protocol for advanced in-vehicle applications. In:
World congress on intelligent transportation systems (ITS)
2008, New York. IEEE, Washington, DC
41. Schubert R, Richter E, Wanielik G (2008) Comparison and
evaluation of advanced motion models for vehicle tracking.
In: International conference on information fusion 2008,
Cologne, Germany. IEEE, Washington, DC, pp 1–6
42. Bierman GJ (1975) Measurement updating using the U-D
factorization. In: Conference on decision and control 1975,
Houston. IEEE, Washington, DC, pp 337–346
Books and Reviews
Bar-Shalom Y, Li XR, Kirubarajan T (2001) Estimation with applica-
tions to tracking and navigation, 1st edn. Wiley, New York
Blervaque V, Mezger K, Beuk L, Löwenau J (2006) ADAS horizon –
How digital maps can contribute to road safety. In: Valldorf J,
Gessner W (eds) Advanced microsystems for automotive
applications. Springer, Berlin, pp 427–436
Grabham D (2009) Street to screen: how sat-nav maps are made.
SatNav News on www.techradar.com, August 2009
Hager JW, Behensky JF, Drew BW (1989) The universal grids: uni-
versal transverse mercator (UTM) and universal polar stereo-
graphic (UPS). Technical manual DMATM 8358.2, Defense
Mapping Agency (DMA), Fairfax, VA
Misra P, Enge P (2006) Global positioning system: signals, measure-
ments and performance, 2nd edn. Ganga-Jamuna Press, Lincoln
Moran T (2008) Tapping mapping for an extra eye on the road.
Automotive News: Information Technology, Detroit. July 2008

Personal Rapid Transit and Its
Development
SHANNON S. MCDONALD
Southern Illinois University, Carbondale, IL, USA
Article Outline
Glossary
Definition of the Subject
Introduction
The Development of PRT: Personal Rapid Transit
System
Key Sustainable/Energy and Planning/Architectural
Future Directions
Bibliography
Glossary
AGT Automated Guideway Transit – Updating the
original AGT definition (US Congress, Office of
Technology Assessment, 1975), Automated Guide-
way Transit (AGT) is defined as a class of transpor-
tation systems in which fully automated vehicles
operate along dedicated guideways.
APM Automated People Movers – According to the
General Accounting Office (1980), Automated Peo-
ple Movers are driverless vehicles operating on
a fixed guideway. Vehicle capacities range up to
100 passengers and may be operated as single
units or as trains up to 30 miles/h.
ATRA Advanced Transit Association
Dual-mode A transportation system where in one
mode the vehicle operates under its own power
and control, usually on existing streets and in the
second mode it operates under automated control
and/or external power.
FRT Freight Rapid Transit – Characteristics of PRT
but the vehicles are designed to handle freight only.
GRT Group Rapid Transit – which is similar to per-
sonal rapid transit but with higher occupancy vehi-
cles and grouping of passengers with potentially
different origin–destination pairs. As noted in an
early study (US Congress, 1975), the starting capac-
ity for GRT is six passengers per car while the upper
limit is around 16 or 18; there are no clear distinc-
tions between GRP and APM in terms of vehicle
capacities.
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Personal Rapid Transit and Its Development 831
Network system Interconnecting links that form the
layout of transit routes and stops that constitute
the total system – as opposed to a loop or corridor
system
Off-line stations A station design where the vehicle is
removed from the main line for loading or
unloading allowing other vehicles to continuously
flow on the main line.
Point-to-point The vehicle is on demand and takes
the rider from his or her starting point directly
to rider’s destination point with no stops in
between – a nonstop journey bypassing intermedi-
ate stations that relies on the use of offline stations
in a network.
PRT Personal Rapid Transit – The definition of PRT
can be a subcategory of AGT systems that offer on-
demand, non-stop transportation, offline stations
using small, automated vehicles on a network of
dedicated guideways.
Definition of the Subject
Personal Rapid Transit (PRT) is in the class under
Automated Guideway Transit (AGT) that includes
Automated People Movers, sometimes identified as
a subset of APM [1, 2]. Personal Rapid Transit is a
driverless automated transit technology that has a
unique movement pattern for the transit rider.
A rider who uses PRTwill travel point to point – similar
to a cab – in comparison to a “typical” transit system
where the rider stays on the system through many
“unnecessary” stops until reaching their final destina-
tion. PRT is a fixed system, it is transit, with
predetermined stops but how you travel from
Stop A to Stop B is one of the key paradigm changes
for this emerging transit technology as it is not
designed in its fullest implementation as a loop or
corridor system. The on-demand and point-to-point
approach will provide much faster travel time and
more destination choices; and along with its smaller
size (two to four people vehicles), operating system,
and offline stations, it can attract more riders to transit
due to a higher level of service.
PRT was originally designed beginning in the 1950s
for the lower-density transit environments that more
commonly exist in the United States. The Morgantown
# 2012, DOI 10.1007/978-1-4419-0851-3
832 Personal Rapid Transit and Its Development

system at the West Virginia University became the first for all people while improving the living environment.
driverless system placed in operation, 1972/6, and has It has been named one of the twenty-first century top
been in continuous operation, since its opening. It is 20 proven ways to save the earth by the London Times
technically considered a GRT, group rapid transit sys- and the Sustainable Transportation Solution by the
tem, due to the fact that the transit vehicles holds Directorate-General of Energy and Transport for the
20 people rather than the two to four people in European Commission (Fig. 1).
a “true” PRT. The offline stations allow the point-to-
point movement pattern to function along with Introduction
computer control systems; however due to the larger
The history of PRT is a confluence of new communi-
vehicle, many other design possibilities of “pure” PRT
cation technology, planning visions allowing for auto-
did not occur [3].
mated control technology and transit to merge
Currently, a “pure” PRT system, with small vehicles
spurring innovative ideas and solutions to urban trans-
is opening at Heathrow airport connecting a long-term
portation providing totally new paradigms for move-
parking lot with a parking garage linking directly to
ment and urban design. The seeds of the system started
Terminal 5. To park in this lot will be more expensive
in the late 1800s and it sustained itself as an idea well
due to the direct and quick access to the airport. The
into the beginning of the computer age. The early
computerized control systems have dramatically
innovators and adaptors understood the power of the
improved since the Morgantown system allowing the
computer to provide new ways to manage and control
initial vision of PRT to finally be constructed and
operated. The system is designed to be expanded to
50 stations, 350 vehicles, and 30 km of track
connecting many different airports and local associated
functions [4].
PRT not only changes the way transit functions for
the user, but as well can redefine how we design our
built environment, allowing for more frequent smaller
transit stops encouraging walkable design and more
compact development. Mazdar City, a new planned
city in Abu Dhabi will rely on PRT as a major part of
its transportation system. Requiring a much smaller
right-of-way and lighter tracks, the infrastructure for
the system is much less expensive and less visually
obtrusive. Its smaller size not only reduces impact on
the built environment but can also allow for alternative
energy sources to power the system; such as solar and
electric [5].
This system is also the perfect connector system
between other forms of movement: transit, car, air-
plane, bus, bicycle, and walking, and for major activity
centers such as downtowns, hospitals, universities, and
airports, where the more traditional automated people
mover system has been implemented over the years.
The reason why the PRT vision has never died and
many continue to advocate for its use today is due to its
potential to change our living patterns for the better-
ment of our society. This technology has the ability Personal Rapid Transit and Its Development. Figure 1
to allow for more options and flexibility of movement ULTra: PRT sustainable transport (Martin Lowsen of ULTra)

these innovative transit vehicles and in turn our move-
ment patterns within the lower density environments
of the United States.
Several individuals and companies were at the fore-
front of creating PRT systems. Edward O. Haltom
(a contractor) Donn Fichter (a city transportation plan-
ner who believed that automated transit was the right
solution for medium- to low-density populations),
Monocab, TTI, Inc., Alden StaRRcar, Uniflo, Jet Rail,
M.I.T, Bartells, and Dr. Jarold Kieffer each created and
worked with early PRT ideas in the United States. The
Urban Mass Transportation Act of 1964 and Housing
and Urban Development Studies furthered the study
and advanced the reality of the Morgantown system
based upon the Alden StaRRcar. Numerous other inno-
vators were involved in the early explorations, including
General Motors, Raytheon, General Research Corpora-
tion, IBM, MITRE Corporation, Parsons Company, LTV
Aerospace Corporation, Honeywell, Renault Engineer-
ing, Bendix, Ford Motor Company, and Otis Elevator
Company along with Johns Hopkins, Ohio State, Uni-
versity of Minnesota, San Diego State, Battelle Colum-
bus Laboratories, Aerospace Corporation, Jet
Propulsion Lab, and Booz-Allen Applied Research [6, 7].
A series of 17 studies sponsored by Housing and
Urban Development spurred the research and develop-
ment of PRT. The HUD studies were summarized in
a report, Tomorrow’s Transportation by Leon Monroe
Cole [8]. The two most influential studies were one by the
Stanford Research Institute and the other by the General
Research Corporation of Santa Barbara. The Stanford
research paper was on various new concepts from moving
sidewalks to PRT to dual-mode that estimated their eco-
nomic benefits. The GRC study modeled alternative sys-
tems in several actual cities to compare alternative
transport systems to conventional systems [9, 10].
Activity in other countries was also occurring at this
time. Ed Anderson concluded that the stimulus for
these explorations came from the US inventors and/or
the HUD studies. The following foreign companies/
countries/municipalities also advanced the work
in this area: Cabtrack, CVS, Cabinentaxi, Aramis,
Gothenburg, Sweden and Canada. From 1968 to 1971
the not-for-profit Aerospace Corporation proved the
feasibility of operating large PRT networks appropriate
for urban application. The newly formed Department
of Transportation led by Secretary John A. Volpe
Personal Rapid Transit and Its Development 833
wanted a system in operation before 1972, so a large
team was put into place and the system barely opened by
1972 – a very ambitious goal for this fledgling technol-
ogy; however, the Morgantown system is still function-
ing today. The final cost of the system was four times as
planned due to many issues related to design as a result
of creating a larger GRT system along with many other
issues related to this. This and other early system stud-
ies are thoroughly covered in a book written by Cath-
erine G. Burke titled: Innovation and Public Policy: The
Case of Personal Rapid Transit [11].
However, although Morgantown, WVA, was the
only system built for the public and due to its cost
overruns slowed other PRT systems from being devel-
oped in the public domain; the work with and about
PRT continued as the system provides so many bene-
fits. Several test facilities built around the world prov-
ing various aspects of the concept continued and now
we are seeing the benefits of this advancing technology.
This concept for transit innovation has never
died, and numerous attempts have been made to
implement the original PRT vision of small
automated point-to-point vehicles, offline stations
on a dedicated network. Gayle Franzen, Chairman of
the Northeastern Illinois Regional Transportation
Authority initiated a new PRT program in 1990 that
did not proceed.
Morgantown, WVA, is the most important histori-
cal example with the Heathrow airport system by
ULTra coming into operation for the public late in
2010 leading the way into the future. Currently, several
other systems and several test tracks, one in Uppsala,
Sweden, a joint venture between the South Koreans and
the Swedes for the Vectus system, are on line to move
forward. Larry Fabian, ATRA, identifies more than
100 applications along the various spectrums of AGT
technologies around the world including shuttles or
circulators, major activity center circulation, and pub-
lic transit, and now we can add a true PRT application
to the mix [12, 13].
This article will trace the beginning PRT explora-
tions all the way to the first implementation for public
use of the complete paradigm change for transit – PRT.
This has involved a complex interrelationship between
innovative thinking in technology, communication
systems, hardware, architecture, transportation, and
urban design/planning spanning the globe.
834 Personal Rapid Transit and Its Development
The Development of PRT: Personal Rapid Transit
System
History/Overview/Development of Systems
in the United States
The evolution of the PRT transit “idea” to its first real-
world application at Heathrow airport, 2010, is to
understand how a paradigm change in transit can
occur, when the idea has merit, no matter how method-
ical and complex the route. Sending a man to the moon
was easier! Only through the tenacity, belief, and hard
work of many people from multiple disciplines over
decades has this “vision” now occurred, so that the
benefits can be “proven” – that an automated network
of small personal point-to-point vehicles with offline
stations can improve the built environment and as well
improve transit for all people. Several papers and books
have been written about this evolution; one specifically
titled: Some Early History of PRT by Edward J. Ander-
son, University of Minnesota, much of the following
history is a synopsis of this report [7].
A group of “transit” technologies currently known
as Automated People Movers now understood as
a subset of Automated Guided Transit that most of us
have experienced mainly in airports around the world
has been the home for the development of a true PRT
system. These systems have a worldwide presence and
according to Larry Fabain they are currently distributed
as in Fig. 2.
In 1953, PRT was born simultaneously out of prac-
ticality and urban vision by two people who may not
World wide applications of APM
Transit Airport
33% 32%
Institution
35%
Personal Rapid Transit and Its Development. Figure 2
Distributions of APM applications (Based on data by
Fabian 2010 [1])
have initially known each other; Donn Fichter and
Edward O. Haltom. Donn Fichter, a transportation
graduate student in Chicago, IL, and Edward O.
Haltom, a Dallas, TX, contractor who was working on
a monorail system. Mr. Fichter sketched a system called
Veyar and gradually developed a systems concept inte-
grated into a city with the essential PRT ideas, although
he was not a technician. He published a book in 1964
titled Individualized Automated Transit and the City
[6]. Mr. Haltom saw a way to “improve” the monorail
system that he was working on (that has been in exis-
tence since the first monorail constructed in St. Paul,
Minnesota in the 1880s) by thinking of smaller guide-
ways and smaller vehicles and invented the monocab.
This system eventually became a full-scale test track in
California with a totally new switching system owned
by Rohr Corporation where linear induction propul-
sion was added to the system. The patents were pur-
chased by Boeing and the transit system continued to
be developed by UMTA (Urban Mass Transportation
Administration) Advanced Group Rapid Transit until
the mid-1980s.
General Motors Research Laboratories had been
working for the military in the late 1950s and created
a system that could become public transit, called
Hovair. They formed a corporation called TTI, Inc,
Transportation Technology Incorporated. A full-scale
testing occurred in Detroit in 1969 and the company
was eventually owned by Otis Elevator. A second
test track was built in Denver. The system was a com-
bination of air-suspension and the linear induction
motor and was implemented at Duke University Med-
ical Center. Otis also constructed several cable-drawn
versions of this system.
William Alden, a graduate of the Harvard Business
School invented, built, and drove a small electric vehi-
cle that could be driven on public roads and then link
to a guideway. He called it the staRRcar and it is
considered the first dual-mode-system and had an
on-board switching mechanism. A test track was built
in Bedford, Massachusetts in 1968, the Morgantown
system was based on this prototype; however a team of
professionals was formed to implement the concept at
Morgantown, WVA (Fig. 3).
Honeywell also in its military division through the
efforts of Lloyd Berggran wanted to create urban
transportation that could be competitive with the

Personal Rapid Transit and Its Development. Figure 3
William Alden StaRRcar – dual-mode vehicle
(William Alden)
automobile. He also concluded the same set of PRT
characteristics. However, due to his background he
created an independent “pod” with all the active
power and controlling systems as part of the track.
His system was called Uniflo and was designed to be
in an enclosed tube. A full-scale test track was
constructed by Rosemont Engineering.
Jet Rail was designed by George Adams. He
designed, built, and operated a system for Braniff
Airlines at Love Field in Dallas that was less expensive
but similar to Monocab and proved that a lightweight
guideway could be successfully constructed. Cornell
Aeronautic Laboratories in the early 1960s designed
Urbmobile, it was never built but the automatic control
technologists were able to show that short headways
could provide adequate capacity – safely. MIT in the
mid-1960s produced a report call Project Metran that
included the PRT ideas and influenced its continued
development.
Personal Rapid Transit and Its Development 835
A planning director, Robert J. Bartells, and a public
affairs head of a university, Dr. Jarold Kieffer, both
seeking to solve real needs of transit movement for
large numbers of people, independently came to the
same conclusion of the PRT paradigm. Robert J.
Bartells was the director of Planning for the City of
Hartford, CT imagined the PRT concepts and ideas and
explained and supported them as an important plan-
ning director. Dr. Jarold Kieffer, as Head of the School
of Public Affairs at the University of Oregon while on
a vacation at a sky resort also understood the basic
concepts of PRT and has been a continual advocate
for the development of the technology.
US Government Participation
It was only through the actions of the Federal Govern-
ment that the first public system was put into place both
technically advancing and greatly hindering the PRT
vision. Not fully understanding how totally new pub-
lic-use transit paradigms need to develop if they are to
have long-term success, many boasted and believed that
a system could be put into place within 2 years – as we
had won the race to the moon! Public-transit is a much
more complex and integrated effort of many govern-
mental agencies, organizations, and professionals.
Henry S. Reuss of Milwaukee, Wisconsin in the
1960s urged political support for the development of
new transit concepts. He participated in the develop-
ment of the Urban Mass Transportation Act of 1964
where a Section 6 was added focusing on research,
development, and demonstration of new systems for
people and goods that stated:
" The secretary shall undertake a study and prepare
a program of research, development and demonstra-
tion of new systems of urban transportation that will
carry people and goods within metropolitan areas
speedily, safely without polluting the air, and in
a manner that will contribute to sound city planning.
The Program shall (1) concern itself with all aspects of
new systems of urban transportation for metropolitan
areas of various sizes, including technological, financial,
economic, governmental and social aspects; (2) take
into account the most advanced available technologies
and materials; and (3) provide national leadership to
efforts of states, localities, private industry, universities
and foundations [12].
836 Personal Rapid Transit and Its Development
At this time, the US Department of Transportation did
not exist; however, the Urban Mass Transportation Act
established the Urban Mass Transportation Adminis-
tration as a unit of the Department of Housing and
Urban Development. Seventeen studies were autho-
rized and these became known as the HUD studies
and were summarized in a report titled: Tomorrow’s
Transportation. Two studies were very influential:
A study by Stanford Research Institute and the General
Research Corporation (GRC) of Santa Barbara.
Stanford focused on new concepts from moving side-
walks to PRT to dual mode and their economic benefits
while the main focus of GRC was to computer-model
“advanced” systems in real cities and compares them to
conventional transit.
There was a team of 17 specialists from various
fields that contributed to this study identifying Boston,
Houston, Hartford, and Tucson. This study strongly
favored new systems due to population growth and
increased use of automobiles traditional transit could
not continue to meet the demand. This work published
in an article, in 1969, by Scientific American titled
Systems Analysis of Urban Transportation, became
a classic for the technology and brought a national
commitment to develop new technologies through
the voice of Ben Alexander, the chairman of GRC
[8–10].
Once the HUD reports had been released The
Aerospace Corporation, a not-for-profit corporation
created by the United States Air Force, wanted to
use aerospace technology to solve urban problems.
Dr. Jack H. Irving concluded that working on high
capacity PRT was a good focus for their skilled engi-
neers. They invented new devices such as: powering the
vehicles with a pair of linear pulsed direct current
motors in interaction with permanent magnets in the
track. Except at the switching points they used electro-
magnets resulting in a no-moving parts switch.
The motor could be controlled by solid-state circuitry
and was very efficient. They also believed that the
guideway should have minimal visual impact so the
vehicle was supported by two wheels in tandem.
They tested this on a scale model. They were able to
develop the system to a very advanced state and used
computer simulations proving the feasibility of com-
plex short headway networks. They performed analysis
for Los Angeles and Tucson and released their work in
a book titled: Fundamentals of Personal Rapid Transit,
1978 [3] (Fig. 4).
The Morgantown system was awarded and built
during a very complex time for the newly
formed Urban Mass Transportation Administration.
Most importantly a presidential change took place
bringing all new people into a process that they had
no history or detailed understanding of. They wanted
immediate results along with preventing the collapse of
existing transit systems – two competing goals; and of
course they were understaffed.
The Development of the Morgantown GRT
The West Virginia University is located in the hilly,
snowy mountain valley town of Morgantown in the
Monongahela Valley of West Virginia, and its campus
is spread between three different sections of the city.
Buses transported students through the center of the
city along with all other traffic, creating at that time
extreme traffic issues. Professor Samy Elias, Head of the
Industrial Engineering Department, became aware of
PRT and the PRT test tracks and studies. He believed
that this could be a solution for Morgantown. Due to
the support of the city, the University and the West
Virginia congressional delegation money was obtained
for a grant to study three systems. The Alden staRRcar
was chosen as the appropriate solution for Morgan-
town. Due to political connections and pressures, this
project was taken very seriously. So, with the 1972
presidential election in mind, the new Head of the
Department of Transportation, John A. Volpe, stated
that by October 1972 the president should be able to
ride the system and therefore this became a federal
demonstration project and a team needed to be
formed. A contract was signed in December 1970 with
the Jet Propulsion Laboratory in Pasadena, California,
Boeing, Seattle, WA was the vehicle manufacture,
Bendix, Ann Arbor, MI the control system supplier
and F.R. Harris Engineering Company of Stanford,
CT for the design and construction of the
guideway. None of these groups had been working on
PRT and there was no time for analysis or a learning
curve. This of course led to a complicated and
difficult outcome costing four times the original
estimate as all of their expertise did not fit with the
paradigm changing technology. Recently WVU was
 Personal Rapid Transit and Its Development 837

GUIDEWAY
MONORAIL ELECTRO-MAGNETIC SWITCHES
45ft x 14ft Oval Higher Flux Density (25k–40k lines/in.2)
134ft of Guideway, Including Siding for Switching Stability
Aluminum/Steel Construction Vehicle-Actuated
PERMANENT MAGNETS COMMAND & POWER CIRCUITS
Strontium Ferrite Controlled by Console
Flux Density≈20k lines/in.2 Dual Circuits
Main Power (20V d.c.)
Control Commands
Reference Pulse (1920 Hz)
Instrumentation

VEHICLE
Weight ≈10 lbs
BODY
Fiber Glass Laminated Construction
Representative Shape
CAR-BORNE CONTROL SYSTEM
Regulates Coil Current & Sweep Rate
per Control Console Event Command
Closed Loop Velocity Control
Stored Maneuvers
Accelerate, Decelerate
Constant Speed
Emergency Stop
Solid State Components
PROPULSION & BRAKING
Pulsed d.c. Linear Motor
Scaled Motor Current & Thrust
Dual Primary Construction
SUSPENSION
Adaprene Main & Guide Wheels
Rolling Support
Maintain Motor Air Gap
Aluminum Construction

Personal Rapid Transit and Its Development. Figure 4

Aerospace Corporation (Reproduced from [11]. With permission)

awarded a Federal Transit Authority grant that will
be a part of the funds to upgrade the existing system
(Fig. 5).
Expositions and International Conferences
In 1972, UMTA sponsored an international exposition
of four different PRT systems called Transpo72.
Millions of federal dollars were set aside for the four
systems to be constructed and money to be matched by
their respective companies after being chosen. The
companies were TTI, Monocab, Dashaveyor, and
Ford. Each was to exhibit a minimum piece of guide-
way and one station in hopes to produce requests for
capital grants. However, again due to a short time
frame, work to improve the technologies took prece-
dence and those attending the Expo did not understand
how these systems could provide a better service and
successfully integrate into and improve the built envi-
ronment. While no city was ready to put its own funds
into the development of PRT, many cities did want
100% federally funded programs to integrate the new
technology. However the UMTA saw their role as one
to encourage private investment [14].
838 Personal Rapid Transit and Its Development
Personal Rapid Transit and Its Development. Figure 5
Heathrow PRT system ( Stan Young, Kansas DOT, University
of Maryland, Center for Advanced Transportation
Technology)
In the years 1972, 1973, and 1975, three major
international conferences were held by the American
Society of Civil Engineers (ASCE) People Mover Com-
mittee and a volume of published proceedings were
produced. However, by 1973, attendance had peaked
and the organizing committee worked to develop a
permanent organization. In 1976 the ATRA, Advanced
Transit Association was formed. ATRA held its own
conference in 1978 and more research was added to
the body of knowledge. In 1988, ATRA published
a broadly based technical committee report to further
the importance of the PRT concept. ATRA and the
ASCE – People Movers Committees – have continued
their work producing published proceedings and
a journal [15, 16].
Planners’ and Architects’ Visions
While the visual impact of an overhead track had been
identified from the beginning as an important factor in
the acceptance of a PRT system, more emphasis was
placed on the engineering aspects than exploring the
many approaches that could support visual acceptance.
Now, due to advanced technology, the first true PRT
system is partially on grade and partially on elevated
tracks, only where necessary within an airport, while
the Mazdar City PRT system in Abu Dhabi, UAE, will
be completely under the city with the vehicles on
grade – while many lovely integrative solutions can
still be explored! Some architects and architectural
studies appeared during the 1970s and 1980s such as
in the work of Victor Gruen, Paul Rudolph, and Ulrich
Franzen.
Paul Rudolph and Ulrich Franzen were sponsored
by the Ford Foundation and the American Federation
of Arts producing a book in 1974 titled Evolving City –
Urban Design Proposals [17] (Fig. 7). This publication
contained amazing drawings and models for New York
City including POD vehicle types and drawings that
incorporated PRT into an urban setting. Victor Gruen,
known as the shopping center architect, in his Fort
Worth Central Business District project in 1950,
envisioned “small electric vehicles” under the city to
allow full access for all people to the city center. He also
created the first pedestrian-only street in Kalamazoo,
MI, that still exists today. The two concepts of PRT and
walking environments have been connected by archi-
tects and planners since the beginning of PRT research,
due to the ability of PRT to have smaller stations closer
together providing a faster, more accessible, and greater
choice in destination [3] (Fig. 6). PRT was also under-
stood as a transit system of choice when connecting
other forms of transportation such as automobiles and
existing longer-haul transit systems (Fig. 7).
In Great Britain, a steering group for the 1963 UK
Department of Transport produced with Colin
Buchanan as the key author Traffic in Towns [18] that
also became known as the Buchanan Report. He was an
influential planner and this report discussed how
a balance must be found as increasing automobile
ownership was going to challenge the urban environ-
ment. Four key points were identified, two being:
Towns should be created as environmental areas with
quality of living coming first and creating a transit
system that could offer an acceptable alternative. This
balanced, classic, and seminal report is still referred to
today in transportation planning and it is this firm,
headed by his son Malcolm Buchanan, that has led the
PRT initiative at Heathrow Airport.
History/Overview of Systems in Other Countries
In the early years of PRT, the HUD reports generated
the interest in understanding and developing PRT sys-
tems around the world. Great Britain, Japan, Germany,

1/2 mile
A B
a b
Personal Rapid Transit and Its Development. Figure 6
Aerospace Corporation Station placement and walking distance studies (Reproduced from [11]. With permission of
Aerospace Corporation)
Personal Rapid Transit and Its Development. Figure 7
Paul Rudolph and Ulrich Franzen, Evolving City [17]
France, Sweden, and Canada all became involved to
some degree in studying and designing PRT systems.
L. R. Blake wrote an article of his own synthesis of
what he saw occurring with PRT development in
American and designed Autotaxi with the British-
built environment in mind [19]. The company was
started as a private venture, sold to Brush Electric and
Personal Rapid Transit and Its Development 839
S1 A C D
E H
B F I K
S2
G J L S3
then was funded by the British National Research and
Development Board now being called Cabtrack, a
true PRT system. The system was being studied in
a systematic and complete way through a step-by-step
process including a contract to a large British architec-
tural firm to integrate the system into a section of
London as reported in the Architects Journal of May
1971. However, after a new election in Great Britain,
the Cabtrack program was stopped. The results and
reports of this process are of great value and may have
contributed to the success of the Heathrow project.
The Japanese had a full-scale test facility operating
in 1972 outside of Tokyo with 4.8 km of guideway and
60 vehicles. Operating with a 1-s headway and a four-
passenger vehicle, a 1,000-vehicle network was simu-
lated. Although many planning and costing studies
were evaluated including one for the City of Baltimore,
MD, in the United States, several design issues had not
been fully considered such as the guideway, station
design, and cold weather conditions.
In Germany, two companies inspired by the HUD
reports were independently working on PRT concepts.
They pooled their resources and thoroughly analyzed
many alternatives creating a three-passenger cab
supported both above and below a track. They used
linear induction motors, one on each side of the vehicle
and asynchronous control that was more flexible in the
real world. A full-scale test facility was in operation by
1973 and was advertised in 1974 as a successful project.
Called Cabintaxi, it was studied to be implemented in
Freiberg and Hagen, Germany, and was marketed in
840 Personal Rapid Transit and Its Development
the United States. The central business district of
Indianapolis tested the system. A program to build
a demonstration of a 12-passenger version in Hamburg
was underway, as well it was a leading competitor for
the Downtown People Movers Program in Detroit, MI,
sponsored by UMTA, when the economic crisis of 1980
occurred and the German government withdrew its
support. Information can be found about this system
as some believe its potential still exists today [20].
The French PRT system called Aramis began with
Gerard Bardet in 1967 whose patents were bought by
Engins Matra, and in 1970 Matra received a contract.
Orly International Airport became the place for a full-
scale testing, in 1974, the first phase for proof testing
was complete and a contract was awarded for a public
demonstration in a suburb of Paris. However the sys-
tem was based on a platoon concept of grouping
vehicles and was designed as a ring system around
Paris so vehicle size was increased and the project
eventually failed; however, the work with berthing
vehicles at stations provided the facts to show that
loading and unloading at stations did not slow capacity
for the entire system. This project was discussed in
a book titled: Aramis or the Love of Technology [21]
(Fig. 8).
The Swedish City of Gothenburg, since their city
was built on solid rock, were facing a problem of how to
deal with expanding mass transportation. Excited by
the British efforts they did a great deal of research on
PRT systems around the world and decided that the
Personal Rapid Transit and Its Development. Figure 8
Aramis vehicle, France (Reproduced from [11]. With
permission)
technology was not ready for public deployment, but
continued to be interested in the technology as it was
continued to be developed. PRT studies have continued
to this day in Sweden as they now move closer to an
urban application.
Canada also did a comparative study of PRT,
conventional highway and transit technology;
however they were interested in freight as well as people
movement and called PRT “Programmed Modules.”
In 1973 in Ontario, the Urban Transportation Devel-
opment Corporation developed a system; however
due to other influences, the vehicle was designed for
40 passengers with all that that implied and met
with minimal success in reference to the concept of
pure PRT.
After Morgantown
In the late 1970s money was appropriated for a study of
automated transit including PRT for Indianapolis
using the Cabintaxi system developed in Germany.
Several different-sized passenger vehicles were tested
in this comprehensive study with the result that the
smallest vehicle providing the lowest total cost per
passenger mile. This system had the support of busi-
ness, government, and civic organizations. When this
project did not proceed into a real-life application,
Raymond MacDonald and J. Edward Anderson began
a development program in 1981 at the University of
Minnesota that was to take into account all prior
research and work to develop the “perfect” PRT system,
although they concluded that the aerospace PRTsystem
was the “closest to being right” [22].
A company was formed with the assistance of the
University of Minnesota and a Chicago company funded
the work until 1986, when Dr. Anderson moved to
Boston University. Raytheon and local Boston engineers
intrigued by the Chicago-Area Regional Transportation
Authority (RTA) were seeking a new approach. Two
teams were organized developing designs for one to be
chosen for an application. This beginning culminated in
the RTA selecting the Taxi 2000 system with Raytheon to
design, build, and operate the test PRT system in
Rosemount, IL. In 1999, Raytheon terminated its work
on PRT 2000; TAXI 2000 Corporation bought back its
rights and has continued to develop its concept, now
called Skyweb Express.
 Personal Rapid Transit and Its Development 841

The PRT concept continued with 2getthere in the Personal Rapid Transit versus Existing
Netherlands implementing several successful projects, Transportation
all on grade: three considered GRT and one PRT. The
In understanding the paradigm-changing value of the
ParkShuttle technology was applied in the Business
modern pure PRT to the transportation field,
Park Riviam, the Airport Schiphol, and the Antibes
a comparison of PRT to existing systems is crucial.
Demonstration. The CyberCab project, a true PRT
This chart initially from Wikipedia has been adapted
system, was tested at a public Flower Show – Floriade
for this entry [27]:
at Hoofddorph, Netherlands in 2002. Developed by
Personal Rapid Transit versus Existing Transporta-
2getthere, all of these projects were successfully oper-
tion
ated and Riviam continues today as a public form of
transit. 2gethere is now designing the PRT system being
installed as part of the new carbon neutral Masdar City
Similar to ● Vehicles are small – typically
further discussed below [23].
automobiles two to six passengers
Ford appeared on the PRT scene briefly with a study ● Vehicles are individually used,
of a dual-mode system called PRISM that actually is like taxis, and shared only with
a Program for Individual Sustainable Mobility [24]. the passengers of one’s
This concept had some similarities to the NEV choosing
(Neighborhood Electric Vehicle) initiative. It included ● On-demand, around-the-clock
availability
very lightweight small vehicles that are on a controlled
● Travel is point to point, with no
system. It is an all-electric program that would start intermediate stops or transfers
small in a few cities eventually growing into a national
Similar to trams, ● A public amenity (although not
network linking to high-speed rail. The reason for this buses, and monorails necessarily publicly owned)
program was traffic congestion and the data collected shared by multiple users
by The Texas Transportation Institute of 68 US urban ● Passengers embark and
areas. They understood a system like this could increase disembark at discrete and fixed
fuel efficiency, reduce emissions, and be safer as less stations
● Can be elevated or use air
driver error would occur.
rights over existing highways –
The South Koreans had been working since 1990 reducing land usage and
on a PRT system called Vectus and in 2005 began congestion
collaborating with the Swedish to construct a test- Similar to automated ● Fully automated, including
track in Uppsala, Sweden, that is designed for winter people movers vehicle control, routing, and
weather application. The test facility passed safety collection of fares
requirements in 2007, had public access to the system Distinct features ● Vehicle movements are
in 2008 and now several studies for real projects in coordinated, unlike the
both Sweden and South Korea are underway. The test- autonomous human control of
automobiles and bikes
track layout has an outer loop of 300 m, allowing
● Reduced local pollution
speeds up to 12.5 m/s, and a station track of (alternative energy sources
100 m with a two-berth station. The station track is such as electric power and
designed be long enough to allow merge operations at solar)
full speed [25]. ● Small vehicle size allows transit
The British Company ULTra moved forward with infrastructure to be smaller
than other transit modes
its demonstration project in 2002, followed by
● Vehicles travel along a network
a feasibility study in 2004, for implementing PRT at of guideways where
Heathrow airport and is now our first “pure” PRT interconnection and multiple
system in operation. A more detailed discussion station options are standard
follows [26].
842 Personal Rapid Transit and Its Development

Current Systems in Operation or Under Cruz, CA, that is attempting a solar powered system
Construction and Fresno, CA, November 2006, the citizens voted to
spend $36 million to establish a fund to begin a PRT
Currently, two GRT networks are operational, one
effort in their downtown and is currently in the study
PRT system is operational, one PRT in testing, one
phase, the others are in various stages of the process.
full-PRT network is under construction, and several
Tysons Corner, VA, and Amber Glen, OR, are all also
more are in various stages of upgrade, planning, and
looking at PRT. In England, Bath, Cardiff, and Corby
design. Again the chart is adapted and updated from
have been studied for PRT implementation. While in
Wikipedia [27].
the United Arab Emirates, capital city Dubai, Lulu
In the United States many cities, towns, and activity
Island, Abu Dhabi, and Bawadi, Dubai, all are consid-
centers are in various stages of planning with San Jose,
ering PRT systems. Several studies are beginning in
CA, and Ithaca, NY, currently leading the way. San Jose
India and other European countries. The technology
is working through a municipal process while Ithaca,
is now being understood for its benefits to people
NY is studying PRT through a NYSERDA research
around the world [28].
grant. Alameda Point, CA, Mountain View, CA, Santa

Stations/
Location Status System Date Guideway vehicles Notes
Morgantown, Operational WVU PRT 1975 13.2 km 5/73 Up to 20 passengers per vehicle,
West Virginia, USA some rides not point to point
during low usage periods
Schiphol Airport Temporarily Park 1997–2004 2,000 m 2 10/4 Connection from parking to
Amsterdam, operational Shuttle loop airport 24/7 operation
Netherlands network
Rivium Business Operational Park 1999-today 18,000+ m 8/6 Public Transportation to
Park Shuttle II Business Park and City Functions
20 passengers
Hoofddorp, Temporarily CyberCab 2002 700 m 2/25 Four passengers, transport to
Netherlands operational 2GetThere view a flower show from a hill
peak 40 m high
London Heathrow Construction ULTra 2010 3.8 km 3/18 Will be the world’s first true
Airport, UK completed, commercial PRT system, initially
open to connecting Terminal 5 with
public a long-term car park. If
successful, BAA plans to extend
it throughout the airport
Hospital Rovisco Construction Critical 2010 The Hospital Rovisco Pais,
Pais, Portugal completed, Move a center for physical
under testing rehabilitation, uses the Move
between the several buildings of
the hospital
Masdar City, Abu Under 2getthere 2011 Magnets in 83/2,500 Automobiles will be banned, the
Dhabi, UAE construction pavement, only powered transport will be
automated PRT and intercity light rail
driving
Suncheon, In planning Vectus 2013 5 km (3.1 ?/40 Will connect Suncheon to the
Republic of Korea miles) future site of the International
Gardening Festival
 Personal Rapid Transit and Its Development 843

Heathrow Airport PRT The initial system is a 3.9 km (2.4 mile) single
guideway connecting three stations. Having 21 vehicles
Heathrow Airport ULTra PRT will be the first “true”
to travel between the stations takes about 5 min. The
PRT project for public use. It has already won a pres-
system had to be constructed over a very complex site
tigious UK award for transit technology advancement
that included two rivers, seven roads, green-belt land as
and has been called one of the 20 proven ways to save
well as complex airport conditions. If this initial
the earth by the UK Times and also a Sustainable Form
application is successful then it will be extended to
of Transport by the Directorate General for Energy and
include a full network of 350 vehicles, 50 stations, and
Transport of the European Union.
30 km of track linking all airport private and public
Heathrow is the main international airport in the
functions [29].
UK and one of the busiest in the world. The airport is
The ULTra pods can hold four adults and two
committed to PRT as the solution to provide connec-
children and can easily accommodate wheelchairs,
tivity within a very complicated airport system. The
bicycles and safely carry those who are physically chal-
initial application and the pilot program connected
lenged. The pods are battery controlled by a system on
Terminal 5 with a businessman’s parking lot. A higher
the pod. They will save 70% of energy compared to cars
cost to park can be applied to the businessman as the
and consume less than 50% of traditional buses [30].
dramatically reduced time for them from car to termi-
The cost in comparison to light rail and transitional
nal is well worth it (Fig. 9).

Personal Rapid Transit and Its Development. Figure 9

Heathrow Airport PRT Phase One (Martin Lowsen of ULTra)
844 Personal Rapid Transit and Its Development
2.0
MJoule per Passenger km
1.5
1.0
0.5
0.0
London bus
a ULTra
Personal Rapid Transit and Its Development. Figure 10
ULTra: PRT sustainable transport (Martin Lowsen of ULTra)
Personal Rapid Transit and Its Development. Figure 11
Two Heathrow Station Design (Martin Lowsen of ULTra)
automatic people moves is quite less, while a traditional
bus could be less expensive, the level of service and
benefits provided to the community cannot be com-
pared (Figs. 10–12).
Hospital Rovisco Pais, Portugal
This is a PRT system for a hospital setting so that all
within the complex have equal access to full mobility –
PRT is perfect for this. It is a fully automated electric
vehicle, on demand and point to point within a
dedicated right-of-way. The user can have communi-
cation with the system operator while in the
vehicle and can also control the position where the
vehicle stops. The vehicle can accommodate eight
passengers.
$m/mile
Automatic People
30–100
Movers (“APM”)
Light Rail 20–40
ULTra 10–15
Urban Car
Underground b
Key Sustainable/Energy and Planning/
Architectural
Being able to synergize transportation, energy, plan-
ning, and architecture to meet sustainable goals is a
rare confluence of technology/people interconnections.
PRT has always been understood for its benefits in
minimizing energy usage, while also providing the
opportunity for the system to be designed connected
to other sources of energy, new planning strategies and
architectural design. Peter Calthrope, a Yale-educated
architect and sustainability advocate from the 1976
calls it “The ideal transit technology: (a) stations right
where you are, within walking distance, (b) no
waiting.” He has been the supporter of the Alameda
Point Project since 2008 especially with the discussion
of an additional 60,000 housing units to the area. He

Personal Rapid Transit and Its Development. Figure 12
Heathrow expansion (Martin Lowsen of ULTra)
advocated ULTra PRT as it would enhance carpooling
and other forms of transit by allowing all of them to
interconnect. PRT can also extend bicycle commutes
and enhance walkable community design. At Alameda
Point, PRT can assist with interconnections along
the Red and Orange lines between Fruitvale BART,
Downtown/Park Street shopping, Marina Village jobs,
College of Alameda and Rapid Bus transfer, Alameda
Point 53-acre Sports Complex, Alameda Point National
Wildlife Refuge, Bridgeside Shopping Center, Alameda
Point cafes and parks, and Fruitvale Shopping Center.
Suncheon, Republic of Korea
Vectus, a company founded in South Korea that teamed
with the South Koreans to construct a test track in
Uppsula, Sweden, is now in the planning and design
phases for a project in Suncheon, S Korea. The PRT
system was chosen to be constructed at the famous
Suncheon Coastal Wetlands Park. It marks the first
time that the governmental agency has chosen PRT
over conventional transit and the environmental reasons
could not be more appropriate. Including all of the
characteristics that have been mentioned throughout
this paper, the South Koreans chose it because it was
less costly to build, takes less than half of the construc-
tion time and emits no pollutants including CO2 [31].
The first phase of the system will connect the
parking lot to the entrance of the wetlands and will
contain 40 vehicles on a 5 km system. Mr. Dong Hee
Lee, President and Chief Investment Officer of POSCO
Personal Rapid Transit and Its Development 845
stated: “POSCO is on the forefront of the global efforts
to protect the environment while improving the quality
of life”. The City will host the International Garden
Expo in 2013 and there are hopes that the initial system
may even be expanded by this time to connect with the
Central Train Station and the city’s downtown.
Masdar City, Abu Dhabi, UAE
Masdar City has been under development as the
world’s first zero-carbon city. This entire 62 km city is
completely new and has been totally designed from scratch
so that every decision is based upon creating a living place
that interacts with its local environment without adversely
changing it while also improving the lives of the people
who live there. Recently – at the 2009 World Future Energy
Summit – the cyber taxi by 2getthere was acknowledged
as one of the key features in creating such an environ-
ment. They have anticipated 3,000 electric cars with 90
stops connecting light rail, parking, and linking to
strategic walking points to the city above. They will
operate 24 h a day. The vehicles can travel at 40 kmph
and are guided by magnets imbedded in the streets that
interact with an onboard navigation system.
It is anticipated that the longest trip by PRT around
the city will take no more than 10 min. The city is raised
on a podium with all of the vehicles running below the
pedestrian-friendly city. The PRT system is anticipated
to eventually have 3,000 vehicles serving 130,000 trips/
day. The PRTsystem will also operate freight delivery to
the whole city – FRT (Freight Rapid Transit). The FRT
846 Personal Rapid Transit and Its Development

is designed for 5,000 trips a day. Renewable energy in
the form of lithium phosphate batteries allow the vehi-
cles to be charged at the stations and will only require
a 1.5 h charge for 60 km of use.
Masdar City and 2getthere have the exclusive rights
to apply the FROG-technology, the supervisory plan-
ning and control system TOMS, as well as unique
proprietary subsystems and components (such as the
FrogBox and Magnet Ruler – MMS) and a number of
related patents for Automated People Mover Systems.
These are the technologies that allow this system to run
on grade and they have a track record of 22+ years [32].
" The FROG (Free Ranging On Grid) technology creates
intelligent vehicles that can operate in any environment.
The on board FROG-box® controls the vehicle based on
electronic maps (route planning). While driving, the
vehicles measure distance and direction traveled by
counting the number of wheel revolutions and measur-
ing the steering angle (odometry). External reference
points (magnets embedded in the road surface) are
used to correct possible small inaccuracies in reference
to the planned route (calibration). The reliability of the
navigation system has been proven in all previous appli-
cations realized and tested successfully by TNO during
the FMECA safety-procedure for the ParkShuttle Rivium
2008, they issued the initial request that is now in the
final contract negotiation phase. They expect to have an
operational system by the end of 2015 (Fig. 13).
The following is an excerpt from the City of San Jose:
" It will start with a limited demonstration project and
proceed in three phases:
● Phase 1: Preliminary Design – A determination of
feasibility as described above.
● Phase 2: Final Design – Selection of system design
and construction contractor(s); completion and
verification of detailed design.
● Phase 3: Construction, Integration and Testing –
Physical construction of the system on-site culmi-
nating in operational certification.
All typical analysis for transportation infrastructure
such as ridership forecasts, routing determinations,
architectural development, civil engineering design, envi-
ronmental impact assessments, and community outreach
will be a part of the process. While nonconventional
aspects of this project include new usage and operational
models, completely new civil architectures and their
MOVING
WALKWAYS
LIGHT AIR
RAIL RAIL TRAVEL
ACCESSIBLITY----THROUGHPUT

ESCALATORS

application that has been operating since 1999.

ELEVATORS

BUSES

San Jose Airport, San Jose, CA

UNDERSERVED
The City of San Jose has identified the area around the MARKET

Sam Maneta airport as a place where the PRT concept

can assist with linking all of the existing movement
systems such as Caltrain, BART, and VTA Light Rail PEDESTRIAN
AUTOMOBILE
as well as potentially connecting to other places not
directly related to the airport such as North San Jose, as
well as to new growth. They have earmarked $4 million
TRIP DISTANCE
to conduct two simultaneous studies, one technical and
et

et

et

ile

the other with a focis on transportation and urban

ile

ile

ile
fe

fe

fe

m
10

planning. Laura Stuchinksy, a sustainability officer for

10

10

00
10

10

10

the San Jose Department of Transportation and many

other city officials see PRT as a complement to all of
these systems and one that will allow all of them to
function as a complete transportation system for the
airport and local area – a circulator system
interconnecting all the systems. This was noted in
a paper by Young, Miller, and McDonald. In September
Personal Rapid Transit and Its Development. Figure 13
Conceptual comparison of modes in the accessibility –
throughput spectrum from Young, Miller, McDonald. Keys
to innovative transport development. Presented at the
87 Annual Meeting of the Transportation Research Board,
Washington, DC, 2007
 Personal Rapid Transit and Its Development 847

integration with the as-built environment, along with
new elements of the technical system – innovative designs
for propulsion, control, vehicles, guideways, and vehicle
management systems, for example, that may be selected,
are specified and arranged in various ways to provide
varying degrees of utility and capability.
New approaches and guidelines/standards may
need to be developed, as will ridership forecasting and
performance evaluation models. A thorough review of
technical capabilities and maturity level relative to the
resulting performance requirements is required as is an
assessment regarding the ability of the still-small ven-
dor base to deliver such systems.
The CITY has selected two consultants as project
advisors, The Aerospace Corporation (SYSTEMS CON-
SULTANT) and Arup North America Limited (TRANS-
PORTATION CONSULTANT). In general,
TRANSPORTATION CONSULTANT will address all
conventional transportation infrastructure issues.
SYSTEMS CONSULTANT and TRANSPORTATION
CONSULTANT will together address the unconven-
tional project issues; the TRANSPORTATION
CONSULTANT focusing on the civil infrastructure-
related items and the SYSTEMS CONSULTANT focus-
ing on the technical system, as described generally above.
General Approach The two consultants will work
collaboratively as part of an Integrated Project Team
(PROJECT TEAM) led by the CITY. They will share
information and provide input necessary to each
other’s work and project deliverables, integrating their
workflows to achieve the CITY’s project objectives [33].
Future Directions
High-Tech/Low-Tech Connections: Man and
Machine
As PRT becomes more available for public use, the positive
connections between technology, the environment, our
communities, and ourselves – the low-tech/High-tech
connection will be better understood and allowed to flour-
ish. What are the high technologies of the future and how
are they transforming the ways that we travel through
space, place, and time. The test is how many lives can be

The Basics of Several PRT Systems

Seating
Capacity
System Location Active Status (per vehicle)
Morgantown PRT (Boeing) West Virginia Yes In service 8 seated plus
12 standing
ULTra (ATS Ltd) UK Yes In testing for final commissioning 4
Critical Move Portugal Yes In on-site testing 4
2getthere PRT Netherlands Yes 10 vehicles produced for Masdar City 6
Vectus PRT (POSCO) South Korea Yes Full prototype, Sweden; Suncheon, Korea 4
Cabinentaxi [24] Germany No Completed system: 1980s approval for 3,12,18
federal transit programs in Germany and US
PRT2000 (Raytheon) USA No Full prototype 4
Skyweb Express (Taxi2000) Minnesota Yes Partial prototype 3
MISTER Poland Yes Partial prototype 5
JPods USA Yes Mockup 4
SkyTaxi Russia Yes Concept 1,2,4
Launchpoint Technologies USA Yes Concept ?
SPM Maglev [25]
Skycab AB Skycab Sweden Yes Concept ?
SkyTran Unimodal USA Yes Full scale prototype 3
848 Personal Rapid Transit and Its Development
transformed for the better, for greater accessibility and the
long-term sustainability of our planet – these technologies
should be explored, developed, and expanded.
PRT with automated transit technologies will con-
tinue to evolve to become a greater part of our future
built world. They have the ability to transform the way
that we design and build our places for living, just as
positively as the car did over 100 years ago. Following that
technological innovation many peoples lives were
changed and for the betterment of all. Since the late
nineteenth century, world’s fairs have offered futuristic
visions of how people could and would live. Automated
People Movers are among the visions that have become
reality from them. Although most people connect mono-
rails with world’s fairs and amusement parks, there is no
reason to limit advanced transit technology to such envi-
ronments. For many land uses – such as airports, medical
centers, and colleges – automated people movers have
tremendous potential to address congestion, increase
connectivity, mitigate parking shortages, and provide
links to parking and PRT has the most potential within
this category to really transform our lives.
Because of its human scale, PRT allows the pedes-
trian experience to be preserved – minus the traffic and
congestion. It can also provide a more flexible link
between man and machine: because it so small, quiet,
and light (compared with conventional transit), PRT
can be adapted to a wider array of environments.
Although PRT does not eliminate the need for parking
in many situations, it does offer opportunities to change
the parking paradigm. In an automated parking facility,
for example, PRT vehicles could easily change direction
or move between levels within a small space [34].
PRTsystems can provide excellent transit coverage for
urban or suburban areas; such a system would be ideal, for
example, in locations such as Destiny USA – the environ-
mentally sensitive entertainment, shopping, and technol-
ogy project in Syracuse, New York. PRT would be equally
appropriate in more traditional urban locations, particu-
larly as a link between parking and other land uses. As part
of a larger effort to explore the transformation of office
parks into transit villages, Cities21, a nonprofit research
organization based in Palo Alto, California, is completing
an EPA-funded study of the environmental benefits of
a PRT system within the Hacienda Business Park [35].
In the 1960s, William Alden designed the StarrCar,
a vehicle that could function independently, on the road,
or on dedicated tracks, as part of a transit system. At the
time, technology had not yet caught up to Alden’s vision,
although part of Alden’s concept was the basis for the
Morgantown PRTsystem. Today, however, with advanced
automotive technology, Alden’s vision is almost within
reach. Within a system that allowed vehicles to be taken
off the tracks as needed, for example, PRT vehicles may be
used to pick up drivers at their parking spots and take
them to their next destination. The parking facility could
both provide space for traditional vehicles, and become
an active participant in new approaches to transit and
movement! (As automotive technologies advance, mov-
ing in the direction of “cars that drive themselves,” it may
be possible to drop off a car at a garage and have the car
park itself as you connect to the PRT system. As PRT,
automated vehicles, and other high-tech transportation
options become integrated into the overall transporta-
tion system, they will affect parking garage design in
ways that cannot yet be imagined [36]).
Beauty, technology, function, engineering, and civic
purpose: the pieces are all in place. But determining the
solutions of the future – whether urban, suburban, or
rural – will mean connecting, at a deeper level, to the
ways in which people relate to the world around them. It
will be through these deeper connections that new solu-
tions emerge. Horst Bredekamp, in an exploration of the
phenomenon of the Kunstkammer – the cabinets of art
and curiosities that were the precursors to modern
museums – notes that
" No one wants to return to the deliberate chaos of the
Kustkammer as museums. But the boundaries between
art, technology, and science are beginning to break
down in a similar manner as has been demonstrated
by the Kunstkammer. In view of this fact, their lessons
of visual association and thought processes which pre-
cede language systems take on a significance which
might even surpass their original status. Highly tech-
nological societies are experiencing a phase of Coper-
nican change from the dominance of language to the
hegemony of images [36].
In looking toward the future of PRT, advances in
new transit technology have been hindered by a lack of
focus on research and development in this area. How-
ever, in fits and starts, and often with private money
and due to the perseverance of innovators and vision-
aries, PRT is just starting on its path to its full potential,

finding its way into our world of public transit linking
with all existing forms of movement.
One of key area of future advancement is the con-
trol software that allows for flexibility and openness so
that the complex and time-consuming tasks that will
allow a large fully operational city PRT system to func-
tion safely and quickly can be applied to public transit.
Systems of this type are being implemented in Auto-
mated Guided vehicles for warehouses and manufac-
tures as well as being used successfully in airport
luggage systems and emerging PRT systems.
The motivation behind PRT is to provide not just
an advanced commuter transit experience but to pro-
vide a type and level of service that meets the needs of
those who cannot or do not have access to the auto-
mobile, a truly new way to move and live in our mod-
ern world. It is a transit system that attempts to put
people and cities first through technology while mak-
ing an improvement to the quality of urban life.
The theory and reality of PRTshows this technology
to be the transit technology of the future. From 2002 to
2005 the EDICT project funded by the European Union
and involving 12 research organizations concluded that
PRT [37]:
● Would provide future cities “a highly accessible, user-
responsive, environmentally friendly transport system
which offers a sustainable and economic solution”
● Could “cover its operating costs, and provide a return
which could pay for most, if not all, of its capital costs”
● Would provide “a level of service which is superior to
that available from conventional public transport”
● Would be “well received by the public, both public
transport and car users”
The risks of being the first to implement the system
are being removed and so we should see the technology
move forward and provide the level of service to all users
of transit that has been envisioned since its beginnings.
A review of all of the literature shows many detailed
specific studies about networks, walkabilty, systems con-
cepts, human affect, and proposals that indicate all that
is claimed about PRT is real so now according to Rachel
Liu [1] all that is required are:
" multiple business models so that its many applications
can find funding. As concluded in a recent PRT study
(Carnegie and Hoffman 2007), AGT possesses the virtue
Personal Rapid Transit and Its Development 849
of sustainability due to its small footprint, lower cost
and lower impact on the environment. On the other
hand, its small size and low-key profile have fostered
a large number of applications worldwide without gar-
nering any major headlines, which may, however, sup-
press its potential as a unique solution to urban
circulation and congestion problems.
Along with:
on-going PRT studies reveals that the specifications
of technology and assessment of costs may be rela-
tively straightforward, but quantifying benefits associ-
ated with the implementation of a transportation
project and evaluating the market conditions are com-
plex. There are a number of analytical tools to assign
a dollar value to benefits; however, some impacts such
as congestion relief, safety improvements, or air quality
improvements are often difficult to quantify financially.
Other qualities, such as aesthetic appearance, may not
even be quantifiable. Environmental and societal
impacts are often referred to as “external” effects of
transportation activities since they are not directly
reflected in monetary costs and benefits of project
implementation. By externalizing these factors, bene-
fit/cost analyses often do not capture the full value of
beneficial impacts. However, the significance of all
impacts, both positive and negative, needs to be con-
sidered in the decision-making process.
Bibliography
Primary Literature
1. Liu R (Rachel) (2010) Spectrum of automated guideway transit
(AGT) technology and its applications. In: Kutz M (ed) Handbook
of transportation engineering. McGraw-Hill, New York
2. Committee of Automated People Mover Standards
(2006) Automated people mover standards, Part 3. American
Society of Civil Engineers, Reston
3. Irving J (ed) (1978) Fundamentals of personal rapid tansit. DC
Heath, Lexington
4. Morgantown PRT: http://transportation.wvu.edu/prt
5. Gunter M (7 Mar 2008) Building the world’s cleanest city.
Fortune Magazine
6. Fichter D (1964) Individualized automated transit and the city.
BH Sikes, Providence, RI
7. Anderson JE (1996) Some lessons from the history of personal
rapid transit. Paper presented at the international conference
on personal rapid rransit (PRT) and other emerging transpor-
tation systems, Minneapolis, Minnesota, 18–20 Nov 1996.
http://advancedtransit.org/doc.aspx?id=1025
850 Personal Rapid Transit and Its Development
8. Cole LM, Merritt HW (eds) (1968) Tomorrow’s transportation:
new systems for the urban future, US Department of
Housing and Urban Development, Office of Metropolitan
Development Chapters available online: “Summary,” pp. 1–5;
“Recommended Future Systems,” pp. 58–77. http://faculty.
washington.edu/jbs/itrans/tomtran.htm.
9. Henderson C et al (1968) Future urban transportation systems:
descriptions, evaluations and programs, Final report, Volumes
1 and 2. Prepared for the US Department of Housing and
Urban Development by the Stanford Research Institute
10. Systems Analysis of Urban Transportation Systems (1969)
Scientific American 221:19–27
11. Burke C (1979) Innovation and public policy: the case of
personal rapid transit. Lexington Books, D.C. Heath
12. Andreasson I (2000) Innovative transit systems, VINNOVA:
Swedish Agency of Innovation Systems. http://vinnova.se/
publ/list_vr.htm#03 as well as Vectus web site http://www.
vectusprt.com/
13. Lowson M ULTra Urban Light Transport, Advanced Transport
Systems. http://www.atsltd.co.uk/
14. United States Congress, Office of Technology Assessment (June
1975) Automated guideway transit: an assessment PRT and other
new systems. US Government Printing Office, Washington, DC
15. Merritt HW (1993) Reflections on the New Systems Study Pro-
ject, pp 35–59 in Automated People Movers IV: Proceedings of
the 4th international conference on automated people movers,
American Society of Civil Engineers, Irving, Texas, 18–20 Mar
1993. http://faculty.washington.edu/jbs/itrans/reflec2.htm
16. US Congress, Office of Technology Assessment (June 1975)
Automated guideway transit: an assessment of PRT and other
new systems. Prepared for the Senate Committee on Appro-
priations, Transportation Subcommittee. http://ntl.bts.gov/
data/OTA/7503/7503.htm
17. Wolf P (1975) Evolving city. American Federation for the Arts,
New York
18. Reports of the Steering Group (1963) Traffic in towns. H.M.
Stationery Office, London and reprint Penguin Books in asso-
ciation with HMSO
19. Blake LR (1966) A public transport system using four-
passenger, self-routing cars, Inst. Mech. Eng. convention on
guided land tTransport, vol 181, Pt. 3 G
20. Tadi R, Utpal D (1997) Detroit downtown people mover: ten
years after. paper presented at the 6th international confer-
ence on automated people movers, Las Vegas, NV
21. Latour B (1996) Aramis or the love of technology. Harvard
University Press, Harvard
22. MacDonald R, Anderson JE (1978) Transit systems theory.
Lexington Books, D.C. Heath. 21st century personal rapid
transit, Self-publish
23. 2getthere: http://www.2getthere.eu/
24. Proposal for Independent Sustainable Mobility (PRISM)
(pdf). Ford Advanced Research Division (2003) http://faculty.
washington.edu/jbs/itrans/PRISMGPCPaper.pdf. Accessed 3
Aug 2010
25. http://www.vectusprt.com/
26. Heathrow airport web site: http://www.heathrowairport.com/
portal/controller/dispatcher.jsp?CiID=e993760079e1d110
VgnVCM10000036821c0a____&ChID=10b35109350d3110
VgnVCM10000036821c0a____&Ct=B2C_CT_PRESS_RELEAS
E&CtID=a22889d8759a0010VgnVCM200000357e120a____
&ChPath=Home%5EHeathrow%5EHeathrow+press+releases
27. http://en.wikipedia.org/wiki/Personal_rapid_transit
28. Multiple website concerning PRT http://www.princeton.edu/
alaink/Orf467F04/NJ%20PRT%20Final%20Small.pdf
29. Lowson M (2004) (doc) A new approach to sustainable trans-
port systems. 13th World Clean Air and Environmental Protec-
tion Congress London, 22–27 Aug 2004. http://www.ultraprt.
com/media/papers/
30. Rodgers L (2007) Are driverlss pods the future. http://news.
bbc.co.uk/2/hi/uk_news/7148731.stm. Accessed Aug 2010
31. http://www.vectusprt.com/center/news_view.php
32. Masdar: http://www.masdar.ae/en/home/index.aspx
33. San Jose: http://www.sanjoseca.gov/transportation/Support
Files/admin/RFI_Auto mated_transit_networkQA.pdf
34. Cities21: http://www.cities21.org/cms/
35. McDonald S (2007) The parking garage: design and
evolution of a modern urban form. Urban Land Institute,
Washington, DC
36. Bredekamp H (1995) The lure of antiquity and the cult of the
machine: the Kunstkammer and the evolution of nature, art,
and technology. M. Wiener Publishers, Princeton, pp 113
37. EDICT: EDICT Final Report (pdf) from cardiff.gov.uk
Books and Reviews
Advanced Transit Association (1988) Personal rapid transit:
another option for urban transit? J Adv Transport
2(38):192–314
Anderson JE (July 2009). An intelligent transportation network system:
rationale, attributes, status, economics, benefits, and courses of
study for engineers and planners. PRT International, LLC
Anderson J, Romig S (1974) Personal rapid transit II. Audio-Visual
Extension, University of Minnesota, Minneapolis, MN
Carnegie JA, Hoffman PS (2007) Viability of personal rapid transit
in New Jersey. New Jersey Department of Transportation,
New Jersey, 141p
Institute for Transportation in Cooperation with the Advanced
Transit Association (1966–present) Advanced Transport Quar-
terly, Wiley-Blackwell
Muller P (2007) A personal rapid Transit/Airport Automated People
Mover Comparison” Proceedings of the 29th international air
transport conference. American Society of Civil Engineers.
http://www.prtconsulting.com/
Richards B (2001) Future transport in cities. Spon Press, London
Schneider JB (2011) Innovative Transportation Systems. This web
site offers a wealth of information, including summaries and
excerpts from many other sources of information. http://
faculty.washington.edu/jbs/itrans/
Warren R (1998) The urban oasis: Guideways and greenways in
the human environment. McGraw-Hill, New York
 PHEVs and BEVs in Coupled Power and Transportation Systems 851

PHEVs and BEVs in Coupled Power reducing the current dependency on oil-based fuels.
Should the gasoline cost increase in the future, Plug-
and Transportation Systems in Hybrid Electric Vehicles (PHEVs) and Battery Elec-
MLADEN KEZUNOVIC1, S. TRAVIS WALLER2 tric Vehicles (BEVs) will become the economical choice
1
Department of Electrical and Computer Engineering, for transportation. Widespread adoption of PHEVs/
Texas A&M University, College Station, TX, USA BEVs will also improve air quality and carbon foot-
2
Department of Civil Engineering, The University of print, since point source pollution is easier to control
Texas at Austin, Austin, USA than mobile source pollution. This level of control is
essential for effective implementation of carbon cap-
and-trade markets, which should spur further innova-
tion. In USA, sales of Hybrid Electric Vehicles (HEVs)
Article Outline
have grown 80% each year since 2000, proving that
Glossary PHEVs/BEVs are likely an eventual reality that must
Definition be dealt with [1]. The implications of this reality will be
Introduction highly dependent on the policies in place to use
Policy Issues PHEVs/BEVs to the benefit of the transportation and
Modeling of Complex Systems power systems, as well as the drivers, industry, and
Benefits public at large.
Conclusions and Future Research Beyond fuel costs and sustainability, the primary
Acknowledgments concern of the transportation sector is congestion.
Bibliography In 2005, congestion was estimated to cost the US
economy $78.2 billion in wasted time and fuel [2]. If
Glossary PHEV/BEV drivers are given appropriate incentives
(e.g., strategically placed energy exchange stations),
BEV Battery electric vehicle.
traveler behavior (e.g., choice of routing, departure
DSM Demand side Management; utility-sponsored
time, and destination) impacting congestion may be
programs to influence the time of use and amount
affected.
of energy use by select customers.
In addition, the power industry is currently chal-
G2V Grid-to-vehicle; using the electrical grid to
lenged to maintain reliability of operation while
charge the battery of a vehicle.
expanding the grid to meet growing demand. Large
HEV Hybrid electric vehicle.
blackout such as the northeastern one in 2003 may
OM Outage management; set of manual and/or auto-
create loses in billions of dollars [3]. Introducing the
mated procedures used by operators of electric dis-
renewable resources to meet growing demand requires
tribution systems to assist in restoration of power.
energy storage to deal with interfacing [4]. If proper
PHEV Plug-in hybrid electric vehicle.
policy is in place PHEVs/BEVs can provide a promising
V2B Vehicle-to-building; exporting electrical power
solution acting as mobile decentralized storage (MDS)
from a vehicle battery into a building.
of electrical energy. In this capacity, PHEVs/BEVs can
V2G Vehicle-to-grid; exporting electrical power from
serve in two modes: grid-to-vehicle (G2V) and vehicle-
a vehicle battery to the electrical grid.
to-grid (V2G), each providing benefits to the power
system operation. The G2V mode can be used to charge
Definition
PHEVs/BEVs at reduced cost when the power system
With the price of oil peaking in the recent past close to load is reduced and generation capacity is abundant,
the once unimaginable $150 per barrel and the threat of such as during night time. The V2G mode may be used
global climate change increasingly acknowledged, the when demand is high or supply is accidentally lost since
transportation sector is employing a number of new the stored electric energy can be released from PHEVs/
technologies that will enhance energy security by BEVs in an aggregated way, which will offer major

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
852 PHEVs and BEVs in Coupled Power and Transportation Systems
contributions to regulation service and spinning
reserves, as well as load-shedding prevention.
The mobility of the energy storage in PHEVs/BEVs
allows for strategic placement of the distributed gener-
ation source to optimize power system needs.
Figure 1 illustrates the spatial and temporal coupling
of the power and transportation systems through show-
ing an example of a PHEV/BEV driver’s route, highlight-
ing destinations where the driver could potentially
engage in G2V or V2G activity. Options for meeting
selected criteria for electricity and transportation
networks simultaneously are numerous. Developing
policy strategy requires understanding of trade-offs
involved in pursuing certain solutions at the expense of
others. The all-encompassing theoretical framework for
such studies to the best of our knowledge is not available.
Traditionally, scientists have adopted a divide and
conquer approach to understanding complex phenom-
ena. Unfortunately, systems with emergent dynamics
that are dominated by contextual interactions are not
well suited to this classical approach (e.g., [5–7]). In
such cases, directly addressing the couplings of system
components may actually hasten progress. While this
linkage presents new opportunities to improve the
functioning and capacity utilization of each system, it
7:30 AM Work
5:00 PM Work
7:00 AM Kid’s
School
6:40-7:00 PM Grocery
5:50-6:30 PM Gym
6:00 AM Home
10:30 PM Home
5:30 PM Home
PHEVs and BEVs in Coupled Power and Transportation
Systems. Figure 1
Temporal and spatial dimensions of plug-in opportunities
also raises the spectrum of increasing dynamic com-
plexity and cascading failures across systems.
In this entry, several open policies and research
goals will be discussed, which facilitate optimizing the
integration of the transportation systems and the
behavior of its travelers with the electricity systems
and behavior of its end-customers. PHEVs/BEVs
based demand side management (DSM) and outage
management (OM) are also presented as an application
of PHEVs/BEVs using in the coupled power and trans-
port system.
Introduction
The impacts PHEVs/BEVs will have on transportation
systems, power systems, and air quality are very com-
plex. Studies conducted to date on this topic make
many assumptions to simplify the problem. As stated
in the definition, the problem space must be treated as
one large complex system in order to capture emergent
behavior.
The complexity of the issues involved in studying
PHEVs/BEVs and their interaction with electricity and
transportation networks is shown in Fig. 2, where sev-
eral disciplines that need to be involved in researching
this multidisciplinary problem are shown.
Recent analyses confirm the feasibility of the grid-
to-vehicle (G2V) and vehicle-to-grid (V2G) concepts
[8–13]. The Electric Power Research Institute
Policy
Economics
Engineering
Technology
Societal Impacts Environment
PHEVs and BEVs in Coupled Power and Transportation
Systems. Figure 2
Illustration of multidisciplinary nature of problem

speculates that V2G could reduce the requirement for
global, central-station generation capacity by up to
20% by the year 2050 [14]. Several studies omit any
consideration of vehicle locations and desired activity
patterns and assume a percentage of vehicles are
plugged in and available when estimating the benefits
to the grid and to drivers [8, 10, 11].
Many researchers have investigated the various
potential benefits and implementation issues of the
V2G concept. Kempton and Tomić studied the funda-
mentals of using PHEVs/BEVs for load leveling, regu-
lation, reserve, and other purposes [15, 16]. Hadley and
Tsvetkova analyzed the potential impacts of PHEVs on
electricity demand, supply, generation, structure,
prices, and associated emission levels in 2020 and
2030 in 13 regions specified by the North American
Electric Reliability Corporation (NERC) [17].
Meliopoulos et al. considered the impacts of PHEVs/
BEVs on electric power network components [18].
Anderson et al. performed the case studies of PHEVs/
BEVs as regulating power providers in Sweden and
Germany [19]. Guille and Gross presented a proposed
framework to effectively integrate the aggregated bat-
tery electric vehicles into the grid as distributed energy
resources [20]. The combined impact PHEVs/BEVs
make on both electric power system and transportation
network has not been explored as much. When consid-
ering the role of PHEVs/BEVs as dynamically
configurable (mobile) energy storage, the potential
impacts on both electricity and transportation net-
works may become quite diverse. The flow of traffic is
an important factor in deciding the flow of electric
power that could be utilized from PHEVs/BEVs.
Correlating the movement of people to the movement
of the power load offers new opportunities in the
smart grid.
One of the major advantages of PHEVs/BEVs is
their usefulness as an MDS. MDS is a revolutionary
concept because currently the power grid has no stor-
age except for 2.2% of its capacity in pumped storage
[11]. Without significant and reliable storage of energy,
maintaining grid stability and reliability under the
growing electricity demand is a complex problem. Util-
ities may contract with others to provide power in any
one of the four types of markets: base-load power, peak
power, spinning reserves, and regulation services.
Several studies have shown that PHEVs/BEVs can
PHEVs and BEVs in Coupled Power and Transportation Systems 853
provide ancillary services (spinning and regulation)
at a profit [8, 10, 11]. Spinning reserves receive pay-
ment for providing continuous capacity regardless of
whether energy is provided, and receive further pay-
ment if called on to feed energy into the grid. Regula-
tion services feed a nominal amount of energy into the
grid, and receive payment for reducing or increasing
their energy consumption as needed. In the case of
PHEVs/BEVs, being plugged-in in a predictable way
means that capacity is available to feed into the system
if called upon. PHEVs/BEVs are particularly well suited
for regulation services since the impact on vehicle’s
energy resources may be zero.
The pricing of V2G and G2V services is expected to
cause a fundamental shift in the behavior of PHEVs/
BEVs drivers. Further research is needed to investigate
the exact nature of this shift; however, if the pricing
schemes are developed with both the power system and
transportation system in mind, then PHEVs/BEVs
could help solve problems plaguing the traffic network,
particularly congestion. The pricing scheme should
also consider air quality impacts caused by charging
at different times in the day. As mentioned earlier, MDS
will allow for renewable energy to be used more effi-
ciently. There will however remain times of the day
more dominated by “dirty” fuels than others.
As observed, the body of research literature related
to the multidimensional impact of PHEVs/BEVs is
quite small. The remainder of this section will focus
separately on the dual problems of improving the sta-
bility and reliability of the electrical grid and improving
the efficiency of the roadway network.
Stability and Reliability of the Electric Grid
Stability and reliability of the US electric grid have
become issues of increasing concern since the occur-
rence of several blackouts in the 1990s (Western Inter-
connect in 1994 and 1996, and the Eastern Interconnect
in 1999) and system deregulation. The devastating
impact of the northeast blackout from August 14,
2003 reminded that the situation with the grid is only
worsening and not improving. Here, a stable system is
defined as one in which the phase and frequency of
power generation units are constant. Ability of the
system to maintain the state of equilibrium during
normal and abnormal conditions is a measure of
854 PHEVs and BEVs in Coupled Power and Transportation Systems

300

Transmission loading relief

250

procedures
200

Data from North American electric reliability council

150

100

50
a 0
0.009
Monthly average frequency

0.007

0.005
errors

0.003

0.001

–0.001
1999 2000 2001 2002 2003
Order 888 Date
b goes into effect

PHEVs and BEVs in Coupled Power and Transportation Systems. Figure 3

Illustration of multidisciplinary nature of problem [21]

stability. Reliability is defined as the ability of the sys-
tem to meet unexpected demand and respond to fail-
ures. Ability of the system to deliver electricity to
customers within the accepted standards, which may
be affected by the failure rate, repair rate, or duration of
loss, is a measure of reliability. Figure 3 illustrates the
worsening stability problem. Order 888 in Fig. 3 relates
to the Open Access to Transmission issued by Federal
Regulatory Commission in 1996, which is the result of
an authorization passed by the Congress as a part of the
Energy Policy Act of 1992.
A major challenge in achieving these goals (stability
and reliability) is the lack of energy storage. Figure 4
depicts the peaking structure of an example power load
over the course of 1 day. In this example, demand grows
rapidly starting at 6 a.m. and begins to decline after
hitting a peak around 3 p.m. This peaking phenome-
non is especially important to consider given that dif-
ferent energy sources are available at different times of
day. For example, wind energy is most widely available
at night when the demand for power is the lowest.
While it may seem intuitive that a flat demand curve
is the ideal, this is not necessarily true. More research is
required to determine if parts of the system (e.g., trans-
formers) require time to cool down. The large scale use
of energy storage would significantly help meeting the
stability and reliability needs, including managing the
load variations shown in Fig. 4.
Efficiency of the Roadway Network
Congestion is a problem not only in the electricity grid
network, but also in the roadway network. Vehicle
miles traveled (VMT) has risen consistently since the
advent of the automobile, with dips when gasoline
prices rise quickly (See Fig. 5 for the VMT trend since
1992). If the transportation sector is shifted to an
alternative fuel source (i.e., electricity) with greater
price stability, and especially if the source of the fuel
is renewable, then VMT is expected to continue to
increase into the foreseeable future. While mobility is
an indicator of economic success, the expansion of
a roadway system is limited by available space and
finances. Roadway network efficiency is further
constrained by the individual autonomy of drivers
who act in their self-interest instead of the interest of
 PHEVs and BEVs in Coupled Power and Transportation Systems 855

Total installed capacity

Peaking plants

Peak day
load shape
Valley-filling

Seasonal average
load shape

Fossil generation

Renewables and hydro

Nuclear
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

PHEVs and BEVs in Coupled Power and Transportation Systems. Figure 4

Illustrative peaking of electricity load [12]

270
260
250
240
230
220
Billions

210
200
190
180
170
160
150
Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07

PHEVs and BEVs in Coupled Power and Transportation Systems. Figure 5

US highway VMT [23]

the system (see [22] for a theoretical description of system via methods such as pricing and technol-
traveler behavior). ogy, but few solutions proposed offer a case
Extensive research has been conducted on even close to being as comprehensive as PHEVs/
improving the efficiency of the transportation BEVs.
856 PHEVs and BEVs in Coupled Power and Transportation Systems
Policy Issues
The policy issues presented here are centered on incen-
tives to help industry develop and bring new value
to end users of electricity and transportation networks,
and society at large, while encouraging competition
and development of new business opportunities.
Improve Electric Grid Performance
Widespread deployment of PHEVs/BEVs will allow for
increased energy storage, and improved reliability and
stability of the electric grid. Linking the transportation
and power systems through PHEVs/BEVs will allow for
electrical energy storage on a scale much larger than is
currently feasible. The additional energy capacity will
be directly proportional to the penetration of PHEVs/
BEVs into the automobile market, and modeling (see
the modeling section) is needed to determine the exact
increase in capacity across the space and time
dimensions.
The new mobile storage can only benefit the electric
grid if it is available at the right time and place to service
the grid when needed. To determine PHEVs/BEVs’
demand for electric energy across space and time, travel
patterns must be considered. Figure 1 shows an example
of such a pattern, highlighting several destinations
where a driver could potentially engage in G2V or V2G.
Stability and reliability issues were mentioned ear-
lier. V2G is poised to greatly aid the grid in becoming
more reliable and stable because vehicles are only in use
for a small portion of each day (average daily travel
time person in 2001 was 82.3 min [24]). During the
remainder of the day, the vehicles can be plugged in and
provide services (e.g., ancillary or regulatory).
This approach requires a policy shift to allow use of
the MDS for energy to maintain stability and reliability.
Also, policy that encourages utilities to cooperate with
the PHEV/BEV owners or aggregators and provide
tariff incentives for their participation in programs
aimed at demand and distributed generation manage-
ment and optimization is missing at the moment.
Enhance Penetration of Renewable Energy Sources
to Improve Energy Security
Increasing energy capacity by using PHEVs/BEVs
as MDS will allow for increased investment in
renewable energies by alleviating concerns related
to the temporarily highly variable nature of solar (day-
time) and wind (primarily nighttime). Using renewable
energy has benefits not only for the environment
and air quality, but also for energy security by
reducing reliance on the supply from oil producing
countries.
This approach requires a policy shift to allow and
encourage large scale use of the MDS for energy to
support interfacing of renewable generation.
Reduce and Redistribute Pollution in the Electric
Grid and Transportation Network
By shifting the source of pollution away from vehicles,
PHEVs/BEVs will change the transportation-based air
pollution problem from a mobile source issue to
a point source issue. This redistribution of pollution
will likely have the effect of reducing pollution because
point sources are much easier to control and some
already have emission caps in place. Hadley [25]
conducted initial research into the potential air quality
impacts of PHEVs/BEVs, describing the impacts of
G2V charging on air quality, and considering the
types of power generation that are typically used at
different times of day (e.g., coal-fired generation is
prevalent at night in some regions).
Some policies are already in place to ensure that the
redistribution of pollution that will occur with wide-
spread deployment of PHEVs/BEVs will actually lead
to a reduction in pollution. Further policy analysis is
needed to ensure that V2G and G2V services are incen-
tivized to occur at times when it will result in the
maximum improvement in air quality.
Create New Markets and Further Deregulate Existing
Markets
PHEVs/BEVs are poised to open new markets and
increase opportunities in existing ones. Carbon-trade
markets should be aided because they facilitate the
change of the transportation-based air pollution prob-
lem from a mobile source issue to a point source issue
(as described earlier). While point sources of pollution
are much easier to control, if they are nonrenewable,
they will likely need to trade carbon credits to counter
the increased emissions.

PHEVs/BEVs will also create new modes for partic-
ipation in the electricity markets. There will be oppor-
tunities for businesses to act as Qualified Scheduling
Entities (QSE) to the electric utility by facilitating V2G/
G2V interactions. Such a QSE that aggregates across
vehicles is necessary because any one vehicle’s contri-
bution will be too small to allow it to participate
directly in the market. PHEVs/BEVs will likely function
akin to small generators as a distributed energy
resource.
Policy that enhances market development and
deregulation allowing a new type of QSE to bid in
a variety of markets is needed to facilitate the aggre-
gated use of PHEVs/BEVs in “transportation-energy”
markets.
Plan and Develop Energy Exchange Stations
Energy exchange stations (for G2V and V2G) could take
one of at least two forms. The first, the way considered
by most electric vehicle research to date, assumes that
individual drivers plug in and charge their vehicle over
a period of several hours. Some examples of potential
charging station locations are shopping malls, recrea-
tional areas, schools, and of course homes.
Further, rather than requiring drivers to plug into
the grid and wait several hours to charge their batteries,
battery exchange locations could be as ubiquitous as
gas stations and automatically exchange discharged
batteries with fully charged batteries. Charging
PHEVs in this way has benefits for drivers because the
process takes only a few minutes as opposed to several
hours. Also, this system would require a leasing system
for batteries similar to the system in place for leasing
cell phones, alleviating driver concerns about battery
life. The benefit for utilities is that control over charg-
ing and servicing the grid is centralized.
In reality, charging (G2V) and discharging (V2G)
services will likely be based on a hybrid of the two
methods mentioned above (individual drivers plugging
into the grid and stations designed to exchange batte-
ries). Depending on the pricing structure in place, it
may make sense for drivers to exchange batteries dur-
ing long drives and plug in to a household plug at
night. Incentive structures will need to be developed
that consider the different players – energy exchange
stations and individual drivers.
PHEVs and BEVs in Coupled Power and Transportation Systems 857
The temporal and spatial aspects of the activity
patterns travelers choose (see, e.g., Fig. 1) adds a layer
of complexity to the problem of locating charging
stations to link the transportation and energy systems.
This requires both micro (neighborhood, city, and
metropolitan area) and macro (region, state, and
nation) driver behavioral dynamics to be studied in
detail. If appropriate incentives are developed, drivers
could be encouraged not only to act in a way that best
serves the grid, but also to act in a way that best serves
the transportation system. The incentives could be
passive such as pricing electricity for planned contri-
bution at the location of charging facilities (either
stations or induction charging embedded in the road-
way), or active such as pricing electricity based on
congestion in both the power grid and local transpor-
tation system. Cognitive and behavioral research is
needed to determine the appropriate incentives.
Policy that addresses the planning requirements
for charging stations and regulates emerging
energy exchange markets is needed. Comprehensive
policy that develops joint electricity and transpor-
tation programs for incentivizing drivers to partic-
ipate in the transportation and electricity grid
optimization are not yet proposed or even clearly
defined.
Modeling of Complex Systems
To develop policy strategies that allow for faster and
more significant penetration of PHEVs/BEVs, research
is needed to model the interactive performance of two
complex systems, power and transportation, linked
through the behavior of individual vehicle operators,
where this linkage is determined by the location of
interface infrastructure. The behavior of travelers
defines the required inputs into power modeling since
time-dependent PHEVs/BEVs locations are critical.
Every aspect of this meta-system enterprise (power,
transport, consumer choice, and infrastructure devel-
opment) is interlinked, therefore fully understanding
policy issues is quite challenging. This section explores
each aspect of the modeling approach beginning with
transportation modeling, and then power systems
modeling, then modeling the role of human agents,
and finally determining economic feasibility (see
Fig. 2 for illustration).
858 PHEVs and BEVs in Coupled Power and Transportation Systems

Transportation Modeling locations. This inference directly relates to the number

of PHEVs/BEVs present at a specific power grid node,
Travel models typically contain demand and supply
which will be related to the node’s self-admittance
components. While most demand models used in prac-
described in the next subsection on power systems
tice are static and consider each leg of a trip separately,
modeling. Consideration of multiple classes of trav-
activity-based models are gaining momentum.
elers, PHEVs/BEVs and non-PHEVs/BEVs, will be
Lemoine et al. [1] illustrate the problems that PHEVs/
critical until PHEVs/BEVs reach high percentage
BEVs could pose if proper incentives are not given to
penetration.
ensure that energy exchange occurs at times beneficial
It has been long understood that through pricing-
to the grid. Activity-based travel models are better
based incentives, the system-level performance of
suited for PHEVs/BEVs modeling because they recog-
transportation networks can be greatly improved. The
nize that travel arises from a fundamental need to
entire field of congestion pricing (e.g., [33, 34])
participate in activities, and thus the models capture
addresses this fact. For instance, PHEVs/BEVs provide
trip-chaining behavior (e.g., home to work to grocery
a novel opportunity to achieve gains in controlling and
to home). Other benefits of activity-based models are
managing congestion in transportation systems
the incorporation of intra-household interactions,
through an incentive based approach that persuades
interpersonal and intrapersonal consistency measures,
users to act in an altruistic manner. Further, such
consideration of space-time constraints on activities
incentives provide a unique opportunity (and com-
and travel, and emphasis on individual level travel
plexity) in that dual objectives must be balanced:
patterns (as opposed to monitoring aggregate travel
improving the efficiency of the transportation as well
demands). A number of micro-simulation platforms
as that of the power system. For the transportation
that employ the activity-based paradigm of transpor-
system, incentives influence route, departure, as well
tation demand forecasting have been developed in the
as destination choice. Incentives change the fundamen-
last 5 years (e.g., [26–28]).
tal costs traveler’s associate with their choices and a new
On the supply side, conventional techniques of trip
general cost dynamic equilibrium emerges (for normal
assignment are static in nature, and consider vehicle
operating states). This requires a further broadening
flows aggregated over one or several hour time periods.
of the previously mentioned integrated modeling
The limitations of the static assignment procedures and
approach to include generalized costs as well as hetero-
the increase in computing capacity have allowed the
geneous values of time.
field to move toward more behaviorally realistic
Clearly, there will be significant uncertainty in the
dynamic traffic assignment (DTA) models. DTA tech-
model inputs that must be built-in to ensure that the
niques offer a number of advantages including captur-
policy recommendations work well for a wide range of
ing the spatial and temporal evolution of traffic
potential future outcomes. A vast amount of research has
dynamics across the transportation network, superior
already been performed on stochastic transportation
capability to capture traffic congestion buildup and
modeling both on the demand and supply side [35–40].
dissipation, and explicitly representing the route-
choice effect of external dynamic prices and other
Power Systems Modeling
costs and incentives. A number of simulation-based
DTA modules have been developed in recent years The planning, design, and operation of modern power
[29–32]. The above mentioned features of DTA make systems call for extensive and detailed simulation.
it an ideal choice for modeling the network congestion Models used to simulate power system behavior
patterns induced by PHEVs/BEVs usage and their depend on the purpose and uses. When considering
impact on other vehicles. the need of studying PHEVs/BEVs impact on power
Travel models produce numerous outputs, metrics, system, different levels of modeling are required.
and system properties. Of critical importance for At the macro level, the power system planning
connecting the transportation and energy models are related to the uses of PHEVs/BEVs requires under-
predictions regarding time-dependent vehicle standing of the generation, storage, and load

characteristics, as well as power flow projections
impacted by the anticipated use of PHEVs/BEVs.
A stochastic nature of PHEV/BEV use in the multiple
possible roles will require advanced probabilistic
methods for power flow analysis, as well as stochastic
optimization related to operation and investment plan-
ning of dispersed generation [41, 42]. Enhanced
modeling techniques must be developed for PHEV/
BEV behavior as a load to assess dynamic stability of
the power system operating in G2V mode [43]. Hadley
[25] used the Oak Ridge Competitive Electricity Dis-
patch (ORCED) model to simulate PHEV/BEV elec-
tricity demand. It did not directly include transmission
and distribution impacts, but discussed the issues of
increased continuous transmission. Also, power system
contingency analysis must be improved to account for
the dynamic nature of both temporal and spatial prop-
erties of PHEVs/BEVs. In the V2G mode, PHEVs/BEVs
may impact power grid operation in many different
roles, both as energy storage used to improve perfor-
mance of renewable energies such as wind and solar
[44], as well as a market participant through aggregated
distributed generation [10, 11, 45]. While it has been
recognized that PHEVs/BEVs can be used for regula-
tion services [10, 11], some studies also suggested the
PHEVs/BEVs use for peak power “shaving” services
[46]. A customized modeling tool that allows examin-
ing the potential impacts of large scale deployment of
PHEVs/BEVs on a given electricity system, such as the
“PHEV-load” tool developed by the National Renew-
able Energy Laboratory (NREL) may be needed [47].
At the micro level, the PHEV/BEV powertrain sys-
tem itself, which is a very complex dynamic electrome-
chanical system, may be studied. Specialized modeling
and simulation tools, such as Argon National
Laboratory’s (ANL’s) Powertrain System Analysis
Toolkit (PSAT) are well suited for such an analysis
[48]. This toolkit allows detailed modeling of charging
and discharging dynamics of PHEVs/BEVs, which is
crucial when defining properties of PHEVs/BEVs as
loads, energy storage, or generation, as discussed
above, Other ANL’s tools such as GCtool, GREET,
and AirCred may also be needed to asses other
impacts [48].
The impact of PHEVs/BEVs ranges from the macro
to micro scales, both in size and time. Different power
system states (steady state, dynamic, and transient)
PHEVs and BEVs in Coupled Power and Transportation Systems 859
may need to be represented in a framework using
different types of mathematical formulations (wave-
forms, phasor, and algebraic). This leads to a new
requirement for developing a method for linking dif-
ferent modeling techniques for accurate and efficient
simulation when representing large scale penetration
of PHEVs/BEVs as generators, storage elements, or
loads [49].
Human Behavior Modeling
The widespread adoption of PHEVs/BEVs will place
human vehicle operators at the intersection of power
and transportation systems. Thus, it is critical to
understand human decision making in the context of
PHEV/BEV usage and how behavior can be shaped by
incentive structures and training interventions. The
large disparate group of decision makers includes not
only drivers, but also utilities, battery exchange loca-
tion coordinators, and fleet managers. Cognitive
research will be critical to not only to understand and
optimize human decision making involving PHEVs/
BEVs, but to also increase the rate of PHEV/BEV
adoption.
Route planning for any type of vehicle is an example
of a dynamic decision task [50]. Choosing a route
requires a series of interrelated decisions that occur in
a changing and uncertain environment. PHEVs/BEVs
introduce a number of additional decision elements,
such as whether to draw energy from the grid or deliver
energy to the grid at destinations with facilities allowing
such interfaces. Complicating this decision process, G2V
costs and V2G credits vary through time and are not
perfectly predictable from the driver’s perspective.
One successful framework for modeling human
performance in dynamic decision making tasks is
reinforcement learning (RL) [51, 52]. The theory of
RL comprises an array of techniques for learning
temporally extended tasks in dynamic environments.
An agent is assumed to be immersed in its environ-
ment, with some number of actions available to be
taken at any given time. The chosen action has an effect,
depending on the current state of the environment, the
immediate reward (or punishment) the agent receives,
and the future state of the environment. Thus actions
can influence situations and rewards arbitrarily far
into the future, and successful performance hinges on
860 PHEVs and BEVs in Coupled Power and Transportation Systems
effective planning and coordination of extended
sequences of actions [53].
Previous research demonstrates that RL agents and
humans are more likely to discover the underlying
structure of a task when state cues are present that
allow for generalization [54]. A state cue in the context
of PHEV/BEV decision making would include observ-
able properties – such as time of day, weather condi-
tions, and congestion – that enable prediction of G2V
credits and V2G costs. State cues play a critical role in
shaping learning and it has been shown that variability
in state predictors disrupts performance more than
equivalent variability in the reward structure [55].
Further research is needed to examine how variability
in state cues and reward structure affects PHEV/BEV
route selection. Establishing how PHEV/BEV driver
performance (with respect to improving conditions
on the grid and transportation system) declines with
variability in state cues is important because transient
changes in incentives could have negative, unintended
consequences, making it difficult for people to acquire
the basic pricing contingencies. Research is also needed
to find the best methods for PHEV/BEV operators
or aggregators to learn about incentive structures. Var-
ious types of feedback are available (e.g., Reward Only,
First Error), and the optimal approach should be deter-
mined via experiment.
Determining Economic Feasibility
To take advantage of the proposed “transportation-
energy” markets, interface infrastructure – the facilities
that will bridge the two systems and serve as energy
transfer points – must be developed and planned.
While prevalence of PHEVs/BEVs in the future is
unknown, their ultimate value can only come if the
interfaces are in place. This leads to a situation where
the demand for PHEVs/BEVs depends on the infra-
structure supply, which in turn is defined by the
demand. The traditional project valuation models fall
short of accounting for this feedback loop.
Developing interface infrastructure is a uniquely
challenging problem because the equipment must not
only adjust energy flow over time, but the location of
transfer points must be determined to maximize long-
term value and minimize risks. Technology adoption,
incentives, and system interdependencies all play a role.
To maximize the value of developing interface
infrastructure in a particular location, two aspects of
the problem must be considered: (1) the value created
to the grid by using PHEVs/BEVs for regulation ser-
vices, and (2) the value of the activity-based travel
patterns that could include a visit to the interface
infrastructure. The former value can be explicitly deter-
mined, but unless the latter is considered and travelers
are enticed to use the new infrastructure, the value to
the grid will not be achieved. Typically, the traveling
public selects route and activity patterns without con-
sidering energy exchange opportunities. New method-
ologies and modeling techniques must be developed
for valuing interface infrastructure given its depen-
dence on traveler behavior.
Unlike most past research into making investment
decisions for infrastructure projects that focus on
a single system (e.g., [56, 57]), the problem posed
here must consider the interdependencies between sev-
eral systems as well as the rate of technology adoption
(availability of PHEVs/BEVs to use this facility and
generate value). In fact, this problem exhibits both
spatial network effects and strategic “bandwagon” net-
work externalities (see seminal contributions in this
area by Rohlfs [58], Farrell and Saloner [59], and
David and Greenstein [60]).
It is clear that in the face of this bandwagon effect, the
value of deferral flexibility is marginal. Hence, the project
developer action space should consider actions that
promote early adoption without fully committing to
irreversible capital expenditures. Stochastic modeling
approaches could be useful here to consider that the
outcome and uncertainty space of the valuation problem
is decision dependent (see, e.g., [61]).
Benefits
This section aims at demonstrating the potential ben-
efits of PHEVs/BEVs that may be used to feed power
back to home or office building, which is known as
“Vehicle-to-Building” (V2B) operation. The new
parking facility called “smart garage” is introduced
and its eclectic power capacity is discussed. Based on
the availability analysis of smart garages, a strategy to
adopting PHEV/BEV uses in the V2B mode under peak
load and outage condition is proposed. V2B approach
considers PHEVs/BEVs as a generation resource for the
 PHEVs and BEVs in Coupled Power and Transportation Systems 861

buildings at certain periods of time via bidirectional PHEVs and BEVs in Coupled Power and Transportation
power transfers, which could increase the flexibility of Systems. Table 1 DSM benefits to customer, utility, and
the electrical distribution system operation. It is society [64]
expected that V2B operation will improve the reliability
Societal
of the distribution system, provide extra economic Customer benefits benefits Utility benefits
benefits to the vehicle owners, and reduce the home
Satisfy electricity Reduce Lower cost of
or building electricity purchase cost based on the demands environmental service
demand side management and outage management impacts
programs with customer incentives. Reduce/stabilize Conserve Improved
costs resources operating
efficiency
Demand Side Management (DSM)
Improve value of Protect global Flexibility of
For electric utility, DSM is defined as “Utility- service environment operation
sponsored programs to influence the time of use and
Maintain/improve Maximize Reduced capital
amount of energy use by select customers,” which lifestyle customer needs
includes peak clipping, valley filling, load shifting, stra- welfare
tegic conservation, strategic load growth, and flexible
load shape [62]. However, for utility end-user (cus-
tomer), DSM is often understood to include two com- lower than when the batteries are discharged, the
ponents: energy efficiency (EE) and demand response battery storage may be used to offset high cost during
(DR). EE is designed to reduce electricity consumption the peak demand.
during all hours of the year; DR is designed to change
on-site demand for energy in intervals and associated Outage Management (OM)
timing of electric demand by transmitting changes in
Another important benefit of V2B is using the battery
prices, load control signals, or other incentives to end
energy storage in PHEVs/BEVs as an emergency
users to reflect existing production and delivery costs
backup power for the commercial facility/building,
[63]. By cooperative activities between the utility and
which increases the reliability of the power supply for
its customers to utilize DSM, it will provide the benefits
that load.
to the customer, utility, and society as a whole, which is
An outage is typically caused by several unplanned
summarized in Table 1 [64].
events, and a timely detection and mitigation of such
In the V2B option, the owners will plug in their
situations is a real concern for the utility. Outage man-
vehicles during the day at their final destination for
agement system helps the operators to locate an outage,
a given time frame. As an example, this may be either
repair the damage, and restore the service. Outage
at their workplace (central business district) or at the
management must be performed very quickly to reduce
place of their study (university). The destinations,
outage time. Recently completed project proposes an
either parking lots or parking garages next to the
optimal fault location scheme which will help the oper-
buildings, are assumed to be equipped with
ator to find the faulted section very quickly [65]. In this
a bidirectional charger and controller. The parking
entry, the restoration strategy under an outage will be
facility should allow either charge or discharge mode
mainly discussed.
for the car batteries when necessary. The idea is that
The following types of outages and studies about
the parking facility can offer an aggregation service
the impact of PHEVs/BEVs adoption are considered:
for charging the batteries when the building demand
is lower than its peak load and discharge the batteries (a) Outage beyond the distribution system: These may
to partially supply the building to reduce the peak be caused by generator failure, fault in transmis-
demand during a high demand. This mode will be sion line, or substation busbar. Usually spinning
considered as DSM by V2B. Considering the electric- reserves are kept for these circumstances. From the
ity rate when the vehicle batteries were charged is previous studies it is concluded that PHEVs/BEVs
862 PHEVs and BEVs in Coupled Power and Transportation Systems
can be a candidate solution for spinning reserves
(as the traditional fastest acting spinning reserve
generators are highly costly while PHEVs/BEVs
qualify for fast response with lesser cost).
One may consider using a real-time security
constrained optimal power flow under the contin-
gencies to calculate the amount of PHEV/BEV
battery capacity required for a certain location at
a specific instance.
(b) Outage in distribution system: These may be
caused by fault inside the distribution system
and can be mitigated by precise spatial adjustment
of PHEV/BEV battery generation that may be used to
feed electricity locally during and after outage.
To propose the restoration strategy where PHEV/
BEV batteries are used to mitigate an outage condition,
the information about events (where the fault is located
and how the impact will propagate) and the location of
the battery storage need to be correlated. Thus, a spatial
as well as temporal analysis should be performed.
The restoration strategy can be executed in the
following steps:
1. Detect a fault.
2. Estimate the location of the fault.
3. Analyze the amount of battery generation required
and the availability of PHEVs/BEVs that can pro-
vide an alternative generation support in the vicin-
ity of the faulted area until the faulted section is
repaired. This will also consider the generation
duration requirement (i.e., time to repair the
faulted section).
4. Schedule the aggregated PHEVs/BEVs generation
optimally. This is a multi-objective optimization
problem which can be formulated to minimize the
distance between location of the fault and available
PHEVs/BEVs battery generation locations as well as
minimize the operating cost under system opera-
tion and security constraints.
Garage Location and Charging/Discharging
Infrastructure
Commercial and public parking garages in a central
business district (CBD) provide thousands of parking
spaces for commuters and visitors. After penetrating
the conventional vehicle market, owners of PHEVs/
BEVs will be using these parking garages, which may
provide an aggregated service to act as an electric power
source or storage.
Figure 6 shows a simple transportation network
with smart garage building. As a smart garage is
constructed, PHEV/BEV drivers have two options: pro-
ceed to final destination directly or park at the smart
garage and walk to the destination along walking links.
Drivers in transportation network select parking garage
based on the location and financial incentives (less
parking fee), which can be modeled as traffic assign-
ment problem. Demand of smart garage (number of
parked PHEVs/BEVs) calculated from the traffic
assignment problem would vary by the location and
incentive of the smart garage.
Electric power capacity of smart garage is estimated
based on demand of smart garage. Demand of smart
garage building is not constant. Generally, the demand
of smart garage building during the day would be
higher than during the night, similar to the demand
structure for a conventional garage as shown in Fig. 7.
Due to the versatility, electric power capacity needs to
be defined in two parts: for periodic service and for
continuous service as in Fig. 7. The available electric
power estimated based on the demand of smart garage
can be used for determining the support service that
can be provided during outage management and
demand side management in vehicle-to-building
(V2B) mode.
Origin
Smart garage Destination
P
Driving link
Incentive Walking link
PHEVs and BEVs in Coupled Power and Transportation
Systems. Figure 6
Simple transportation network with smart garage
 PHEVs and BEVs in Coupled Power and Transportation Systems 863

Number of vehicles
Number of vehicles

For periodic service

For continuous service

4 8 13 18 23 3
Hour

PHEVs and BEVs in Coupled Power and Transportation Systems. Figure 7

Demand of smart garage for a day

Case Study ● Maximum capacity of each vehicle is 10 kWh (very

conservative for BEVs).
Test cases for two scenarios are studied: demand side
● The batteries in PHEVs/BEVs are drained by an
management using V2B mode during peak power
average of 4.0 kWh by the driving cycle used.
demand and outage management using V2B mode
during faults. When PHEVs/BEVs are on site, the building can
charge the batteries during the morning hours (lower
electricity price) and drain the batteries by an equal
Demand Side Management During Peak Power
amount during afternoon hours (higher electricity
Demand In this case, a large commercial building is
price). Thus the owner of PHEV/BEV will have the
analyzed to demonstrate the potential savings using
required energy in his/her battery to make sure the
demand side management based on V2B operation.
driving cycle to return home can be met. Figure 8
Itron, Inc. prepared a technical survey for the
shows the impacts of charging PHEVs by faster charg-
California Energy Commission (CEC), which modeled
ing methods (AC Level 3 or DC charging). It will
difference commercial sectors, including large office
elevate the peak demand to 1.86 MW of the office
building [66]. The load shapes include typical day,
building since the faster charging method will cause
hot day, cold day, and weekend for each of four seasons.
a large load in a short period (10–15 min), which is not
According to the definition used in this report, large
recommend for either utilities or customers.
office buildings are defined as premises with total floor
Figure 9 shows the change in the load shape for the
area equal or larger than 30,000 sq ft. The largest
typical summer day by using the AC Level 1 charging
electric end-uses in this building type are interior light-
method defined by the Society for Automotive Engi-
ing, cooling, office equipment, and ventilation [66].
neers (SAE) J1772 [67]. The load curve was changed by
The summer typical load shape for a large office
shifting the afternoon peak load to the morning off-
building is selected for our case study. The single build-
peak load when charging and discharging PHEVs/
ing demand is obtained from the results reported in the
BEVs. Considering the rate structures for peak and
literature [66]. The following assumptions are taken:
off-peak load in commercial buildings, peak load
● The studied building is 450,000 sq ft. shifting using V2B mode may provide the electricity
● There are up to 80 PHEVs/BEVs that arrive at 8 a.m. bill saving. Further study could be conducted to show
and are available for the entire day. the total saving expressed in dollars.
864 PHEVs and BEVs in Coupled Power and Transportation Systems

1800

1600

1400

1200

1000
kW
800

600

400

200 Without BEVs / PHEVs load

With BEVs / PHEVs load
0
0 5 10 15 20
Hour

PHEVs and BEVs in Coupled Power and Transportation Systems. Figure 8

Impacts of faster charging PHEVs/BEVs on load demand

1800

1600

1400

1200

1000
kW

800

600

400

200 Without BEVs / PHEVs shifting

With BEVs / PHEVs shifting
0
0 5 10 15 20
Hour

PHEVs and BEVs in Coupled Power and Transportation Systems. Figure 9

Peak load shifting with PHEVs/BEVs for a typical summer daily load

Outage Management During Faults The proposed
restoration scheme was tested on a small distribution
system (IEEE 37 node radial test feeder [68]).
This is an actual feeder located in California, which
consists of several unbalanced spot loads. The nominal
voltage is 4.8 kV.
Figure 10 shows the test feeder with smart garages at
some nodes.
The following assumptions are taken:
● Three nodes are specified as smart garages (nodes
718, 735, and 740).
● Maximum capacity of each vehicle is 10 kWh.
● Discharge vehicles with state of charge (soc >70%.
● PHEV/BEV tariff for charging is 5 c/kWh and for
discharging is (15–40) c/kWh (depending on
 PHEVs and BEVs in Coupled Power and Transportation Systems 865

799
724
Case 1
722
707
712
701

742 713 704

720
705 702

714
706

729 744 727

703 718
725

728

Case 2 730 Smart garage

732 708 709

731

736
733 775
Smart garage

710 734
740

735
737 738 711 741

Building
Smart garage

PHEVs and BEVs in Coupled Power and Transportation Systems. Fig. 10

Diagram of test feeder with smart garages

PHEVs and BEVs in Coupled Power and Transportation Systems. Table 2 Case study 1: results for PHEV/BEV generation
scheduling
Node 718 Node 735 Node 740
Ph-1 (kW) Ph-2 (kW) Ph-3 (kW) Ph-1 (kW) Ph-2 (kW) Ph-3 (kW) Ph-1 (kW) Ph-2 (kW) Ph-3 (kW)
0 0 411 300 300 300 427 339 380
Total cost for three phases is $733.2/h
866 PHEVs and BEVs in Coupled Power and Transportation Systems
PHEVs and BEVs in Coupled Power and Transportation
Systems. Table 3 Case study 2: results for PHEV/BEV
generation scheduling
Node 735 Node 740
Ph-1 Ph-2 Ph-3 Ph-1 Ph-2 Ph-3
(kW) (kW) (kW) (kW) (kW) (kW)
300 127 300 51 0 81
Total cost for three phases is $221.35/h
different garages). Discharging tariff for node 718 is
40 c/kWh, for node 735 is 30 c/kWh, and for node
740 is 25 c/kWh.
Under normal operating condition, node no. 799
acts as an infinite bus and all the loads are fed through
it. Two different outage cases are studied:
1. Case 1: Fault on or beyond node 799: PHEVs/BEVs
at nodes 718, 735, and 740 were scheduled to satisfy
all the loads on the feeder. Table 2 shows the case
results.
2. Case 2: Fault on line segment 703–730: Node 799
will supply all the loads beyond this line segment.
PHEVs/EBVs at nodes 735 and 740 will be sched-
uled to satisfy the island created by a fault on line
703–730. Table 3 shows the case results.
Conclusions and Future Research
The policy implications of widespread PHEV/BEV
deployment in the energy and transportation systems
are explored. Previous research has approached the
problem from selected angles, making many simplify-
ing assumptions. Some thoughts on how the problem
may be approached from a non-myopic perspective are
provided.
In summary, numerous policy shifts are needed to
realize the full potential of PHEVs/BEVs, and the coop-
eration of the transportation and energy sectors is vital.
If policies such as the ones outlined in this paper are
adopted, PHEVs/BEVs can provide many benefits to the
electric grid in terms of reliability and stability by acting
as mobile decentralized storage and allowing for vehicle-
to-grid and grid-to-vehicle services. PHEVs/BEVs will
also allow for enhanced penetration of renewable energy
resources such as wind and solar, which will also aid with
energy security by reducing dependence on foreign
sources of oil. Important benefits can be made to air
quality through transferring pollution from numerous
mobile sources to fewer point sources that are easier to
control and may participate in cap-and-trade markets.
In addition to the carbon market, new markets will be
created in power systems due to the potential for
PHEVs/BEVs (or aggregators of PHEVs/BEVs) to par-
ticipate, particularly with ancillary and regulation ser-
vices. Lastly, charging stations must be planned and
developed carefully to allow for flexibility in driver
options and optimal performance of the transportation
and electricity networks.
The proposed multi-layered modeling framework
considers the spatial and temporal nature of the system
interactions. PHEV/BEV time-dependent travel pat-
terns are outputs of a transportation model and inputs
to power systems model. The framework also includes
cognitive behavior modeling for the purposes of devel-
oping appropriate incentives to encourage drivers to
behave in a way that improves the efficiency of the
transportation and energy systems.
The potential benefits of using PHEVs/BEVs as
dynamically configurable dispersed energy storage
that can serve as load or generation in a power system
as needed is discussed. If serving in G2V as well as V2B
mode and if aggregated, PHEVs/BEVs may play
a major role in both the electricity and the transporta-
tion networks. Selecting garage location and charging/
discharging infrastructure needs special attention from
the transportation system demand point of view. For
demand side management in electricity networks, the
use of PHEVs/BEVs to create a peak load shifting
strategy can reduce the electricity purchase cost for
the customer and vehicle owner. For outage manage-
ment in electricity networks, the use of PHEVs/BEVs to
generate power during outage restoration stage is
envisioned by solving a multi-objective optimization
problem of merit-order scheduling of PHEVs/BEVs
under operating constraints.
In recent years, Smart Grid revolution has begun
with the sponsorship and involvement from govern-
ment, businesses, utilities, and other stakeholders,
especially with the development and integration of
renewable energy resources. Envisioning the longer-
term impact, if there is enough aggregated PHEV/
BEV vehicles, such as a fleet, they can serve as backup
generation and storage for renewable energy in smart

grid applications. Many other functions of the future
electricity network may be affected when PHEVs/BEVs
act as dynamically configurable energy storage, which
may have profound impact on the transportation
networks as well. Better understanding of the role of
PHEVs/BEVs in coupled power and transportation
systems will be beneficial to transform existing power
grid into the Smart Grid, a power system that is more
efficient, reliable, resilient, and responsive.
Acknowledgments
The author wishes to acknowledge numerous col-
leagues and graduate students who contributed to the
findings of this entry: Dr. Bradley Love, and Dr.
Jennifer Duthie from The University of Texas at Austin,
as well as Dr. Ivan Damnjanovic, and graduate students
Mr. Chengzong Pang, Ms. Papiya Dutta, and Mr. Seok
Kim from Texas A&M University. Funding for this
study came from the National Science Foundation
through I/UCRC grant for the Center for “PHEVs/
BEVs: Transportation and Electricity Convergence,”
and another NSF I/UCRC grant for the “Power Systems
Engineering Research Center.”
Bibliography
Primary Literature
1. Lemoine DM, Kammen DM, Farrell AE (2008) An innovation
and policy agenda for commercially competitive plug-in
hybrid electric vehicles. Environ Res Lett 3(1):014003
2. Schrank D, Lomax T (2007) The 2007 urban mobility report.
Texas Transportation Institute
3. U.S.-Canada Power System Outage Task Force (2004) Final
report on the August 14, 2003 Blackout in the United
States and Canada: causes and recommendations. http://
www.nerc.com
4. Burges K, Twele J (2005) Power systems operation with high
penetration of renewable energy – the German case. In: 2005
international conference on future power systems,
Amsterdam, pp 1–5
5. Kelso JAS (1995) Dynamic patterns: the self-organization of
brain and behavior. MIT Press, Cambridge, MA
6. Turvey MT, Moreno M (2006) Physical metaphors for the men-
tal lexicon. Ment Lexicon 1:7–33
7. Van Orden GC, Holden JG, Turvey MT (2003) Self-organization
of cognitive performance. J Expr Psychol Gen 132:331–350
8. Brooks AN (2002) Vehicle-to-grid demonstration project: grid
regulation ancillary service with a battery electric vehicle. AC
Propulsion, California
PHEVs and BEVs in Coupled Power and Transportation Systems 867
9. Duncan R, Osborne MJ (2005) Report on transportation con-
vergence, Austin energy
10. Kempton W, Tomic J, Letendre S, Brooks A, Lipman T (2001)
Vehicle-to-grid power: battery, hybrid, and fuel cell vehicles as
resources for distributed electric power in California. Report #
UCD-ITS-RR-01-03, Electric transportation program
11. Kempton W, Tomic J (2005) Vehicle to grid implementation:
from stabilizing the grid to supporting large-scale renewable
energy. Power Sources 144(1):280–294
12. Kintner-Meyer M, Schneider K, Pratt R (2007) Impacts assess-
ment of plug-in hybrid vehicles on electric utilities and
regional U.S. power grids, part 1: technical analysis
13. Solomon J, Vincent R (2003) Development and evaluation of
a plug-in HEV with vehicle-to-grid power flow. Final report,
A.C. Propulsion
14. Letendre S, Perez R, Herig C (2002) Battery-powered, electric-
drive vehicles providing buffer storage for PV capacity value.
In: Proceedings of the 2002 American solar energy society
annual conference, Boulder
15. Kempton W, Tomić J (2005) Vehicle-to-grid power fundamen-
tals: calculating capacity and net revenue. J Power Source
144(1):268–279
16. Kempton W, Tomić J (2005) Vehicle-to-grid implementation:
from stabilizing the grid to supporting large-scale renewable
energy. J Power Source 144(1):280–294
17. Hadley SW, Tsvetkova A (2008) Potential impacts of plug-
in hybrid electric vehicles on regional power generation. Oak
Ridge National Laboratory, Oak Ridge, TN, ORNL/TM-2007/
150. http://apps.ornl.gov/~pts/prod/pubs/ldoc7922_regional_
phev_analysis.pdf
18. Meliopoulos S, Meisel J, Cokkinides G, Overbye T (2009)
Power system level impacts of plug-in hybrid vehicles, PSerc
project T34 final report #09-12. http://www.pserc.wisc.edu/
documents/publications/reports/2009_reports/
19. Andersson SL, Elofsson AK, Galus MD, Goransson L,
Karlsson S, Johnsson F, Andersson G (2010) Plug-in hybrid
electric vehicles as regulating power providers: case
studies of Sweden and Germany. Energy Policy 38(6):
2751–2762
20. Guille C, Gross G (2009) A conceptual framework for the
vehicle-to-grid (V2G) implementation. Energy Policy
37(11):4379–4390
21. Lerner EJ (2003) What’s wrong with the electric grid? The
industrial Physicist, Oct/Nov 2003
22. Wardrop JG (1952) Some theoretical aspects of road traffic
research. Proc Inst Civil Eng 1(2):325–378
23. U.S. Department of Transportation, Federal Highway Admin-
istration, Office of Highway Policy Information (2007) Traffic
volume trends. http://www.fhwa.dot.gov
24. Toole-Holt L, Polzin SE, Pendyala RM (2005) Two minutes per
person per day each year: exploration of growth in travel time
expenditures. Transp Res Rec 1917:45–53
25. Hadley SW (2006) Impact of plug-in hybrid vehicles on the
electric grid. Oak Ridge National Laboratory, for the U.S.
department of energy, ORNL/TM-2006/554
868 PHEVs and BEVs in Coupled Power and Transportation Systems
26. Bhat CR, Guo JY, Srinivasan S, Sivakumar A (2004) A compre-
hensive econometric microsimulator for daily activity-travel
patterns. Transp Res Rec 1894:57–66
27. Vovsha P, Petersen E, Donnelly R (2004) Impact of intra-
household interactions on individual daily activity-travel pat-
terns. In: Proceedings of the 83rd annual meeting of the
transportation research board, Washington, DC
28. Bowman JL, Bradley MA (2005) Activity-based travel forecast-
ing model for SACOG, technical memos numbers 11–11.
http://jbowman.net
29. Waller ST, Ziliaskopoulos AK (1998) A visual interactive system
for transportation algorithms. In: Proceedings of the 78th
annual meeting of the transportation research board,
Washington, DC
30. Taylor NB (1990) CONTRAM 5: an enhanced traffic assignment
model. TRLL research report 249
31. Ben Akiva M, Bierlaire M, Koutsopoulos H, Mishalini R, DynaMIT
(1998) A simulation-based system for traffic prediction
and guidance generation. Presented at TRISTAN III, Delft,
The Netherlands
32. Mahmassani HS, Hu TY, Jayakrishnan R (1992) Dynamic traffic
assignment and simulation for advanced network informatics
(DYNASMART). In: Proceedings of the second international
capri seminar on urban traffic networks, Italy
33. Arnott R, Small K (1994) The economics of traffic congestion.
Am Scientists 20:123–127
34. Hearn DW, Ramana MV (1998) Solving congestion toll pricing
models. In: Marcotte P, Nguyen S (eds) Equilibrium and
advanced transportation modeling. Kluwer, Boston, pp 109–124
35. Morrison SA (1986) A survey of road pricing. Transp Res 20A
(2):87–97
36. Waller ST, Schofer JL, Ziliaskopoulos AK (2001) Evaluation with
traffic assignment under demand uncertainty. Transp Res Rec
1771:69–75
37. Karoonsoontawong A, Waller ST (2006) Dynamic continuous
network design problem: linear bi-level programming and
metaheuristic approaches. Transp Res Rec 1964:104–117
38. Ukkusuri S, Tom VM, Waller ST (2007) Robust network design
problem under demand uncertainty. Comput-Aided Civ
Infrastruct Eng 22:6–18
39. Waller ST, Ziliaskopoulos AK (2006) A chance-constrained
based stochastic dynamic traffic assignment model:
analysis, formulation and solution algorithms. Transp Res C:
Emerg Technol 14(6):418–427
40. Duthie J, Unnikrishnan A, Waller ST (2006) Network evaluation
with uncertain and correlated long-term demand. In: Proceed-
ings of 85th transportation research board meeting,
Washington, DC
41. Stefopoulos G, Meliopoulos APS, Cokkinides G (2005) Proba-
bilistic power flow with non-conforming electric loads. Int
J Electr Power Energy Syst 27(9–10):627–634
42. Gollmer R, Neise U, Schultz R (2007) Risk modeling via stochas-
tic dominance in power systems with dispersed generation.
Technical report, Department of mathematics, University of
Duisburg, Germany
43. Choi BK, Chiang HD, Li YH (2006) Measurement-based
dynamic load models: derivation, comparison, and validation.
IEEE Trans Power Syst 21:1689–1697
44. Short W, Denholm P (2006) A preliminary assessment of
plug-in hybrid electric vehicles on wind generation markets.
Technical report, NREL/TP-620-39729
45. Tomic J, Kempton W (2007) Using fleets of electric
drive vehicles for grid support. J Power Sources 168(2):
459–468
46. Kempton W, Kubo T (2000) Electric-drive vehicles for peak
power in Japan. Energy Policy 28:9–18
47. Denholm P, Short W (2006) An evaluation of utility system
impacts and benefits of optimally dispatched plug-in hybrid
electric vehicles. Technical report, NREL/TP-620-40293
48. Argon National Laboratory Tools. http://www.transportation.
anl.gov/software/PSAT/index.html
49. Kasztenny B, Kezunovic M (2000) A method for linking differ-
ent modeling techniques for accurate and efficient simulation.
IEEE Trans Power Syst 15(1):65–72
50. Busemeyer JR (2002) Dynamic decision making. In: Smelser NJ,
Baltes PB (eds) International encyclopedia of the social and
behavioral sciences, vol 6. Elsevier, Oxford, pp 3903–3908
51. Fu WT, Anderson JR (2006) From recurrent choice to skill
learning: a reinforcement-learning model. J Exp Psychol Gen
135:184–206
52. Gray WD, Sims CR, Fu W-T, Schoelles MJ (2006) The soft
constraints hypothesis: a rational analysis approach to
resource allocation for interactive behavior. Psychol Rev
113:461–482
53. Sutton RS, Barto AG (1998) Reinforcement learning: an intro-
duction. MIT Press, Cambridge, MA
54. Gureckis TM, Love BC (2009) Short term gains, long term pains:
reinforcement learning in dynamic environments. Cognition
113:293–313
55. Gureckis TM, Love BC (2009) Learning in noise: dynamic
decision-making in an uncertain environment. J Math Psychol
53:180–193
56. Zhao T, Fu CC (2006) Infrastructure development and expan-
sion under uncertainty: a risk-preference-based lattice
approach. ASCE J Constr Eng Manag 132(6):620–625
57. Zhang Z, Damnjanovic I (2006) Quantification of risk cost
associated with short-term warranty-based specifications for
pavements. Transp Res Rec 1946:3–11
58. Rohlfs JH (1974) A theory of interdependent demand for
a communication service. Bell J Econ Manag Sci 5(1):16–74
59. Farrell J, Saloner G (1985) Standardization, compatibility and
innovation. Rand J Econ 16:70–82
60. David PA, Greenstein S (1990) The economics of compatibility
standards: an introduction to recent research. Econ Innovat
New Tech 1:3–41
61. Jonsbraten TW, Wets RJB, Woodruff DL (1998) A class of sto-
chastic programs with decision dependent random elements.
Ann Oper Res 82:83–106
62. Gellings CW (1985) The concept of demand-side management
for electric utilities. Proc IEEE 73(10):1468–1470

63. NERC (2007) Data collection for demand-side management
for quantifying its influence on reliability: results and recom-
mendations. In: North American electric Reliability Corpora-
tion, Princeton. http://www.nerc.com/docs/pc/drdtf/NERC_
DSMTF_Report_040308.pdf
64. IIEC (2006) Demand side management best practices guide-
book or pacific island power utilities. International Institute
for Energy Conservation, Washington, DC. www.sidsnet.org/
docshare/other/20070110DSMBestpractices.pdf
65. Kezunovic M, Ward J et al (2009) Integration of asset and
outage management tasks for distribution systems. PSerc
project T36 final report #09-11. http://www.pserc.wisc.edu/
documents/publications/reports/2009_reports/
66. Itron, Inc. (2006) California commercial end-use survey:
consultant report. CEC-400-2006-005, California Energy Com-
mission. http://www.energy.ca.gov/2006publications/CEC-
400-2006-005/CEC-400-2006-005.PDF
67. SAE (2010) Recommended practice for electric vehicle and
plug in hybrid electric vehicle conductive charger coupler,
SAE standard J1772, Jan 2010
68. Radial Test Feeders – IEEE distribution system analysis sub-
committee. http://ewh.ieee.org/soc/pes/dsacom/testfeeders.
html
Books and Reviews
Dowds J, Hines P et al (2010) Plug-in hybrid electric vehicle
research project: phase two report. Burlington, VT, Transpor-
tation research center. UVM TRC report #:10–001. http://www.
uvm.edu/~trans. . .reports/UVM-TRC-10-001.pdf
Galus MD, Andersson G (2008) Demand management of
grid connected plug-in hybrid electric vehicles (PHEV). In:
Energy 2030 conference, ENERGY 2008. IEEE, 17–18 Nov
2008, pp 1–8
PHEVs and BEVs in Coupled Power and Transportation Systems 869
Galus MD, Andersson G (2009) Integration of plug-in hybrid elec-
tric vehicles into energy networks. PowerTech, 2009 IEEE
bucharest. June 28–July 2 2009, pp 1–8
Galus MD, Andersson G (2009) Power system considerations of
plug-in hybrid electric vehicles based on a multi energy carrier
model. In: Power & energy society general meeting, PES ‘09.
IEEE. 26–30 July 2009, pp 1–8
Kempton W, Udo V et al (2008) A test of vehicle-to-grid (V2G) for
energy storage and frequency regulation in the PJM system.
University of Delaware, Pepco Holdings, Inc, PJM interconnect,
and Green Mountain College. http://www.magicconsortium.
org/_Media/test-v2g-in-pjm-jan09.pdf
Markel T, Simpson A (2005) Energy storage systems considerations
for grid-charged hybrid electric vehicles. In: Vehicle power and
propulsion, 2005 IEEE conference. 7–9 Sept 2005. http://www.
nrel.gov/vehiclesandfuels/vsa/pdfs/38538.pdf
Morrow K, Karner D et al (2008) Plug-in hybrid electric vehicle
charging infrastructure review, Idaho National Laboratory.
Report #: INL/EXT-08-15058. http://avt.inel.gov/pdf/phev/
phevInfrastructureReport08.pdf
ORNL (2010) Plug-in hybrid value proposition study final report,
Oak Ridge National Laboratory (ORNL). Report #: ORNL/TM-
2010/46. http://www.sentech.org/phev/pdfs/PHEV%20Value
%20Proposition%20Study%20Final%20Report%20Draft.pdf
Sullivan JL, Salmeen IT et al (2009) PHEV marketplace penetration:
an agent based simulation. University of Michigan, Ann
Arbor, Transportation Research Institute. Report #: UMTRI-
2009-32. http://deepblue.lib.umich.edu/bitstream/2027.42/
63507/1/102307.pdf
Turton H, Moura F (2008) Vehicle-to-grid systems for sustainable
development: an integrated energy analysis. Technol Fore-
casting Soc Change 75(8):1091–1108
Wynne J (2009) Impact of plug-in hybrid electric vehicles on
California’s electricity grid. Masters of environmental manage-
ment degree, Duke University
870 Plug-in Hybrid Electric Vehicles

Plug-in Hybrid Electric Vehicles vehicle powered by gasoline and diesel engine alone
[1]. The HEV also has high vehicle performance and
MEHRDAD EHSANI more user acceptability than pure battery powered
Department of Electrical & Computer Engineering, electric vehicle (EV). However, all of the HEV still
Texas A&M University, College Station, TX, USA comes from burning fossil fuel, gasoline, or diesel. On
the other hand, the EV has curtain advantages over HEV,
Article Outline mostly zero emission, independence from petroleum,
and perhaps low operating cost. However, the major
Glossary
disadvantage of EV is the range limitation and long
Definition of the Subject
battery charging time. It should be noted that only
Introduction
fraction of battery energy is used in a conventional
Statistics of Daily Driving Distance
HEV, in which the variation of the battery state of charge
Motor and Battery Power
(SOC) is limited to a narrow band [1]. This fact implies
Energy Consumption in Typical Driving Cycles
that most of the battery energy just stays there unused.
Operation Strategies
Using most of the battery energy, drawn from the
Battery Design
utility grid, to displace part of petroleum fuel is the
Summary and Future Directions
major advantage of plug-in hybrid electric vehicles.
Bibliography
The plug-in hybrid electric vehicle (PHEV) is a form
Glossary of hybrid electric vehicle, which is specifically designed
and operated to fully use the energy in the battery
AER All-electric range for period of distance of pure EV operation. It has
CD Charge depletion the advantage of being both pure EV and an HEV.
CDR Charge-depletion range When the battery is in high charge state, after overnight
CS Charge sustaining charging, the vehicle may be operated in pure EV
DoD Depth of discharge mode: charge-depletion (CD) mode. In this mode, all
EV Electric vehicle the traction energy is supplied by the battery. When the
HEV Hybrid electric vehicle state of charge of the battery reaches a certain low level,
PHEV Plug-in hybrid electric vehicle such as 30%, depending on the characteristics of the
SOC State-of-charge battery, the engine is started and the vehicle goes to
battery charge-sustaining (CS) mode. In the CS mode,
Definition of the Subject the energy in the battery is maintained around this level
The plug-in hybrid electric vehicle (PHEV) is vehicle till the end of the trip, at which time the battery is
that has the capability of accepting part of its propul- charged to its full state through utility grid [2–6].
sion energy from the electric utility grid. However, like Thus, the PHEV shares the combined operation
the conventional hybrid electric vehicles, it also draws characteristics of pure EV and CS HEV. However, the
part of its traction energy from its fuel tank. This will advantages of PHEV over pure EV and HEV are (1) dis-
afford the plug-in hybrids dual fuel flexibility, electric placement of significant amount petroleum fuel by
charge, and fossil fuel. Depending on the design of the electric energy that comes from utility grid, (2) longer
vehicle drive train and its operation, the plug-in hybrid range than a pure EV, (3) no fuel consumption and
vehicle can behave as pure electric, pure engine, or emission during pure EV mode, (4) and lower overall
hybrid vehicle. fuel consumption and emission.

Introduction Statistics of Daily Driving Distance

It is now well recognized that the hybrid electric vehicle As mentioned above, using the energy stored in the
(HEV) is much more efficient and cleaner than the battery from the utility grid to displace part of

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
 Plug-in Hybrid Electric Vehicles 871

petroleum fuel is the major feature of the plug-in
hybrid electric vehicles. The fraction of the petroleum
fuel displaced by electricity depends mostly upon the
amount of electrical energy drawn from the utility grid.
That is, the energy capacity of the battery and the total
driving distance, which is usually the diurnal driving
distance, and electrical traction power usage profiles
which is related to the drive cycle features and vehicle
control strategies. For optimal PHEV design, especially
the battery size, understanding of diurnal driving dis-
tance in some detail is very helpful.
Figure 1 shows a histogram of diurnal driving dis-
tance distribution and their cumulative frequency in
1995, from National Personal Transportation Survey
(NPTS) data [2]. The cumulative frequency or utility
factor in Fig.1 represents the percentages of the total
driving time (days) during which the daily driving
distances are less than or equal to the said distance on
the horizontal axis. Figure 1 reveals the fact that about
half of daily driving distance is less than 40 miles
(64 km). If a vehicle is designed to have 40 miles
(64 km) of pure EV range, that vehicle will have half
of its total driving distance in pure EV mode. Even if
the daily traveling distance is beyond 40 miles, there is
still a significant amount of petroleum fuel that can be
displaced by electricity, due to the pure EV driving
taking a large portion of the daily traveling. Research
also shows that even if the pure EV range is less than 40
miles, such as 20 miles (32 km), the portion of petro-
leum fuel that can be displaced in normal daily driving
is still very significant [2].
Motor and Battery Power
A PHEV consists of a considerable amount of pure EV
operation in which the electric motor and the battery
are the sole power plant and the energy source for the
vehicle. The motor and battery power must be large
enough to stratify the peak power requirement of vehi-
cle. Otherwise, the vehicle cannot accomplish the
demanded driving cycle.
The traction power of a vehicle, measured on the
drive wheels, can be expressed as:


V 1 dV
Pt ¼ Mgfr þ ra CD Af V 2 þ Md þ Mgi ðkW Þ
1000 2 dt
ð1Þ
where M is the vehicle mass in kg, V is vehicle speed in
m/s, g is gravity acceleration, 9.81 m/s2, ra is the air
mass density, 1.205 kg/m3, CD is the aerodynamic drag
coefficient of the vehicle, Af is the front area of the
vehicle in m2, d is the rotational inertia factor, dV/dt
is the acceleration in m/s2, and i is the grade of road. In
standard driving cycles, road is flat with i = 0.

90
Percentage of trips
80
Cumulative frequency
70

60

50

40

30

20

10

0
0 16 32 48 64 80 96 112 128 144 160 176 192 208 224 240 km
0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 miles
Trip distance

Plug-in Hybrid Electric Vehicles. Figure 1

Diurnal driving distance distribution and cumulative factor
872 Plug-in Hybrid Electric Vehicles

100

Power on the drive Vehicle speed

80

(km/h)
60
40
20
0
30
20
wheel (kW)

10
0
−10
−20
0 1 2 3 4 5 6 7 miles
0 1.6 3.2 4.8 6.4 8.0 9.6 11.2 km
Driving distance

Plug-in Hybrid Electric Vehicles. Figure 2

Vehicle speed and traction power in FPT 75 urban driving cycle

Plug-in Hybrid Electric Vehicles. Table 1 Vehicle param- sure that the motor can produce this peak power at this
eters used in power computation vehicle speed. Similarly, the peaking power of the bat-
Vehicle mass (kg) 1,700 tery should include the losses in the electric motor,
Rolling resistance coefficient 0.01
power electronics, and the mechanical transmission.
Suppose the efficiencies of the motor and power elec-
Aerodynamic drag coefficient 0.3
tronics are 0.85 and 0.95, respectively. Then, the peak
2
Front area(m ) 2.2 power of the battery must be around 34.7 kW for this
Rotational inertia factor 1.05 example. Table 2 lists the motor power and the battery
power in FTP75 urban, highway, LA92 and US06 driv-
ing cycles.

Energy Consumption in Typical Driving Cycles

Figure 2 is a diagram showing vehicle speed and
traction power measured on the drive wheels versus the
travel distance in the FTP 75 urban driving cycle. The
vehicle design parameters used in the computation are
listed in Table 1.
Figure 2 indicates that the peaking traction power
on the drive wheels is around 25 kW. However, there
are power losses from the battery to the drive wheels. In
order to meet the drive cycle power requirement, the
motor output power should be designed to include the
power losses from the motor shaft to the drive wheels.
Suppose the efficiency from motor shaft to the drive
wheels is 90%, then the motor shaft power rating is
round 28 kW. It should be noted that this motor power
required is related to the vehicle speed at which this
peak power occurs. For example, the peaking power in
Fig. 2 occurs at the vehicle speed of 50 km/h
(31.25 mph). In the motor power design, one must be
The all-electric range (AER) of a PHEV is determined
by the energy capacity of the battery and the energy
consumption in drive cycle. The amount of energy
consumed in a typical drive cycle can be obtained by
integrating (1) over the driving time period, as shown
in Fig. 3, in which no regenerative braking is included.
Considering the energy losses in the power electronics,
motor, and transmission, the usable energy in the bat-
tery for 20 and 40 miles (32 and 64 km) of AER in
typical drive cycles is listed in Table 3.
In vehicle design, a proper reference drive cycle
should be selected. An aggressive drive cycle, such as
US06, needs a large motor drive and battery which also
leads to good vehicle acceleration and gradeability per-
formance. On the other hand, a mild drive cycle, such
as FTP75, needs a small motor drive and battery, but
also leads to a sluggish vehicle performance.
 Plug-in Hybrid Electric Vehicles 873

Plug-in Hybrid Electric Vehicles. Table 2 Powers of motor and energy storage in typical driving cycles
FTP 75 urban FTP 75 highway LA92 US06
Motorpower@vehicle speed 28 @50 km/h 32 @72 km/h 55@57 km/h 98 @117 km/h
(31 mph) (57 km/h) (36 mph) (73 mph)
Energy storage power 35.7 39 68.5 121

20
Energy consumption on the drive wheel (k Wh)

18
LA92 US06
16

14

12 FTP75
urban
10

8
FTP75
6 highway

0
0 10 20 30 40 50 60 miles
0 16 32 48 64 80 96 km

Plug-in Hybrid Electric Vehicles. Figure 3

Energy consumption on the drive wheels versus driving distance in typical driving cycles

Plug-in Hybrid Electric Vehicles. Table 3 Energy con- AER Mode Operation Strategy
sumption in typical driving cycles
The principle of this operation strategy is to use the
FTP 75 FTP 75 energy of the battery exclusively. One possibility is to
Urban highway LA92 US 06 allow the driver to manually select between the CS HEV
20 miles (32 km) 5.2 5.14 7.29 8.4 mode and the pure EV mode. The availability of AER
40 miles (64 km) 10.4 10.28 14.58 16.8 with sufficient range allows the vehicle to be driven in
areas where emissions are restricted. This strategy pro-
vides flexibility for the driver to choose the time when
the pure EV mode is used. For example, in a trip that
includes a distance where pure EV operation is required,
the driver can select the pure EV mode just prior
Operation Strategies
to entering this area in order to have sufficient range.
A PHEV may operate in an all-electric range (AER) In other places, the vehicle may be operated in pure EV
mode or a blended EV/HEV operation strategy. mode or CS HEV mode, depending on the charge status
874 Plug-in Hybrid Electric Vehicles

in the battery and the power demand. In normal condi-
tions where the trip does not have an imperative pure EV
operation, the driver may select pure EV mode at start of
the drip in order to fully use the energy of the battery to
save petroleum fuel, until the charge in the battery
reaches its design specified level at which the CS HEV
mode will start automatically.
The energy drawn from the battery in the pure EV
mode can be expressed as
Z T
Ed ¼ Pd dt; ð2Þ
0 power (full open throttle), the engine is operated
where Pd is the discharging power of the battery, T is
the total driving time, Pd is the traction power. During
braking, regenerative braking may be used and part of
the braking energy is recovered and restored to the
battery. The stored energy can be expressed as:
Z T
Ec ¼ Pc dt; ð3Þ
0 Pmot ¼ Ptra  Peng ;
where Pc is the charging power of energy storage during
braking. It should be noted that regenerative braking
power is smaller than the total braking power of
the vehicle and only part of braking energy can be
recovered [7–9]. The SOC of the battery can be
expressed as
Ed  Ec
SOC ¼ SOC0  ; ð4Þ
Et
where SOC0 is the initial value of the SOC that may be
equal to 1, and Et is the total energy of the energy
storage with SOC = 1. When the SOC of the battery
drops to a specified valued (0.3 for example, depending
on the operation characteristics of the battery), the CS
mode starts.
In the CS mode, the SOC of the battery should be
maintained around the specified value. Here, an engine
control constrained to on/off control strategy is used,
as described below.
1. When the acceleration pedal traction power
command is greater than the engine maximum
with full open throttle and the electric motor
supplies the rest of the power to the drive
train. That is,
Peng ¼ Peng max ; ð5Þ
and
ð6Þ
where Peng is the engine power, Peng-max is the engine
maximum power with full open throttle, Pmot is the
motor traction power, and Ptra is the commanded
traction power of the driver through the accelera-
tion pedal.
2. When the commanded traction power is smaller
than the engine maximum power with full open
throttle, the engine and motor operation depends
on the SOC of the battery, as described below and
depicted in Fig. 4.

PEH
Engine power

a
A PEM
b
SOCH
B
PEL
c
SOCM
C
SOCL

a Engine rpm b SOC

Plug-in Hybrid Electric Vehicles. Figure 4

Constraint engine on/off control strategy, (a) engine operating regions, (b) energy storage

(a) When the commanded traction power,
represented by point A in Fig.4a, falls in the
engine power range of high to medium thresh-
old (labeled by PEH and PEM in Fig. 4a), and the
battery SOC is lower than SOCL, the engine is
operated at point a (maximum engine power)
so that it has excess power to charge the energy
storage. Otherwise if SOC is greater than SOCL,
the engine is operated to produce power that is
equal to the commanded power (point A).
(b) When the commanded traction power,
represented by point B in Fig.4a, falls in the
engine power range of medium to low
(between PEM and PEL) as shown in Fig. 4a,
and if battery SOC is lower than SOCL, the
engine is operated at point a with full open
throttle so as to quickly raise the battery SOC
with a large charging power (PEaPB, where
PEa is the engine power at point a and PB is
traction power at point B). Otherwise if the
battery SOC is higher than SOCL but lower
than SOCM, the engine is operated at point b
and the relatively small engine power is used to
charge the battery (PEbPB). However, if the
battery SOC is higher than SOCM, engine alone
traction is used, e.g., the engine is controlled to
produce power just equal to the commanded
traction power and the battery is neither
charged nor discharged.
(c) When the commanded traction power,
represented by point C in Fig.4a, falls in the
low power range (below PEL as shown in
Fig. 4a) and if the battery SOC is below SOCL,
the engine is operated at point b to produce
large battery charging power (PEbPC).
Whereas if the battery SOC is in a range larger
than SOCL, but lower than SOCM, the engine is
operated at point c, and the charging power of
energy storage is PEcPC. However, if the battery
SOC is higher than SOCM, the engine is shut
down to avoid low engine operating efficiency,
and the vehicle is operated with pure EV mode.
3. When braking is applied, the electric motor is
always operated in regenerative braking [7–9].
The thresholds of the engine power (PEH, PEM, and
PEL as shown in Fig.4a) are related to the engine fuel
Plug-in Hybrid Electric Vehicles 875
consumption characteristics, and should be set such that
the battery SOC is maintained in a region lower than
SOCH and higher than SOCL. The thresholds of battery
SOC (SOCH, SOCM, and SOCL) should be set such that
it has sufficient power to support motor operation.
The control strategy discussed above is only one of
many possible options. Other control strategies may be
used.
A PHEV which has the parameters listed in Table 1
has been simulated by using simulation software devel-
oped by The Advanced Vehicle Systems Research Program
at Texas A&M University. In this simulation, a 400 W
electric load for the vehicle accessories is added. The total
energy in the fully charged battery is 10 kWh. The simu-
lation sequentially ran nine cycles of FTP75 urban drive
cycle, and the pure EV mode was started at the beginning
of the simulation until the SOC reached about 30%,
beyond which, charge-sustained mode was started.
The simulation results in FTP75 urban drive cycle
are shown in Figs. 5, 6, 7. The all-electric range is
around 42 km in which about 7 kWh of electric energy
is consumed. The engine operating points are placed
within the favorite region of the engine fuel consump-
tion map as shown in Fig. 6. The fuel and electric
energy consumption are shown in Fig. 7. It can be
seen that when the traveling distance is less than four
cycles (42 km or 26 miles), the vehicle operates in pure
EV mode and completely displaces the petroleum fuel
with electricity. The total electric energy consumed is
about 7.1 and 15.5 kWh per 100 km, or 4.05 miles/
kWh. With the increase in the total traveling distance,
the percentage of the fuel displacement decreases since
the charge-sustained operation takes a larger percent-
ages of the travel distance. For nine sequential drive
cycle (96 km or 60 miles), the fuel and electrical energy
consumptions are about 3.2 L/100 km or 74 mpg, and
7.42 kWh/100 km or 8.43 miles/kWh.
Blended Control Strategy
Unlike the AER control strategy, the blended control
strategy uses both engine and motor for traction with
charge-depleting (CD) mode until the battery SOC
reaches a specified low level, beyond which the vehicle
operates in the charge-sustained (CS) mode.
In the CD mode, both the engine and the motor
may operate at the same time. The range before
876 Plug-in Hybrid Electric Vehicles

energy storage Motor power Engine power

40

(kW)
20

0
50

(kW)
0

−50
10
Energy in the

power (k Wh)

50

0
0 20 40 60 80 100 km
0 12.5 25 37.5 50 62.5 miles
Traveling distance

Plug-in Hybrid Electric Vehicles. Figure 5

Profile of engine power, motor power, and energy in the energy storage along with traveling time in FTP75 urban
driving cycle

Power(kW)
7 13 19 25 31 37 43 49 55 61 67
Maximum torque
180 with full throttle
bsfc (g/kWh) and
160 fuel efficiency

140 250 (32.7%)

.5%)
260 (31
.3%)
270 (30
Torque (Nm)

120 .2%)
280 (29
100 )
6.4%
310 (2
)
80 .4%
(23
350
60 5%)
(20.
400
)
40 6.4%
500 (1
0 (1 3 .6 %)
60 .7%)
20 700 (11 .18%)
1000 (8

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Speed (rpm)

Plug-in Hybrid Electric Vehicles. Figure 6

Engine operating points on the fuel consumption map in FTP75 urban drive cycle

entering CS mode is longer than that with AER control one in which the engine and motor alternatively propel
strategy. New control strategies are needed to control the vehicle with no battery charging from the engine.
the engine and motor to meet the load demand. There The engine is constrained to operate in its optimal fuel
are many possible control strategies. The following is economy region. The details are described as follows.
 Plug-in Hybrid Electric Vehicles 877

20

k Wh/ 100km
15

10
Total k Wh

5
Liters /100km
Total fuel liters
0
0 1 2 3 4 5 6 7 8 9 Cycle number
0 10.6 21.2 31.9 42.5 53.1 63.8 74.4 85.0 95.6 km
0 6.64 13.3 19.9 26.6 33.2 39.9 54.7 53.1 59.8 miles
Cycle number and traveling distance

Plug-in Hybrid Electric Vehicles. Figure 7

Fuel and electric energy consumption in FTP75 urban drive cycle

Maximum torque
with full throttle
Iso-fuel
consumption
curves
En
gin
eo
EngineTorque

per Top torque

atin
g boundary
Fue

are
a
l co
nsu

Bottom
mp

torque
tion

boundary
incr
eas
e

Engine speed

Plug-in Hybrid Electric Vehicles. Figure 8

The operation area of the engine in the CD mode

Figure 8 schematically shows the engine operating
area. When the requested engine torque is larger than
the top torque boundary, the engine is controlled to
operate on this boundary and the remaining torque is
supplied by the electric motor. When the requested
engine torque falls in the area between the top and
bottom boundaries, the engine alone propels the vehi-
cle. When the requested engine torque is below the
bottom torque boundary, the engine is shut down
and the electric motor alone propels the vehicle. In
this way, the engine operation is constrained to within
its optimal fuel economy region. Since there is no
battery charging from the engine, the energy level in
the battery continuously drops toward its specified low
level. Then the vehicle goes into CS mode, in which the
constrained engine on/off control strategy, or some
other control strategy, is used as discussed above.
The example vehicle mentioned above has been
simulated with the control strategy discussed above in
nine sequences of FTP 75 urban driving cycle. The
results are presented in Figs. 9, 10, 11. Figure 9 shows
that the battery energy continuously drops to the
878 Plug-in Hybrid Electric Vehicles

energy storage Motor power Engine power

40

(kW)
20

0
50

(kW)
0

−50
10
Energy in the

power (kWh)

5.0

0
0 20 40 60 80 100 km
0 12.5 25 37.5 50 62.5 miles
Traveling distance

Plug-in Hybrid Electric Vehicles. Figure 9

Profile of engine power, motor power, and energy in the energy storage along with traveling time in FTP75 urban drive cycle

Power(kW)
7 13 19 25 31 37 43 49 55 61 67
Maximum torque
180 with full throttle
bsfc(g/kWh) and
160 fuel efficiency
250 (32.7%)
140 )
260 (31.5% )
270 (30.3%
Torque (Nm)

120 )
280 (29.2%
)
100 6.4%
310 (2

80 %)
(23.4
350
%)
60 (20.5
400
)
6.4%
40 500(1
.6 % )
600(13 7%)
20 700 (11.
1000 (8.18%)

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Speed (rpm)

Plug-in Hybrid Electric Vehicles. Figure 10

Engine operating points on the fuel consumption map in FTP75 urban drive cycle

specified value during CD mode, at which point the As indicated in Fig. 10, the engine operating points
engine joins the effort traction occasionally, operating overlap the favored region of the engine fuel con-
in its optimal region as shown in Fig. 8 but does not sumption map. Figure 11 indicates that the fuel
charge the battery. Figure 9 shows that the CD range is consumption in the CD range (0–53 km) is around
longer than the all-electric range as shown in Fig. 5. 1 L/100 km (235 mpg) and electric energy

15
k Wh/ 100 km
10
Total k Wh
Liters/100 km
0
0 1 2 3 4 5
0 10.6 21.2 31.9 42.5 53.1
0 6.64 13.3 19.9 26.6 33.2
Cycle number and traveling distance
Plug-in Hybrid Electric Vehicles. Figure 11
Fuel and electric energy consumption in FTP75 urban drive cycle
consumption is around 12.5 kWh/100 km (5 mile/
kWh). However, with the increase in the travel distance,
the fuel liters per 100 km increases and kWh per 100 km
deceases since the ratio of the charge-sustained opera-
tion distance to the total trip distance increases.
Battery Design
The amount of energy in the battery determines the all-
electric range (AER) or the charge-depletion range
(CDR). It is also closely related to fuel consumption,
fuel displacement, initial cost, and operating costs.
Through simulation, similar to those performed
above, the usable energy in the battery can be deter-
mined. The total energy capacity can be obtained from
Eusable
Ec ¼ ; ð7Þ
SOCtop  SOCbottom
where the Eusable is the usable energy in the battery
consumed in the AER or CDR modes, SOCtop is the
top SOC with fully charged battery, which usually
equals 1, and SOCbottom is the SOC of the battery at
which the operation mode is switched from AER or
CDR modes to CS mode. In the example above, the
usable energy is about 7 Wh (refer to Figs. 5 and 9).
Suppose that the SOC operating window is 0.7 (from 1
to 0.3). Then, the total energy capacity of the battery is
about 10 kWh.
Plug-in Hybrid Electric Vehicles 879
el liters
Total fu
6 7 8 9 Cycle number
63.8 74.4 85.0 95.6 km
39.9 54.7 53.1 59.8 miles
It should be noted that the depth of discharge (DoD)
of batteries is closely related to battery life. Figure 12 illus-
trates the cycle life for NiMH and Li-ion batteries [5]. If
one deep discharge per day is supposed, a total 4,000+
deep charges would be required for a 10–15 year life-
time. With the characteristics shown in Fig. 12, 70%
depth of discharge for NiMH and 50% for Li-ion batteries
may be the proper designs.
In the battery design, it should be ensured that the
energy storage can supply sufficient power to support
the vehicle when its SOC is at a low level (0.3 for NiMH
battery and 0.5 for Li-ion battery, for example, as
indicated in Fig. 12).
Energy/power ratio of a battery is a good measure
of its suitability. The size of the battery will be mini-
mized when its energy/power ratio equals that
required. The energy/power ratio is defined as:
Total Energy
Re=p ¼ ; ð8Þ
Power@operating SOC
In the example vehicle simulated above, the total
energy required is around 10 kWh for NiMH battery
(0.7DoD) and 14 kWh for Li-ion battery (0.5 DoD).
The power required is about 60 kW (refer to Table 2 for
LA92 drive cycle) which is defined at 30% of SOC for
NiMH battery and 50% of SOC for Li-ion battery. The
required energy/power ratio is 0.167 h for NiMH
880 Plug-in Hybrid Electric Vehicles

100
Typical PHEV cycles
during life of vehicle
90

80

70
Depth of discharge (%)
60
NiHM
50
Li-oin
40 Typical HEV
operating window
30

20

10

0
10000 20000 30000 40000
Number of cycles

Plug-in Hybrid Electric Vehicles. Figure 12

Cycle life characteristics of Varta battery technologies [5]

12 kWh of total energy (0.2  60). However, 0.25 h

NiMH, Cobasys
(slightly larger when 0.233 for safety) of SAFT Li-ion
1 Li-ion, SAFT
battery yields a total weight of 58 kg (60/1.03) which
Energy/power ratio (h)

carries 15 kWh of energy. Obviously, the SAFT Li-ion

0.8
battery is superior to the Cobasys NiMH battery. How-
0.6 ever, other factors have to be considered such as, cost,
safety, etc.
0.4
Summary and Future Directions
0.2
Plug-in hybrid electric vehicle (PHEV) can displace a
0
0.465 1.03 significant amount of petroleum fuel with electric
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 energy. This technology can significantly change the
Specific power (kW/kg)
makeup of transportation fuel supply. Design method-
Plug-in Hybrid Electric Vehicles. Figure 13 ologies presented in this entry can be used to design a
Typical energy/power ratio versus specific power drive train that has a specified all-electric range or
charge-depletion range. The control strategies devel-
oped here can be used in vehicle control to realize
battery at 30% of SOC and 0.233 h for Li-ion battery at all-electric operation, charge depletion, and charge-
50% of SOC. sustained operations. These control strategies can also
Figure 13 shows the energy/power ratio versus the operate the engine always within its low fuel consump-
specific power of NiMH and Li-ion batteries [5]. tion region, thus, yielding high overall efficiency. These
A 0.2 h of energy/power ratio (slightly larger than design methodology and control strategies have been
0.166 h for safety) for Cobasys NiMH battery yields validated by simulation of passenger car driving in FTP
a total weight of 129 kg (60/0.465), which carries 75, a typical urban drive cycle.

Plug-in hybrid vehicles are now being introduced
by manufacturers, such as the Chevrolet Volt by GM. It
is anticipated that PHEVs will occupy a larger segment
of the automotive market in the coming years. These
will take various forms and design philosophies to fit
the market demands, price points, and fuel and envi-
ronmental requirements.
Bibliography
1. Ehsani M, Gao Y, Gays SE, Emadi A (2005) Modern electric,
hybrid electric and fuel cell vehicle – Fundamentals, theory
and design. CRC Press
2. Simpson A (2006) Cost-benefit analysis of plug-in hybrid elec-
tric vehicle technology. Presented at the 22nd international
battery, hybrid and fuel cell electric vehicle symposium and
exhibition (EVS-22), Yokohama, 23–28 Oct 2006
3. Gonder J, Markel T (2007) Energy management strategies for
plug-in hybrid electric vehicles. SAE paper # 2007-01-0290,
Detroit
Plug-in Hybrid Electric Vehicles 881
4. Markel T, Simpson A (2005) Energy storage systems consider-
ations for grid-charged hybrid electric vehicles. Presented in
vehicle power and propulsion, 2005 IEEE conference, Chicago,
7–9 Sep 2005
5. Markel T, Simpson A (2006) Plug-in hybrid electric vehicle
energy storage system design. Presented at advanced auto-
motive battery conference, Baltimore, 17–19 May 2006
6. Markel T, Wipke K (2001) Modeling grid-connected hybrid
electric vehicles using ADVISOR. Presented in the, The 16th
annual battery conference on application and advances, Long
Beach, California
7. Gao Y, Ehsani M (2001) Electronic braking system of EV and
HEV – Integration of regenerative braking, automatic braking
force control and ABS. SAE J Passenger Cars Electron Electr
Syst 110. Paper No. 2001-01-2478
8. Gao Y, Chen L, Ehsani M (1999) Investigation of the effective-
ness of regenerative braking for EV and HEV. SAE J Passenger
Cars 108. Paper No. 1999-01-2901
9. Gao Y, Chu L, Ehsani M (2007) Design and control principle of
hybrid braking system for EV, HEV and FCV. In: IEEE on vehicle
power and propulsion conference, Arlington, 9–12 Sep 2007
882 Regenerative Braking
Regenerative Braking
YIMIN GAO
Department of Electrical and Computer Engineering,
Texas A&M University, College Station, TX, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Braking Energy
Hybrid Brake System
Future Directions
Bibliography
Glossary
Antilock braking system (ABS) Safety system that
prevents the wheels on a vehicle from locking up
while braking.
Brake-by-wire Technology to replace the traditional
mechanical and hydraulic control systems with
electronic control systems using electromechanical
actuators and human–machine interfaces.
Brake controller A microchip-based device which
receives brake strength command from driver and
feedback signals and generates control signals to
control the mechanical and electric brake systems.
Braking energy The energy consumed by the brake
system of a vehicle.
Hybrid brake system A vehicle brake system that uses
both regenerative braking and mechanical fictional
brake.
Regenerative braking Braking device of a vehicle
which can absorb vehicle braking energy and store
the absorbed braking energy into an energy storage
and then uses it for later traction.
Definition of the Subject
Regenerative brake system is a newly developed brake
system used in electric, hybrid electric, and fuel cell
vehicles which can convert part of braking energy into
electric energy using an electric motor/generator. For
braking safety, the traditional mechanical brake is still
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
required. The electric and mechanical brake systems
constitute a hybrid braking system. In the design and
control of such a system, the braking force distribution
on front and rear wheels has to meet the braking
regulation for braking safety, and meantime,
a braking force distribution on mechanical and electric
brake system design should be able to recover as much
braking energy as possible.
Introduction
One of the most important features of electric, hybrid,
and fuel cell vehicles is the ability of recapturing part of
braking energy, which can save significant amount of
fuel, especially for the vehicles driving mostly with
stop-and-go pattern.
Brake system of a vehicle is most important for
vehicle safety, which has to meet the demands of
quickly slowing down the vehicle and maintaining the
vehicle stability. In normal driving, braking force
requirements widely vary from slightly retarding vehi-
cle to emergency braking, depending on the traffic
situation. Heavy braking needs a huge braking force
that regenerative brake cannot produce, consequently,
a mechanical brake has to be engaged. This forms a
hybrid brake system. As in a hybrid propulsion system,
there are many options in configurations and control
strategies. An advanced hybrid braking system may be
a braking-by-wire system with intelligent control.
Braking Energy
Percentage of Braking Energy to Propelling Energy
Braking force has to be applied to a vehicle while slowing
down its speed or descending a grade. The kinetic
energy (1=2M V 2 ) or potential energy (M g h, where,
h is the vertical height of the vehicle descent) is dissi-
pated through vehicle brake system. For relative flat
roads, most of the braking energy is consumed for
slowing down the vehicle speed, especially, driving in
urban areas. Thus, discussion in this section focuses on
the urban stop-and-go driving pattern. The stop-and-
go driving pattern is interpreted by typical driving
cycles, such as those of EPA FTP 75 urban, LA 92, US
06, ECE-15, and New York city. Of course, other
driving cycles can also be used.
# 2012, DOI 10.1007/978-1-4419-0851-3
 Regenerative Braking 883

100

Vehicle speed (km/h)

80

60

40

20

0
0 200 400 600 800 1000 1200 1400
1.5
Propelling energy
Energy (kWh)

on wheels
1

0.5
Braking energy
0
0 200 400 600 800 1000 1200 1400
Running time (sec)

Regenerative Braking. Figure 1

Significant percentage of propelling energy consumed in braking [1]

While a vehicle driving on a flat road, the braking
force measured on drive wheels for developing
a deceleration j (m/s2) can be expressed as
Fb ¼ Mdj  Fr  Fw ;
where, M is the vehicle mass in kg, d is the equivalent
mass factor, which is a factor that equivalently converts
the rotating inertia into translational inertia, Fr is the
rolling resistance, and Fw is the aerodynamic drag of
the vehicle. Correspondingly, the braking power can be
expressed as
V
Pb ¼ ðMdj  Fr  Fw Þ; ðkW Þ
1000
where, V is vehicle speed in m/s. The energy consumed
Z can be obtained as
by braking
Eb ¼ Pb dt: ð3Þ
Figure 1 shows the propelling and braking energy
profiles of 1,500 kg passenger car while driving in EPA
FTP 75 urban driving cycle. It can be seen that, propelling
energy consumed on the wheels in a whole cycle is about
1.13 kWh, in which, about 0.63 kWh is consumed in
braking. The braking energy is about 55.4% of the total
propelling energy which comes from the power sources of
the vehicle. Table 1 lists the energy scenario of the
1,500 kg passenger car in the driving cycles of FTP 75
urban, LA92, US06, ECE-15, and New York city. The data
in Table 1 shows a significant percentage of the
energy consumed by braking. It also hints the importance
ð1Þ
of an effective regenerative brake system in energy saving.
Braking Energy Distribution over Vehicle Speed
Understanding the braking energy distribution over
vehicle speeds is helpful for an effective generative
brake system design. For example, in regenerative
brake system design, attempts may be made for maxi-
ð2Þ mizing regenerative braking capability in the speed
range in which most significant braking energy
appears. In other speed ranges with less braking energy,
other braking performance may be the major concern
without significant scarification of energy recovery.
In calculation of the braking energy distribution
over vehicle speeds, the whole speed range is divided
into a serial of speed interval, DVj (j = 1, 2, . . .,n). The
braking energy in the speed interval DVj can be
obtained by integrating the powers falling into the
speed interval DVj, with respect to time, as
Z
j
DEb ¼ Pbj dt; ð4Þ
DVj
884 Regenerative Braking

Regenerative Braking. Table 1 Braking energy scenario of 1,500 kg passenger car in FTP 75 urban drive cycle [1]
FTP75 urban LA92 US06 ECE-15 New York city
Maximum speed (km/h) 86.4 107.2 128.5 120 44.6
Average speed (km/h) 27.9 39.4 77.5 49.9 12.2
Traveling distance per cycle (km) 10.6 15.7 12.8 7.95 1.9
Propelling energy (kWh)
Per cycle 1.129 2.356 2.266 0.969 0.296
Per km 0.106 0.15 0.177 0.122 0.156
Braking energy (kWh)
Per cycle 0.625 1.337 0.923 0.330 0.243
Per km 0.059 0.087 0.072 0.042 0.127
Percentage of braking energy to propelling energy 55.4 58.0 40.7 34.1 81.9

16

14
Brake energy distribution on vehicle speed, %

12

10

0
0 10 20 30 40 50 60 70 80 90 100
Vehicle speed (km/h)

Regenerative Braking. Figure 2

Braking energy density over vehicle speeds in FTP75 urban driving cycle

j
where DEb and Pbj are the braking energy and braking speed range of less than 15 km/h and larger than 50 km/h.
power in the speed interval of DVj. Figure 2 shows It has been known that electric motor has a poor perfor-
a vehicle braking energy distribution diagram over vehicle mance for generation in low speed due to its low back
speeds in FTP 75 urban driving cycle. It can be seen that electromagnetic force (EMF). Thus, in regenerative brake
most of the braking energy are consumed in the speed system design, regenerative braking may be cancelled and
range of 15–50 km/h and very small amount falls into the only mechanical braking is applied at low speed, which
 Regenerative Braking 885

Regenerative Braking. Table 2 Braking energy scenario 12

of 1,500 kg passenger car in FTP 75 urban drive cycle [1]

10
FTP 75
urban LA92 US06 ECE-15 NY city

Braking energy, %
8
Braking energy 10.93 5.51 3.27 4.25 21.32
@ speeds 6
<15 km/h (%)
4

will not sacrifice much braking energy recovery. Table 2 0

shows the braking energy consumed in the speed range 0 5 10 15 20 25
Braking power (kW)
of less then 15 km/h in typical driving cycle.
Regenerative Braking. Figure 3
Braking Energy Distribution over Braking Power Braking energy distribution over braking power in FTP75
urban driving cycle
Braking energy distribution over braking power can sup-
ply useful information for the power capacity design of an
electric, especially, for mild hybrid vehicle, in which
a small electric motor is usually used and for battery Regenerative Braking. Table 3 Braking range in which
design, so that they can recover most of the braking energy. 85% of braking energy falls dissipated in typical urban
Similar to the braking energy distribution on driving cycles [1]
vehicle speeds, the braking power is divided into
FTP 75
a series of power intervals DPj and the braking energy
urban LA92 US06 ECE-15 NY city
in a power interval DPj is calculated by
Z Power range <14.4 <44.5 <46.5 <33.5 <18.5
j in which 85%
DEb ¼ Pbj dt; ð5Þ
of total energy
DPj falls (kW)
j
where DEb and Pbj are the braking energy and braking
power interval of DPj. Figure 3 shows the braking
power distribution over braking powers of a 1,500 kg
passenger car, in which 85% of braking energy falls into controller controls a brake regulator for producing
the power range of less than 15 kW. Table 3 gives the a correct thrust force for the brake actuator.
power ranges in which 85% of the braking is consumed The brake actuator pushes the master cylinder rod.
in the driving cycles of FTP 75 urban, LA92, US06 A brake power buffer may be used for supplying
ECE-15, and New York city. brake thrust force and a brake power source
powers the brake system. This system may be
pneumatic, which includes an air solenoid as the
Hybrid Brake System
brake actuator, air pressure regulator as the brake
A brake-by-wire system is conceptually illustrated in regulator, a compressed air storage as the brake power
Fig. 4. A brake pedal simulator is connected to a brake buffer, and an air compressor as the brake
pedal. The brake simulator is a mechanical or hydraulic power source.
device that produces a “braking feel” for the driver. A brake position sensor is also attached to the brake
Instead of generating a braking force, the brake pedal to sense its position. Based on receiving braking
simulator generates a braking command signal which command and feedback speed signals, the computing
represents driver’s desired braking force. The braking algorithm embedded in the brake controller generates
886 Regenerative Braking

Position
sensor
Braking
controller Brake actuator
Master Brake
cylinder Pedal pedal
simulator

Brake Brake Brake

regulator power power
buffer source

Mechanical
brake
Wheel speed signals

controller

Motor
controller

Brake Brake Speed

Control Control Control Control
clipper plate sensor
valve valve valve valve
Axle

Motor and
controller

Batteries

Regenerative Braking. Figure 4

A brake-by-wire hybrid braking system

control commands to the electric motor and the
pressure regulator. In the case of failure of the electric
system, the brake simulator can directly apply its force
to the master cylinder to generate a mechanical
braking force.
The mechanical brake controller operates very
similar to the controller of a conventional anti-lock
brake system (ABS), which operates the control values
to produce correct mechanical brake torque to each of
the wheels to preventing wheels from lockup.
The braking controller functions for correctly
allocating total braking force to electric regenerative
braking and mechanical braking. There are several
control strategies for this purpose.
A simple control strategy is shown in Fig. 5, where,
front wheels are braked by both regenerative and
mechanical brakes, and rear wheels only by mechanical
brake. This control strategy is explained as below.
With a slight braking strength, for example, less
than 0.15 g, only regenerative braking on front
 Regenerative Braking 887

Braking force on the rear wheel, kN

5
Mechanical
μ=1 brake
.0
0.9

j/g
4
0.8

=1
d
0.7

.0
0.9
0.6

0.8
3 I

0.7
0.6
0.5 c

0.5
0.4
0.4
2

μ = 1.0
0.3 0.3
0.2

0.2 b
1 Mechanical brake
0.1

0.1 +
Fbr

Electrical brake
a
0
0 2 4 6 8 10 12 14
Fbf-mech Fbf-regen Braking force on the front wheel, kN

Regenerative Braking. Figure 5

Braking forces varying with braking strength

0.8
d
0-a-b-c-d: Total braking force on front wheels
0-a-e-f-g: Regenerative braking forces on front wheels c
0.7 0-h-i-c-d: Mechanical barking force on front wheels
Braking force ratios to the vehicle weight

0-h-j-k-m: Mechanical braking forces on rear wheels

0.6

b
0.5

0.4
i

0.3

m
0.2 k
a e

0.1
j
h f g
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Braking deceleration rate, g

Regenerative Braking. Figure 6

Braking force ratios relative to the vehicle weight versus braking strength [1]

wheels is applied. This operation is simulating the driving time [1]. Since no mechanical braking is
engine retarding effect. Actually, in normal driving, applied to both the front and the rear wheels, the
the braking strength falls in this range in most of braking torque generated by the electric machine may
888 Regenerative Braking

be commanded by the position of the brake pedal. When strong braking is required, both mechanical
However, at low speed, the mechanical braking may and regenerative brakes are needed to be applied to
be needed due to the weak regenerative braking capa- front wheels and only mechanical braking is applied
bility. Furthermore, the electric motor should have to rear wheels. The braking force distribution on the
sufficient torque to fulfill this braking action. front and rear wheels follows the lines of a–b–c–d and
mechanical braking distribution follows the straight
line a–d. In the braking strength range of 0.15–0.6 g,
Regenerative Braking. Table 4 The percentage of the
the regenerative braking part is a constant as shown in
total braking energy available for recovering
Fig. 6. When the braking strength is greater than
FTP 75 New 0.6 g, the regenerative braking is gradually reduced
urban LA 92 US 06 York ECE-15 to zero at a high braking strength of 0.9 g. At high
Percentage of 89.69 82.92 86.55 76.16 95.75 braking strength, mechanical brake is more reliable.
the total braking Figure 6 shows an example of total braking force
energy available
scenarios. Table 4 shows the calculation results of the
for recovering
percentage of the braking energy that can be recovered

0.4

r- lines
Ratio of braking force on rear wheels to vehicle weight, Fbr /Mg

0.35
j= f- lines
j=

1.0
0.9
j=

g
j=

0.8

0.3
0.7

g
j=

βm
0.6
j=

g
0.5
j=

0.25
0.4

g
g

μ= I
0.9 μ=
μ=0 1.0
μ=0 .8
j=

0.2 .7
μ = 0.6
0.3

μ = 1.0
g

μ = 0.9

μ = 0.5
μ = 0.8
j=

μ = 0.7

0.15 μ = 0.4 e βr
0.2

μ = 0.5

μ = 0.6
g

μ = 0.3
μ = 0.4

βhb-max
0.1
μ = 0.2
μ = 0.3

b
a
j=

μ = 0.2
0.1

0.05 c
g

ECE R13
d

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
FRegen-max Ratio of braking force on front wheels
to vehicle weight, Fbf / Mg
FRegen-max

Regenerative Braking. Figure 7

Illustration of braking force distribution on the front wheels (electrical + mechanical) and rear wheels

by this control strategy in typical urban driving
cycles [1].
A control strategy that focuses on maximum regen-
erative braking may be applied. This braking control
strategy follows the rule of allocating the total braking
force to front wheels as much as possible within
the range stipulated by regulation, for example, ECE
R13 regulation. The ECE R13 regulation stipulates
a braking force distribution zone, which is in between
curves I and ECE R13 as shown in Fig. 7. For control
simplicity, a straight line, bhb-max may replace the
ECE R13 curve.
For certain braking strength commanded by the
driver (e.g., 0.5 g) on a road with adhesive coefficient
(e.g., 0.6), the braking force distribution can be at any
point in the segment of b–e as shown in Fig. 7.
Obviously, operating at point b, front wheels obtain
its maximum braking force. In this case, the available
regenerative braking force on the front wheels is the
segment of a-b, depending on the motor torque capac-
ity. The mechanical brake works at point a. If the road
coefficient increases to 0.7 and still with braking
strength 0.5 g, the braking force distribution can be at
any point between points c–e. Obviously, the maxi-
mum available braking force on the front wheels is at
point c, limited by the ECE R13 regulation.
The mechanical brake works at point d and available
regenerative braking force is the segment of c–d.
This control strategy can be summarized as a rule
such that, if the road adhesion and motor torque
permit, operate the total braking force distribution on
the line bhb-max. Obviously, line bhb-max stands for the
extreme case. In practice, a margin would be set,
for example, line br as shown in Fig. 7.
Regenerative Braking 889
Future Directions
A fully electronic controlled, brake-by-wire hybrid brak-
ing system is still under development. In this system, the
electric motor/generator and mechanical braking force
on individual wheels can be independently and coordi-
nately controlled. While driving on any road, the road
adhesive capacity can be fully used for quickly stopping
the vehicle, maintaining the vehicle stability, and mean-
while recovering braking energy as much as possible. The
braking system is coordinately controlled with steering
system for vehicle driving stability.
Bibliography
Primary Literature
1. Ehsani M, Gao Y, Emadi A (2010) Modern electric, hybrid
and fuel cell vehicles-fundamentals, theory and design. CRC,
Boca Raton
Books and Reviews
Cikanek SR, Bailey KE (1995) Energy recovery comparison between
series and parallel braking system for electric vehicle using
various drive cycles, DSC vol 56/DE, 86. Advanced Automotive
Technologies, American Society of Mechanical Engineers
(ASME), New York, pp 17–31
Gao H, Gao Y, Ehsani M (2001) Design issues of the switched
reluctance motor drive for propulsion and regenerative
braking in EV and HEV. Proceedings of the SAE 2001 future
transportation technology conference, Costa Mesa, CA,
paper No. 2001-01-2529
Gao Y, Chu L, Ehsani M (2007) Design and control principle of
hybrid braking system for EV, HEV and FCV, 2007 IEEE VPPC
Shu J, Zhang Y, Yin CL (2009) Longitudinal control of hybrid electric
buses using traction motor and pneumatic braking system.
WSEAS Trans Circuits Syst 8(11):873–882
890 Sustainable Transportation

Sustainable Transportation Introduction

MEHRDAD EHSANI The development of internal combustion engine (ICE)

Department of Electrical & Computer Engineering, vehicles, especially automobiles, is one of the greatest
Texas A&M University, College Station, TX, USA achievements of modern technology. Automobiles
have made great contributions to the growth of
modern society by satisfying many of its needs for
Article Outline
mobility in everyday life. The rapid development of
Glossary the automotive industry, unlike that of any other
Definition of the Subject industry, has prompted the progress of human society
Introduction from a primitive one to a highly developed industrial
Sustainable Transportation: Definition society. However, the large number of automobiles in
Sustainable Transportation: Solutions use around the world has caused and continues to
Sustainable Transportation: Types of Vehicles and cause serious problems for the environment and
Technologies human life. Air pollution, global warming, and the
Bibliography rapid depletion of the Earth’s petroleum resources
are now problems of paramount concern [1]. Around
Glossary the world, there were about 900 million cars and light
trucks on the road in 2009; they burn over 260 billion
Conventional vehicle A vehicle with an internal
gallons of gasoline and diesel fuel yearly [2, 3]. These
combustion engine alone.
numbers are increasing rapidly. According to the Wall
Electric vehicle A vehicle with an electric motor and
Street Journal, there could be one billion cars on the
a battery bank for propulsion.
road by 2020; Forrester Research puts the number at 1.2
Hybrid vehicle A vehicle having a combination of
billion but it could be higher [4]. According to the
internal combustion engine and electric motor for
Institute for the Analysis of Global Security, world oil
propulsion.
consumption will grow by 60% between now and 2020,
Hydrocarbons Organic fuels consisting mainly of
with India and China leading the charge (China is
hydrogen and carbon atoms from fossil or other
predicted to overtake the United States as the world’s
sources.
biggest car market in 20 years; India could follow
Sustainable Compatible with long-term life on
10 years later) [4].
earth.
Transportation Vehicle and fuel technology
Air Pollution
complex.
At present, all vehicles rely on the combustion of
hydrocarbon fuels to derive the energy necessary for
Definition of the Subject
their propulsion. Combustion is a reaction between the
Sustainable transportation, as used in this entry, refers fuel and the air that releases heat and combustion
to vehicle and fuel technologies whose availability and products. The heat is converted to mechanical power
environmental impacts are compatible with sustained life by an engine and the combustion products are released
and economic development on the planet. The sustain- to the atmosphere. A hydrocarbon is a chemical
able vehicle technologies focus on the fuel efficiency, compound with molecules made up of carbon and
operation, environmental impact, and availability of hydrogen atoms. Ideally, the combustion of
the raw materials for fabrication. The sustainable a hydrocarbon yields only carbon dioxide and water,
fuel technologies focus on long-term availability and which do not harm the environment. Indeed, green
environmental impact. These two foci are intimately plants “digest” carbon dioxide by photosynthesis.
coupled and make sustainable transportation highly Carbon dioxide is a necessary ingredient in vegetal
multidisciplinary. life. Animals do not suffer from breathing carbon

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3

dioxide unless its concentration in air is such that
oxygen is almost absent.
Actually, the combustion of hydrocarbon fuel in
combustion engines is never ideal. Besides carbon
dioxide and water, the combustion products contain
a certain amount of nitrogen oxides (NOx), carbon
monoxides (CO), and unburned hydrocarbons (HC),
all of which are toxic to human health.
Nitrogen Oxides Nitrogen oxides (NOx) result from
the reaction between nitrogen in the air and oxygen.
Theoretically, nitrogen is an inert gas. However, the
high temperatures and pressures in engines create
favorable conditions for the formation of nitrogen
oxides. Temperature is by far the most important
parameter in nitrogen oxide formation. The most
commonly found nitrogen oxide is the nitric oxide
(NO), although small amounts of nitric dioxide
(NO2) and traces of nitrous oxide (N2O) are present.
Once released in the atmosphere, NO reacts with
oxygen to form NO2. This is later decomposed by the
sun’s ultraviolet radiation back to NO and highly
reactive oxygen atoms that attack the membranes of
living cells. Nitrogen dioxide is partly responsible for
smog; its brownish color makes smog visible. It also
reacts with atmospheric water to form nitric acid
(HNO3), which dilutes in rain. This phenomenon is
referred to as “acid rain” and is responsible for the
destruction of forests in industrialized countries [1].
Acid rain also contributes to the degradation of
historical monuments made of marble [1].
Carbon Monoxide Carbon monoxide results from
the incomplete combustion of hydrocarbons due to
a lack of oxygen [1]. It is a poison to humans and
animals that breathe it. Once carbon monoxide
reaches the blood cells, it fixes to the hemoglobin in
place of oxygen, thus diminishing the quantity of
oxygen that reaches the organs and reducing the phys-
ical and mental abilities of the affected living beings
[1]. Dizziness is the first symptom of carbon mon-
oxide poisoning, which can rapidly lead to death.
Carbon monoxide binds more strongly to hemoglo-
bin than oxygen. The bonds are so strong that
normal body functions cannot break them. Persons
intoxicated by carbon monoxide must be treated in
Sustainable Transportation 891
pressurized chambers, where the pressure makes
carbon monoxide–hemoglobin bonds easier to
break.
Unburned Hydrocarbons Unburned hydrocarbons
are a result of the incomplete combustion of hydrocar-
bons [1, 2]. Depending on their nature, unburned
hydrocarbons may be harmful to living beings [2].
Some of these unburned hydrocarbons may be direct
poisons or carcinogenic chemicals such as particulates,
benzene, or others. Unburned hydrocarbons are also
responsible for smog: the sun’s ultraviolet radiations
interact with the unburned hydrocarbons and NO in
the atmosphere to form ozone and other products.
Ozone is a molecule formed of three oxygen atoms.
It is colorless but very dangerous, and poisons as it
attacks the membranes of living cells, thus causing
them to age prematurely or die. Toddlers, older people,
and asthmatic humans suffer greatly from exposure to
high ozone concentrations. Annually, deaths from high
ozone peaks in polluted cities are reported [3].
Other Pollutants Impurities in fuels result in the
emission of pollutants. The major impurity is sulfur
which is mostly found in diesel and jet fuel, but also in
gasoline and natural gas [1]. The combustion of sulfur
(or sulfur compounds such as hydrogen sulfide) with
oxygen releases sulfur oxides (SOx). Sulfur dioxide
(SO2) is the major product of this combustion. Upon
contact with air, it forms sulfur trioxide, which
later reacts with water to form sulfuric acid, a major
component of acid rains. It should be noted that sulfur
oxide emissions originate from transportation sources
but also largely from the combustion of coal in power
plants and steel factories. In addition, there is debate
over the exact contribution of natural sources such as
volcanoes.
Petroleum companies add chemical compounds to
their fuels in order to improve the performances or
lifetime of engines [1]. Tetraethyl lead, often referred
to simply as “lead,” was used to improve the knock
resistance of gasoline and therefore allow better
engine performance. However, the combustion of this
chemical releases lead metal, which is responsible for
a neurological disease called “saturnism.” Its use is now
forbidden in most developed countries and it has been
replaced by other chemicals [1].
892 Sustainable Transportation

Global Warming Residential

19%
Transportation
Global warming is a result of the “greenhouse effect” 32%
induced by the presence of carbon dioxide and other
gases, such as methane, in the atmosphere. These gases
trap the sun’s infrared radiation reflected by the
Commercial
ground, thus retaining the energy in the atmosphere 15%
and increasing the temperature. An increased Earth
temperature results in major ecological damages to its
ecosystems and in many natural disasters that affect
human populations [2].
Among the ecological damages induced by Industrial
global warming, the disappearance of some endan- 34%
gered species is a concern because it destabilizes the
Sustainable Transportation. Figure 1
natural resources that feed some populations. There
Carbon dioxide emission distribution from 1980 to 1999
are also concerns about the migration of some spe-
cies from warm seas to previously colder northern
seas, where they can potentially destroy indigenous
species and the economies that live off those species. 2000
Industrial
This may be happening in the Mediterranean Sea, 1800
CO2 emission in Million Metric Tons

where barracudas from the Red Sea have been

1600
observed.
Natural disasters command our attention more 1400
Transportation
than ecological disasters because of the amplitude of 1200
Residential
the damages they cause. Global warming is believed to
1000
have induced a meteorological phenomenon such as
“El Nino,” which disturbs the South Pacific region and 800
regularly causes tornadoes, inundations, and dryness. 600 Commercial
The melting of the polar icecaps, another major result
400
of global warming, raises the sea level and can cause
the permanent inundation of coastal regions, and 200
sometimes of entire countries. 0
Carbon dioxide is the result of the combustion of 1975 1980 1985 1990 1995 2000
hydrocarbons and coal. Transportation accounts for Year
a large share (32% from 1980 to 1999) of carbon
Sustainable Transportation. Figure 2
dioxide emissions. The distribution of carbon dioxide
Evolution of CO2 emission
emissions is shown in Fig. 1 [4].
Figure 2 shows the trend in carbon dioxide
emissions. The transportation sector is clearly now be largely responsible for the increase in global
the major contributor of carbon dioxide emissions. It Earth temperature observed during the last decades
should be noted that developing countries are rapidly (Fig. 3). It is important to note that carbon dioxide is
increasing their transportation sector, and these indeed digested by plants and sequestrated by the
countries represent a very large share of the world oceans in the form of carbonates. However, these
population. Further discussion is provided in the next natural assimilation processes are limited and cannot
subsection. assimilate all of the emitted carbon dioxide, resulting
The large amounts of carbon dioxide released in in an accumulation of carbon dioxide in the
the atmosphere by human activities are believed to atmosphere.
 Sustainable Transportation 893

Global temperature changes (1861–1996) EPA

0.6 0.33
0.4 0.22
0.2 0.11
0 0

ΔC°
−0.2 −0.11
ΔF°

−0.4 −0.22
−0.6 −0.33
−0.8 −0.44

−1 −0.56
61
71
81
91
01
11
21
31
41
51
61
71
81
91
18
18
18
18
19
19
19
19
19
19
19
19
19
19
Year
Source: IPCC (1995) updated

Sustainable Transportation. Figure 3

Global earth atmospheric temperature

Petroleum Resources
Sustainable Transportation. Table 1 Proved petroleum
The vast majority of fuels for transportation are liquid reserves in 2000
fuels originating from petroleum. Petroleum is a fossil Proved reserves in 2000 in R/P
fuel, resulting from the decomposition of living matters Region billion tons ratio
that were imprisoned millions of years ago (Ordovi- 8.5 13.8
North America
cian, 600–400 million years ago) in geologically stable
South and Central 13.6 39
layers. The process is roughly the following: the living
America
matters (mostly plants) die and are slowly covered by
sediments. Over time, these accumulating sediments Europe 2.5 7.7
form thick layers, and transform to rock. The living Africa 10 26.8
matters are trapped in a closed space, where they Middle East 92.5 83.6
encounter high pressures and temperatures, and slowly
Former USSR 9.0 22.7
transform into either hydrocarbons or coal, depending
Asia Pacific 6.0 15.9
on their nature. This process took millions of years to
accomplish. This is what makes the Earth’s resources in Total world 142.1 39.9
fossil fuels finite.
The proved reserves are “those quantities that
geological and engineering information indicates with
reasonable certainty can be recovered in the future
from known reservoirs under existing economic and The oil extracted nowadays is the easily extractable
operating conditions” [5]. Therefore, they do not oil that lies close to the surface, in regions where the
constitute an indicator of the Earth’s total reserves. climate does not pose major problems. It is believed
The proved reserves, as they are given in the British that far more oil lies underneath the crust of the Earth
Petroleum 2001 estimate [5], are given in billion tones in such regions like Siberia, or the American and
in Table 1. The R/P ratio is the number of years that the Canadian Arctic. In these regions, the climate and
proved reserves would last if the production were to ecological concerns are major obstacles to extracting
continue at its current level. This ratio is also given in or prospecting for oil. The estimation of the total
Table 1 for each region [5]. Earth’s reserves is a difficult task for political
894 Sustainable Transportation

Sustainable Transportation. Table 2 US Geological 25000

Oil consumption in thousand barrels per day OIL

Survey estimate of undiscovered oil in 2000
North America
Undiscovered oil in 2000 in 20000
Region billion tons
Asia Pacific
North America 19.8
South and Central 14.9 15000
Western Europe
American
Europe 3.0
10000
Sub-Saharan Africa and 9.7 Eastern Europe
and Former USSR
Antarctic
Middle East and North 31.2 5000
South & Central America
Africa Middle East
Former USSR 15.7 Africa
0
Asia Pacific 4.0

80

82

84

86

88

90

92

94

96

98
19

19

19

19

19

19

19

19

19

19
World (potential 98.3 (91.5)
growth)
Sustainable Transportation. Figure 4
Oil consumption per region

and technical reasons. A 2000 estimation of the

undiscovered oil resources by the US Geological Survey
is given in Table 2 [6].
Although the R/P ratio does not include future
discoveries, it is significant. Indeed, it is based on
proved reserves, which are easily accessible to this day.
The amount of future oil discoveries is hypothetical,
and the newly discovered oil will not be easily accessi-
ble. The R/P ratio is also based on the hypothesis
that the production will remain constant. It is obvious,
however, that consumption (and therefore the produc-
tion) is increasing yearly to keep up with the growth of
developed and developing economies. Consumption
is likely to increase in gigantic proportions with the
rapid development of some largely populated coun-
tries, particularly in the Asia Pacific region. The plot
of Fig. 4 shows the trend in oil consumption over
the last 20 years [7]. Oil consumption is given in
1,000 barrels per day (one barrel is about
8 metric tons).
Despite the drop in oil consumption for Eastern
Europe and the former USSR, the world trend is clearly
increasing, as shown in Fig. 5. The fastest growing
region is Asia Pacific, where most of the world’s
population lives. An explosion in oil consumption is
to be expected, with a proportional increase in
pollutant emissions and CO2 emissions.
Induced Costs
The problems associated with the frenetic combustion
of fossil fuels are many: pollution, global warming, and
foreseeable exhaustion of resources, among others.
Although difficult to estimate, the costs associated
with these problems are huge and indirect [8], and
may be financial, human, or both.
Costs induced by pollution include, but are not
limited to, health expenses, the cost of replanting
forests devastated by acid rain, and the cost of
cleaning and fixing monuments corroded by acid
rain. Health expenses probably represent the largest
share of these costs, especially in developed countries
with socialized medicine or health-insured
populations.
Costs associated with global warming are difficult
to assess. They may include the cost of damages caused
by hurricanes, lost crops due to dryness, damaged
properties due to floods, and international aid to
relieve the affected populations. The amount is poten-
tially huge.
Most of the petroleum-producing countries are not
the largest petroleum-consuming countries. Most of
the production is located in the Middle East, while
most of the consumption is located in Europe, North
America, and Asia Pacific. As a result, consumers have
 Sustainable Transportation 895

80000

Oil consumption in thousand barrels per day

70000

60000

50000

40000

30000

20000

10000

0
80

82

84

86

88

90

92

94

96

98
19

19

19

19

19

19

19

19

19

19
Year

Sustainable Transportation. Figure 5

World oil consumption

to import their oil and depend on the producing Importance of Different Transportation
countries. This issue is particularly sensitive in the Development Strategies to Future Oil Supply
Middle East, where political turmoil affected the oil
The number of years that oil resources of the Earth can
delivery to Western countries in 1973 and 1977. The
support our oil supply completely depends on the new
Gulf War, Iran-Iraq war, and the constant surveillance
discovery of oil reserves and cumulative oil production
of the area by the United States and allied forces
(as well as cumulative oil consumption). Historical
come at a cost that is both human and financial. The
data show that the new discovery of oil reserves grows
dependency of Western economies upon a fluctuating
slowly. On the other hand, the consumption shows
oil supply is potentially expensive. Indeed, a shortage in
high growth rate, as shown in Fig. 6. If oil discovery
oil supply causes a serious slowdown of the economy,
and consumption follow the current trends, the world
resulting in damaged perishable goods, lost business
oil resource will be used up by about 2038 [9, 10].
opportunities, and the eventual impossibility to run
It is becoming more and more difficult to discover
businesses.
new reserves of petroleum under the Earth. The cost of
In searching for a solution to the problems associ-
exploring new oil fields is becoming higher and higher.
ated with oil consumption, one has to take into
It is believed that the scenario of the oil supply will
account those induced costs. This is difficult because
not change much if the consumption rate cannot be
the cost is not necessarily asserted where it is gener-
significantly reduced.
ated. Many of the induced costs cannot be counted in
As shown in Fig. 7, the transportation sector is the
asserting the benefits of an eventual solution. The
primary user of petroleum, consuming 49% of the oil
solution to these problems will have to be economi-
used in the world in 1997. The patterns of consumption
cally sustainable and commercially viable without
of industrialized and developing countries are quite
government subsidies in order to sustain itself in the
different, however. In the heat and power segments of
long run. Nevertheless, it remains clear that any solu-
the markets in industrialized countries, nonpetroleum
tion to these problems – even if it is only a partial
energy sources were able to compete with and
solution – will indeed result in cost savings, which will
substitute for oil throughout the 1980s; and by 1990,
benefit the payers.
896 Sustainable Transportation

2500
Discovered reserves
(Remaining reserves +
cumulative consumption )

Remaining reserves, total reserves and

culumative consumption from 1970, Gb
2000

1500 Remaining
reserves

1000

500
Cumulative
consumption
0
1970 1980 1990 2000 2010 2020 2030 2040 2050
Year
The year of oil
supply ends

Sustainable Transportation. Figure 6

World oil discovery, remaining reserves, and cumulative consumption

50
Transportation
45 Other
40
Million Barrels per day

35

30
25
20

15
10

5
0
90 97 05 010 015 020 90 97 05 010 015 020
19 19 20 2 2 2 19 19 20 2 2 2
Industrialized Developing

Sustainable Transportation. Figure 7

World oil consumption

oil consumption in other sectors was less than in the from 1997 to 2020, 10.7 million barrels per day are
transportation sector. attributed to the transportation sector, where few alter-
Most of the gains in worldwide oil use occur in the natives are economical until late in the forecast.
transportation sector. Of the total increase (11.4 million In the developing countries, the transportation
barrels per day) projected for the industrialized countries sector also shows the fastest projected growth in

petroleum consumption, promising to rise nearly to
the level of non-transportation energy use by 2020. In
the developing world, however, unlike in industrialized
countries, oil use for purposes other than transporta-
tion is projected to contribute 42% of the
total increase in petroleum consumption. The growth
in non-transportation petroleum consumption in
developing countries is caused in part by the substitu-
tion of petroleum products for noncommercial fuels
(such as wood burning for home heating and cooking).
Improving the fuel economy of vehicles has crucial
impact on the oil supply. So far, the most promising
technologies are hybrid-electric vehicles and fuel cell
vehicles. Hybrid vehicles, using current internal
combustion engines (ICEs) as their primary power
source and batteries/electric motor as the peaking
power source, have much higher operation efficiency
than those powered by ICE alone. The hardware and
software of this technology are almost ready for
industrial manufacturing. On the other hand, fuel cell
vehicles, which are potentially more efficient and
cleaner than hybrid-electric vehicles, are still in
the laboratory stage and it will take a long time to
overcome technical hurdles for commercialization.
Figure 8 shows the generalized annual fuel
consumptions of different development strategies of
next-generation vehicles. Curve a-b-c represents the
2.2
Generalized annual oil consumption
2 80
1.8
1.6
1.4
b 30
1.2
a d
1
0.8
0 10 20
Sustainable Transportation. Figure 8
Comparison of the annual fuel consumption between
different development strategies of the next-generation
vehicles
Sustainable Transportation 897
annual fuel consumption trend of current vehicles,
which is assumed to have a 1.3% annual growth rate.
This annual growth rate is assumed to be the annual
growth rate of the total vehicle number. Curve a-d-e
represents a development strategy in which conven-
tional vehicles gradually become hybrid vehicles during
the first 20 years, and after 20 years, all the vehicles will
be hybrid vehicles. In this strategy, it is assumed that
the hybrid vehicle is 25% more efficient than a current
conventional vehicle (25% less fuel consumption).
Curve a-b-f-g represents a strategy in which, in the
first 20 years, fuel cell vehicles are in a developing
stage while current conventional vehicles are still on
the market. In the second 20 years, the fuel cell vehicles
will gradually go to market, starting from point b and
becoming totally fuel cell powered at point f. In this
strategy, it is assumed that 50% less fuel will be
consumed by fuel cell vehicles than that by current
conventional vehicles. Curve a-d-f-g represents the
strategy that the vehicles become hybrid in the first
20 years, and fuel cell powered in the second 20 years.
Cumulative oil consumption is more meaningful
because it involves the annual consumption and the
time effect, and is directly associated with the reduction
of oil reserves as shown in Fig. 6. Figure 9 shows
the scenario of generalized cumulative oil consump-
tions of the development strategies mentioned above.
100
90
a-b-c
c
Cumulative oil consumption
70 a-b-f-g
60
e 50
40
a-d-e a-d-f-g
20
g
10
f 0
0 10 20 30 40 50 60
30 40 50 60 Years
Sustainable Transportation. Figure 9
Comparison of the cumulative fuel consumption between
different development strategies of the next-generation
vehicles
898 Sustainable Transportation
Although fuel cell vehicles are more efficient than
hybrid vehicles, the cumulative fuel consumption by
strategy a-b-f-g (a fuel cell vehicle in the second
20 years) is higher than strategy a-d-e (a hybrid vehicle
in the first 20 years) within 45 years, due to the time
effect. From Fig. 8, it is clear that strategy a-d-f-g (a
hybrid vehicle in the first 20 years and a fuel cell vehicle
in the second 20 years) is the best. Figures 6 and 9 reveal
another important fact: Fuel cell vehicles should not
rely on oil products because of the difficulty of future
oil supply 45 years later. Thus, the best development
strategy of next-generation transportation would be to
commercialize hybrid-electric vehicles immediately,
and at the same time, do the best to commercialize
nonpetroleum fuel cell vehicles as soon as possible.
Sustainable Transportation: Definition
In the opinion of some, urban transport systems based
around the automobile have proved unsustainable,
consuming excessive energy, affecting the health of
populations, and delivering a declining level of service
despite increasing investments. The sustainable
transport movement focuses on solutions to these
problems [5]. The term “sustainable transport” came
into use as a logical follow-on from sustainable devel-
opment, and is used to describe modes of transport,
and systems of transport planning, which are
consistent with wider concerns of sustainability.
Sustainable transport systems make a positive
contribution to the environmental, social, and
economic sustainability of the communities they
serve. There are many definitions of the sustainable
transport, and of the related terms sustainable
transportation and sustainable mobility. One such def-
inition, from the European Union Council of Ministers
of Transport, defines a sustainable transportation
system as one that:
● Allows the basic access and development needs of
individuals, companies, and society to be met safely
and in a manner consistent with human and
ecosystem health, and promotes equity within and
between successive generations
● Is affordable, operates fairly and efficiently, offers
a choice of transport mode, and supports
a competitive economy, as well as balanced regional
development
● Limits emissions and waste within the planet’s
ability to absorb them, uses renewable resources at
or below their rates of generation, and uses
nonrenewable resources at or below the rates of
development of renewable substitutes, while
minimizing the impact on the use of land and the
generation of noise [6]
Sustainable Transportation: Solutions
There is a need to develop and accelerate the use of
clean and efficient vehicle technologies, as well as
renewable fuels, which together could dramatically
reduce the demand for petroleum products [7]. This
can be done in the following ways:
● Development of hybrid-electric and plug-in
hybrid-electric vehicles can provide significant
improvement in fuel economy and petroleum
displacement.
● Researchers are looking to make batteries more
affordable and recyclable, as well as enhance battery
range, performance, and life.
● Alternative fuels (ethanol blends, biodiesel,
hydrogen, electricity, propane, and compressed
natural gas) can reduce oil consumption.
● Reducing vehicle weight directly improves vehicle
efficiency and fuel economy, and can potentially
reduce vehicle operating costs. The introduction
of cost-effective, high-strength materials can signif-
icantly reduce vehicle weight without compromis-
ing safety.
● Improved combustion technologies and optimized
fuel systems can improve near- and mid-term fuel
efficiency.
Sustainable Transportation: Types of Vehicles
and Technologies
There are three main types of vehicles: combustion,
hybrid, and electric vehicles. The combustion vehicle
is one in which the only source of power comes from
using an internal combustion engine (ICE). The hybrid
vehicle or hybrid-electric vehicle (HEV) is composed of
the combination of an electric motor and a combustion
engine for propulsion, as where the electric vehicle
(EV) only uses one or more electric motors for
propulsion.

Conventional Combustion Vehicles
A vast majority of current transportation vehicles use
ICEs for propulsion. The ICE is a power plant in which
the power for propulsion comes from the combustion
of a fuel with an oxidizer within a chamber [8]. The fuel
typically used for this process is petrol or gasoline
which is derived from fossil fuels. This liquid hydro-
carbon fuel has a high energy density (approximately
34.2 MJ/L), therefore giving the ICE a high power-to-
weight ratio. From these key advantages over other
types of vehicles also come the disadvantages of
increased air pollution from exhaust emissions, an
average efficiency of 18–20%, and the continuing use
of fuel and production of emissions during times of no
movement. ICEs are typically classified by the fuel
they use:
● Petroleum: Gasoline, diesel, liquefied petroleum gas
(LPG), compressed natural gas, jet fuel, etc.
● Coal: Methanol, gasoline, and diesel made from
carbon using the Fischer-Tropsch process
● Biofuels and vegoils: Biobutanol, biodiesel,
bioethanol, biomethanol, biogas, peanut oil, etc.
A specific type of ICE that is becoming popular in
the transportation market is the flex-fuel vehicle (FFV).
FFVs are an alternative fuel vehicle with an internal
combustion engine designed to operate on more than
one fuel, usually gasoline blended with either ethanol
or methanol fuel, and stored in the same tank [10].
Flex-fuel engines are capable of burning any
proportion of the resulting blend in the combustion
chamber as fuel injection and spark timing are adjusted
automatically according to the actual blend detected by
electronic sensors. Like conventional gasoline vehicles,
FFVs have a single fuel tank, fuel system, and engine.
The most common commercially available FFV in the
world market is the ethanol FFV, with around
18 million automobiles and light duty trucks on the
roads by 2009, and concentrated in four markets, Brazil
(9.3 million), United States (around 8 million), Canada
(600,000), and Europe (181,458 led by Sweden).
The FFV’s fuel economy on E85 (85% ethanol, 15%
gasoline) is somewhat less than when operating on
gasoline alone, but has lower greenhouse gas emissions
than that of the gasoline-operated ICE. Using the
GREET model (greenhouse gases, regulated emissions,
Sustainable Transportation 899
and energy use in transportation), Argonne National
Laboratory indicates in “Ethanol, The Complete
Energy Lifecycle Picture” that greenhouse gas emis-
sions are reduced approximately 15% when ethanol
produced from corn is used. Using E85 also reduces
CO2 emissions and provides significant reductions in
emissions of many harmful toxins, including benzene,
a known human carcinogen. However, E85 also
increases emissions of acetaldehyde – a toxic pollutant.
Additional testing is being completed to expand the
understanding of the emissions impacts of E85 [1].
The operating fuel economy of a vehicle depends on
a number of factors, which include: fuel consumption
characteristics of the engine, gear number and ratios,
vehicle resistance, vehicle speed, and operating condi-
tions. The fuel economy of an ICE is usually evaluated
by the amount of fuel per kWh energy output, which is
referred to as the specific fuel consumption (g/kWh).
The fuel consumption is quite different from one
operating point to another, shown in Fig. 10. Vehicle
fuel economy can be calculated by finding the load
power and the specific fuel consumption for the engine.
The engine power output is always equal to the
resistance power of the vehicle, that is:


V dV
Pe ¼ Ff þ Fw þ Fg þ Mv d

t dt
The engine speed, related to vehicle speed and gear
ratio can be expressed as:
30Vig i0
Ne ¼
prd
After the engine power and speed are determined,
the value of the specific fuel consumption, ge, can be
found from Fig. 10. The time rate of fuel consumption
can be calculated by:
Pe ge
Qfr ¼ ð1=hÞ;
1000gf
where ge is the specific fuel consumption of the engine
in g/kWh and gf is the mass density of the fuel in kg/L.
The total fuel consumption within a total distance, S, at
a constant cruising speed, V, is obtained by:
Pe ge S
QS ¼
1000gf V
900 Sustainable Transportation

Maximum engine power

100

Optimum
80 operation

5
5
28
26
line

0
32
Engine power (kW)

* 55 0
2 35 0
60 40
0
50
0
60
40
7 0
0
800

20 Engine-specific fuel
consumption (g/kWh)

0
1000 2000 3000 4000 5000
Engine speed (rpm)

Sustainable Transportation. Figure 10

Fuel economy characteristics of a typical gasoline engine

Highest gear can be improved with a higher gear transmission or

I/100 km mpg
30
2nd highest gear
60 a continuously variable transmission.
It should be noted that because of the complexity of
25 50 vehicle operation in the real world, fuel consumption at
Fuel consumption

20 40 constant speed cannot accurately represent fuel

I/100 km consumption for a vehicle under real driving
15 30
conditions. Various drive cycles have been developed
10 20 to simulate real driving conditions. The drive cycles are
mpg usually represented by the speed of the vehicle along
5 10
with the relative driving time. To calculate fuel
0 0 consumption in a drive cycle, the total fuel consump-
50 100 150
Vehicle speed (km/h)
tion can be obtained by the summation of fuel
consumption in each time interval, Dti :
Sustainable Transportation. Figure 11
Fuel economy characteristics of a typical vehicle at X Pei gei
Qtc ¼ Dti ;
constant speed i
1000gf

At high speeds, the fuel consumption increases
because the aerodynamic resistance power increases
with the speed cubed. The fuel economy of the vehicle
can be enhanced due to the reduced engine speed at
a given vehicle speed and increased gear ratio and can
be seen in Fig. 11.
Figure 12 indicates that the engine has a much
lower operating efficiency in low gear than in high
gear. For this reason, the fuel economy of a vehicle
where Pei is the average power of the engine in the ith time
interval in kW, gei is the average specific fuel consump-
tion of the engine in the ith time interval in g/kWh, and
Dti is the ith time interval in hours. Figures 13, 14, and
15 show the EPA FTP75 urban and highway drive
cycles, fuel economy for urban drive cycle, and fuel
economy for highway drive cycle, respectively. From
these one can see the typical operating points and fuel
consumption for this particular drive cycle.
 Sustainable Transportation 901

Maximum engine power

80

70 Optimum
economy
line

5
Engine power output (kW)

60 Highest

5
26

28
5 gear
25

0
32
0
50 35
0
40
40 0
50
0
60
30 Engine-
700
800 specific
20 fuel
consumption
10 2nd (g/kWh)
highest gear
0
1000 2000 3000 4000 5000
Engine speed (rpm)

Sustainable Transportation. Figure 12

Operating point of an ICE at constant speed with highest gear and second highest gear

100
Speed (km/h)

Urban driving

50

0
0 200 400 600 800 1000 1200 1400

100
Speed (km/h)

50
Highway driving

0
0 100 200 300 400 500 600 700 800
Driving time (sec)

Sustainable Transportation. Figure 13

EPA FTP75 urban and highway drive cycles

Conventional vehicles with ICEs provide good
performance and long driving range by utilizing the
high energy-density advantages of petroleum
fuels. However, conventional ICE vehicles bear the
disadvantages of poor fuel economy and environmen-
tal pollution. The main reasons for their poor fuel
economy are as follows:
● Engine fuel efficiency characteristics mismatched
with the real operation requirements
● The dissipation of vehicle kinetic energy during
braking especially while operating in urban areas
● Low efficiency of hydraulic torque converter in
the transmission of current automobiles in
stop-and-go driving patterns
902 Sustainable Transportation

Fuel economy:
12.8 I/100 km (18.5 mpg) Maximum engine power

80
Optimum
Engine power output (kW) operating line
Actual
60

5
operating

26

28
5 0
points 25 32 0
35 0
40
40 0
50
600 0
70 0
80
20
Specific fuel
consumption (g/kWh)
0
1000 2000 3000 4000 5000 6000

Sustainable Transportation. Figure 14

Fuel economy and engine operating points for EPA FTP75 urban and highway drive cycle overlapped on engine fuel
consumption characteristics map

Fuel economy:
10.5 I/100 km (22.5 mpg) Maximum engine power
80
Optimum
operation
Engine power output (kW)

Actual line
5

60
5
26

28

operating
0

5
32

points 25 0
35
0
40
40 0
50
600 0
70 0
80
20
Specific fuel
consumption (g/kWh)
0
1000 2000 3000 4000 5000 6000
Engine speed (rpm)

Sustainable Transportation. Figure 15

Fuel economy and engine operating points for EPA FTP75 highway drive cycle overlapped on engine fuel consumption
characteristics map

The effort to improve the fuel economy of vehicles
has been an ongoing process in the automobile indus-
try. This can be done in various ways: reducing vehicle
resistance, improving ICE operating efficiency, and
properly matching transmission and advanced drive
trains. The use of light materials and advanced
manufacturing technologies can reduce the weight of
vehicles and in turn reduce the rolling resistance and
internal resistance during acceleration. This will reduce
the demanded power from the ICE. The use of
advanced materials in tire production is another
important method of reducing the rolling resistance
of vehicles. By introducing a flow-shaped body style
with a smooth surface, the aerodynamic resistance can

be reduced causing an improvement in fuel economy at
higher speeds. Proper construction of the transmission,
good lubrication, and proper adjustment and tighten-
ing of moving parts in the transmission can help
improve the transmission efficiency. The parameters
of the transmission, especially the gear number and
gear ratios, have considerable influence on the operat-
ing fuel economy as described previously. In the design
of the transmission, the parameters should be
constructed such that the engine will operate close
to its fuel optimum region. Improving engine
operation efficiency has the potential to contribute to
the improvement of vehicle fuel economy. There are
many effective advanced techniques, such as accurate
air/fuel ratio control with computer-controlled fuel
injection, high thermal isolated materials for reducing
thermal loss, varying ignition-timing techniques, active
controlled valve and port, etc. Advanced drive trains
developed in recent years, such as new power plants,
various hybrid drive trains, etc., may greatly
improve the fuel economy of vehicles. Fuel cells have
higher efficiency and lower emissions than conven-
tional internal combustion engines. Hybridization of
a conventional combustion engine with an advanced
electric motor drive may greatly enhance the overall
efficiency of vehicles.
Hybrid-Electric Vehicles
A hybrid-electric vehicle, HEV, is one that has two or
more energy sources and energy converters. In this case
the HEV has an electrical energy source (batteries and
electric motor) and a mechanical energy source (ICE).
Advantages of HEVs
Use less fuel, get better mileage, cost
less to run
Provide a quieter, better-quality ride
Have more efficient engines
Emit fewer pollutants (because they use
less fuel and some run on alternative
fuels)
Emit fewer greenhouse gases that con-
tribute to global warming
Reduce our dependence on foreign oil
and enhance national security
Sustainable Transportation. Figure 16
Advantages/disadvantages of HEVs
Sustainable Transportation 903
Base power for the HEV may be provided by the ICE,
a Stirling engine, fuel cell, etc. The design of the base
power source, ICE or fuel cell, for the HEV will be
much smaller than that of a single power train design
because the base power source does not have to be
designed to provide the dynamic power. The dynamic
power is to be provided by the second energy source,
most commonly, electrochemical batteries and electric
motors. The dynamic power may also be provided by
an electric motor powered by ultracapacitors, flywheels
(mechanical batteries), and their combinations. Some
advantages and disadvantages of HEVs are shown in
Fig. 16. The currently available HEVs are the Toyota
Prius, Honda Insight, Ford Escape, and Chevrolet Volt.
Conventional vehicles with ICEs provide good
performance and long operating range by utilizing the
high energy-density advantages of petroleum fuels.
However, conventional ICE vehicles bear the disadvan-
tages of poor fuel economy and environmental
pollution. Battery-powered electric vehicles (EV), on
the other hand, possess some advantages over
conventional ICE vehicles, such as high energy
efficiency and zero environmental pollution. However,
the performance, especially the operation range per
battery charge, is far less competitive than ICE vehicles,
due to the lower energy content of the batteries vs. the
energy content of gasoline. HEVs have the advantages
of both ICE vehicles and EVs and overcome each of
their individual disadvantages.
The architecture of a hybrid vehicle is loosely
defined as the connection between the components
that define the energy flow routes and control ports.
There are three types of HEVs:
Disadvantages of HEVs
Cost more to buy
Have expensive batteries that might
wear out before vehicle does
Less safe (batteries may pose a danger
to people unfamiliar with them, like
mechanics or rescue workers at an
accident scene)
Create hazardous waste (used batteries
need to be recycled or disposed of in a
safe manner)
904 Sustainable Transportation
● Parallel: The gas tank supplies fuel for the ICE and
the batteries provide power to the motor. Both
power the transmission, which provides propulsion
for the vehicle.
● Series: The ICE does not directly propel the vehicle
but instead turns a generator, which sends power to
either the motor or the batteries.
● Series–parallel: It has the flexibility to operate in
either a series or parallel mode.
The series drive train is fully independent when
decoupled from the driven wheels, Fig. 17. Therefore,
it can be operated at any point on its torque – speed
characteristic map, and can potentially be operated
solely within its maximum efficiency region. Because
electric motors have near-ideal torque – speed charac-
teristics for traction – they do not need a multi-gear
transmission. Therefore, their construction is greatly
simplified and the cost is reduced. Furthermore,
instead of using one motor and a differential
gear, two motors may be used, each powering a single
wheel. This provides speed decoupling between the
two wheels like a differential but also acts as a limited
slip differential for traction control purposes. Simple
control strategies may be used as a result of the
Fuel tank
Engine
Gene- Recti-
rator fier
Engine oerating region
Power
Speed
Battery
Configuration of a series hybrid electric drive train
Sustainable Transportation. Figure 17
Series HEV configuration
mechanical decoupling provided by the electrical
transmission. A disadvantage of the Series HEV is
that the energy from the engine is converted twice
(mechanical to electrical in the generator and electrical
to mechanical in the traction motor). The inefficiencies
of the generator and traction motor add up and the
losses may be significant. Also, the generator adds
additional weight and cost to the vehicle. Since the
traction motor is the only power plant for propulsion
it must be designed to meet the maximum require-
ments for performance.
A parallel hybrid drive train is a drive train in which
the engine supplies its power mechanically to the
wheels like in a conventional ICE-powered vehicle,
Fig. 18. It is assisted by an electric motor that
is mechanically coupled to the transmission. The
mechanical coupling of parallel hybrids may be
a torque or speed mechanical coupling leaving room
for several types of configurations. The torque coupling
adds the torques of the engine and the electric motor
together or splits the engine torque into two parts:
propelling and battery charging. If a multi-gear
transmission is used, the performance and overall
efficiency of the drive train may be superior to other
designs, because two multi-gear transmissions provide
Torque
Speed
Tractive effort
Motor Traction Mech.
controller motor trans.
Vehicle Speed
DC
Traction
Battery Battery charger
charger
 Sustainable Transportation 905

Fuel tank

Final drive
and differential

Engine

Mechanical
coupling
Mechanical
transmission

Battery
Motor
controller

Battery
charger Traction

Battery charger

Configuration of a Parallel hybrid electric drive train

Sustainable Transportation. Figure 18

Parallel HEV configuration

more opportunities for both the engine and electric
traction system (electric machine and batteries)
to operate in their optimum region. This design also
provides great flexibility in the design of the engine and
electric motor characteristics. The multi-gear transmis-
sion also adds complexity to the drive train, which may
not be desired. The main advantage of the hybrid drive
train with speed coupling is that the speeds of the two
power plants are decoupled; therefore, the speed of
both power plants can be chosen freely. This advantage
is important to power plants such as the Stirling engine
and the gas turbine engine, in which their efficiencies
are sensitive to speed and less sensitive to torque. Also,
the electric machines and the eventual differential gear
system occupy considerable space and may reduce the
available passenger and luggage space.
Another subtype added to the hybrid market is
the plug-in hybrid-electric vehicle (PHEV), Fig. 19.
The PHEV is usually an ICE – electric (parallel or
series) hybrid with increased energy storage capacity
(usually Li-ion batteries). It may be connected to the
utility grid after driving to avoid charging using
the ICE. The sudden push for this architecture stems
from the fact that the advantages of the battery-
powered EV and the charge-sustained HEV are
combined. The biggest advantage of this is that one
can charge the batteries from the utility grid as opposed
to using fuel and the ICE. As of 2007, most plug-in
hybrid conversions of conventional hybrids utilize the
series–parallel architecture. PHEVs use no fossil fuel
during their all-electric range and produce lower
greenhouse gas emissions if their batteries are charged
from renewable electricity. In most cases these vehicles
will be charged by a fossil fuel–based generating plant,
causing the emissions to be shifted to the generating
plants, making them localized points of pollution.
PHEVs also eliminate the problem of “range anxiety”
associated with all-electric vehicles, because the
combustion engine works as a backup when the
batteries are depleted. Other benefits include improved
national energy security, fewer fill-ups at the filling
station, convenience of home recharging, and the
906 Sustainable Transportation
controller
Brake
controller
Chemicals
storage
Energy
controller
Motor
controller
Vehicle
controller
Engine
Engine
controller
Hybrid drive train
Trans.
Sustainable Transportation. Figure 19
PHEV configuration
opportunity to provide emergency backup power in
the home. Some of the issues that surround the
PHEV are design and trade-offs against battery life,
capacity, heat dissipation, weight, costs, and safety.
Another major hurdle to overcome is the introduction
of new infrastructure to accommodate these vehicles.
Even if this infrastructure is installed, the issue of
utilization may affect the electric grid if the vehicles
are charged during peak hours.
The major component of HEVs, PHEVs, and EVs is
the energy storage system. This technology represents
one of the critical barriers to the development and
marketing of cost-competitive HEVs and PHEVs [9].
The energy storage requirements for these two types of
vehicles are somewhat different. HEVs require energy
storage devices that can deliver high power pulses. For
HEV applications, the goal is to develop cells that
provide peak power of 1,000 W/kg or greater, have a
cycle life of at least 300,000 shallow cycles, and have
a calendar life of 15 years. PHEVs require devices that
can both store significant energy and deliver high
power pulses. PHEVs will require batteries that can
deliver significant energy (several kWh) for several
Hybrid
energy
inputs
Electrical
Hybrid
F energy
storage
Motor
Mechanical
coupling
Trans.
Hybrid
brake
thousand discharge cycles from an almost full charge
to a lower state of charge. It has been suggested that
a PHEV battery would operate in a charge-depleting
hybrid mode from about 90% of full charge to about
25% of full charge. Once the battery reaches this lower
state of charge, it will function in a manner similar to
the battery in an HEV and must be able to sustain
200,000–300,000 shallow, high power cycles with
a 15-year calendar life. The specific energy (Wh/kg) of
most current batteries is not sufficient to meet the
weight and volume goals for PHEVs.
Electric Vehicles
Electric vehicles (EVs) use an electric motor for trac-
tion, and chemical batteries, fuel cells, ultracapacitors,
and/or flywheels for their corresponding energy
sources. A typical configuration is shown in Fig. 20.
The drive train consists of three major subsystems:
electric motor, energy source, and auxiliary
components. EVs produce zero emissions during
propulsion, do not use any petroleum onboard, have
a high efficiency, and have a quiet and smooth
 Sustainable Transportation 907

Electric propulsion subsystem

Wheel
Brake
Electronic
Vehicle Electric Mechanical
power
controller motor transmission
converter
Accelerator
Wheel

Energy Auxiliary power

management Energy power steering
unit source supply unit
Steering
Energy Wheel
Hotel climate
refueling control unit
unit
Energy source
subsystem Auxiliary subsystem

Mechanical link
Electric link
Control link

Sustainable Transportation. Figure 20

Typical EV configuration

operation. The major issue with EVs is the energy and
power density of the batteries.
The overall system design of EVs is based on the
type of market the designer is looking to enter into.
Each market has different performance criteria and
some are shown below:
● On-road personal transportation: Speed, comfort,
load capacity, and range
● Service vehicle (mail): Speed, acceleration, and range
● City taxi: High speed, high acceleration, and range
● Trucks: Speed, acceleration, high load capacity, and
range
● Off-road vehicles: High gradeability, high load
capacity, and range
Each of these designs has to be designed around the
energy storage component. This component energy
density will determine the number of units necessary
to fulfill the most important design criteria of all for
EVs, the range. All of these performance criteria are
related to the power of the drive train and the energy
stored in the energy storage component. For EV drive
train design, proper motor power rating and transmis-
sion parameters are the primary considerations to meet
the performance specification (acceleration time,
maximum speed, and gradeability). The design of all
these parameters depends mostly on the speed–power
(torque) characteristics of the traction motor, shown
in Fig. 21.
Fuel cell vehicles (FCVs) look like conventional
vehicles from the outside, but inside they contain
technologically advanced components not found in
today’s vehicles. The most obvious difference is the
fuel cell stack that converts hydrogen gas stored
onboard with oxygen from the air into electricity to
drive the electric motor that propels the vehicle. A well
to wheel energy analysis, Fig. 22, performed for DOE
shows the typical efficiency of this process. Fuel cells are
electrochemical devices that convert chemical energy in
fuels into electrical energy directly, promising power
generation with high efficiency and low environmental
impact [2]. Yet the biggest issue is the onboard storage
of hydrogen and how to get the hydrogen onboard
safely. Though fuel cells could, in principle, process
a wide variety of fuels and oxidants, of most interest
today are those fuel cells that use common fuels (or
their derivatives) or hydrogen as a reductant, and
ambient air as the oxidant.
908 Sustainable Transportation

80 400

70 350
Power
60 300

Motor torque (Nm)

Motor power (km)
50 250
Torque

40 200

30 150

20 100

Base
10 speed 50

0
0 1000 2000 3000 4000 5000
Motor speed (rpm)

Sustainable Transportation. Figure 21

Typical variable speed electric motor characteristics

Energy Energy Power Energy

Resource Recovery Processing Reformer Conversion Management to Wheels

Methanol Methanol
Conversion/ steam Fuel Cell Overall
storage Reformer 50-55% Efficiency
76-81% 80-85% 20-28%
Natural Extraction/
Motor/
Gas Cleaning/
Controller/
Delivery
Accessories
88%
75-84%
Hydrogen
Overall
Conversion/ Fuel Cell Efficiency
storage 50-60% 22-33%
68-75%

Gasoline Motor/
Processing/ Autothermal Controller/ Overall
Crude Extraction Fuel Cell
Delivery Reformer Accessories Efficiency
Oil 96% 50-55%
88% 75-80% 75-84% 24-31%

Overall
Gasoline IC Vehicle (current)
Efficiency
18%
15%

Sustainable Transportation. Figure 22

“Well to wheel” efficiency for various vehicle scenarios
 Sustainable Transportation 909

Most fuel cell power systems are comprised of three
main components: unit cells, stacks, and balance of
plant components. The unit cell is where the electro-
chemical reaction takes place. If you string numerous
of these unit cells together you get what is called a fuel
cell stack. The unit cells are combined by electrically
connecting the cells to construct the desired electrical
output capacity. The balance of plant components are
used to provide feedstream conditioning (including
a fuel processor if needed), thermal management, and
electric power conditioning among other ancillary and
interface functions.
In a typical fuel cell, fuel is fed continuously to the
anode (negative electrode) and an oxidant (often
oxygen from air) is fed continuously to the cathode
(positive electrode). The electrochemical reactions take
place at the electrodes to produce an electric current
through the electrolyte, while driving a complementary
electric current that performs work on the load.
Although a fuel cell is similar to a typical battery in
many ways, it differs in several respects. The battery is
an energy storage device in which all the energy
available is stored within the battery itself (at least the
reductant). The battery will cease to produce electrical

Sustainable Transportation. Table 3 Summary of major differences of fuel cell types

PEFC AFC PAFC MCFC SOFC

Electrolyte Hydrated Mobilized or Immobilized liquid Immobilized Perovskites
polymeric ion immobilized potassium phosphoric acid in liquid molten (ceramics)
exchange hydroxide in asbestos SiC carbonate in
membranes matrix LiAlO2
Electrodes Carbon Transition metals Carbon Nickel and nickel Perovskite
oxide and
perovskite/
metal cermet
Catalyst Platinum Platinum Platinum Electrode Electrode
material material
Interconnect Carbon or metal Metal Graphite Stainless steel or Nickel,
nickel ceramic, or
steel
Operating 40–80 C 65–220 C 205 C 650 C 600–1,000 C
temperature
Charge carrier H+ OH H+ CO3 O
External Yes Yes Yes No, for some No, for some
reformer for fuels fuels and cell
hydrocarbon designs
fuels
External shift Yes, plus Yes, plus purification to Yes No No
conversion of purification to remove CO and CO2
CO to hydrogen remove trace CO
Prime cell Carbon based Carbon based Graphite based Stainless based Ceramic
components
Product water Evaporative Evaporative Evaporative Gaseous product Gaseous
management product
Product heat Process gas + Process gas + electrolyte Process gas + liquid Internal Internal
management liquid cooling circulation cooling medium or reforming + reforming +
medium steam generation process gas process gas
910 Sustainable Transportation
energy when the chemical reactants are consumed
(i.e., discharged). A fuel cell, on the other hand, is an
energy conversion device to which fuel and oxidant are
supplied continuously. In principle, the fuel cell
produces power for as long as fuel is supplied [2].
Table 3 shows typical fuel cell types and materials and
operating characteristics.
In parallel with the classification by electrolyte,
some fuel cells are classified by the type of fuel used.
Direct alcohol fuel cells (DAFC) or, more commonly,
direct methanol fuel cells (DMFC) use alcohol with-
out reforming. Mostly, this refers to a PEFC-type fuel
cell in which methanol or another alcohol is used
directly, mainly for portable applications. Direct car-
bon fuel cells (DCFC) use solid carbon (presumably
a fuel derived from coal, pet-coke, or biomass)
directly in the anode, without an intermediate gasi-
fication step. Concepts with solid oxide, molten car-
bonate, and alkaline electrolytes are all under
development. The thermodynamics of the reactions
in a DCFC allow very high efficiency conversion.
Therefore, if the technology can be developed into
practical systems, it could ultimately have
a significant impact on coal-based power generation.
The most common is the hydrogen fuel cell which
uses hydrogen as its fuel and oxygen (usually from
air) as its oxidant.
Battery electric vehicles (BEVs) are propelled by an
electric motor (or motors) powered by rechargeable
battery packs. BEVs are becoming more and more
attractive with the advancement of new battery
technology (Li-ion) that have higher power and energy
density (i.e., greater possible acceleration and more
range with less batteries) and higher oil prices [11].
Electric motors have several advantages over internal
combustion engines (ICEs). For example, electric
motors convert 75% of the chemical energy from the
batteries to power the wheels as where the ICE can only
convert 20% of the energy stored in gasoline. EVs emit
no tailpipe pollutants, although the power plant
producing the electricity may emit them. If the
electricity used to charge the EV comes from renewable
then there would be no air pollutants. Using electricity
as the propulsion fuel would reduce energy dependence
on fossil fuels. Electric motors provide quiet, smooth
operation and stronger acceleration and require less
maintenance than ICEs.
EVs face significant battery-related challenges in
the areas of driving range, recharge time, battery cost,
and size and weight. The typical driving range for
current EVs are about 100–200 miles before needing
to be recharged as where typical gasoline vehicles can
go well over 300 miles before refueling. This could
possibly be overcome by setting up more charging
stations, but in order to fully recharge the battery, it
can take anywhere from 4 to 8 h. Even a “quick charge”
to 80% can take up to 30 min. Also the cost of these
batteries is relatively expensive and may need replace-
ment one or more times over the life of the vehicle.
The amount and weight of these batteries has
improved over the years, but they still take up
a significant amount of vehicle space depending on
the type of performance that the battery pack
was designed.
Bibliography
1. U.S DOE, Office of the Energy Efficiency and Renewable Energy
(2010) DOE/GO-102010-3002
2. EG&G Technical Services Inc for DOE (2004) Fuel cell
handbook, 7th edn
3. http://en.wikipedia.org/wiki/Automobiles
4. Oliver R (2007) All about: hybrid transportation. CNN.com/
world business. 16 Sep 2007
5. http://en.wikipedia.org/wiki/Sustainable_transport
6. Todd Litman, Sustainable transportation and TDM online TDM
encyclopedia. http://www.vtpi.org/tdm/tdm67.htm
7. Department of Energy: Vehicle Technologies Program,
Advancing new vehicle technologies and fuels. www.vehicles.
energy.gov
8. http://en.wikipedia.org/wiki/Internal_combustion_engine
9. http://www.science.energy.gov/sbir/solicitations/Energy_Storage
10. http://en.wikipedia.org/wiki/Flexible-fuel_vehicle
11. http://en.wikipedia.org/wiki/Battery_electric_vehicle
12. Brill M, Dorau T Life-cycle Analysis. ICF International
13. http://www.gdrc.org/uem/lca/life-cycle.html
14. Cohn R (1994) Electric vehicle life-cycle analysis. Master of
Science thesis, MIT
15. Samaras C, Meisterling K (2008) Life-cycle assessment of
greenhouse gas emissions from plug in hybrid vehicles.
Environ Sci Technol 42:3170–3176
16. ISO 14040, International Standards Organization (2006)
17. Meier P (2002) Life-cycle assessment of electricity generation
systems and applications for climate change policy analysis.
PhD dissertation, University Wisconsin-Madison
18. Ehsani Y et al (2010) In: Ehsani M, Gao Y, Gay SE, Emadi A
(eds) Modern electric, hybrid electric, and fuel cell vehicles –
Fundamentals, theory, and design, 2nd edn. Boca Raton,
CRC Press

Sustainable Vehicle Fuels, Well-to-
Wheel Analysis
AMGAD ELGOWAINY, MICHAEL WANG
Center for Transportation Research, Argonne National
Laboratory, Argonne, IL, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Biofuel Pathways
Hydrogen Pathways with Fuel-Cell Vehicles
Electricity Pathways with Phevs and Bevs
Future Directions
Bibliography
Glossary
All-electric range (AER) The driving range of a plug-
in hybrid electric vehicle (PHEV) in charge-
depleting (CD) mode.
Battery-powered electric vehicle (BEV) A vehicle that
stores electricity in rechargeable battery packs and
uses an electric motor instead of an internal com-
bustion engine for propulsion.
Biofuels Liquid fuels (such as ethanol and biodiesel)
that are derived from bio-feedstock sources such as
corn, cellulosic biomass, and soybeans.
Charge depleting (CD) In charge-depleting mode,
PHEV uses the electricity in the battery pack to
propel the vehicle until the battery reaches
a defined minimum state of charge, at which the
battery is considered depleted.
Charge sustaining (CS) In charge-sustaining mode,
PHEV uses the onboard fuel and internal combus-
tion engines to propel the vehicle and to sustain the
state of charge of the battery pack.
Fuel-cell vehicle (FCV) A vehicle that converts hydro-
gen to electricity by using fuel cells, which powers
an electric motor to propel the vehicle.
Greenhouse gases (GHGs) Gases in the atmosphere,
such as carbon dioxide, nitrogen oxides, and meth-
ane, that absorb and trap solar radiation in the
infrared frequency range, thus elevating the
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 911
temperature of the atmosphere in a process similar
to the greenhouse concept.
Hybrid electric vehicle (HEV) A vehicle that com-
bines an internal combustion engine (ICE) with
a battery and an electric motor to propel the vehicle
and achieve improved fuel economy.
Internal combustion engine vehicle (ICEV) A vehicle
that uses an internal combustion engine (ICE) to
propel the vehicle.
Plug-in hybrid electric vehicle (PHEV) A hybrid elec-
tric vehicle with a rechargeable battery, which can
be charged from a wall outlet that is connected to
the electrical grid.
Pump-to-wheels (PTW) Represents the vehicle’s
operation stage of the fuel cycle.
Well-to-pump (WTP) Represents the fuel cycle stage
that incorporates the feedstock and fuel production
and processing, as well as fuel delivery or energy
transmission.
Well-to-wheels (WTW) Represents the combined
well-to-pump and pump-to-wheels stages of the
fuel cycle.
Definition of the Subject
As the global population and economy continue to
grow, so, too, will the demand for energy. The trans-
portation sector has been relying solely on petroleum,
consuming more than 50% of the global world oil
production [1]. In terms of demand, the United States
is the top oil-importing country. Two major issues
facing the transportation sector in the United States –
and other major countries as well – are energy security
and environmental sustainability. Improvements in the
energy efficiency of vehicles and the substitution of
petroleum fuels with alternative fuels can help slow
the growth in the demand for petroleum oil and mit-
igate the increase in greenhouse gas emissions associ-
ated with the use of petroleum. Biofuels, hydrogen, and
electricity are being promoted as alternatives to petro-
leum, and their use may help cut greenhouse gas emis-
sions. Although using biofuels in vehicles produces
greenhouse gas emissions, using hydrogen and electric-
ity as vehicle fuels produces no such emissions. Such
use of fuel in vehicles to power the wheels is known as
the pump-to-wheels (PTW) stage of the fuel cycle. On
the other hand, the production of biofuels results in
# 2012, DOI 10.1007/978-1-4419-0851-3
912 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
Fuel:
Feedstock:
Production,
Production,
Transportation,
Transportation,
Distribution,
and Storage
and Storage
Well-to-Pump Stages
Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 1
WTW stages of alternative fuel and vehicle systems
sequestration of CO2 from the atmosphere during the
growth of the bio-feedstock sources, while hydrogen
production and electricity generation result in green-
house gas emissions because they are mostly produced
from fossil sources. Such fuel production from the
primary feedstock sources, along with the associated
transportation and distribution activities, is known as
the well-to-pump (WTP) stage of the fuel cycle. Com-
bining the WTP stage with the PTW stage provides
a well-to-wheels (WTW) metric that facilitates the com-
parison of alternative fuels from an environmental
sustainability perspective on a common basis.
The terms WTW, fuel cycle, cradle-to-grave, and
life cycle are used interchangeably in the literature for
life-cycle assessments. However, the term WTW
is more pertinent to transportation fuels for use in
vehicles. Figure 1 shows a schematic of the WTW
stages for fuel use in vehicles. This chapter highlights
the key stages and factors that impact the WTW
results of using biofuels, hydrogen, and electricity
and examines the potential reduction of WTW
energy use and greenhouse gas emissions associated
with using biofuels in internal combustion
engine vehicles (ICEVs), hydrogen in fuel-cell vehi-
cles, and electricity in plug-in hybrid electric
vehicles (PHEVs) and battery-powered electric vehi-
cles (BEVs), in comparison to using gasoline in
ICEVs.
Introduction
In 1990, the world consumed 366
109 GJ of energy; in
2006, this total consumption was increased by 36% to
499
109 GJ. As the world economy and population
continue to grow, total energy use worldwide is
projected to increase to 716
109 GJ by 2030, which
Vehicle Operation:
Vehicle Refueling,
Fuel Combustion/Conversion,
Fuel Evaporation,
and Tire/Brake Wear
Pump-to-Wheels Stages
is a 44% increase over consumption in 2006 [1].
Figure 2 shows the contribution of different types of
primary energy to the total energy consumption world-
wide. Of the five primary types of energy, oil accounts
for the largest share, coal the second largest, and natu-
ral gas the third largest. In 1990, worldwide, oil
accounted for 39% of total energy use, coal for 26%,
and natural gas for 22%. In 2030, the shares are
projected to be 32, 28, and 23% for oil, coal, and
natural gas, respectively.
The transportation sector is the largest oil-
consuming sector worldwide. In 2005, the transporta-
tion sector consumed 52% of total worldwide oil
production. In 2030, the transportation sector is
projected to consume 58% of that total. In the United
States, the transportation sector already accounts for
approximately two-thirds of oil consumption. More
significantly, the US transportation sector relies solely
on oil (over 95% of transportation energy is from oil).
The transportation sector is the major contributor to
the fast growth of worldwide demand for oil. One issue
under debate is whether the world will run out of oil
and, if so, when. What is not under debate is the
geographic distribution of oil resources and oil
demand – both are undoubtedly unbalanced.
Figure 3 shows the geographic distribution of the
proved world oil reserve as of the end of 2007 [2]. This
oil reserve distribution results in concentrated oil pro-
duction in a few regions. In 2005, total world oil
production was 82 million barrels a day (MM b/d).
In 2030, oil production is projected to be 103 MM b/d
[1]. Of the total production, the 13 members of the
Organization of the Petroleum Exporting Countries
(OPEC) produced 35.3 MM b/d in 2005 and
are projected to produce 47.7 MM b/d in 2030.
Other non-OPEC countries (including Russia,
 Sustainable Vehicle Fuels, Well-to-Wheel Analysis 913

250
Oil

World Marketed Energy Use by Fuel Type,

Natural Gas Oil
Coal
200 Nuclear Coal
Renewables
Billion GJ/year
150 Natural Gas

100
Renewables

50 Nuclear

0 2006
1990 2000 2010 2020 2030

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 2

Shares of primary energy types to the total world energy consumption [1]

800
World Proved Oil Reserves at End 2008

600
Billion Barrels

400

200

0
Asia North South & Africa Europe & Middle
Pacific America Central Eurasia East
America

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 3

Geographic distribution of world proved oil reserve [2]

Latin American countries, non-OPEC African coun- geographic distribution of world oil consumption in
tries, and Caspian area countries) produced 18 MM 2005 and 2030 [1]. The geographic mismatch between
b/d in 2005 and may produce 31.3 MM b/d in 2030. oil production and oil consumption has prompted
These countries together produced 53.3 MM b/d in major oil-consuming countries to pursue policies
2005 and will produce 79 MM b/d in 2030, account- of achieving energy independence by reducing the
ing for 65% and 77% of total production in 2005 amount of oil imported, especially when oil
and 2030, respectively. To help put oil production prices rose to 140 US dollars a barrel in the summer
in the context of consumption, Figure 4 shows the of 2007.
914 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
North America
Non-OECD Asia
OECD Europe
OECD Asia
Central and South America
Middle East
Non-OECD Europe and Eurasia
Africa
0 5
Million Barrels Oil Equivalent per Day
Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 4
Geographic distribution of world oil consumption [1] (OECD is Organization for Economic Cooperation and Development)
In addition to the oil production-consumption
issue, the effect of using oil on the environment is an
ongoing concern. More specifically, global greenhouse
gas emissions have grown from 28.7 gigatonnes (Gt) of
CO2 equivalents (CO2-eq) in 1970 to 49 Gt in 2004,
which is a 70% increase in 34 years [3]. With a busi-
ness-as-usual scenario, the Intergovernmental Panel on
Climate Change (IPCC) projects that global green-
house gas emissions could increase by up to 36.7 Gt
of CO2-eq between 2000 and 2030. On the other hand,
stabilization of global atmospheric CO2 concentration
at 490–590 ppm requires reduction of global green-
house gas emissions in 2050 by as much as 60% relative
to the 2000 global greenhouse gas emissions, which is
an enormous global challenge. In 2004, the global
transportation sector alone emitted 6.3 Gt of CO2-eq.
While the global transportation sector accounts for
only 13% of total global emissions of greenhouse
gases, it accounts for 23% of global energy-related
greenhouse gas emissions [3]. In developed countries
such as the United States, the transportation sector
contributes 26% of total greenhouse gas emissions
and more than 33% of energy-related greenhouse gas
emissions [4]. Global transportation greenhouse gas
emissions could increase to 9.1 Gt by 2030 and 12 Gt
by 2050.
As a result of concerns about energy security
and greenhouse gas emissions associated with the
transportation sector, many nations have mandated
higher standards for vehicle’s fuel economy and are
2005
2030
10 15 20 25 30 35
exploring the use of alternative fuels in order to
contain growth in the demand for petroleum oil
and mitigate the projected increase in greenhouse
gas emissions.
In the United States, biofuels, hydrogen, and elec-
tricity have been promoted for their potential to help
reduce petroleum use and greenhouse gas emissions. In
general, they can be produced regionally and locally
from various sources to provide fuels for motor vehicle
use. While electricity and most biofuels can be deliv-
ered to energy storage devices in vehicles by using the
existing delivery infrastructure, the use of hydrogen
requires the development of a delivery and refueling
infrastructure before large commercial deployment of
fuel-cell vehicles can be realized.
For the WTW analysis of alternative vehicle tech-
nologies powered by biofuels, hydrogen, and electricity,
the Greenhouse gases, Regulated Emissions, and
Energy use in Transportation (GREET) model, devel-
oped by Argonne National Laboratory, has been
employed to estimate the energy use and greenhouse
gas emissions associated with the production, trans-
portation, and consumption of these alternative fuels
[5]. The energy and emissions accounting starts from
the primary energy source (well) and ends with the
vehicle’s operation stage (wheels).
The following sections present the important
stages, results, and key issues associated with the
WTW stages of biofuels, hydrogen, and electricity as
alternative transportation fuels.
 Sustainable Vehicle Fuels, Well-to-Wheel Analysis 915

Biofuel Pathways flexible-fuel vehicles (FFVs), as in the United States and

currently in Brazil. Ethanol is usually transported
The carbon in biofuels is from the air during biomass
separately to distribution centers where it is blended
growth – thus, biofuels have the potential to reduce
with gasoline.
greenhouse gas emissions significantly. However, the
Pure biodiesel (B100) can be used in compression-
life cycle of biofuel contains such activities as fertilizer
ignition engines (diesel engines). However, there are
production, farming, biofuel production, and biofuel
issues associated with the use of pure or a higher per-
combustion, which may consume fossil energy and
centage of biodiesel in a blend with petroleum diesel.
produce greenhouse gas emissions. Life-cycle analysis
Using B100 requires modifying diesel vehicles in addi-
(LCA) has been conducted to examine biofuel energy
tion to meeting special storage requirements to ensure
and environmental effects, and its methodologies have
fuel stability. In most cases, and with adequate biodiesel
advanced in the past 20 years. In the early years of
fuel quality, biodiesel blends with petroleum diesel of
examining biofuels, the so-called energy balance
up to 20% by volume (B20) can be used in diesel
(energy contained in biofuels minus fossil energy con-
vehicles without vehicle modifications.
sumed to make them through the whole life cycle) was
Ethanol is currently produced from the fermenta-
estimated for biofuels, especially for corn-based etha-
tion of starches and sugars in corn, sugarcane, cassava,
nol [6, 7]. LCA models have been developed since the
wheat, sugar beets, and other crops. In the United
early 1990s, and detailed LCAs have been conducted to
States, which is the largest producer of ethanol, ethanol
examine the energy benefits of using biofuels and their
has been produced from corn since 1980. In Brazil,
emission effects on the environment – especially corn
sugarcane ethanol has been produced for almost
based and cellulosic ethanol in comparison to petro-
100 years. Recently, China and Southeast Asia began
leum fuels [5, 8, 9]. While LCA results of biofuels have
to produce fuel ethanol from cassava (China also pro-
generally shown energy and greenhouse gas benefits of
duces a significant amount of corn ethanol). In Europe,
biofuels relative to petroleum fuels, the magnitude of
ethanol is produced from corn, wheat, and sugar beets.
the benefits is determined by the types of feedstocks
Cellulosic biomass, such as crop residues (e.g., corn
and production technologies. In addition, LCA results
stover), forest residues, and energy crops (e.g., switch-
are influenced heavily by decisions in LCAs regarding
grass [SWG]), is being considered for producing cellu-
the system boundary of a given analysis and the
losic ethanol.
method of dealing with coproducts of biofuels,
Biodiesel is produced from vegetable oils and ani-
among many other factors. The current focus of eval-
mal fats via a transesterification process. In the United
uations of the effects of using biofuel on energy and
States, biodiesel is produced from soybeans. In Europe,
environmental is on indirect effects, such as land-use
biodiesel is produced primarily from rapeseeds. In
changes from biofuel production, and on other envi-
Southeast Asia (particularly in Malaysia), biodiesel is
ronmental sustainability issues, such as water con-
produced from palm oil.
sumption, biodiversity, and soil erosion, among many
Besides the fermentation and transesterification
others [10–12].
processes, there are many other technology paths avail-
At present, two major biofuels are produced world-
able for producing biofuels. For example, cellulosic
wide: ethanol and biodiesel. Ethanol is used in spark-
biomass can be gasified to produce synthetic gas (syn-
ignition engines (or gasoline engines) from low-level
gas). Syngas can be then used to produce Fischer-
blends, such as E10 (10% of ethanol and 90% gasoline
Tropsch (FT) diesel via the FT synthesis process or
by volume), to high-level blends, such as E85 in
ethanol via fermentation of syngas. Renewable hydro-
the United States and E100 (pure ethanol) in Brazil.
carbon fuels (such as gasoline and diesel) could be
While low-level ethanol blends can be used in gasoline
produced from vegetable oils and animal fats via the
vehicles without vehicle modifications, use of high-
hydrogenation process. Cellulosic biomass can also be
level blends requires modifying gasoline vehicles to
pyrolyzed to produce a bio-oil. The bio-oil can then be
make so-called dedicated ethanol vehicles, as was
used to produce gasoline and diesel. Butanol, a fuel
done in Brazil in the 1970s to late 1990s, or to make
with higher volumetric energy content than ethanol,
916 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
could be produced from sugars via fermentation pro-
cesses. Recently, interest has heightened in producing
hydrocarbon fuels from algae.
Biofuel Life-Cycle Analysis System Boundary
Figure 5 shows the life-cycle analysis boundary defined
for the corn ethanol pathway in GREET. For other
biofuel cycles, the boundary is defined in similar
ways. In particular, the corn ethanol life cycle includes
fertilizer manufacture, corn farming, ethanol produc-
tion, and ethanol use in vehicles. All transportation
activities involved in moving goods from one location
to another (such as corn movement from farms to
ethanol plants) are included. Coproduct distillers’
grains and solubles (DGS) and their emission effects
are also included in the life-cycle analysis. Most
recently, potential direct and indirect land use changes
associated with large-scale corn ethanol production
have begun to be included as well.
An LCA of biofuels is usually compared to an LCA
of such baseline fuels as petroleum gasoline and diesel,
CO2 in the
CO2 via
atmosphere
photosynthesis
Energy inputs
for farming
Fertilizer Carbon in kernels
N2O emissions il
so
from soil and in
e ct
water streams ng bon
a r
Ch ca
Indirect land use
Direct land changes for other
use change crops and in other
regions
Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 5
Life-cycle analysis system boundary for corn-based ethanol
and so the life-cycle system boundary needs to be
defined for gasoline and diesel. To make the compari-
son between biofuels and petroleum fuels valid, the
system boundary between them needs to be defined
as consistently as possible. Figure 6 shows the LCA
system boundary usually defined for petroleum gaso-
line and diesel.
As Fig. 6 shows, the life cycle of petroleum fuels
begins with petroleum recovery in oil fields and ends
with gasoline and diesel combustion in motor vehicles.
Besides production-related activities, all transporta-
tion-related activities to move goods from one location
to another (such as crude oil from oil fields to petro-
leum refineries) are included. Again, infrastructure-
related activities (such as construction of drilling rigs
and petroleum refineries) are not included in the LCA
of petroleum gasoline and diesel. Oil exploration,
which occurs well before oil recovery, is also usually
not included in petroleum fuel LCAs. Recently,
attempts began to be made to include such indirect
effects for petroleum life cycle as military operations to
secure access to Middle East oil [13].
CO2 emissions
during fermentation CO2 emissions
from ethanol
combustion
Carbon in
ethanol
ice
pr DGS
via
ive al
n
fe sig
Ef
Conventional Animal
Feed Production
 Sustainable Vehicle Fuels, Well-to-Wheel Analysis 917

Energy Sources
Emissions to Water
and Land

Coal Natural Gas Oil Renewables Emissions to Air,

including CO2

Exploration and Transportation Petroleum Gasoline Transport Gasoline

Recovery of Crude Refining and Distribution Combustion

Crude Oil Co-reactants,

catalysts

Military Operations

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 6

Life-cycle analysis system boundary for petroleum gasoline and diesel

Life-Cycle Energy Use and Greenhouse Gas Emission plants. Recently, the size of dry milling plants has
Results of Key Biofuel Pathways approached that of wet milling plants. Some of the
newly built dry milling plants produce more than 100
Corn and Cellulosic Ethanol Since the beginning of
million gallons of ethanol per year.
the US corn ethanol program in 1980, production of
Corn ethanol plants require a large amount of
US corn ethanol has risen to 13 billion gallons in 2010
steam for fermentation and distillation. Natural gas
[14]. The US 2007 Energy Independence and Security
(NG) is the primary process fuel, and coal is used
Act established a goal to produce 15 billion gallons per
in wet milling plants to generate steam. On average,
year of corn ethanol by 2015. The corn ethanol industry
80–90% of US corn ethanol capacity is fueled with NG
expanded quickly to reach that goal.
and the remaining 10–20% with coal.
Historically, corn ethanol has been produced from
Since the beginning of the US corn ethanol program,
both dry and wet milling plants with different front-
the energy use intensity of corn ethanol plants has
end milling technologies and with different coprod-
been reduced from more than 19.5 mega Joules (MJ)
ucts. Wet milling plants were built in large sizes and
per liter (70,000 British thermal units [Btu] per gallon)
with the flexibility of producing multiple coproducts,
of ethanol [6] to less than 8.4 MJ (30,000 Btu) [15].
but they required significant capital investments. Dry
This greater than 57% reduction in energy intensity has
milling plants initially were built in small sizes and with
been achieved by high ethanol yield, increased produc-
a single coproduct (i.e., DGS) and required lower cap-
tion of wet DGS in lieu of dry DGS, and better process
ital investments. Since 2000, all newly built corn etha-
designs, all of which were driven by the economics of
nol plants in the United States have been dry milling
ethanol plant operation.
918 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
Ethanol
Production
45% N2O/CO2 from
Corn Field
27%
Fertilizer and
Chemical
Corn Farming Production
8% 15%
Corn &
Ethanol
Transportation
5%
Sustainable Vehicle Fuels, Well-to-Wheel Analysis.
Figure 7
Shares of greenhouse gas emission sources for corn
ethanol
Corn farming requires large amounts of nitrogen
fertilizer and fuels, although since the 1970s, usage
intensities for chemicals and fuels associated with US
corn farming have been reduced significantly. For
example, US corn productivity in terms of corn yield
per unit of fertilizer input to farms has increased by
88% between 1970 and 2005 [15]. This was accom-
plished by a continuous increase in corn yield per
unit of land without a corresponding increase in fertil-
izer use. Mainly because the corn yield per unit of land
was increased, farming energy use per unit of corn
yielded was reduced by 50% between 1996 and 2005,
according to the three most recent farming energy
expenditure surveys conducted by the US Department
of Agriculture.
Figure 7 shows shares of greenhouse gas emissions
by key activities for corn ethanol. Ethanol plants are by
far the largest source of greenhouse gas emissions. N2O
emissions from nitrogen fertilizer nitrification and
denitrification in cornfields (and with a small amount
of CO2 emissions from lime in cornfields) are the
second largest source of greenhouse gas emissions.
Greenhouse gas emissions from nitrogen fertilizer and
other chemical plants (such as phosphorous fertilizer,
potash fertilizer, and lime) and from farming energy
consumption are significant contributors as well.
Various cellulosic biomass feedstocks could be used
for ethanol production, including crop residues, such
as corn stover, wheat straw, and rice straw; forest resi-
dues; dedicated energy crops, such as switchgrass,
miscanthus, willow trees, and hybrid poplars; and
municipal solid wastes. The LCAs of cellulosic ethanol
that have been completed at Argonne National Labo-
ratory include ethanol from corn stover, forest residues,
and switchgrass.
Wu et al. examined these cellulosic ethanol path-
ways [16]. Corn stover is usually left in cornfields for
soil protection and as a nutrient supplement for the
next growing season. Extensive research has been done
to examine how much stover can be removed from
cornfields without causing soil quality to deteriorate.
Within the LCA context, the operation of collecting and
transporting corn stover from fields to cellulosic etha-
nol plants needs to be taken into account. In addition,
the nutrients removed from cornfields as corn stover is
removed need to be supplemented during the next
season for growing crops. These factors were considered
in Argonne’s LCA for the corn stover-to-ethanol path-
way. As for the forest residue-to-ethanol pathway, major
activities for this pathway include stumping, collecting,
and transporting forest wastes from fields to ethanol
plants. In fact, the amount of diesel fuel used for these
activities could be significant [16].
Switchgrass can be farmed as a dedicated energy
crop. Managed switchgrass farms may require fertilizer
applications in order to maintain a desirable yield per
unit of land, although the amount of fertilizer used is
less for switchgrass farming than for corn farming.
Also, if switchgrass farming occurs in arid regions (such
as the US Pacific Northwest), irrigation may be also
required. If switchgrass is grown on marginal land or
unmanaged prairie land, the growth of switchgrass
could, indeed, help increase soil carbon content,
resulting in additional greenhouse gas emission
reductions.
In ethanol plants, cellulosic biomass undergoes
a pretreatment process so that cellulose and hemicellu-
lose can be broken down into simple sugars for hydro-
lysis and further fermentation. The lignin portion of
the biomass cannot be fermented. Because of its high
energy content, lignin can be used as a process fuel in
cellulosic ethanol plants to provide the needed steam.
In fact, mass balance calculations indicate that the
amount of lignin available in cellulosic ethanol plants
can exceed the amount of lignin needed for steam
generation. Combined heat and power systems are
proposed to generate both steam and electricity in

cellulosic ethanol plants. Some of the generated elec-
tricity can be exported to the electric grid to displace
conventional electric power generation, which offers
additional reductions in greenhouse gas emissions.
Figure 8 shows life-cycle CO2-eq emissions of
greenhouse gases for six types of corn ethanol, and
three types of cellulosic ethanol. Here, the effects of
corn ethanol on greenhouse gas emissions include
emissions from direct and indirect changes in land
use changes, which are subject to great uncertainties
(see section Direct and Indirect Land Use Changes).
Corn ethanol plants using coal do not offer greenhouse
gas emissions reductions. Average wet milling plants
(with 73% of process fuel being natural gas) offer
moderate reductions in greenhouse gas emissions.
The lower greenhouse gas emissions benefits associated
with wet milling ethanol are the result of higher energy
use in wet milling plants and lower emission credits
with wet milling coproducts (the system boundary
expansion method has been used to estimate emission
credits of ethanol’s coproducts). On the other hand,
dry milling plants (except those using coal) offer sig-
nificant greenhouse gas emissions reductions, espe-
cially when wet DGS is produced and a significant
120
100
80
Life-Cycle GHG (gCO2e/MJ)
60
40
20
0
−20
−40
Fuel Use
−60
Fuel Production
−80 Life Cycle
−100
Plant Type Wet Mill Dry Mill Wet Mill
Process Fuel Coal
Co-product Avg.
Feedstock Gasoline
Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 8
Life-cycle greenhouse gas emissions of petroleum gasoline, corn ethanol, and cellulosic ethanol (g CO2-eq per MJ)
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 919
amount of energy use can be avoided. On the other
hand, the three types of cellulosic ethanol offer large
reductions in greenhouse gas emissions because the
steam required in cellulosic ethanol plants is generated
with lignin of biomass and electricity is exported to
displace conventional electricity generation [15].
Sugarcane Ethanol Wang et al. evaluated the poten-
tial reduction in greenhouse gas emissions associated
with Brazilian sugarcane ethanol used in both Brazil
and the United States [17]. Similarly, Loarie et al.
evaluated sugarcane ethanol production and use in
Brazil [18].
The sugarcane-to-ethanol cycle includes fertilizer
manufacture, sugarcane farming, ethanol production,
and ethanol use in vehicles. Traditionally, sugarcane is
harvested by laborers, comprising the so-called manual
harvest with open-field burning. Brazil is on a trend to
replace manual harvests with mechanical harvests
using farming machinery. Brazilian sugarcane mills
produce ethanol and sugar. The split between the two
products depends on market demand and prices of
both. Bagasse, the leftover material after juice extrac-
tion, is burned in sugar mills to generate steam and
Avg. Dry Mill Dry Mill .
Coal Avg. Avg. Avg. NG .
DDGS Avg. Avg. Avg. WDGS .
Corn F.Res. SWG Stover
920 Sustainable Vehicle Fuels, Well-to-Wheel Analysis

Feedstock Farming Feedstock Transport Fuel Production Fuel Transport Fuel Combustion

Fertilizer and Electricity

Chemical Export
Production
EtOH Transport EtOH Transport
Sugar Cane Sugar Cane EtOH Vehicle
from Brazil to and Distribution
Farming Transport Production Operation
U.S. in U.S.

Seedling
Growth Equipment Mill
Manufacturing Construction
Machinery
Manufacturing

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 9

Life-cycle analysis system boundary of sugarcane ethanol life-cycle analysis

N2O emissions Ethanol

from transportation
sugarcane Fertilizer 13%
Sugarcane
fields production
farming
22% 14%
10%

Ethanol plant
Sugarcane field 9%
burning
25%

CO2 emissions
from lime
Sugarcane
application
transportation
4%
3%

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 10

Shares of greenhouse gas emissions by key activities of the sugarcane ethanol life cycle (Notes: Two renewable diesel
production technologies [the Canadian SuperCetane and the U.S. UOP technologies] are presented: The four bars for each
fuel group represent four methods of dealing with coproducts of biodiesel and renewable diesel)

power. A significant amount of electricity is exported
from sugar mills. Figure 9 shows the system boundary
for an LCA of sugarcane ethanol in the GREET model.
Figure 10 shows the shares of greenhouse gas emis-
sions by key activities of the sugarcane ethanol life
cycle. CH4 and N2O emissions from open-field burning
in sugarcane plantations alone are responsible for 25%
of total greenhouse gas emissions for sugarcane etha-
nol. Overall, the five major contributors to sugarcane
ethanol greenhouse gas emissions are open-field burn-
ing, N2O emissions from sugarcane fields, fertilizer
production, greenhouse gas emissions from sugarcane
ethanol transportation, and farming energy consump-
tion. Sugarcane ethanol plants generate the least
amount of greenhouse gas emissions since combustion
of bagasse for steam and power generation returns CO2
untaken during sugarcane plant growth back to the air.
Using sugarcane ethanol achieves reductions in green-
house gas emissions of 75–80%, which is similar to the
reductions in greenhouse gas emissions associated with
using cellulosic ethanol.
Biodiesel and Renewable Diesel from Soybeans
Biodiesel is produced from seed oils or animal fats via

the transesterification process. In the United States,
most biodiesel is produced from soybean oil. The pro-
duction volume of biodiesel in the United States
increased dramatically between 2000 and 2008 [19].
However, since then, US biodiesel production declined
because of high soybean feedstock costs, limited bio-
diesel demand in the United States, and the EU’s policy
restriction of biodiesel import from the United States.
It remains to be seen if these conditions will change.
New process technologies based on hydrogenation
to convert seed oils and animal fats to renewable diesel
with properties similar to those of petroleum diesel
fuels have recently emerged [20]. Huo et al. conducted
an LCA of soybean-based biodiesel and renewable die-
sel in the United States. Soybean farming in the US
Midwest is usually rotated with corn farming [20].
Because the soybean plant’s legume has the ability to
fix nitrogen in the soil, soybean farming requires much
less nitrogen fertilizer than corn farming does, which
helps increase the energy and emission benefits of soy-
bean-based biodiesel and renewable diesel.
Before production of biodiesel or renewable diesel,
soybeans are crushed to separate soy meals and soy oil.
Renewable diesel
Biodiesel (superCetane)
0%
−20%
−40%
−60%
−80%
−100%
−120%
−140%
Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 11
Reductions in greenhouse gas emissions associated with soybean-based biodiesel and renewable diesel (relative to
petroleum diesel, on a per-unit energy basis) [20]
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 921
Soy meals are a high-value animal feed product. On
a mass basis, 82% of soybeans end up in soy meals and
the remaining 18% in soy oil. Soy oil is then used to
produce biodiesel or renewable diesel.
In biodiesel plants, glycerin, a specialty chemical, is
produced together with biodiesel. On a mass basis, 82%
of soy oil ends up in biodiesel and 18% in glycerin. In
renewable diesel plants, fuel gas, heavy oils, and pro-
pane are also produced.
Figure 11 presents LCA results of reductions in
greenhouse gas emissions associated with the use of
biodiesel and renewable diesel. In general, biodiesel
and renewable diesel can reduce greenhouse gas emis-
sions by more than 60% relative to petroleum diesel. As
the chart shows, the methods used in LCAs to address
coproducts have a significant effect on LCA results for
biodiesel and renewable diesel.
Outstanding Issues of Biofuel Life-Cycle Analysis
This section presents three outstanding issues that are
in debate and that can significantly affect LCA results
for biofuels.
Renewable diesel (UOP)
Displacement Energy allocation
Market allocation Hybrid
922 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
Direct and Indirect Land-Use Changes Although
the land-use change issue for biofuels was not new,
Searchinger et al. were the first to develop quantitative
results for this issue and advocated against certain types
of biofuel on the basis of their results [10]. Conceptu-
ally, the production of biofuels will require that bio-
mass feedstocks are grown on land. Growth of a given
feedstock on a piece of land changes the original land
use pattern of that piece of land. This land-use change
is referred to as a direct land-use change. Direct land use
changes are identifiable and measureable since such
changes can be directly observed and attributed to
biofuel production. On the other hand, use of agricul-
tural commodities (such as corn) causes an imbalance
between supply and demand of agricultural commod-
ities, which can trigger increases in commodity prices.
The price increase signal can have ripple effects, causing
cultivation of additional land for growing agricultural
commodities somewhere in the world. This land-use
change is referred as to an indirect land use change. As
one can expect, indirect land use changes are supposed
to be caused by increased commodity prices. Indirect
land use changes may be simulated with computable
general equilibrium (CGE) models to take into account
all interrelationships among all economic sectors and
activities via the price elasticity of commodities.
Searchinger et al. used the FAPRI (Food and Agricul-
tural Policy Research Institute) model (which is
a partial, not general, equilibrium model) to develop
estimates of direct and indirect land use changes [10].
The GTAP (Global Trade Analysis Project) model
developed by Purdue University is a general equilib-
rium model that has been used by several organizations
to address land-use change issues.
After publication of the Searchinger study, the sim-
ulation and data problems of that study were identified,
as were the problems of land-use change modeling by
CGE models in general. Major efforts have been made
to expand and upgrade these models to better simulate
land-use change issues for biofuels [21–24].
Carbon emissions (carbon sequestration, in some
cases) from land-use changes are determined by using
changes in aboveground biomass and belowground
biomass and soil carbon contents of different types of
land cover. Even though data sources regarding these
are available, they are not comprehensive enough to
cover all major types of land cover in different global
regions. Often, scarce data from a small set of regions
are applied to different global regions [25]. In addition,
soil carbon content data cover only a shallow depth of
soil (such as the top 30 cm). Adequate, comprehensive
data on aboveground biomass, belowground biomass,
and soil carbon content will require considerable effort
to collect and accumulate for use.
Coproduct Treatment in Biofuel Life-Cycle Analyses
As the results in Fig. 11 for biodiesel and renewable
diesel show, the results of LCA for biofuels can vary
significantly, depending on the method selected in
dealing with biofuel coproducts.
Wang et al. examined methodologies for dealing
with coproducts in biofuel LCAs [26]. They explored
five methods. First, with the mass-based allocation
method, energy and emission burdens of a given bio-
fuel pathway are allocated among all products
according to their mass output shares. This allocation
method is based on the presumption that energy use
and emissions are somewhat related to the amount of
mass processed. This method is widely used in LCAs of
consumer products and embedded in some generic
LCA models. The method is applicable as long as all
products are used for their mass values (e.g., a kilogram
of steel for use). However, this method becomes prob-
lematic when products have distinctly different uses.
For example, in the cases of sugarcane ethanol and
cellulosic ethanol production, electricity is coproduced
but cannot be allocated by mass.
Second, with the energy-content-based method, the
energy and emission burdens of a given fuel production
pathway are allocated among products according to
their share of energy output. The energy outputs of all
products are calculated by using the amount of prod-
ucts and their energy content (usually heating content
of the products). This method is applicable where most
of the products, if not all, are used for their energy
content. The method becomes problematic when prod-
ucts have distinctly different uses. For example, starch-
based ethanol plants produce ethanol and animal feeds.
Even though animal feeds have energy content, they are
used because of their significant nutritional values, not
their heating values, which are on par with conven-
tional animal feeds (such as corn and soybean meal).
Third, the market-value-based method allocates
energy and emission burdens based on economic

revenue shares of individual products. The economic
revenue of a given product is calculated from the prod-
uct yield of a given pathway and the price of the product.
Economists generally advocate use of this method. In
fact, some LCA applications of general equilibrium
models adopt this method. This method assumes that
activities and decisions are driven by economics, and
thus burdens should be disbursed according to eco-
nomic benefits. One unique advantage of this method
is that it normalizes all products to a common basis –
their economic values. However, in practice, this method
is subject to great fluctuations in product prices.
Fourth, the process-purpose-based method esti-
mates energy use and emissions of individual processes
in a fuel production facility. The energy use and emis-
sions of a given process are allocated to a given product
if the purpose of that process is solely for the produc-
tion of the given product. An example is the dryer in
a corn ethanol plant. The dryer is installed to dry DGS.
Thus, energy use and emissions from the dryer opera-
tion are allocated to DGS. However, in many cases,
individual processes in a facility may produce multiple
products, creating a need to allocate energy and emis-
sions of a given process among all products from the
process. Furthermore, this method requires energy and
emission data at the process level, not at the facility
level, which may not be available to researchers for
many biofuel facilities. Even if the process-purpose-
based method is applied to a given facility, the activities
upstream of the facility still need to be allocated. For
example, this method can be used to allocate energy use
and emissions of corn ethanol plants between ethanol
and DGS. But the allocation of energy use and emis-
sions of corn farming between ethanol and DGS still
needs to be decided. This decision, in turn, might be
based on the mass-, energy-content-, or market-based
method.
Fifth, with the displacement method (also called the
“system boundary expansion method”), the products
that are to be displaced by nonfuel coproducts are
determined first. Energy and emission burdens of pro-
ducing the otherwise displaced products are then esti-
mated. The estimated energy and emission burdens are
credits that are subtracted from the total energy and
emission burdens of the biofuel production cycle. While
the displacement method is generally advocated
for LCAs, its implementation poses some major
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 923
challenges. The method requires conducting LCAs for
the conventional products that will be displaced, which
could be time- and resource-intensive. Another major
problem with the displacement method is that
when nonfuel products are a large share of the total
output, the method generates distorted LCA results for
fuels [26].
Whether a given method can be uniformly and
blindly recommended for LCA studies is far from set-
tled. Consistency of coproduct method choices for
evaluation of different biofuel production pathways
may not serve the purpose of providing reliable LCA
results well. Transparency of LCA methods is impor-
tant in LCA studies, and sensitive cases with multiple
coproduct methods may be warranted in LCA studies
where coproducts can significantly impact study
outcomes.
Other Environmental Sustainability Issues Envi-
ronmental sustainability issues of biofuel production
and utilization are now an important topic. Such issues
include freshwater consumption for feedstock growth
and biofuel production, effects on soil erosion of grow-
ing certain feedstocks, implications of feedstock
growth on biodiversity, and effects of producing and
using biofuels on air pollution and its health-related
effects. So far, these issues have not been addressed
systematically on the life-cycle basis. Nonetheless,
such issues as freshwater consumption were estimated
by examining key (but not all) stages of the biofuel life
cycle. In addition, some of the issues (such as biodiver-
sity) are difficult to address quantitatively. Eventually,
these issues should be addressed along the entire life
cycle of biofuels (i.e., all stages of the life cycle should be
considered). More importantly, these issues should
be addressed on a comparative basis so that biofuels
can be compared with baseline petroleum gasoline
and diesel for relative environmental sustainability
implications.
Wu et al. recently estimated the consumptive water
requirements of ethanol and gasoline production [12].
For biofuel production, the key determinants are feed-
stock and the amount of irrigation water needed to
generate reasonable yields. For gasoline production,
the key determinants are the characteristics of individ-
ual oil reservoirs, the recovery technology used, and the
degree of produced water recycling. On average, corn
924 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
ethanol production tends to consume more water on
a life-cycle basis than does cellulosic ethanol produc-
tion. Net water use for cellulosic ethanol production is
comparable to that of gasoline. Biofuels production
exhibits significant regional differences in water use.
Consumptive water use for corn ethanol production
varies significantly in the major US corn-growing
regions. Producing a liter of corn ethanol can consume
as little as 10 or as much as 324 L of water, depending
on the amount of irrigation water used for corn grow-
ing. On average, more than half of US corn ethanol is
produced at a water use rate of 10 L of water per liter of
ethanol. Switchgrass-based cellulosic ethanol produc-
tion, when the switchgrass is grown in its native habitat
in the United States, can consume from 1.9 to 9.8 L of
water per liter of cellulosic ethanol, depending on
process technology. In comparison, net water use to
produce a liter of gasoline varies from less than 3 L to
nearly 7 L.
Hydrogen Pathways with Fuel-Cell Vehicles
Hydrogen is a clean fuel that can be converted into
electricity with high efficiency in fuel cells to power
electric motors for vehicle propulsion. Using hydrogen
in FCVs produces no tailpipe emissions. Furthermore,
hydrogen can be produced from diverse feedstock
sources, including nonfossil and renewable sources.
This mitigates the risk of future dependence on
a single source of fuels to satisfy the transportation
needs, as is the case with the current global dependence
on petroleum fuels.
Fuel-cell vehicles (FCVs) are being promoted pri-
marily because of their high efficiency that is not lim-
ited by Carnot theoretical heat engine efficiency and
because of their zero or near-zero emissions. Govern-
ments and private industries are actively investigating
market barriers and vehicle and fuel infrastructure
constraints so that the introduction of FCVs will be
successful when the technology is ready for the mass
market. FCVs are based on proton exchange membrane
(PEM) technology of fuel cells, which requires hydro-
gen as a fuel. The PEM fuel cell generates electricity
from the hydrogen fuel through an electrochemical
conversion process and stores the generated electricity
in a battery. The battery then powers an electric motor
to propel the vehicle.
The delivery infrastructure for large-scale deploy-
ment of FCVs is not in place. This represents a major
challenge to a large-scale commercialization of FCVs.
Another challenge for hydrogen FCV technology is stor-
ing a sufficient amount of hydrogen aboard the vehicle in
a practical amount of space for an acceptable vehicle
range and at a reasonable cost. A variety of hydrogen
storage technologies are being researched and developed.
These include physical storage approaches – modest-
pressure gas (350 bar), high-pressure gas (700 bar), and
cryo-compressed hydrogen storage – as well as material-
based storage systems. Material approaches include
metal hydrides; sorbents, such as carbon and metal
organic frameworks (MOFs); and chemical hydrides.
The storage system technology directly impacts the
WTW energy efficiency, greenhouse gas emissions, and
costs of hydrogen and fuel-cell vehicles [27].
Although hydrogen-powered FCVs offer zero vehic-
ular emissions, production of hydrogen – such as from
natural gas (NG) or grid electricity as hydrogen is being
currently produced – can generate a significant amount
of emissions and suffer significant energy penalty. Such
energy penalty and emissions are accounted for in the
WTP stage of the fuel cycle as explained earlier.
Hydrogen Fuel-Cycle Pathways
Central Hydrogen Production Via Steam Methane
Reforming Steam methane reforming (SMR) is
a process in which high-temperature steam (700–
1,000 C) is used to produce hydrogen from natural
gas. In the first stage of the process, methane reacts
with steam in an endothermic reaction under 3–25-bar
pressure in the presence of a catalyst to produce hydro-
gen, carbon monoxide, and a relatively small amount of
carbon dioxide. Subsequently, the carbon monoxide
and steam are reacted by using a catalyst to produce
carbon dioxide and more hydrogen. Carbon dioxide
and other impurities are typically removed from the gas
stream by using pressure swing adsorption (PSA), leav-
ing essentially pure hydrogen. The efficiency range of
hydrogen production via SMR is between 68% and
75% [28].
Central Hydrogen Production Via Wind Energy
The desire to minimize greenhouse gas emissions asso-
ciated with hydrogen production for FCVs is likely to

increase interest in producing hydrogen via the elec-
trolysis of water by using renewably generated electric-
ity. Wind power has entered the mainstream utility
market because currently available government incen-
tives make it competitive with conventional alterna-
tives. Without a major breakthrough or shift in
incentives, wind is likely to remain the lowest-cost
source of renewable electricity. Production of hydrogen
from wind power helps alleviate the intermittent prob-
lem of wind power. An annual average wind speed
greater than 6 m/s (13 mph) is required for utility-
scale wind power plants. Producing 1 kg of gaseous
hydrogen approximately requires 50 kWh of electricity,
while the lower heating value (LHV) of hydrogen is
33.3 kWh/kg. In such case, the electrolyzer efficiency is
67% [29].
Distributed Hydrogen Production Via Electrolysis
Hydrogen can be produced at the refueling station by
electrolysis of water with grid electricity. Because elec-
tricity transmission and distribution infrastructure is
already in place throughout most countries, this pro-
duction option helps avoid long-distance transporta-
tion of hydrogen from central production locations.
However, if hydrogen is produced via electrolysis at
refueling stations for the large-scale FCV market,
a large amount of electricity could be required. In
such case, electricity distribution lines to refueling sta-
tions will probably need to be upgraded to allow suffi-
cient power for hydrogen production. The energy use
and the greenhouse gas emissions associated with
hydrogen production depend on the electrolyzer’s effi-
ciency and the technologies used for grid power gener-
ation, respectively. The electrolyzer efficiency ranges
from 67% to 76% [28].
Compression of Hydrogen Hydrogen compression
is needed for pipeline delivery, for loading into tube
trailers, and for dispensing into FCVs. The power
demand of a compressor can be reduced by
maintaining the gas temperatures as low as possible.
Thus, low compression power can be achieved through
multistage compression with intercooling between
stages. Considering the thermodynamic properties of
hydrogen, and with isentropic efficiencies in the range
of 60–80%, the allowable pressure rise per stage of
compression is in the range of 2.0–2.5. For such
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 925
pressure ratios and with intercooling, stage outlet tem-
peratures are low enough that perfect gas relationships
adjusted with compressibility factor provide
a reasonable estimate of compression power require-
ments as follow:



 1 k
Power ¼ ðZÞð m ÞðRÞðTÞðnÞ

k1
"
ðk1 #
Poutlet nk Þ

1 ;
Pinlet
where Z is the mean compressibility factor
m is the mass flow rate
R is the universal gas constant
T is the inlet gas temperature
n is the number of stages
Z is the isentropic efficiency of compression
k is the ratio of specific heats
Poutlet is the compressor discharge pressure and
Pinlet is the compressor inlet pressure
The equation assumes the intercooler outlet tem-
peratures are equal to the ambient temperature [30].
The greenhouse gas emissions associated with com-
pression of hydrogen depend on the compression pres-
sure ratio and efficiency, as well as the technologies
used to generate the electricity for compression.
Liquefaction of Hydrogen One of the most efficient
methods to package hydrogen molecules at high energy
density for transportation and storage purposes is to
liquefy hydrogen after its production. Liquefaction
includes gas compression, cooling with water, and
precooling with liquid nitrogen to drop the hydrogen
below its inversion temperature (200 K). However,
liquefying hydrogen requires cooling to temperatures
below 20 K. Cooling to that temperature requires sig-
nificant electrical energy use at approximately 10–13
kWh per kg of hydrogen. It is estimated, however, that
advanced liquefaction plants could reduce electrical
energy requirements to 8 kWh/kg [30]. The greenhouse
gas emissions associated with liquefying hydrogen
depend on the liquefier efficiency and the technologies
used to generate the electricity required for liquefac-
tion. Once liquefied, hydrogen can be efficiently
transported, stored, and dispensed into FCVs in liquid,
cryo-compressed, or gaseous form. However, liquid
926 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
hydrogen is subject to significant boil-off losses if
stored for a long period. Because of these losses, liquid
hydrogen pathways generally consume more energy
and produce more emissions than gaseous hydrogen
pathways.
Hydrogen Consumption by FCVs
The key parameter determining the hydrogen consump-
tion by FCVs is the vehicle’s fuel economy. The testing of
the fuel economy of the first- and second-generation
FCVs, through the Department of Energy (DOE)’s Con-
trolled Hydrogen Fleet and Infrastructure Validation
and Demonstration Project, resulted in an adjusted
fuel-economy range of 42–57 miles/kg hydrogen for
the four teams for first-generation vehicles, as compared
to 43–58 miles/kg for second-generation vehicles [31].
More recently, FCVs produced by Honda and Toyota
achieved fuel economies at 60 mi/kg and 70 mi/kg,
respectively. The fuel economy of FCVs depends greatly
on the fuel cell system and electric motor efficiencies.
The DOE targets a peak efficiency of 60% for fuel cells,
which has already been demonstrated by many FCV
OEMs. Currently, most of the research is focused on
reducing the cost of the fuel cell system.
Hydrogen Delivery and Storage in FCVs
The transition to hydrogen-powered vehicles
requires the development of hydrogen infrastruc-
ture, including refueling stations. Distributing
hydrogen depends on how it is packaged for dis-
pensing into hydrogen vehicles. Packaging affects
not only density (weight/volume) but also the oper-
ation of potential delivery modes and onboard stor-
age, a problem that has been called the “grand
challenge” of the hydrogen economy. Energy-dense
packaging is a critical issue for hydrogen fuel and
was a major focus of the DOE’s hydrogen and fuel
cells initiative [44]. Although high-pressure gaseous
storage is currently closest to meeting the DOE goal
of 8 wt.% hydrogen, other storage technologies are
also being investigated (e.g., metal hydrides, carbon
nanostructures, and carriers like methanol or other
hydrogen-rich chemicals). Depending on the success
of these efforts, hydrogen could be delivered as a high-
pressure gas, as a cryogenic liquid, or at standard tem-
perature and pressure via a solid or liquid carrier.
The associated energy use associated with each mode
of delivery/dispensing could be quite different [32].
Among the options that are being used and pro-
posed for onboard hydrogen storage in FCVs are
350-bar and 700-bar gaseous storage, as well as super-
critical cryo-compressed storage of hydrogen. The
350-bar gaseous storage option requires less energy
for hydrogen compression into the vehicle’s tank, but
it results in low energy density and reduced vehicle
travel range between refills. The 700-bar gaseous stor-
age option improves the vehicle’s travel range but
requires higher compression energy, in addition to
cooling to 40 C to attain a fill rate comparable to
that of current gasoline vehicles (approximately at
1.6 kg/min). The compression energy requirements
for 350-bar and 700-bar gaseous dispensing options
are 2 kWh/kg and 3 kWh/kg, respectively. The cooling
to 40 C results in an additional 0.25 kWh per kg of
hydrogen for the 700-bar dispensing option. The cryo-
compressed option requires delivering hydrogen in
liquid form, which incurs a large energy penalty at the
liquefaction plant, as discussed earlier. Currently, most
OEMs are favoring the 700-bar storage option.
WTW Results of Hydrogen Fcvs
Figure 12 shows the reduction in the WTW total energy
use for various hydrogen production pathways relative
to the conventional gasoline ICE vehicle pathway.
These energy use results reflect the WTP energy use
and the fuel economy of FCVs. The reduction in the
total energy use is directly proportional to the WTW
efficiency of the pathway. Central production of hydro-
gen from natural gas via steam methane reforming with
350-bar dispensing provides a reduction in total energy
of 39% compared to the conventional gasoline path-
way. The reduction is attributed mainly to the high
efficiency of fuel cells, which results in a fuel economy
ratio of 2.3 relative to the baseline gasoline’s vehicle fuel
economy. The reduction in total energy for the same
production pathway drops to 36% with the 700-bar
dispensing option because of the additional compres-
sion and cooling requirements, as discussed earlier. If
liquefaction is used to facilitate high-energy-density
packaging for more cost-effective delivery and
improved FCV travel range, the reduction in total
energy for the natural gas SMR pathway fades to 7%,
 Sustainable Vehicle Fuels, Well-to-Wheel Analysis 927

40%
30%
20%
10%
0%
−10%
−20%
−30%
−40%
−50%
Natural Gas Natural Gas Natural Gas Wind Electrolysis-
(350 bar) (700 bar) (Liquid) (700 bar) US mix
(700 bar)

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 12

Reduction in WTW total energy (relative to gasoline) for various hydrogen production pathways

0%
−10%
−20%
−30%
−40%
−50%
−60%
−70%
−80%
−90%
−100%
Natural Gas Natural Gas Natural Gas Wind Electrolysis-
(350 bar) (700 bar) (Liquid) (700 bar) US mix
(700 bar)

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 13

WTW petroleum reduction (relative to gasoline) for various hydrogen production pathways

relative to baseline gasoline vehicles. This reduction is
due to the high energy use associated with the lique-
faction process. Central production of hydrogen via
electrolysis using electricity from renewable sources,
such as wind, results in a reduction in total energy
similar to that of the natural gas SMR pathway at
39%. For the wind-to-hydrogen pathway, the account-
ing of energy use starts with the energy in electricity
because the wind power is available in abundance and
will be lost if not utilized. Therefore, the conversion
efficiency of wind power to electricity impacts the
economics rather than the sustainability of such
conversion. However, when hydrogen is produced via
electrolysis at distributed locations (i.e., at a refueling
station), it uses electricity from the grid. Considering
the US grid mix with a combined fossil energy effi-
ciency of 40%, an electrolyzer efficiency of 71%, and
a compression efficiency of 91%, the WTW total energy
use for this pathway exceeds that of the baseline gaso-
line by 32%, as shown in Figure 12.
Figure 13 shows the reduction in the WTW petro-
leum energy use for various hydrogen production
pathways relative to a conventional gasoline ICEV
pathway. All hydrogen production pathways virtually
928 Sustainable Vehicle Fuels, Well-to-Wheel Analysis

eliminate the dependence on petroleum fuels for vehi- Hydrogen Life-Cycle Analysis Issues and
cle operation. The WTW efficiency of the pathways Uncertainties
implied in Fig. 12 does not impact the degree of petro-
The production and delivery of hydrogen for dispens-
leum reduction simply because the feedstock sources in
ing in FCVs may result in significant energy penalty
these pathways are nonpetroleum, except for a small
and greenhouse gas emissions for some pathways. Nat-
share of electricity power generation (2%) in the US
ural gas is currently being used for commercial pro-
mix that is produced from residual oil.
duction of hydrogen. While natural gas is the least
Figure 14 shows the reduction in the WTW green-
carbon intensive option among all fossil fuels and is
house gas emissions for various hydrogen production
abundant at a lower cost than petroleum fuels on a
pathways relative to a conventional gasoline ICEV
per-unit-energy basis, its potential for greenhouse gas
pathway. The reduction in greenhouse gas emissions
emissions reduction by producing hydrogen for FCVs
shows a trend similar to the reduction in total energy in
is limited to less than 40% relative to gasoline ICEVs.
Fig. 12, but with renewable sources (i.e., wind)
Furthermore, packaging hydrogen at high energy den-
exhibiting significantly more reductions in greenhouse
sity through the increased compression to 700 bar and
gas emissions compared to fossil sources of hydrogen
cooling, or through liquefaction, further reduces the
production (i.e., natural gas) and the US grid mix.
efficiency of the hydrogen’s WTW pathway and its
While natural gas SMR production pathway provides
potential greenhouse gas emissions benefits. Distrib-
reductions in WTW greenhouse gas emissions of 49%
uted production of hydrogen via electrolysis at
and 46% for the 350-bar and 700-bar dispensing
refueling sites would rely on the local generation mix
options, respectively, the wind pathway provides
of electricity. If the electricity is generated from such
a reduction of 89% in greenhouse gas emissions relative
fossil sources as coal or natural gas, the resulting WTW
to the baseline gasoline pathway. Hydrogen production
efficiency and greenhouse gas emissions of the hydro-
via electrolysis using the US average grid mix results in
gen via electrolysis pathway will be worse than those of
a 40% increase in greenhouse gas emissions relative to
the baseline gasoline ICEVs. However, producing
gasoline, mainly because of the low WTP efficiency for
hydrogen from nonfossil sources (such as wind elec-
this pathway and the high dependence of the US grid
tricity) reduces the greenhouse gas emissions signifi-
on fossil sources – especially coal – for power
cantly compared to gasoline. Further technological
generation.

60%
40%
20%
0%
−20%
−40%
−60%
−80%
−100%
Natural Gas Natural Gas Natural Gas Wind Electrolysis-
(350 bar) (700 bar) (Liquid) (700 bar) US mix
(700 bar)

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 14

WTW reductions in greenhouse gas emissions (relative to gasoline) for various hydrogen production pathways

advances in the production of hydrogen from nonfossil
sources, as well as developing a viable delivery infra-
structure and improving FCV onboard storage tech-
nologies, are needed to facilitate a meaningful
reduction of greenhouse gas emissions relative to
today’s gasoline ICEVs.
Electricity Pathways with Phevs and Bevs
Electricity can be used in PHEVs and BEVs. Since US
electricity production is largely independent of petro-
leum fuel use, PHEVs and BEVs have been touted for
their potential to reduce petroleum use and greenhouse
gas emissions through gains in efficiency and use of
electricity. The PHEV category can cover a wide variety
of options with respect to technical attributes, such as
the battery chemistry, the amount of grid electricity
that can be stored in the battery, and the powertrain
and fuel choices. In addition, the driving behavior of
consumers could also significantly affect the energy use
and the greenhouse gas effects of PHEVs. The magni-
tude of energy and greenhouse gas emission benefits of
PHEVs and BEVs is affected by the type of electricity
generation used for the vehicle’s recharging and the
vehicle’s energy efficiency.
Many LCA studies have been completed for PHEVs
and BEVs, as well as for conventional gasoline ICEVs
and gasoline hybrid electric vehicles (HEVs) [33–38].
These studies concurred that the battery operation
(also known as charge-depleting [CD] operation) of
PHEVs and BEVs reduces petroleum use since the
electricity generation in the United States (as in most
of the world’s large economies) is powered by
nonpetroleum fuels. However, these studies produced
mixed results when estimating the greenhouse gas
emissions of PHEVs and BEVs. These results are attrib-
uted to the difference in the assumptions associated
with efficiencies of vehicle powertrains, on-road
adjustment factor for fuel and electricity consumption,
and electricity generation mix and efficiencies.
PHEVs are similar to regular HEVs, except that
PHEVs employ a bigger battery, which is recharged
through a wall outlet by drawing electricity from the
grid. They have the potential to reduce the transpor-
tation system’s dependence on petroleum and the
emissions of greenhouse gases by using off-peak excess
electricity-generation capacity and increasing the
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 929
vehicle’s energy efficiency. The reduction in petro-
leum use by PHEVs increases with a corresponding
increase in their electric range, which is proportional
to the size of the battery employed. Gains in fuel
economy by PHEVs and BEVs relative to conventional
vehicles are another key factor determining their
benefits in terms of energy use and greenhouse gas
emissions.
While all previous studies predicted significant
reductions in petroleum energy with the use of
PHEVs and BEVs, they predicted mixed results for
greenhouse gas emissions from these vehicles. For
example, some studies estimated that PHEVs and
BEVs (recharging from the US average mix) generate
fewer greenhouse gas emissions than gasoline ICE vehi-
cles, but they generate greenhouse gas emissions equal
to (or more) than those of gasoline HEVs [33–35]. In
contrast, other studies estimated that PHEVs operating
in CD mode can outperform HEVs in terms of reduc-
tions in greenhouse gas emissions if 75% or more of the
required electricity is generated from natural gas com-
bined-cycle (NGCC) [36] and produce lower green-
house gas emissions, even with coal-based electricity,
compared with gasoline ICE vehicles [38]. A recent
study by Argonne National Laboratory conducted
detailed dispatch modeling simulations in different
US utility service areas with different charging scenar-
ios for PHEVs and explored “real-world” driving
energy use for PHEVs vs. gasoline vehicles [38]. The
key factors impacting the WTW analysis of PHEVs and
BEVs and the findings of the Argonne study are
discussed next.
PHEV Fuel-Cycle Pathways
Since a PHEV consumes fuel and electricity, its path-
way consists of two parallel paths for these two energy
sources. The fuel path includes the recovery of the
feedstock (e.g., crude), the transportation of the feed-
stock, the production of the fuel (e.g., refining of crude
to gasoline), and the transportation of the fuel to the
pump. The electricity path includes the recovery,
processing, and transportation of the fuel used for
electricity generation (e.g., natural gas, coal, and ura-
nium), the technology used for electric power genera-
tion (e.g., steam power plant, natural gas combustion
turbine), the transmission of the electricity to the wall
930 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
outlet, and the charging of the vehicle’s battery. The
fuel and electricity production and transmission rep-
resent the WTP stage of the pathway, while the vehicle’s
consumption of fuel and electricity represents the PTW
stage.
A PHEV charges the battery to a high state of charge
(SOC) (e.g., 90%). Then, the vehicle operates in a CD
mode by using the stored electricity in the battery until
it reaches a low SOC (e.g., 30%). Once the battery
reaches the low SOC threshold, the PHEV operates in
a charge-sustaining (CS) mode, which is similar to the
operation of regular HEVs [39]. This operation strat-
egy allows the vehicle to operate as a zero-emission
vehicle in CD operation. However, battery cost and
PHEV performance requirements have led automakers
to consider a “blended” CD mode, through which the
engine is intermittently turned on, resulting in an
increase in the CD range by utilizing the electric
powertrain and the engine simultaneously (blended
operation).
Fuel and Electricity Consumption by PHEVs
Since the automotive industry has yet to produce
PHEVs and BEVs, the fuel and electricity consumption
of PHEVs and BEVs are estimated by employing vehicle
simulation models, such as Argonne’s Powertrain Sys-
tem Analysis Toolkit (PSAT). PSAT can simulate dif-
ferent vehicle configurations through a standard or
custom driving cycle to produce the fuel consumption
by these vehicle technologies. PHEV designs in the
Argonne study covered PHEV10 (i.e., 10 miles of elec-
tric range), PHEV20, PHEV30, and PHEV40. The
higher electric range the vehicle has, the less petroleum
fuel the vehicle will need to consume. Nevertheless,
increased electric range requires a bigger battery,
resulting in increased cost and weight of the vehicle.
The high cost of batteries led vehicle designers to
explore ways to extend the electric range through
a “blended” operation of the engine with the electric
motor. However, the intermittent operation of the
engine in blended mode has the potential to create
bursts of emissions when the engine comes on, making
it necessary to develop modifications of the emissions
control systems in order to meet emissions standards.
The blended operation of the engine and the electric
motor extends the vehicle’s electric range significantly
while allowing for more efficient utilization of the
battery.
One major complication associated with PHEVs is
how to rate these vehicles with respect to their electric-
ity and fuel consumption. The EPA develops on-road
fuel economy estimates that appear on the window
stickers of all new cars and light trucks sold in the
United States [40]. The purpose of the fuel economy
label is to assist consumers in comparing the fuel econ-
omy of different vehicles as they make their purchase
decision [41]. However, the current methodology for
rating fuel economy is not adaptable to rating PHEVs
because of electricity consumption and intermittent
engine use in blended CD operation. Furthermore,
fuel economy expressed as miles per gallon (MPG)
leads customers to falsely believe that the amount of
gasoline consumed by an automobile decreases as
a linear function of a car’s MPG [42]. The actual rela-
tionship is curvilinear, as shown in Fig. 15. The figure
shows that a 10-mpg improvement at lower fuel econ-
omy produces much higher fuel consumption savings
compared to the same improvement at higher fuel
economy. For example, increasing fuel economy
from 10 to 20 mpg results in savings of 5 gal over
100 miles of traveling distance, while an equivalent
increase from 40 to 50 mpg results in savings of a mere
0.5 gal over the same traveling distance. Critics of fuel
economy labeling argue that representing fuel effi-
ciency in terms of amount of gasoline consumed for
a given distance (e.g., gallons per 100 miles), which is
commonly used outside of the United States, would
make the benefits of greater fuel efficiency more trans-
parent [42]. More recently, the New York State Senate
passed a bill requiring state car dealers to post a chart
converting MPG to gallons per 1,000 miles (the average
monthly driving distance) in 5-mpg increments. The
bill has to clear the State Assembly before it could be
signed into law. Reporting gallons per 100 miles for fuel
consumption and kWh per 100 miles for electricity
consumption on the window sticker of PHEVs could
be easily converted to the cost per 100 miles since the
prices per gallon of gasoline and per kWh of electricity
are usually known in the locality surrounding the
customer traveling. Table 1 shows the fuel and electric-
ity consumption of gasoline ICEVs, HEVs, and PHEVs,
as well as BEVs based on recent analysis by
Argonne [38].
 Sustainable Vehicle Fuels, Well-to-Wheel Analysis 931

Fuel Consumption (Gallons/100 miles)

12.0

10.0

8.0

6.0

4.0

2.0

0.0
0 10 20 30 40 50 60 70 80 90 100
Fuel Economy (mpg)

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 15

Relationship between fuel economy and fuel consumption

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Table 1 Fuel and electricity consumption for alternative gasoline
vehicle technologies
PHEV10 PHEV20 PHEV30 PHEV40 BEV
ICEV HEV CD CS CD CS CD CS CD CS CD
Gallons per 100 miles 3.52 2.41 1.35 2.24 1.24 2.25 0.28 2.91 0.16 2.94 N/A
kWh per 100 miles N/A N/A 22.6 N/A 22.1 N/A 38.7 N/A 39.0 N/A 40.4

PHEV Electric Demand (Load) Profiles and predict the marginal electricity generation mix
Electricity Generation Mix for Battery Recharging dispatched for battery recharging in that area. The
marginal electricity generation mix for battery
PHEVs will draw electric energy from the electric grid.
recharging is a major factor impacting the WTW results
The extent of electricity demand can be estimated by
of PHEVs and BEVs.
examining patterns of vehicle usage and estimating the
The generation mix at the time of charging is
potential number of PHEVs that will be plugged in.
a strong function of the time of day, time of year,
The daily electricity demands for various PHEVs can be
geographic region, vehicle and charger design, base
estimated by analyzing the following four factors:
and vehicular load growth patterns, and the associated
(1) daily vehicle usage, (2) pattern of vehicle arrival at
generation expansion in the years before the charging
home at the end of the last trip, (3) number of PHEVs
event of interest.
of different electric ranges that will be plugged in each
Figure 16, developed by Shelby and Mui, is an
day, and (4) amount of electric power that will be
illustration of the diurnal peaks of demand for
drawn by each PHEV and the time required for charg-
a hypothetical summer day [43]. Sharp summer peaks
ing. For estimating electricity demand by PHEVs, the
are caused by the demand for air-conditioning,
number of PHEVs that will be on road in a given year
although such peaks typically occur in the late after-
needs to be estimated.
noon and early evening. However, demand is at
Once the PHEV load profile is established,
a minimum overnight when businesses are closed,
a dispatch model simulating the electric power gener-
lights are off, and air-conditioning load is at its lowest.
ation and transmission system in a utility area can
932 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
Electricity Demand
6 a.m. Noon Midnight Noon
Time of Day
Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 16
Typical summer electric load profile and dispatch scheme for many U.S. utilities [43]
As the demand for electricity increases, additional gen-
erating units are dispatched to meet the load. When
a PHEV charger is activated, it causes additional load
on the marginal generator (i.e., the last unit brought on
line). When that unit reaches full capacity, another unit
is brought on line as the marginal unit, and so forth.
Therefore, when a large number of PHEVs are added
to a system, several additional generation units may
be required to meet the charging load. Consequently,
the energy use and emissions of those units are allo-
cated to the PHEV charging load. In an extensive
interconnected region, transmission constraints can
develop so that several geographically separated gener-
ating units must operate at part load to meet an
increasing demand.
Seasonal load variations also affect the mix of units
brought on line to meet the PHEV charging demand.
High summer electric loads are typical for most of
the United States, reflecting power demand for air-
conditioning. Electric heating loads tend to increase
off-peak demands and may compete with the off-peak
charging of PHEVs during the winter season.
The vehicle design characteristic with the greatest
influence on PHEV charging load is battery capacity,
which is related to the electric range and weight of the
Peaking Generation, typically simple cycle
oil and gas fired turbines. Capacity factors
for peaking generation are less than 3 − 5%
Intermediate Load Generation, typically
gas-fired combined cycle systems. Units in
this range offer fast ramp rates and short
start-up times.
Base Load Generation, typically coal fired.
Base load generation has some ability to
follow loads, though ramp rate and minimum
nighttime load constraints exist.
Nuclear
Hydro
vehicle. It is most commonly assumed that the charger
will operate at normal household power levels, typically
110 V and no more than 20 A. A sport utility vehicle
(SUV) type of PHEV may require larger batteries than
a compact or sedan type of PHEV. To charge these
batteries in a reasonable length of time, more charging
current is required. Higher charging current could be
accomplished with a charger operating on 220 V at
30 A. Single-phase, 220-V service is available to all
residential customers, but such service typically will
require professional installation of additional circuit
breakers, lines, and a dedicated outlet. The benefit of
reduced charging time comes at an additional cost of
the higher demand.
The inventory of units available for PHEV charging
is slowly changing as old units are retired or are refitted
with new environmental controls and as new units are
constructed in anticipation of increasing demand. Also,
existing units may change place in the dispatch order as
they age or as new plants come on line. Although
commercial introduction of PHEVs is expected in
2010, the demand for PHEV charging will not likely
be substantial for a long time because of the low fleet
turnover rate (approximately 7–8% per year in the
United States).
 Sustainable Vehicle Fuels, Well-to-Wheel Analysis 933

Planning to expand generation must take into
account both the extent of likely demand growth and
the daily and seasonal dynamic structure of the projected
demand. Relatively constant loads are best served by
large base-load units with low fuel and variable costs.
Daily peak loads may best be served by units with low
fixed (investment) costs, such as gas turbines. However,
lower fuel-cost options, including hydro, will be applied
to peak loads if capacity is available. The generation
mix applied at a specific time is predicted by dispatch
models. The dispatch models match available capacity
to the dynamic load by using cost, emissions, or other
criteria to optimize the system. Dispatch models also
take reliability and scheduled plant outages into
consideration.
Combining Charge-Depletion and Charge-
Sustaining WTW Results for Phevs
To combine the fuel and electricity pathways together
for PHEVs, PHEV miles with electricity vs. with
fuels must be estimated. The split depends on daily
miles driven by individual drivers and the electric
ranges of different PHEVs. With national daily mile
distribution, Figure 17 shows the miles that could be
driven with electricity for given electric ranges of
PHEVs on the basis of US national household travel
survey data [38].
WTW Results of PHEVs and BEVs
Figure 18 shows the WTW results of gasoline PHEVs,
in addition to gasoline ICEVs, gasoline regular HEVs,
and BEVs. The implication of the electricity genera-
tion mix resides in the carbon intensity of the used
fuel and the efficiency of the generation technology.
The energy use associated with electricity generation
and greenhouse gas emissions increase progressively
as the marginal mix becomes less efficient and dom-
inated by coal or residual oil. PHEVs recharging from
NGCC generation produce greenhouse gas emissions
comparable to or less than those of regular HEVs,
while PHEVs recharging from coal-based power gen-
eration produce greenhouse gas emissions compara-
ble to or slightly higher than those of ICEVs. Usually,
more petroleum savings and greater reductions in
greenhouse gas emissions are realized with an
increase in a vehicle’s electric range when the gener-
ation mix is independent of oil. Furthermore, when
the charging of PHEV occurs from a low-carbon
generation mix (e.g., NGCC) or from a renewable
source, PHEVs with high electric ranges provide sig-
nificantly more petroleum savings and reductions in
greenhouse gas emissions. BEVs extend the savings in
petroleum energy and greenhouse gas emissions
further and represent the upper limit in potential
benefits from PHEVs.

100%
90%
% of Daily VMT on CD Mode

80%
70%
60%
50%
40%
30%
20%
10%
0%
0 10 20 30 40 50 60 70 80 90 100
Miles Driven on CD Mode

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 17

Percentage of daily VMT available for substitution by a PHEV in CD mode [38]
934 Sustainable Vehicle Fuels, Well-to-Wheel Analysis

PHEV40
1.2 PHEV30
PHEV20

PHEV10
GHG Emissions (relative to conventional gasoline vehicle)

Conventional
1 Gasoline Vehicle

0.8
Regular HEV

0.6

Coal
0.4
NGCC

Renewable
0.2

0
0 0.2 0.4 0.6 0.8 1 1.2
Petroleum Use (relative to conventional gasoline vehicle)

BEV

Sustainable Vehicle Fuels, Well-to-Wheel Analysis. Figure 18

WTW petroleum use and greenhouse gas emissions for CD operation of gasoline vehicles (conventional, HEVs, PHEVs, and
BEVs)

Electric Drive Technology Life-Cycle Analysis Issues uncertainties impact fuel and electricity consumption
and Uncertainties during vehicle operation, which, in turn, impacts the
WTW results.
Electricity and Fuel Consumption of PHEVs Two
main factors impact the electricity and fuel consump-
Electricity Generation Mix for Recharging PHEVs
tion of PHEVs: (1) vehicle configuration and (2) driv-
Generation expansion planning, which optimizes
ing conditions. The vehicle configuration could assume
changes to the generator inventory, is a complex process
a variety of power ratings for the battery, engine, and
that takes into account load growth projections, the
electric motor, as well as a variety of control strategies
technical performance characteristics of current and
used in these powertrain components. Driving condi-
future generator options, and known and potential
tions include driver aggressiveness, extent of air-
changes in regulations. Existing and evolving policies
conditioning use, and cold-weather operation. Other
aimed at reduction of electricity demand, greenhouse
uncertainties include the rate of technological advance-
gas emissions, and other criteria pollutants (e.g., Energy
ment in the critical components of each vehicle tech-
Efficiency Resource Standard [EERS] and Renewable
nology (e.g., batteries and chargers), which affects the
Portfolio Standard [RPS] in many states, mercury regu-
efficiency of battery charging and discharging. Such
lations, and Clean Air Interstate Rule [CAIR]) impact

the economics and ultimately the retirement of existing
generation technologies, as well as the selection of
future generation technologies. The final inventory in
one or two decades, or more in the future, would likely
be substantially different under these constraints than
it would be in a business-as-usual case. Generation
expansion may also be influenced by the PHEV charg-
ing demand itself, and this charging demand is likely to
increase along with a general increase in transportation
energy demand. Thus, projections of generation expan-
sion become linked to projections of transportation
demand. Such transportation demand is determined
by many uncertain factors, which include the number
of PHEVs and BEVs that will be on the road in a given
region during a given year, the number of charges per
vehicle per day, the time of day for each charging event,
and the rate at which charging occurs.
Share of National Vehicle Miles Traveled (VMT)
Powered by Electricity The “share of miles electri-
fied,” which directly impacts petroleum energy savings,
depends on many factors with considerable uncer-
tainty. Adoption of PHEVs will be slow because of
low vehicle-fleet turnover rate. Furthermore, not every-
one will purchase PHEVs because they will likely com-
plement, rather than displace, HEVs. Also, PHEVs will
vary in terms of their electric range capabilities and will
have different configurations of the electric machine,
battery, and engine. The nominal electric range will not
be realized because of variations in driving conditions,
driver characteristics, accessory use, and other param-
eters. The various control strategies for utilizing the
engine and the electric motor could result in a myriad
of VMT shares in CD operation. Batteries may be
charged more than once per day and may not be fully
discharged before recharging. Such factors impact the
share of miles electrified, which determines the per-
centage of petroleum displacement, and impact the
PHEV load profile, which determines the marginal
generation mix and the corresponding greenhouse gas
emissions profile of PHEV in different regions.
Future Directions
In the 2007 US Energy Independence and Security Act,
the US Congress established goals of producing
15 billion gallons of corn-based ethanol a year by
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 935
2015 and additional 21 billion gallons of advanced
biofuels a year by 2022. In 2010, 13 billion gallons of
corn ethanol were already produced. The United States
is on the way to exceed the goal of 15 billion gallons of
corn ethanol by 2015, despite the current debate
whether corn ethanol incentive offered by the federal
government should continue or not. On the other
hand, the US government has invested through R&D,
grants, and loans in developing and building cellulosic
ethanol plants. It is anticipated that some cellulosic
ethanol plants will be in operation in the next few
years with significant amount of cellulosic ethanol to
be produced. At present, the E10 market (10% ethanol
blended into gasoline) is facing an ethanol demand
saturation point (the so-called blending wall issue).
The US EPA recently determined that ethanol blends
higher than E10 could be used in gasoline vehicles that
were produced after 2000. This finding addresses the
blending wall issue for the near future. In the long term,
flexible-fuel vehicles (FFVs) using ethanol up to E85
can expand the ethanol market. In addition, new bio-
fuel R&D efforts are being made to develop technolo-
gies that produce gasoline- or diesel-like liquid biofuels
(the so-called drop-in fuels). Drop-in fuels based on
such biofuel technologies as pyrolysis and gasification,
among other technologies, can overcome the fuels
infrastructure and vehicles incomparability problems
related to ethanol. Finally, public and private R&D
efforts are being made to develop algae-based biofuels.
Algae have a significantly high theoretically yield,
between 10 and 100 times more fuel per unit area
than other biofuel crops, and can be grown without
affecting fresh water resources. The key R&D stages
for algae-based fuels include algae growth, algae
drying, algal oil extraction, and biofuel production.
The key challenges to producing algae-based fuels
include the capturing and transportation of CO2 to
the algae farm, the drying of the algae, and the high
investment cost.
Many countries developed plans to commercialize
fuel-cell vehicles. Initiated in 2010, the H2 Mobility is
a commercialization plan for hydrogen fuel-cell vehi-
cles in Europe, which aims at identifying cost-effective
funding mechanisms to deploy hydrogen infrastructure
and FCVs, and to accelerate vehicles’ market transfor-
mation in Europe. The stakeholders in Europe joined
efforts to build up a hydrogen fueling infrastructure
936 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
and establish Germany as lead market for FCVs in
Europe. Similarly, stakeholders in Japan announced
an initiative to introduce FCVs in the Japanese market
and to develop a hydrogen supply infrastructure
starting with 100 hydrogen refueling stations by 2015.
California developed a plan for the deployment of
hydrogen FCVs and refueling stations. The plan called
for the development of 40 refueling stations statewide
in anticipation of 50,000 FCVs planned for deployment
by major car manufacturers in California by 2017. The
deployment of FCVs in California will partially satisfy
the zero-emission vehicle (ZEV) mandate, which aims
at meeting California’s mobile source emission reduc-
tion goals by requiring large automakers to produce
certain percentages of zero-emission vehicles for sale in
California. Since 2003, the US DOE has been funding
a research and development program for fuel cells with
a goal of achieving 60% peak efficiency for PEM fuel
cells at a cost of $30/kW and a durability of 5,000 h of
operation. While the PEM fuel cells achieved the effi-
ciency and durability targets, the cost target has yet to
be attained, mainly because of the expensive platinum
that is being used as a catalyst. R&D efforts are under-
way to develop catalysts for PEM fuel cells using less
expensive materials such as doped carbon nanotubes.
The US DOE is also funding R&D efforts to ultimately
achieve onboard storage targets of 2.5 and 2.3 kWh of
hydrogen per kg and per liter of FCV’s storage system,
respectively, as well as a traveling range of 300 miles.
DOE has also set a pretax cost target in the range of $2
to $4 per kg of delivered hydrogen to be competitive
with gasoline ICEVs on a cost-per-mile basis over the
vehicle’s life. Gas and energy companies are exploring
ways to produce and deliver hydrogen at a competitive
cost while satisfying the vehicle’s onboard storage
requirements and fill rate protocols.
In 2008, President Obama stated his goal of putting
one million electric vehicles on the road by 2015. To
reach that goal, $2.4 billion was allocated in the Recov-
ery Act investment to support research and develop-
ment of electric vehicles. Additionally, The US
government initiated a tax credit of $7,500 per electric
vehicle (some states such as California offer additional
credits) as an incentive for purchasing BEVs and
PHEVs. The US Department of Energy (DOE) devel-
oped an extensive energy storage R&D program with
the goal of developing a battery that could power
a vehicle 300 miles on a single charge at a cost of
$300/kWh, a durability of 15 years, and an energy
density of 150 Wh per kg of the battery system. While
great progress has been made in advanced battery tech-
nology, these targets have yet to be achieved or verified.
Meanwhile, automakers are developing vehicles that
are lighter in weight with aerodynamic designs to
decrease the required battery capacity for a given vehi-
cle’s range. Furthermore, the US federal government as
well as many US state governments have enacted poli-
cies and provided incentives to encourage electric
power generation from renewable sources such as
wind, solar, and geothermal. Other countries such as
Canada and Germany provide tax credits for wind
generation, with guaranteed grid access at prices
above average electricity prices. This will reduce the
impact of electrifying the transportation system on
the environment. The US government is also providing
funds to support building of electric vehicle infrastruc-
ture and accelerate the installation of charging stations.
To manage the added electricity demand created by
electric vehicles, the US DOE supports the develop-
ment of the smart grid concept, which involves
upgrading the existing electric grid to employ real-
time, two-way communication technologies that
allow consumers to connect directly with their electric-
ity providers.
Bibliography
1. Energy Information Administration, the U.S. Department of
Energy (2009) International energy outlook 2009. DOE/EIA-
0484, Washington, DC
2. BP Statistical review of world energy (2009) British Petroleum,
London
3. Intergovernmental Panel on Climate Change (2007) Climate
change 2007: mitigation: technical summary, working group
III report. Cambridge University Press, Cambridge
4. U.S. Environmental Protection Agency (2008) Inventory of
U.S. greenhouse gas emissions and sinks: 1990–2006.
Washington DC
5. Wang MQ (1996) Development and use of the GREET model to
estimate fuel-cycle energy use and emissions of various trans-
portation technologies and fuels. ANL/ESD-31. Center for
Transportation Research, Argonne National Laboratory,
Argonne, IL
6. Chambers RS, Herendeen RA, Joyce JJ, Penner PS (1979) Gaso-
hol: does it or doesn’t it produce positive net energy? Science
206:789–795

7. Pimentel D, Patzek TW (2005) Ethanol production using corn,
switchgrass, and wood. Biodiesel production using soybean
and sunflower. Nat Res Res 14:65–76
8. Delucchi MA (1991) Emissions of greenhouse gases from use
of transportation fuels and electricity, vol. 1. ANL/ESD/TM-22.
Center for Transportation Research, Argonne National Labo-
ratory, Argonne, IL
9. Wang MQ, Saricks C, Wu M (1997) Fuel-cycle fossil energy use
and greenhouse gas emissions of fuel ethanol produced from
U.S. Midwest corn. Prepared for Illinois Department of Com-
merce and Community Affairs by Center for Transportation
Research, Argonne National Laboratory, Argonne, IL
10. Searchinger T, Heimlich R, Houghton RA, Dong F, Elobeid A,
Fabiosa J, Tokgoz S, Hayes D, Yu TH (2008) Use of U.S. crop-
lands for biofuels increases greenhouse gases through emis-
sions from land use change. Science 319(5867):1235–1238
11. Kim H, Kim S, Dale B (2009) Biofuels, land use change, and
greenhouse gas emissions: some unexplored variables. Envi-
ron Sci Technol 43:961–967
12. Wu M, Mintz M, Wang M, Arora S (2009) Consumptive water
use in the production of ethanol and petroleum gasoline.
Environ Manage 44:981–997
13. Liska AJ, Yang HS, Bremer VR, Klopfenstein TJ, Walters DT,
Erickson GE, Cassman KG (2009) Improvements in life-cycle
energy efficiency and greenhouse gas emissions of corn-
ethanol. J Ind Ecol 13:58–74
14. Renewable Fuels Association (2011) 2011 ethanol industry
outlook: growing innovation. Washington DC
15. Wang M, Han J, Haq Z, Tyner WE, Wu M, Elgowainy A (2011)
Energy and greenhouse gas emission effects of corn and
cellulosic ethanol with technology improvements and land
use changes. Biomass Bioenergy 35:1885–1896
16. Wu M, Wang M, Huo H (2006) Fuel-cycle assessment of
selected bioethanol production pathways in the United
States. ANL/ESD/06-7. Center for Transportation Research,
Argonne National Laboratory, Argonne, IL
17. Wang M, Wu M, Huo H, Liu J (2008) Well-to-wheels energy use
and greenhouse gas emissions of Brazilian sugarcane ethanol
production simulated by using the GREET model. Int Sugar J
110(1317):527–545
18. Loarie SR, Lobell DB, Asner GP, Mu Q, Field CB (2011) Direct
impacts on local climate of sugar-cane expansion in Brazil. Nat
Clim Change Lett 1:105–109. doi:10.1038/NCLIMATE1067
19. National Biodiesel Board (2009) Estimated U.S. biodiesel pro-
duction. http://www.biodiesel.org/pdf_files/fuelfactsheets/
Production_Graph_Slide.pdf. Accessed 3 March 2009
20. Huo H, Wang M, Bloyd C, Putsche V (2008) Life-cycle assess-
ment of energy and greenhouse gas effects of soybean-
derived biodiesel and renewable fuels. ANL/ESD/08-2. Center
for Transportation Research, Argonne National Laboratory,
Argonne, IL
21. CARB (California Air Resources Board) (2009) Proposed regu-
lation for implementing low carbon, volume1, staff report:
initial statement of reasons, Sacramento, CA
Sustainable Vehicle Fuels, Well-to-Wheel Analysis 937
22. EPA (U.S. Environmental Protection Agency) (2010) Regula-
tions of fuels and fuels additives: changes to renewable fuel
standard program; final rule, Federal Register March 26, vol 5,
No 58, pp14670–14904
23. Hertel TW, Golub AA, Jones AD, O’Hare M, Plevin RJ, Kammen DM
(2010) Effects of US maize ethanol in global land use and green-
house gas emissions: estimating market-mediated responses.
Bioscience 60:223–231
24. Tyner WE, Taheripour F, Zhuang Q, Birur D, Baldos U (2010)
Land use changes and consequent CO2 emissions due to
US corn ethanol production: A comprehensive analysis,
Department of Agricultural Economics, Purdue University,
Indiana
25. IPCC (Intergovernmental Panel on Climate Change)
(2006) 2006 IPCC guidelines for national greenhouse gas
inventory: vol 4 — agriculture, forestry, and other land use.
Hayama, Japan
26. Wang M, Huo H, Arora S (2010) Methodologies of dealing with
coproducts of biofuels in life-cycle analysis and consequent
results within the U.S. context. Energy Policy. doi:10.1016/j.
enpol.2010.03.052
27. Paster MP et al (2011) Hydrogen storage technology options
for fuel cell vehicles: Well-to-wheel costs, energy efficiencies,
and greenhouse gas emissions, Int J Hydrogen Energy,
doi:10.1016/j.ijhydene.2011.07.056
28. Wang M (2002) Fuel choices for fuel-cell vehicles: well-to-
wheels energy and emission impacts. J Power Sources
112:307–321
29. Elgowainy A, Marianne Mintz M, Steward D, Sozinova O,
Brown D, Gardiner M (2010) Liquid hydrogen production
and delivery from a dedicated wind power plant. Center for
Transportation Research, Argonne National Laboratory,
Argonne, IL
30. Chen TP, Paster M, Gardiner M, Elgowainy A, Ringer M, Kelly B,
Lasher S, McKenney K (2008) Hydrogen delivery infrastructure
options analysis: Task 2, appendix B, http://www1.eere.energy.
gov/hydrogenandfuelcells/pdfs/delivery_infrastructure_analysis.
pdf. Accessed on 31 Aug 2011
31. Wipke K, Sprik S, Kurtz J, Ramsden T (2010) Learning demon-
stration interim progress report. Technical Report NREL/TP-
560-49129. NREL, CO
32. Mintz M, Gillette J, Elgowainy A, Paster M, Ringer M,
Brown D, Li J (2006) A hydrogen delivery scenario analysis
model to analyze hydrogen distribution options: HDSAM,
Paper No. 06–3085, Transportation Research Record series,
TRB
33. Kromer MA, Heywood JB (2007) Electric powertrains: oppor-
tunities and challenges in the U.S. light-duty vehicle fleet.
Publication No. LFEE 2007–03 RP, May 2007. Massachusetts
Institute of Technology (MIT), Cambridge MA
34. Thomas CE (2009) Cost-benefit analyses of alternative light-
duty transportation options for the 21st century. National
Hydrogen Association Conference, Columbia, SC, 31 March
2009
938 Sustainable Vehicle Fuels, Well-to-Wheel Analysis
35. National Research Council (2009) Transitions to alternative
transportation technologies — plug-in hybrid electric vehi-
cles. The National Academies Press, Washington DC. ISBN
0-309-14851-0
36. Passier G, Conte FV, Smets S, Badin F, Brouwer A, Alaküla M,
Santini D, Alexander M (2007) Status overview of hybrid and
electric vehicle technology. MON-RPT-033-DTS-2007-02955,
Final Report Phase III, Annex VII, IA-HEV, IEA, TNO Report.
Delft, Netherlands. IEA, 6 Dec 2007
37. Electric Power Research Institute and Natural Resources
Defense Council (2007) Environmental assessment of plug-in
hybrid electric vehicles. Vol 1: Nationwide greenhouse gas
emissions. EPRI Report# 1015325, July 2007. EPRI, USA
38. Elgowainy A, Han J, Poch L, Wang M, Vyas A, Mahalik M,
Rousseau A (2010) Well-to-wheels analysis of energy use and
greenhouse gas emissions of plug-in hybrid electric vehicles.
ANL/ESD/10-1. Center for Transportation Research, Argonne
National Laboratory, Argonne, IL
39. Shidore N, Bohn T, Duoba M, Lohse‐Busch H, Sharer P (2007)
PHEV ‘all electric range’ and fuel economy in charge
sustaining mode for low SOC operation of the JCS VL41M
Li‐ion battery using battery HIL. In: Proceedings of the
electric vehicle symposium 23, Anaheim, CA, 2–5 December
2007
40. U.S. Environmental Protection Agency (2006) Fuel economy
labeling of motor vehicle revisions to improve calculation of
fuel economy estimates, Final Technical Support Document,
EPA420-R-06-017, Dec 2006
41. Walsh K (2009) Comments on proposed rulemaking to estab-
lish light duty vehicle greenhouse gas emission standards and
corporate average fuel economy standards. Fisker Automo-
tive, Inc. Anaheim CA, 3 Dec 2009
42. Larrick RP, Soll JB (2008) ECONOMICS: the MPG illusion, policy
forum. Science 320(5883):1593–1594
43. Shelby M, Mui S (2007) Plug-in hybrids: a scenario analysis.
Presented at the 86th Annual Meeting of the Transportation
Research Board, Washington, DC, 22 Jan 2007
44. U.S. Department of Energy, Office of Hydrogen, Fuel Cells
and Infrastructure Technologies (2007) Multiyear Research,
Development and Demonstration Plan. http://www1.eere.
energy.gov/hydrogenandfuelcells/mypp/. Accessed on 31
Aug 2011
 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV
Switched Reluctance Motor Drives
for Propulsion and Regenerative
Braking in EV and HEV
MEHRDAD EHSANI
Department of Electrical & Computer Engineering,
Texas A&M University, College Station, TX, USA
Article Outline
Glossary
Definition of the Subject
Introduction
Advantage of the Extended Constant Power Range for
Propulsion
Extended Constant Power Range for Regenerative
Braking
Switched Reluctance Motor Drive for Extended
Constant Power Operation
Future Directions
Bibliography
Glossary
Drive cycle A profile of vehicle speed versus time
Extended constant power range Operation of the
electric motor beyond its rated speed by attempting
to keep its power constant
Power electronic converter The electronic circuit for
controlling voltages and currents for power control
Regenerative braking The use of electric machine as
generator for vehicle braking force
Switched reluctance motor A form of variable reluc-
tance motor with discrete air gap magnetic fluxes
and currents
X factor The ratio of maximum to rated motor
speeds
Definition of the Subject
Electric and hybrid electric vehicles (EV and HEV) are
considered an essential element of sustainable energy
and transportation complex. This is mainly due to their
high efficiency and low emissions. The specific
technology and characteristics of the traction motor
are essential for the performance and efficiency of the
EV and HEV. Various motor technologies have been
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
939
considered and used in EV and HEV drives over the
past 100 years. This choice was mainly governed by the
available motor and electronic controls available at
the time and also on the deeper understanding of the
nature of electrical vehicle traction. The historical
progression of the traction motor technologies has
roughly been dc motors, induction motors, and lately,
permanent magnet motors.
This article introduces a new basis for the selection
of vehicle traction motor technology. On this basis, the
modern switched reluctance motor drive technology is
shown to be very attractive for future EV and HEV
applications.
Introduction
Electric and hybrid electric vehicles are attractive as
candidates for sustainable transportation. This is
mainly due to their high efficiency and low emission.
In EV and HEV, the electric motor is used for both
propulsion and regenerative braking. Therefore, the
selection of an electric motor for the EV and HEV
should be based on both its propulsion and regenera-
tive braking characteristics.
Various motor technologies have been considered
and used in EV and HEV drives over the past 100 years.
This choice was mainly governed by the available motor
and electronic control technologies at the time and
also on the understanding of the nature of electrical
vehicle traction. The historical progression of the
traction motor technologies has roughly been dc
motors, induction motors, and lately, permanent magnet
motors.
Vehicle dynamics, such as initial acceleration,
indicates the power rating of the propulsion motor. It
has been shown by our research group that a motor
with an extended speed range at constant powercan
minimize the power rating of the propulsion system
for achieving the same acceleration performance [1].
Furthermore, a motor with an extended constant power
range can also recover more kinetic energy during
regenerative braking. Therefore, a motor with extended
constant power speed range is advantageous for EV and
HEV traction application.
For the above reasons, the switched reluctance motor
(SRM) is considered a promising candidate for traction
# 2012, DOI 10.1007/978-1-4419-0851-3
940 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV

in EV and HEV. In addition to having a wide speed
range, especially at constant power, these motors have
a rugged and low-cost structure, reliable drive power
converter topology, and high efficiency over a wide
range of speed. However, in order to obtain an
extended constant power range, the SRM must be
especially designed. Special control design will then
meet the requirements for recovering most of the
kinetic energy during regenerative braking.
This article introduces the concept of extended
speed range at constant power and its benefit in vehicle
traction application. Then, the advantage of switched
reluctance motor for meeting the required extended
speed range is introduced.
Advantage of the Extended Constant Power
Range for Propulsion
The initial vehicle acceleration requirement is the main
determinant of the power rating of the propulsion
system. A traction motor with constant power over an
extended speed range will result in the minimal power
rating of the propulsion system for a specified initial
acceleration. This has been proven by our group at
Texas A&M University and is shown in [1] and briefly
reviewed in this section.
Figure 1 illustrates a typical torque-speed character-
istic of any electric motor. In this figure, Vrm, Vrv , and
Vmax represent the rated speed of the motor, the rated
Constant
Torque
Force
speed of the vehicle, and the maximum speed of the
vehicle, respectively. When the vehicle accelerates from
zero to Vrv (for example, 100 km/h or 60 mph), the
maximum traction force on the driven wheels, Fm, is
limited to a constant design value up to speed Vrm.
Above this speed, the maximum traction force is lim-
ited by Pm/V, where Fm and Pm represent the maximum
traction force and power of the motor, respectively.
It can be shown that the motor power rating can be
represented by:
m
Pm ¼ ðVrm 2 þ Vrv 2 Þ ð1Þ
2tf
where m stands for vehicle speed.
Minimizing this motor power with the design
value, Vrm means
@Pm =@Vrm ¼ 0 ð2Þ
Therefore,
Vrm ¼ 0 ð3Þ
Equation 3 indicates that the power rating of
the propulsion system will be minimized for a given
design specification of going from standstill to Vrv in tf
seconds. This means that the motor runs entirely in
constant power regime. However, it is physically
impossible to run a motor entirely in the constant
power range, from zero speed to its maximum speed.
Therefore, a motor with a long speed range in constant
power is preferred for minimizing the power rating of
the propulsion system.
Here a realistic vehicle example to illustrate the
point is given. The parameters of the vehicle are
shown in Table 1.

Constant Switched Reluctance Motor Drives for Propulsion and

Power Regenerative Braking in EV and HEV. Table 1 Parame-
ters of the studied vehicle
Vehicle mass (kg) 1,500
Vrm Vrv Vmax
Radius of the wheel (m) 0.2794
Vehicle Speed
Aerodynamic drag coefficient 0.3
Switched Reluctance Motor Drives for Propulsion and Rolling resistance coefficient 0.009
Regenerative Braking in EV and HEV. Figure 1 Front area (m )2
2.0
Torque-speed characteristic of the electric motor in terms 3
Air density (kg/m ) 1.225
of traction force and vehicle speed

Different traction motor technologies, with differ-
ent ratios of maximum speed to rated speed, Vmax/Vrm
(referred to as the X factor), are used for accelerating the
above vehicle from standstill to 100 km/h (62.5 mph) in
15 s. Ratios of 2:1, 3:1, 4:1, 5:1, and 6:1 are considered.
The required motor and vehicle drive train power rat-
ings for each motor are shown in Table 2.
Extended Constant Power Range for
Regenerative Braking
In EV and HEV, when the electric braking system
can provide high enough braking force, the mechanical
braking is not used, and all the kinetic energy can
be recovered by regenerative braking. Otherwise,
the mechanical braking system has to be added to
Switched Reluctance Motor Drives for Propulsion and
Regenerative Braking in EV and HEV. Table 2 Required
power ratings of the propulsion system versus the X factor
of the motor for achieving the same acceleration
X ratio 2:1 3:1
Power rating (kW) 90 66
140
120
100
Vehicle speed, km/h
80
60
40
20
0
0 100
Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 2
Vehicle speed variation in US06 drive cycle
Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV 941
provide enough braking force on the wheels. Thus,
some kinetic energy is dissipated in the form of heat
(Fig. 2).
To find the optimal motor torque-speed character-
istic for braking, the required braking force distribu-
tion of the above vehicle at different speeds is studied
on a standard vehicle drive cycle called US06 (com-
bined, high-speed, and high-acceleration driving cycle)
and is plotted in Fig. 3. The torque-speed characteris-
tics of motors with different X factors that are capable
of accelerating the vehicle from standstill to 100 km/h
in 15 s are also plotted in Fig. 3.
Figure 3 shows that more braking force is used at low
speeds and less braking force is used at high speed on
this drive cycle. Figure 3 also shows that when the
X factor is less than 4:1, the motor cannot provide
enough braking force at low speeds. Therefore, more
mechanical braking has to be used to slow down the
vehicle, and more kinetic energy is lost in the form of
heat. In this example, when the X factor is higher than
4:1, the motor can provide adequate braking force at
both low and high speeds. Therefore, more of the
4:1 5:1 6:1
kinetic energy can be recovered by regenerative brak-
ing. Figure 4 shows the kinetic energy that is not recov-
57.8 54 52.3
ered for each motor X factor.
200 300 400 500 600
Time, Sec
942 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV

6
6:1

5
5:1

4:1
Braking Force, kN 4
3:1
2:1
3

0
0 20 40 60 80 100 120 140 160 180 200
Vehicle Speed, km/h

Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 3
Braking force distribution at different speeds in US06 drive cycle and the braking force-speed profile of the motors with
different X factors capable of achieving the same initial acceleration

50
Energy Unecovered(KJ) in One Drive Cycle

45
40
35
30
25
20
15
10
5
0
1:1 2:1 3:1 4:1 5:1
X Ratio

Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 4
Energy not recovered in one drive cycle by motors with different X factors

To obtain the required extended constant power An example of a motor technology with limited constant
range, a multi-gear transmission may be used in combi- power range is the permanent magnet motor. However,
nation with a motor with a narrow constant power range. the multi-gear transmission complicates the structure
 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV 943

Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 5
Cross sections of rotor and stator of a 4-phase 8/6 SRM

2-phase, 4-phase,
a 4 rotor poles/2 stator poles b 8 rotor poles/6 stator poles

3-phase, 5-phase,
c 6 rotor poles/4 stator poles d 10 rotor poles/8 stator poles

Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 6
Geometries of various switched reluctance motors

and lowers the efficiency of the propulsion system.
Further, if automatic transmission is used, the poor effi-
ciency of the hydraulic torque converter further aggra-
vates the power efficiency of the vehicle traction system.
Switched Reluctance Motor Drive for Extended
Constant Power Operation
The switched reluctance motor can more easily have
the necessary extended constant power range without
944 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV

the need for a multi-gear transmission. Therefore, the
SRM is an attractive candidate for the traction motor in
EV and HEV.
Mechanical Structure of SRM
The switched reluctance motor has salient poles both
on the stator and rotor. It has concentrated windings
on the stator and no winding on the rotor. In Fig. 5,
cross sections of stator and rotor of a 4-phase 8/6 SRM
is shown. The geometries of a few other types of SRM
are shown in Fig. 6.
Electrical Structure of SRM
Figure 7 shows one common structure of the power
electronic converterconnected to a 4-phase SRM. Each
leg of the converter drives one phase of the SRM in
a sequential manner, based on the instantaneous posi-
tion of the rotor.
Figure 8 shows the ideal phase inductance,
current, and torque of the SRM. Positive (motoring)
torque is produced if the phase is excited when the
phase inductance is increasing as the rotor rotates.
Negative (generating) torque is produced if the

B⬘ C⬘

A⬘ F A
F

C B

Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 7
Construction of SRM drive

Ideal Phase
Inductance

Ideal Phase Current

(Motoring Operation)

Ideal Torque
(Motoring Operation)

Ideal Phase Current

(Regeneration Operation)

Ideal Torque
(Regeneration Operation)

Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 8
Idealized inductance, current, and torque profiles of the SRM
 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV
τ=2π/Nr, Nr: Number of rotor poles;
α: Rotor pole arc;
β: Stator pole arc;
θ: Rotor position angle.
Phase inductance
θ1
τ−α
Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV. Figure 9
Inductance profile of the SRM
phase is excited when the phase inductance is
dropping as the rotor moves. This implies that the
position information is necessary for control of the
SRM drive.
SRM can inherently operate with extremely long
constant power range. This long constant power
range makes SRM highly favorable for vehicle traction
application. In addition, recently developed full speed
sensorless control technology will make SRM a more
attractive candidate for military vehicle traction appli-
cation [2–8].
Figure 9 shows the details of phase inductance
versus rotor position angle of the SRM. By properly
choosing small pole arcs, large air gap at unaligned
position, and small air gap at aligned position, one
can make the inductance profile of the SRM have
a large dead zone at the unaligned position while
making the SRM able to provide the required rated
power. In vehicle traction application, this large
dead zone in the inductance profile enables large
phase advancing control of SRM so that enough
phase current, hence enough torque, can be built
in the SRM when the rotor speed is above the base
speed. Some examples of the traction SRM drive for
achieving extended constant power speed range are
presented in [9].
945
θ2 θ3 θ4 θ5
Min(α,β) | α−β | Min(α,β) θ
τ
The steep SRM inductance slope, shown in Fig. 9,
can be used to produce a high electric torque both
in propulsion and regenerative braking operation.
In regenerative braking operation, when the motor
speed is below the base speed, the phase current can
be easily adjusted to produce the required braking
torque. One can find the relation between the phase
current and the electric torque with an artificial neural
network using the data obtained from numerical sim-
ulation. Neural networks can be used to find the rela-
tion between the turn-off rotor position and the phase
current amplitude and torque using the data obtained
from numerical simulation. This neural-network-
based SRM control strategy for regenerative braking
in EV and HEV is presented in [10].
Future Directions
The advantage of the motor with an extended constant
power speed range for both propulsion and regenerative
braking operation in electric and hybrid electric
vehicle applications was presented. The switched reluc-
tance motor is an attractive candidate for the traction
motor of EV and HEV due to its inherent capability of
having an extended constant power range, with
proper design and control. The basic design and
946 Switched Reluctance Motor Drives for Propulsion and Regenerative Braking in EV and HEV
control strategies of the SRM drive for achieving the
extended constant power range in both propulsion and
regenerative braking operation were also presented in
this article.
At the present, the SRM technology suffers from
a few drawbacks. The most important of these is the
dearth of manufacturers and suppliers. The others are
the inherent noise and vibrations that these motors
produce. However, these problems have been shown
to be surmountable.
In time, rare earth permanent magnets, used in
permanent magnet motors, will become more scarce
and expensive, and the advantages of the SRM for
traction will become better understood by the vehicle
designers and manufacturers. This may lead to the
evolution of EV and HEV traction motor technology
to switched reluctance motors.
Bibliography
1. Ehsani M, Rahman KM, Toliyat HA (1997) Propulsion system
design of electric and hybrid vehicles. IEEE Trans Ind Electron
44(1):19–27
2. Ehsani M (1990) Position sensor elimination technique for the
switched reluctance motor drive, US Patent 5072166
3. Ehsani M (1994) Phase and amplitude modulation techniques
for rotor position sensing in switched reluctance motors,
US Patent No. 5,291,115
4. Ehsani M, Husain I, Mahajan S, Ramani KR (1994) New modula-
tion encoding techniques for indirect rotor position sensing in
switched reluctance motors. IEEE Trans Ind Appl 30(1):85–91
5. Dunlop GR, Marvelly JD (1987) Evaluation of a self commuted
switched reluctance motor. In: Proceeding of 1987 electric
energy conference, Adelaide, pp 317–320
6. Ehsani M, Husain I (1992) Rotor position sensing in switched
reluctance motor drives by measuring mutually induced volt-
ages. In Conference Record of 1992 IEEE Industry Application
Society Annual Meeting, vol 1, Houston, pp 422–429
7. Lumsdaine A, Lang JH (1990) State observer for variable reluc-
tance motors. IEEE Trans Ind Electron 37(2):133–142
8. Bass JT, Ehsani M, Miller TJE (1987) Simplified electronics
for torque control of sensorless switched reluctance motor.
IEEE Trans Ind Electron 34(2)
9. Rahman KM, Fahimi B, Suresh G, Rajarathnam AV, Ehsani M
(2000) Advantages of switched reluctance motor applications
to EV and HEV: design and control issues. IEEE Trans Ind Appl
36(1):111–121
10. Gao H, Gao Y, Ehsani M (2001) A neural network based
SRM control strategy for regenerative braking in EV and HEV.
To be presented in IEEE International Electric Machine Design
and Control Conference 2001, Massachusetts, Cambridge,
June 2001
11. Ehsani M, Gao Y, Butler KL (1999) Application of electrically
peaking hybrid (ELPH) propulsion system to a full-size passen-
ger car with simulated design verification. IEEE Trans Veh
Technol 48(6):1779–1787
12. Ehsani M et al. (2001) Design issues of the switched reluctance
motor drive for propulsion and regenerative braking in EV and
HEV. In: Proceedings of the SAE-FTT’01, Costa Mesa
13. Gao Y, Rahman KM, Ehsani M (1997) Parametric design of
the drive train of an electrically peaking hybrid (ELPH) vehicle,
SAE International SP-1234, Paper No. 970294
14. Gao Y, Chen L, Ehsani M (1999) Investigation of the effective-
ness of regenerative braking for EV and HEV. SAE International
SP-1466, Paper No. 1999-01-2910, August 1999
 Transit-Oriented Development and Land Use 947

Transit-Oriented Development and travel, including transit usage, and that promotes
environmental conservation and which can include,
Land Use but is not limited to, transit-oriented development.
ROBERT CERVERO Transit value capture Securing profits from increases in
Department of City and Regional Planning, land values conferred by public transit investments to
University of California, Berkeley, CA, USA help finance capital costs, including fixed guideways,
rolling stock, and neighborhood improvements.
Article Outline
Definition of the Subject
Glossary
Definition of the Subject Transit-oriented development, or TOD, is a term used
Introduction to describe the physical integration and linkage of
The Scope of TOD public transportation investments and urban land
Bus-Based TOD: Bogotá and Curitiba development at or near a station. TOD typically takes
America’s TOD Success Story: Arlington County, the form of compact, mixed land-use development
Virginia with high-quality pedestrian environments that is con-
The Transportation and Environmental Benefits of centrated around rail transit nodes [1]. TODs can also
TOD occur around the stations and stops of busways and bus
TOD and Ridership at the Workplace services that operate on exclusive, dedicated lanes,
TOD and Parking often referred to as Bus Rapid Transit (BRT).
TOD Implementation Tools The chief objective of TOD is to encourage larger
Central-City Redevelopment Around Streetcar Lines shares of motorized trips to be taken by public trans-
The Challenges of Mixed-Use TODs portation and as a result to reduce traffic congestion
Financing TODs and improve environmental conditions, including air
Future Directions quality (since the private automobile is the primary
Bibliography source of ozone, carbon monoxide, and particulate
matter emissions in many cities of the world). Second-
Glossary ary objectives of TOD might include revitalizing
a long-distressed urban district, encouraging the pro-
Bus rapid transit High-quality bus services that mimic
duction of more affordable housing and curbing the
some of the mobility benefits of exclusive-guideway
consumption of agricultural land and open space.
rail transit services but at a fraction of the cost, includ-
ing busways, dedicated-lane services, and buses that
Introduction
operate with special features like signal prioritization
and queue-jumper lanes at busy intersections. TOD is often viewed as an antidote to car-dependent
Joint development Development activities on or near sprawl. By attracting a mix of residences, businesses,
a transit agency’s property that generates revenue shops, and civic activities within a quarter-mile walking
and ridership benefits. distance of an urban railway station, proponents argue
Public transportation Common-carrier forms of that TOD can draw people to transit and thus relieve
mass transportation available as a mobility service traffic congestion, improve air quality, and contribute
to the general populous, including heavy rail, light toward climate stabilization. The station and its imme-
rail, and bus transit. diate surroundings can also serve as the hub of
Self selection Willful decisions of residents to move a community – a focal point for organizing the regener-
into particular neighborhoods for lifestyle reasons ation of stagnant neighborhoods and in the case of
and thereby have a bearing on travel behavior. greenfields, the design and construction of new ones.
Smart growth Compact, mixed-use urban develop- Thus TOD is championed for both its environmental
ment that encourages alternatives to automobile and its place-making benefits. Of course, the two are not

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
948 Transit-Oriented Development and Land Use
unrelated. Traffic-snarled districts hardly make attrac-
tive, livable places. And as discussed later, places that are
conducive to transit riding can help free-up traffic jams.
TOD has gained popularity in the United States, the
world’s most car-dependent country, for three key rea-
sons. One, it is arguably the most cogent, understand-
able form of “smart growth.” Citizens, politicians, and
city-planners alike understand that if there is a logical
place to concentrate urban growth, it is in and around
transit stations. Second, demographic and lifestyle
trends are working in favor of TOD. Living around
transit stations appeals to growing numbers of
Americans, like childless couples, Generation X’ers,
and empty-nesters who value convenience and access
to high-quality transit and who place a premium on
being in a walkable community with urban amenities.
Recent studies have rated TOD as a top real estate
investment, estimating that as many as one- third of
newly formed households in rail-served US metropol-
itan areas are receptive to transit-oriented living [2].
For many, TOD residency is a quality-of-life issue. As
one study put it, “TODs are all about creating choice
for people, about housing, lifestyle and travel”[3].
Third, more and more public policy initiatives favor
TODs. As the federal, state, and local government
levels, funding programs and incentives in the form of
regulatory relief are favoring public and private invests
in and around transit stations.
The Scope of TOD
A national study in the United States found that in 2003
about 100 of the nation’s 3,300 fixed rail transit sta-
tions, or around 3%, could be considered TODs – in
the sense all had a mix of land uses, embodied urban
designs that created safe and attractive walking envi-
ronments, and met minimum density thresholds (at
least 12 dwelling units and 50 workers per net acre of
land) [4]. Far more common than TODs, however, are
what have been called “Transit adjacent developments,”
or TADs. TADs comprise buildings that are near transit
stops but fail to induce residents, workers, and shop-
pers to patronize transit because of their designs and
physical orientations. A mid-rise office building with
standard parking supplies whose main entrance faces
away from the nearby train station, forcing patrons to
take a round-about path, is more TAD than TOD. By
one definition, TAD is “physically near transit (but)
fails to capitalize upon this proximity. . .(It) lacks any
functional connectivity to transit – whether in terms of
land-use composition, means of station access, or site
design” [5].
The trade-off between creating a TOD versus a TAD
reflects the inherent conflict between the transit station
as “nodes” versus “places.” On the one hand, stations
are logistical nodes wherein cars, buses, taxis, delivery
trucks, pedestrians, and cyclist converge for accessing
transit and allowing intermodal transfers. On the other
hand, stations and their environs are places for creating
or rebuilding community hubs. Whenever the logistical
needs of a station win out, the resulting road designs
and parking layouts often detract from the quality of
walking, creating more of a TAD than a TOD. This has
often been the case in and around suburban rail transit
stations in the United States.
On the global stage, TOD is most fully developed in
Europe, and in particular Scandinavia. In cities like
Copenhagen, Denmark, and Stockholm, Sweden, cor-
ridors for channeling overspill growth from the urban
centers were identified and reserved early in the plan-
ning process, and rail infrastructure was built, often in
advance of demand, to steer growth along desired
growth axes [6]. As importantly, greenbelt wedges set
aside as agricultural preserves, open space, and natural
habitats were also designated and accordingly major
infrastructure was directed away from these districts. In
the case of Stockholm, the last half-century of strategic
regional planning has given rise to a regional settlement
and commutation pattern that has substantially
lowered car-dependency in middle-income suburbs.
Stockholm planners focused on creating jobs-housing
balance along rail-served axial corridors. This in turn
has produced directional-flow balances. During peak
hours, 55% of commuters are typically traveling in one
direction on trains and 45% are heading in the other
direction [6]. Moreover, Stockholm’s transit modal
share is nearly twice that found in bigger rail-served
European cities like Berlin and even higher than inner
London’s market share. In fact, Stockholm is one of the
few places where automobility appears to be receding.
Between 1980 and 1990, it was the only city in a sample
of 37 global cities that registered a per capita decline in
car use – a drop off of 229 annual kilometers of travel
per person [7].
 Transit-Oriented Development and Land Use 949

Bus-Based TOD: Bogotá and Curitiba much bigger Rio de Janeiro and São Paulo [6]. It also
boasts the cleanest air among any Brazilian city over
While rail-based TODs are most prevalent worldwide,
1 million inhabitants, despite being a provincial capital
in good part because rail transit delivers the most
with a sizable industrial sector. The strong, workable
regional accessibility benefits and thus are attractive
nexus that exists between Curitiba’s bus-based transit
to private investors, in recent time TODs are also taking
system and its mixed-use linear settlement pattern
form along Bus Rapid Transit (BRT) corridors. Many
deserves most of the credit.
medium-sized global cities are looking to BRT as the
Bogotá, the Andean capital of Colombia, has gained
most affordable form of high-performance public tran-
global recognition for its highly efficient and produc-
sit investment. BRT aims to achieve the speed and
tive bus rapid transit (BRT) system. For a city of
performance advantages of grade-separated services at
7.5 million inhabitants facing civil conflict and deep
a fraction of the cost by cleverly using bus-based
economic problems, Bogotá’s emergence as one of the
approaches. Among its key features are: exclusivity,
world’s most sustainable metropolises is all the more
notably physical segregation; seamless (same-level)
remarkable. In the late 1990s, Bogotá began operating
transfers; advanced bus technology: clean fuels, light-
a high-speed, high-capacity bus system, called
weight materials, low floors, advanced communica-
Transmilenio, building upon Curitiba’s successes with
tions, docking systems; supportive armature: signal
dedicated busways. A big difference, however, is that
priorities, bus turnouts, curb realignments, automated
Curitiba relies principally upon circular, crosstown bus
vehicle location (AVL) systems, automated routing and
routes to interconnect radial busways. Outside of
dispatching; and expeditious fare collection and
downtown, relatively little was invested in pedestrian
boarding: off-vehicle payment, smart cards. Two note-
and bikeway improvements. Bogotá, on the other
worthy international experiences with BOT and TOD,
hand, actively embraced pedestrian and bicycle access.
both in Latin America, are Curitiba, Brazil and Bogotá,
The 42-km, 3-line Transmilenio busway is the cen-
Colombia.
terpiece of Bogotá’s vast bus network. (The dedicated
The environmental benefits of balancing urban
system will eventually expand to 22 lines covering 391
growth along bus-served linear axes and aggressively
km.) Bus lanes are situated in boulevard medians, with
pursuing a “transit first” policy are underscored by
weather protected, attractively designed stations spaced
experiences in Curitiba, Brazil. Curitiba, widely viewed
every 500 m or so. Because of dual carriageways that
as one of the world’s most sustainable, well-managed
enable buses to overtake each other and high-level plat-
metropolises, is also one of the most accessible –
forms that allow expeditious boardings and alightings,
a product of some 40 years of carefully integrating
Transmilenio has a throughput of some 35,000 persons
urbanization and transportation improvements. By
per direction per hour, a number that matches that of
emphasizing planning for people rather than cars,
many Metrorail system. Some one million passengers
Curitiba has evolved along well-defined radial axes
ride Transmilenio buses each weekday, three times the
that are intensively served by dedicated busways.
ridership of two rail lines in Medellin, Colombia
Along some corridors, streams of double-articulated
(achieved at less than one-fifth of the Medellin Metro’s
buses haul 16,000 passengers per hour, comparable to
construction costs) and according to one study provid-
what much pricier Metrorail systems carry. A design
ing for a social rate of return of 61% [8].
element used to enhance accessibility in Curitiba is the
Particularly important to the transitway has been
“trinary” – three parallel roadways with compatible
Bogotá’s attention to pedestrian and bicycle access, in
land uses. An important benefit of mixed land uses
the form of “green connectors.” Perpendicular and
and transit service levels along these corridors, besides
grade-separated pedways and bikeways connect some
phenomenally high ridership rates, has been balanced,
of the poorest barrios and informal housing settle-
bi-directional flows, ensuring efficient use of available
ments (with highly transit-dependent populations) to
bus capacity, just as in the case of Stockholm. On a per
the busways. Other innovative features of Bogotá’s sus-
capita basis, Curitiba is one of Brazil’s wealthiest city
tainable transport program include license-plate
yet it averages considerably more transit trips than
rationing, parking management, and car-free
950 Transit-Oriented Development and Land Use
districting. Bogotá is an extraordinary example of
matching infrastructure “hardware” with public-policy
“software.” The city boasts Latin America’s most exten-
sive network of cycleways (250 km), the world’s longest
pedestrian corridor (17 km), and the planet’s biggest
Car Free Day (covering an entire city of 35,000 ha).
While Bogotá’s Transmilenio system has not dra-
matically altered the cityscape to date, at least when
compared to places like Curitiba, research shows that
commercial properties have reaped benefits from prox-
imity to busway stations. A hedonic price- model anal-
ysis found a monthly rental discount of 1.87% for every
additional 0.1 km from a BRT station, all else being
equal [9]. This suggests a pent-up market demand for
the accessibility benefits conferred by high-quality bus-
based transit in cities of developing countries. A more
recent analysis found that land-value capitalization
increased as Transmilenio expanded into new neigh-
borhood, revealing the network benefits of adding
regional connectivity to the system [10].
As with many successful transit investments, it has
been the attention to design details, matched by good
macro-scale planning, that has contributed to
Tranmilenio’s success [11]. Car parking is mainly lim-
ited to the end stations of the Transmilenio busway.
Nearly half of the 57 intermediate stations are served by
skywalks/pedestrian overpasses. A phalanx of sidewalks
and bikeways feed into all stations, most embellished by
vegetative landscaping. Some two dozen civic plazas,
pocket parks, and recreational facilities lie within
a half kilometer of busway stops. Today, an estimated
45% of Transmilenio users reach stations by foot or
bicycle [11].
America’s TOD Success Story: Arlington County,
Virginia
No place in the United States has witnessed more mid-
to-high rise, mixed-use development along a rail cor-
ridor over the past three decades than Arlington
County, Virginia [4]. This 26 square mile county just
south of the nation’s capital has experienced
a tremendous increase in building activity since
Washington Metrorail’s 1978 opening: more 25 million
square feet of office space, 4+ million square feet of
retail space, some 25,000 mixed-income dwelling units,
and over 6,500 hotel rooms. Of the nearly 190,000
people today living in Arlington County, in the early
2000s, 26% resided within a Metrorail-served corridor
(roughly a one-quarter mile walkshed of stations), even
though these corridors comprised only 8% of county
land area. If the development added to these two cor-
ridors had been built at suburban density standards,
such as in neighboring Fairfax County, Virginia, seven
times as much land area would have been required [4].
Arlington County’s TODs have hardly been
the result of good fortune or happenstance.
The transformation of once-rural Arlington County
into a showcase of compact, mixed-use TOD has been
the product of ambitious, laser-focused station-area
planning and investment. Prior to Metrorail’s arrival,
Arlington County planners understood that high-
performance transit provided an unprecedented
opportunity to shape future growth and proceeded to
introduce various strategies – targeted infrastructure
improvements, incentive zoning, development prof-
fers, and permissive and as-of-right zoning – to entice
private investments around stations. After preparing
countywide and station-area plans on desired land-
use outcomes, density and setback configurations,
and circulation systems, zoning classifications were
changed and developments that complied with these
classifications could proceed unencumbered. The
ability of complying developers to create TODs
“as-of-right” was particularly important for it meant
developers could line up capital, secure loans, incur
upfront costs, and phase-in construction without the
fear of local government “changing its mind.” Another
key factor was the decision not to align Arlington
County’s Metrorail rail corridor in the median of Inter-
state-66, thus suppressing development potential.
Instead, County officials persuaded the region’s transit
authority to align the corridor in the traditional urban
centers to help jump-start the process of urban
regeneration.
The pay-off of concentrated growth along rail corri-
dors is revealed in Arlington County’s transit ridership
statistics. The County today boasts one of the highest
percentages of transit use in the Washington, DC.
region, with nearly 40% of Metrorail corridor residents
commuting to work by public transit [4]. An important
outcome of promoting mixed-use development along
Arlington County’s rail corridors has been balanced
jobs and housing growth which in turn has produced

balanced two-way travel flows. Counts of station entries
and exits in Arlington County were nearly equal during
peak hours as well as the off-peak. During the morning
rush hours, many of the county’s Metrorail stations are
both trip origins and destinations, meaning trains and
buses are full in both directions. The presence of so
much retail-entertainment-hotel activities along the
County’s Metrorail corridors has further filled trains
and buses during the midday and on weekends. Arling-
ton County averages higher shares of transit boardings
and alightings at its stations in off-peak hours than
other jurisdiction in the region with the exception of
downtown Washington, DC.
The Transportation and Environmental Benefits
of TOD
The litmus test for whether TOD yields environmental
and other societal benefits is the degree to which it
increases transit ridership, and in particular draws for-
mer motorists into trains and buses. Past research in
the United States shows that neighborhoods designed
according to TOD principles are associated with lower
car ownership rates [4, 12], appreciably higher transit
modal splits for commuting [13, 14], and fewer vehicle
trips per day [15]. In California, surveys show that
residents who live near a transit station use transit for
their commutes at a rate four to five times higher than
residents of the same region who don’t live near sta-
tions [14]. This pattern has held steady over time. In
the case of the Pleasant Hill station of the Bay Area
Rapid Transit (BART) system in the San Francisco-
Oakland region, for instance, 47% of station-area res-
idents took transit to work in 1993 [13]. Ten years later,
in 2003, the share was 44% [14].
While TODs bump up the shares of residents’ trips
by transit, more important to many observers – partic-
ularly traffic engineers and elected officials – is the
impact on local traffic conditions. A recent study of
17 TOD projects in five US metropolitan areas (Phila-
delphia, Northeast New Jersey, Washington, DC,
Portland, and the San Francisco Bay Area) uncovered
evidence of “vehicle trip de-generation” [15]. Over
a typical weekday period, TOD housing projects in
these five regions averaged 44% fewer daily vehicle
trips than that estimated by the Trip Generation manual
of the Institute of Transportation Engineers (ITE),
Transit-Oriented Development and Land Use 951
which is widely used as the reference manual for gaug-
ing traffic impacts of new projects. During peak hours,
transit-oriented housing projects generated one-half
the typical number of vehicle trips per dwelling unit.
The higher transit use and lower trip generation
rates among station-area residents are largely a
product of what economists call “self selection.” People
who prefer to take a train to work – whether to avoid
the stress of fighting traffic or to have time to read
a newspaper en route to work – purposely choose to
live near a rail stop, that is, they are predisposed to
transit commuting; they are not “converted” to transit
use simply because of where they live. Residing near
transit is thus a lifestyle choice. A recent study esti-
mated that around 40% of the ridership bonus attrib-
uted to TOD is due to self selection [16].
While TODs generate fewer trips per dwelling unit,
this does not necessarily mean they reduce local traffic
congestion. Indeed, often the opposite holds which
poses a fundamental dilemma – not only for TODs
but for all urban developments that increase average
densities. Invariably, because not all TOD residents
take transit, denser development will congest the
nearby road intersections during peak periods. Expe-
riences show that in settings like the United States,
where the majority of households own cars, a given
percentage increase in density (dwelling units per acre)
will not be matched a similar percentage decline in car
trips per acre, meaning that car-traffic densities invari-
ably rise. Thus in the near term, even if it is well served
by transit, dense development translates into more
traffic congestion. Not-in-my-backyard (NIMBY)
opposition to higher density has stopped TOD plans
around a number of middle-income neighborhoods
served by the heavy-rail BART system in the San
Francisco East Bay, mainly through building height
restrictions. Over the past 20 years, some 2,400 new
households have located within a 5-min walk to the
Pleasant Hill BART station. Once a critical mass of
TOD residents forms, they create neighborhood asso-
ciations which, among other things, stop efforts to add
more development. In Pleasant Hill, plans to build
a massive entertainment complex, with a 20-screen
IMAX movie theater, were scuttled due to neighbor-
hood backlash. The lesson is clear: do not expect to
transform middle-income, stable neighborhoods with
large-scale infill projects, because residents have the
952 Transit-Oriented Development and Land Use
political might to stop such proposals every step of the
way. More acceptable are TOD proposals for transi-
tional neighborhoods, redevelopment districts, or
greenfields with few people around to oppose new
development. Also, large-scale, regional trip genera-
tors that bring outsiders to neighborhoods made up
of professional-class residents will rarely be embraced
by residentially dominated TODs. Only local- and
neighborhood-serving land uses will be welcomed in
such settings.
The downside of preventing new growth around
transit stations is that the development will end up
elsewhere and most likely increase vehicle miles trav-
eled (VMT) – the strongest single correlate of green-
house gas emissions (GHG), air pollution, and energy
consumption in the transport sector. Like toothpaste in
a tube, if development is squeezed from one area, it
simply gets redistributed elsewhere. Shifting growth
from TOD to AOD (automobile-oriented develop-
ment) can reduce local traffic congestion at the cost
of environmental degradation for a region at large.
It can be argued that there is good and bad conges-
tion, just like good and bad cholesterol. Congestion
caused by TOD, many urbanologist would contend,
is, for the most part, good. There is no avoiding the
fact that denser areas have denser traffic. Dense cities
with world-class rail systems, like Paris and London,
are often more traffic-choked than sprawling cities. Yet
these dense cities are also attractive places to live, work,
and visit. Congestion is part of the territory of being an
active, vibrant place. TOD, moreover, offers a relief
valve to congestion. In a TOD, it is easier to predict
when and where congestion will occur, and it is easier
to avoid that congestion, e.g., one can hop on a train or
ride a bike.
TOD and Ridership at the Workplace
While TOD planning and design has traditionally
focused on residential development, unless
nonresidential destinations are also served by high-
quality transit, few TOD residents opt for rail or bus
travel. Studies show that TOD workplaces also yield
ridership dividends, though not because of self-selection
but rather proximity and convenience [17]. Parking pol-
icies and designs are particularly important determinants
of whether those working in TODs ride transit or not.
A study of workers in 10 office buildings near sub-
urban rail stations in California found, on average, 19%
of commute trips were by public transit, with consid-
erable variation [14, 16]. Other factors, however,
weighed heavily on whether TOD workers commuted
by transit. This is reflected by the logit-model results in
Fig. 1. The figures show the likelihood that TOD workers
in California commuted by transit as a function of
parking supplies, employer assistance with transit costs,
and frequency of feeder bus services. With 25 feeder
buses per day, an office setting with 50% more parking
spaces than workers, and no employer help with transit
costs, the model predicts that just 8% of office workers
near a rail station will commute by transit. At the other
extreme, for a worker heading to a station with 400 daily
feeder buses who work for an employer who provides
transit-pass assistance and provides one parking space
for every two workers, the likelihood that he or she will
commute by transit is 50%. Over the range of feeder bus
frequencies, the differential in transit commuting prob-
abilities is 30–40% depending on how generous
employers are in promoting transit (i.e., minimal
parking and help with transit costs) or in accommodat-
ing the automobile (i.e., ample parking and no help with
transit costs). Clearly, successful TODs are far more than
physical design challenges: the policy “software” that
accompanies the built-design “hardware” matters
tremendously.
TOD and Parking
Critics charge that many large-scale housing projects
near urban rail stations are “over-parked” – more
parking is provided than is needed [18]. Excessive
parking can drive up the cost of housing, consume
valuable land near transit, and impose such environ-
mental costs as increased impervious surface areas
(that contribute to urban heat-island effects and
stream-water pollution from oily run-offs). As noted,
transit stations are thought to offer “location effi-
ciency,” enabling residents to get by with fewer cars
than they might otherwise own. Yet lenders and plan-
ners often insist that code-standard parking be pro-
vided in station areas regardless. This can drive up the
cost of station-area housing, thus defeating the purpose
of promoting affordable housing construction near
major transit stops.

0.6
- - - - - - No employer help with transit costs
0.5
Probability Chose Transit
0.4
0.3
0.2
0.1
0
0 25 50
Transit-Oriented Development and Land Use. Figure 1
Influences of employer policies and feeder bus frequencies on transit commuting among office workers in California
A recent study raises some questions whether sub-
urban TODs indeed average less parking, at least in the
United States [15]. In the case of Portland, Oregon,
while average vehicle trip generation rates of TOD pro-
jects were 41% below ITE rates, the average use of
parking spaces was only 11% less. The parking occu-
pancy at three of the 15 surveyed TOD projects in
Portland was actually higher than that predicted by
the ITE Parking Generation manual. In the San
Francisco Bay Area, owning and parking a car was
even more of a necessity. There, TOD parking rates
were equivalent to ITE’s standard of 1.2 spaces per
unit. For all seven TOD housing projects surveyed
near the Fremont BART station, parking levels were
actually higher than the ITE rates – as much as 40%
above.
What might explain high parking rates combined
with high transit ridership rates for TODs? One factor
may be that most cities do not reduce the parking
requirements for TODs which in turn induces car own-
ership. One survey of TODs in California found no
reduction in cities’ parking requirements at seven of
the 11 sites studied [17]. Planners appear to assume
that more transit will not reduce parking demand, and
they may be right. In most suburban TODs, many
Transit-Oriented Development and Land Use 953
Employer help with transit costs
0.5 parking space/worker
1 parking space/worker
1.5 parking spaces/worker
0.5 parking space/worker 1 parking space/worker
1.5 parking spaces/worker
75 100 125 150 175 200 225 250 275 300 325 350 375 400
Feeder Bus Frequency at Nearest Station (buses per day)
residents still need access to a car. They just do not
use them as much. But like most suburbanites, they still
need a car to get to most non-work destinations – the
vast majority of which are located away from rail stops.
While transit-oriented housing might mean that more
trip origins are near rail stops, as long as most destina-
tions are not, TOD residents will still own cars and use
them for shopping, going out to eat, and the like.
The fact that some suburban TODs encourage tran-
sit riding yet fail to prompt residents to reduce car
ownership suggests that TODs might be a natural set-
ting for carsharing. A study of City CarShare members
in San Francisco found that 4 years into the program,
29% of carshare members had gotten rid of one or
more of their cars and 63% lived in zero-vehicle house-
holds [18]. Putting shared-cars in and around TODs
could relieve many households from owning a second
car or a vehicle altogether. Through a combination of
proximity advantages and lifestyle predispositions, liv-
ing near transit can de-generate vehicle trips. And with
the option of carsharing, it might reduce parking
demands as well. This has been the case in cities like
Zurich, Switzerland. Zurich, for example, has the sec-
ond highest per capita transit use in the world (over 600
transit trips per capita per year) and the highest per
954 Transit-Oriented Development and Land Use
capita carsharing participation anywhere (7% of
households are members of Mobility Carsharing
Switzerland) [6]. And in spite of having one of the
world’s highest per capita incomes (on a purchasing
power parity basis), only one out of three households
has an off-street parking space. Zurich also has the
highest commercial real estate prices in Europe (along
Bahnhoffstrasse) and according to the management
consulting firm, Arthur D. Little, ranks number one
in quality of life among global cities [6]. In Zurich,
world-class transit, carsharing, parking limits, and
prosperity go hand-in-hand.
Another policy response might be the introduction
of more flexibility in parking policies for housing near
rail stops. Flexibility can be in the form of enabling
projects to provide below-code parking levels when
justified, e.g., compact projects with short, direct walk-
ing connections to transit and perhaps on-site retail
establishments. Flexible parking standards provide lat-
itude in providing the optimal number of parking
spaces [19]. Flexibility can also take the form of
unbundling the cost of providing parking from the
cost of building (or renting) housing [20]. This would
allow developers to better scale the amount of parking
provided to what each tenant or homeowner is willing
to pay for each car owned, i.e., let the market demand,
rather than a possibly out-dated government fiat,
determine supply. And flexibility can be in the form
of allowing TOD tenants to choose deeply discounted
transit passes for frequent riders instead of a 300 square
foot parking space.
TOD Implementation Tools
Moving from the theory to the practice of TOD is often
fraught with difficulties, especially in car-dependent
countries like the United States. A national survey of
urban planners in US cities with TODs cast light on the
kinds of implementation tools being used to leverage
TOD [4]. Most prominent has been zoning, usually in
the form of overlays. Overlay zones are frequently
introduced on an interim basis to head-off auto-
oriented uses that might compromise a TOD and to
specify desired land uses as of right, such as housing
and convenience shops. For urban TODs, densities of
20–30 dwelling units per residential acre and floor area
ratios (i.e., building area divided by land area) of 1.0
and above are not uncommon. Some of the more
progressive TOD zoning districts, such as those found
in Portland, Oregon, Seattle, San Diego, and Denver,
also lower requirements for car parking and sometimes
even for bicycles. The city of San Diego, for instance,
recommends parking reductions as high as 15% for
urban TODs.
Besides zoning, other tools frequently used in the
United States to encourage TOD include: funding
for station-area planning and ancillary capital
improvements; density bonuses, sometimes used to
encourage affordable housing; and relaxation of
parking standards. Next in the order of frequency of
usage have been land-based tools, like land purchases
on the open market (for land-banking and potential
“deal-making”) and assistance with land assemblage.
For the most part, redevelopment agencies have
applied these tools, meaning their role in leveraging
TOD has been mainly limited to economically
depressed or blighted neighborhood settings. Because
of the higher risk involved, redevelopment tools have
often been accompanied by other funding sources,
sometimes with a dozen or more participants involved
in the process.
Central-City Redevelopment Around Streetcar
Lines
A good example of successfully using zoning tools and
financial incentives to leverage central-city redevelop-
ment in a transit corridor is the Pearl District in down-
town Portland, Oregon. The Pearl District is the most
dramatic transformation of downtown Portland in the
last 20 years. Once home to a large artist community
and an “incubator” for start-up businesses in aban-
doned warehouses, the Pearl District is now an emerg-
ing mixed-use neighborhood of upscale loft housing,
parks, art galleries, boutiques, cafes, and restaurants.
Since 2001, more than 1,700 condominiums and apart-
ments have been built. Moreover, 55% of all develop-
ment in downtown Portland has been within a block of
the streetcar service since its opening [21].
A major catalyst to the transformation of the Pearl
District was the construction of the Portland Streetcar,
the first modern streetcar system to be built in the
United States. The streetcar has been equal parts hous-
ing and transportation tool, as streetcar construction

was explicitly linked to high-density development via
density bonuses and an innovative developer agree-
ment. As a result of this agreement, the average density
of the District is now 120 housing units per acre, the
highest in the city. Moreover, properties near the street-
car line more closely approach permissible densities
than properties situated farther away [21].
Tax incentives were introduced to moderate possi-
ble gentrification effects of streetcar-induced redevel-
opment. Many affordable housing projects in Portland
receive 10-year property tax abatements if they are
within walking distance of a rail line. While the abate-
ments are loosely related to projected price levels and
affordability, their primary purpose is to ensure denser
development (as specified in the developer agreement)
than the market would otherwise support.
Residents in the Pearl District fit the demographic
profile found in other Portland area TODs – childless,
and either young people seeking smaller lofts, older
professionals looking for an urban lifestyle with little
upkeep (“downsizing boomers”), or retiring seniors.
This variety of homeowner types has contributed to
the depth of the market.
The Challenges of Mixed-Use TODs
As dense, mixed-use forms of development, many of
the barriers to TOD are generic to all forms of compact
growth – not-in-my-backyard (NIMBY) resistance,
higher risks and costs, and institutional inertia. Still,
some of the barriers to smart growth are more pro-
nounced when it comes to TOD. Mixed-use develop-
ment is one of them. Mixed land uses, a signature
feature of TODs, pose a host of difficulties not only in
terms of design but also in lining up funding, investors,
and contractors. Planners sometimes impose a design
template of ground-floor retail and upper-level hous-
ing or offices, i.e., vertical mixing – on any and all
development proposals within a TOD. Mixed-use pro-
jects are much trickier to design, finance, and some-
times lease than single-use ones [12]. Finding the right
formula for mixed land uses can be every bit as difficult
as rationalizing parking policies. Vertical mixing is
particularly problematic. Quite often, the ground-
level retail component of mixed-use TODs suffer the
most, in part because they are poorly laid out. Ground-
floor retail, for example, is doomed to fail unless it
Transit-Oriented Development and Land Use 955
opens onto a street with busy foot traffic and conve-
nient car access. Mixed housing-retail projects also
pose unique design challenges. Ground-floor retail
needs greater floor-to-floor height (typically 15–18 ft)
to be marketable, compared with the 8–10 ft between
residential floors. This means the entire ground floor,
including multifamily areas, must have higher ceilings,
which increases project costs. Ground-floor restaurants
pose problems such as where to put the exhaust shafts
for kitchens. The exact size and location of restaurant
space may not be known until leases are signed.
Designers must thus allow exhaust shafts to be put in
several potential locations, which can reduce net leas-
able space. And ground-floor restaurants might be
unappealing to upper-level residences seeking quiet
and privacy in the evening. Local governments need
to be sensitive to these issues and focus more on achiev-
ing a desired land-use mix within a transit station area
as opposed to individual parcels, i.e., pursue “horizon-
tal” neighborhood-scale mixes versus “vertical”
within-building mixes.
Financing TODs
For TODs to take form, there must be money to pay not
only for transit infrastructure, like stations and
parking, but also the armature – civic squares, street-
scape enhancements, pocket parks, expanded trunk-
line sewage and piped water capacity – that surrounds
a TOD station. Joint development is the form of value
capture that has been most widely used to finance
America’s most successful TODs. Under this approach,
a public transit agency partners with a private devel-
oper to build a real-estate project on land or air rights
owned by the transit agency itself. In return, the transit
agency receives either revenue (i.e., revenue sharing) or
passes on part or all of the costs of rail-station and
ancillary construction (i.e., cost sharing). Among the
most common forms of revenue-sharing schemes are
land leases, air-rights development, and station inter-
face or connection-fee programs. Cost-sharing
schemes include sharing construction expenses, incen-
tive-based programs that produce benefits for private
financing (e.g., density bonuses), and joint use of
equipment like ventilation systems.
As a value capture tool for transit, joint develop-
ment has the most appeal in settings where a significant
956 Transit-Oriented Development and Land Use
amount of land is available to a transit agency, prefer-
ably purchased on the open market at a fairly low cost
(which usually means prior to formal announcement
of a new railway project). This has been the case in
Hong Kong where the city’s transit operator, MTR
Corporation, obtains more than half of its revenue
from ancillary real-estate property development [22].
To the degree joint development is limited to
a geographically restricted area, such as one or two
small parcels owned by a transit agency, it fails to
capture value from a broader area benefiting from
new transit services.
Another common way to leverage and fund TODs,
especially in the United States, has been Tax Increment
Financing, or TIF. Under this scheme, incremental
increases in property tax revenues are channeled back
into a district to help revitalize distressed neighbor-
hoods. TIF has been used extensively to encourage
TOD in many parts of the United States. In the city of
Chicago, half of the 129 TIF districts in 2003 contained
a railway station [4]. The state of Pennsylvania applies
TIF in Transit Revitalization Investment Districts
(TRIDs) to promote economic development and
TOD. A criticism of TIF is it diverts tax dollars that
would otherwise accrue to a city’s general treasury and
as thus creates a privileged (i.e., subsidized) zone. Such
equity concerns have limited the application of TIFs to
much of the developing world.
Some TODs have relied on impact fees as their chief
source of funding. These are charges assessed on new
development to defray the cost of expanding and
extended public services. In 1981, the city of San
Francisco, California introduced a Transit Impact
Development Fee (TIDF) levied against new down-
town office buildings to produce income for operating
and maintaining the city’s MUNI transit network,
comprising light rail, diesel bus, trolley bus, and
cable-car services. TIDF revenues produce around $10
million annually. Broward County, Florida, has
a Transit Oriented Concurrency system that similarly
generates income for transit operations by levying
a fixed fee on new development. This program covers
around 30% of annual bus-transit operating and cap-
ital costs for the county [23]. Because limiting impact
fees to new development only can deter investments in
all but the healthiest real-estate markets and inflates the
cost of housing, it has been applied on a limited basis
outside of the United States. In the developing world,
exactions are more common wherein developers cover
the cost of expanding infrastructure (e.g., water and
sewer trunkline capacities) to serve a site.
Future Directions
TODs are a type “smart growth” and sustainable
urbanism that have tremendous growth potential due
to both supportive public policies and market forces.
Global experiences show that the integration of public
transport and land use can yield appreciable sustain-
ability benefits that redound to both the private and
public sectors. In many settings, developers know they
can earn handsome profits building, leasing, and selling
TODs. As long as traffic congestion continues to
worsen in TODs and demographic trends continue to
shift in the direction of smaller, non-traditional house-
holds, there will always be a market demand for TODs.
However, forward-looking and pro-active public-
sector strategic planning will always be a necessity as
well if sustainable and cost-effective TODs are to take
form. In many automobile-reliant cities of the world,
TODs often take the form of a few nodes, or islands, in
a sea of auto-oriented development. Unless there are
enough TODs, linearly aligned along natural travel
corridors to create a critical mass, few transit ridership
(and correspondingly auto-reducing) benefits will
accrue. That is, there need to be enough origin–
destination combinations formed by TODs in non-
downtown settings to draw enough motorists into the
trains and buses so as to put an appreciable dent in car
traffic. Otherwise, stand-alone TODs can very much
backfire by creating dense nodes in otherwise auto-
oriented settings so as to increase traffic congestion
along roads feeding into transit stations.
What is likely on the horizon for particularly pro-
gressive-minded cities might be the emergence of “Green
TODs.” Green TOD is a marriage of TOD and Green
Urbanism. This combination is, potentially, a potent
elixir. TOD works on the VMT-reduction side of shrink-
ing a city’s carbon footprint. Green Urbanism reduces
emissions and waste from stationary sources, in the form
of green architecture and sustainable community
designs. In combination they can deliver energy self-
sufficiency and zero-waste living. Renewable energy
might come from solar and wind as well bio-fuels
 Transit-Oriented Development and Land Use 957

created from human waste and other byproducts. Hammarby’s residents and correspondingly greenhouse
Recycling and reuse of materials, community gardens, gas emissions and energy consumption. In 2008, 25% of
and low-impact building materials further reduce the trips by Hammarby’s residents were by bicycle or walking
carbon footprint of Green TODs (Table 1). and 52% were by transit, much higher than regional
Synergies would like accrue from combining TOD averages [24]. Carbon dioxide emission from private
and Green Urbanism. The higher community densities cars was 52% less for households in Hammarby Sjöstad
needed to fill the trains and buses that serve TODs at the than in planned communities of similar income struc-
same time reduce heating and cooling expenses from ture in the suburbs of Stockholm [24]. And the design of
the embedded energy savings of shared-wall construc- an energy self-sufficient and low-waste community has
tion. Smart electric grids generated by photovoltaic further shrunk the project’s environmental footprint.
panels and wind turbines designed as canopies above Today, residents of Hammerby Sjöstad produce 50% of
stations might also power light-rail cars, plug-in the power they need by turning recycled wastewater and
hybrids, and electric buses. Additionally, the compact, domestic waste into heating, cooling, and electricity.
mixed-use pattern of development around rail stations The future of TOD, worldwide, is undeniably
can support walking, biking, and the use of limited bright. As long as urban ills like traffic congestion and
range private vehicles to reach relatively nearby desti- air pollution – and global concerns like oil dependency
nations, thus helping to grow such infant-industries as and climate change – persist or worsen, TOD will be
electric vehicles. One could imagine a future of hydro- embraced as a desirable form of urbanism. Rapid
gen-fueling and electric-battery swap depots in a green urbanization, combined with equally rapid urban and
community focused on a rail node and for which the inter-city rapid rail construction, in countries like
destinations of many travelers are close by. China could mean more TODs find particular accep-
Hammarby Sjöstad, a brownfield redevelopment in tance in former third-world countries that are rapidly
the city of Stockholm, is an example, par excellence, of industrializing and modernizing.
marrying of TOD and green urbanism. The combination
of railway services, carsharing, and bike-sharing has dra-
matically reduced vehicle-kilometers traveled of Bibliography
1. Cervero R (2008) Transit-oriented development in America:
strategies, issues, policy directions. In: Haas T (ed) New urban-
Transit-Oriented Development and Land Use. Table 1 ism and beyond: designing cities for the future. Rizzoli,
New York, pp 124–129
Forms of emission reductions form green TODs
2. Renne J (2009) From transit-adjacent to transit-oriented devel-
TOD Green urbanism opment. Local Environ 14(1):1–15
3. Urban Land Institute, San Francisco (2008) Bay area TOD mar-
Mobile sources Stationary sources
ketplace. Urban Land Institute, San Francisco, p 1
● Transit design: ● Energy self-sufficient:
4. Cervero R et al (2004) Transit oriented development in
World-class transit (renewably powered –
America: experiences, challenges, and prospects. TCRP report
(trunk and distribution) solar, wind turbines)
102. National Academy Press, Washington
Station as hub ● Zero-waste:
5. Cervero R, Ferrell C, Murpy S (2002) Transit-oriented develop-
● Non-motorized access: (recycle, reuse, methane
ment and joint development in the United States: a literature
(bike paths, pedways) digesters, rainwater
review, research results digest number 52. Transportation
● Bikesharing/carsharing collection for irrigation,
Research Board, Washington
● Minimal parking: and gray water use)
6. Cervero R (1998) The transit metropolis: a global inquiry. Island
(reduced land ● Community gardens:
Press, Washington
consumption, building (compost, canopies)
7. Kenworthy J, Laube L (1999) An international sourcebook of
massing, and ● Buildings:
automobile dependence in cities: 1960–1990. University of
impervious surfaces) Green roofs
Colorado Press, Boulder
● Compact, mixed uses Orientation (optimal
8. Consejo Nacional de Politica Económica y Social
temperatures)
(2000) Sistema de Servicio Publico Urbano de Transporte
Materials (recycled, low
Masivo de Pasajeros de Bogotá. Departamento Nacional de
impact)
Planeación, Documento No. 3093
958 Transit-Oriented Development and Land Use
9. Targa F, Rodriquez D (2004) Analysis of Bogotá’s bus rapid
transit system and its impact on land development. Carolina
Plann J Winter:26–36
10. Rodriquez D, Mojica C (2009) Capitalization of BRT network
expansions effects into prices of non-expansion areas. Transp
Research Part A 43(5):561–575
11. Cervero R, Sarmiento O, Jacoby E, Gomez L, Neiman A (2009)
Influences of built environments on walking and cycling: les-
sons from Bogotá. Int J Sustain Transp 3:203–226
12. Dunphy R, Cervero R, Dock F, McAvey M, Porter D, Swenson C
(2004) Developing around transit: strategies and solutions
that work. Urban Land Institute, Washington
13. Cervero R (1994) Transit-based housing in California: evidence
on ridership impacts. Transp Policy 1(3):174–183
14. Lund H, Willson R, Cervero R (2006) A re-evaluation of travel
behavior in California TODs. J Arch Plann Res 23(3):247–263
15. Cervero R, Arrington G (2008) Vehicle trip reduction impacts of
transit-oriented housing. J Public Transp 11(3):1–17
16. Cervero R (2007) Transit oriented development’s ridership
bonus: a product of self selection and public policies. Environ
Plann A 39:2068–2085
17. Bernick M, Cervero R (1997) Transit villages for the 21st cen-
tury. McGraw-Hill, New York
18. Daisa J (2004) Traffic, parking, and transit oriented develop-
ment. In: Dittmar H, Ohland G (eds) The new transit town: best
practices in transit-oriented development. Island Press,
Washington, pp 114–129
19. Cervero R, Golub A, Nee B (2007) City carShare: longer-term
travel-demand and car ownership impacts. Transp Res Rec
199:70–80
20. Dunphy R, Myerson D, Pawlukiewicz M (2004) Chapter 7: Ten
principles for developing around transit. In: Dunphy R et al
(eds) Developing around transit: strategies and solutions that
work. Urban Land Institute, Washington
21. Shoup D (2005) The high cost of free parking. Planner’s Press,
Chicago
22. Cervero R, Murakami J (2009) Rail + property development in
Hong Kong: experiences and extensions. Urban Stud
46(10):2019–2043
23. Fogarty N, Eaton N, Belzer D, Ohland G (2008) Capturing the
value of transit. Center for Transit Oriented Development,
Oakland
24. Brick K (2008) Summary report: follow-up of environment
impact in Hammarby Sjöstad. Grontmij AB, Stockholm

True Color Night Vision Video
Systems in Intelligent Vehicles
DIRK HERTEL
Quincy, MA, USA
Article Outline
Glossary
Definition of the Subject and Its Importance
Introduction
The Automotive Imaging Application
Human Perception of the Automotive Imaging
Environment
The Automotive Imaging Environment
Extending the Dynamic Range
Extending Spectral Sensitivity into NIR
Adding Color Differentiation to VIS + NIR-Sensitive
Imager
Visual Representation of the VIS + NIR Color
Information
Conclusion
Future Directions
Bibliography
Glossary
Automotive night vision A technical system to
enhance human night vision in situations where
visibility is limited or challenging to the human
visual system, for example in situations where
the range of distances illuminated by headlights
is shorter than the safe braking distance or in
situations where the human eye may be too slow
to adapt to sudden changes in brightness. It
requires an effective human–machine interface,
for example, a visual display to make available
to the driver real-time additional visual
information.
CMOS image sensor An active-pixel image
sensor (APS) consisting of an integrated circuit
with an array of light-sensitive cells (pixels) and
manufactured using complementary metal–oxide–
semiconductor (CMOS) technology. Each pixel
combines a photodetector with an active amplifier
that may perform additional functions such as
M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,
# Springer Science+Business Media New York 2013
Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology,
True Color Night Vision Video Systems in Intelligent Vehicles 959
controlling the pixel response to light and
extending the pixel’s dynamic range.
Driver assistance system (DAS) A technical system to
assist the driver in the task of driving, especially in
situations where the human sensory system (visual,
auditory) or human attention might fail to recog-
nize a dangerous situation, or when the human
response time in the perceptual-motor task of rec-
ognizing a dangerous situation then using vehicle
controls to take corrective action may be too slow to
avoid an accident.
Dynamic range Dynamic range is the ratio between
the largest and smallest possible signal values of
a changeable quantity such as light. For image sen-
sors and cameras, dynamic range refers to input
radiance or luminance signals and is defined as the
ratio between the maximum and minimum signals
where a camera can simultaneously capture detail
carried by small signal variations.
Electromagnetic spectrum, visible (VIS) and near-
infrared (NIR) The human eye can detect the
visible (VIS) portion of the electromagnetic spec-
trum ranging from wavelengths of about 380 to
730 nm. Infrared is electromagnetic radiation with
longer wavelengths than visual, with NIR denoting
the shortest wavelengths of the IR-spectrum
between 700 and 1,400 nm.
Wide dynamic range (WDR) Also referred to as high
dynamic range (HDR), it covers a family of tech-
niques to extend the dynamic range of image sensors
beyond that of an image sensor with a linear relation-
ship between input radiance and digital output signal.
Definition of the Subject and Its Importance
Vision-based driver assistance aims to increase traffic
safety at night by enhancing human night vision in
situations where visibility is limited or challenging to
the human visual system. The night vision cameras
described sense the near-infrared radiation of the car’s
own headlights to extend the range of forward vision,
while, at the same time, discriminating traffic-relevant
colors of signal lights and signage. Adaptive wide-
dynamic-range image sensor technology ensures that
scene detail in both deep shadow and bright highlight is
captured even in situations where the human eye may be
# 2012, DOI 10.1007/978-1-4419-0851-3
960 True Color Night Vision Video Systems in Intelligent Vehicles
blinded or too slow to adapt to sudden changes in bright-
ness. The information from the near-infrared and visible
spectrum is combined into natural-looking color images
and communicated to the driver via on-board visual
displays to enable recognition of the traffic situation.
Introduction
Since the late 1990s, there has been research into driver
assistance systems employing sensors to perceive the
automotive environment and utilize their outputs
to create warning and vehicle control signals. Vision-
based systems are of particular importance because
more than 90% of driving decisions and actions are
based on vision. Although the human visual system
(HVS) is able to adapt to a vast range of lighting
conditions, the rate of visual adaptation is often
outpaced by the rate of change in lighting conditions
while driving, for example when passing blinding head-
lights at night. Camera-based vision systems have the
capability to aid the driver in critical situations by more
rapidly adapting to changes in brightness and intra-scene
dynamic range. They may also extend spectral sensitivity
into the NIR to utilize the invisible portion of the head-
light spectrum or can additionally employ NIR-only
headlights that can be used in high-beam mode without
blinding other drivers. Since the visual channel has the
highest capacity of information transfer compared to
other senses, image displays are used as human–machine
interface to communicate information to the driver.
However, the displayed image must be of sufficient
quality to meet minimum requirements of usefulness so
that scene objects can be seen, and of naturalness so that
scene objects can be instantly recognized. For example,
displayed objects with too little contrast will be
overlooked while objects with unnatural grayscale and
color appearance or distorted perspective may not be
recognized in a timely fashion.
Although the majority of research on driver
assistance systems focuses on image processing to
interpret the information delivered by image sensors,
comparatively little attention is given to the task of
capturing scene detail. Since the camera forms the
first link in the image information processing chain,
its performance is critical; if the intra-scene dynamic
range, the ratio of the brightest to darkest information-
carrying intensity, exceeds the dynamic range of the
image sensor, then scene detail will be lost and cannot
be recovered by image processing.
A model-based methodology is one that:
● Characterizes the automotive imaging environment
(photospace) as wide dynamic range (WDR)
● Defines incremental signal-to-noise ratio (iSNR) as
the criterion for reliably capturing WDR scene detail
● Uses the inherent sensitivity of silicon-based image
sensors to utilize scene information carried by NIR
radiation
● Separates color and NIR information in the image
sensor and WDR image signal processing
● Defines the principles for combining luminance
information from the visual and NIR spectral
ranges with traffic-relevant color information into
a natural and useful image for visual display
The Automotive Imaging Application
Reduced visibility at night substantially increases the
risk of traffic accidents caused by misjudgment of
dark roadways or failure to recognize the presence
of pedestrians and obstacles. If an accident does
occur, passive safety systems such as seat belts and
airbags can reduce the percentage of fatalities to the
vehicle’s occupant, but only to a limited degree [1].
Any further reduction of traffic fatalities requires the
use of active driver assistance systems.
Driver Assistance
Driver assistance can be classified into comfort systems
and safety systems. These systems employ sensors that
take measurements of the automotive environment,
interpret the signals delivered by the sensors, generate
warning signals, and, if the driver takes no action, even
take over vehicle controls. The distinction between
comfort and safety is fluid; parking guidance makes
parallel parking more comfortable but also enhances
safety by detecting persons and objects not visible to
the driver. As an example of active safety, lane-keeping
systems analyze sensor signals to determine if the
vehicle is leaving the lane or roadway, then generate
warning signals and in case of driver inaction steering
signals. Other examples are lane-change assistance
with blind-spot monitoring, collision avoidance, and
driver-drowsiness monitoring.

Vision-Based Driver Assistance Vision-based driver
assistance has a high potential of increasing traffic safety.
First, more than 90% of the information that is critical
for making driving decisions comes through the visual
channel. Second, conditions that impair the flow of
visual information have a significant impact on traffic
safety. Nighttime driving is twice as risky as daytime
driving, even though the volume of traffic is often less
at night. Nighttime accidents tend to be more severe,
with a disproportionate increase of fatalities compared
to the volume of traffic. Nighttime accidents more often
involve pedestrians or cyclists. The illumination of the
roadway may be inadequate. For example, dipped head-
lights stretch over a fixed range of about 180 ft, but the
safe braking distance increases with speed exceeding the
visibility range at speeds greater than 30 mph [2].
Video-based driver assistance systems provide
additional image information in situations where visi-
bility is poor (night vision) or obstructed (side or rear
vision). Applications range from driver assistance with
forward and rear view images displayed on a dashboard
screen to active safety systems where the video footage
is analyzed by software that can issue warnings or
trigger vehicle controls. On the front-end of these
systems are video cameras that must capture scene
details reliably under all circumstances.
Requirements of Vision Systems for Driver
Assistance The basic requirements for a vision-based
driver assistance system are image quality and reliability.
Image Quality Image quality can be defined as “the
degree to which the image can be successfully exploited
by the observer” [3]. An image of “high” quality should
satisfy three criteria: genuineness, usefulness, and
naturalness (GUN-model).
● Genuineness is described as the degree of apparent
similarity of the reproduced image and its viewing
environment with the external reference, that is, the
original image and environment. Genuineness is
related to fidelity, for example between a painting
and its photographic reproduction.
● Usefulness is how relevant the reproduced image
and environment are to the observer and the task
activity. Usefulness is related to the discriminability
of objects in the image, for example obstacles on
a dark roadway.
True Color Night Vision Video Systems in Intelligent Vehicles 961
● Naturalness is described as the degree of apparent
similarity between the reproduced image and envi-
ronment and the viewer’s internal references, that
is, memory prototypes. It is related to recognizabil-
ity of objects in the image, for example whether the
obstacles on a dark roadway are pedestrians or deer.
Night vision systems for driver assistance do not
require genuineness, but they do require both useful-
ness and naturalness. Color and near-infrared (NIR)
can supply important additional information that
increases the usefulness of a night vision image,
though NIR sensitivity may affect color saturation
and thus reduce the naturalness of displayed images
(section “Adding Color Differentiation to VIS + NIR-
Sensitive Imager”).
Image information that is interpreted by software
has to fulfill the criterion of usefulness. Image informa-
tion displayed for viewing has to be natural so that it
can be interpreted instantly. An unnatural image
requires effort by the interpreter and thus can distract,
for example, false colors, inverted grayscale (thermal
images), or distorted perspective (extreme fisheye).
Reliability The main difference between photography
and automotive imaging is that the former is in
most cases controlled by a conscious subject – the
photographer – who selects the frame, knows what is
in the scene, and can avoid extremes (strong light
sources). However, when a camera is fixed in
a vehicle, frame selection and content are arbitrarily
determined by vehicle position, so intra-scene dynamic
range may include, for example, approaching
headlights, glare from other vehicles, tunnel entrances
or exits, or solar glare. Image detail provides
information about the objects in the real world, such
as their location, shape, size, brightness, and color.
If the dynamic range of a camera is too narrow to
accommodate the intra-scene dynamic range,
important object details will be missing from the
corresponding image. Once lost by capture failure,
these details can never be recreated with image
post-processing [4, 5].
Conclusion: Model-Based Approach The design of
automotive vision systems is governed by performance
versus cost trade-offs. Therefore, it is important that
the design is based on a clear understanding of
962 True Color Night Vision Video Systems in Intelligent Vehicles
the nonnegotiable performance thresholds such as
sensitivity, signal-to-noise ratio, and dynamic range.
A model-based design can quantitatively define and
meet the performance requirements of automotive
night vision. This includes models of human color
night vision, the automotive night vision environment
(photospace), and detection of scene detail (signal-to-
noise ratio).
Human Perception of the Automotive Imaging
Environment
The human visual system (HVS) is able to adapt to
a vast range of natural and artificial lighting conditions.
The mechanisms of adaptation are complex and
include the combined responses of iris, lens, retinal
photoreceptors, plus processing in the retina and
visual cortex.
Day and Night Vision
Viewing the environment from a fast moving vehicle
means that the driver’s visual system must adapt to
a visual environment that changes more rapidly than
for a pedestrian or stationary observer. Under daylight
conditions, such changes requiring instant adaptation
include driving into the deep shadow of a bridge or
tunnel, emerging from deep shadow into bright
daylight, and solar glare. At night, the most challenging
situation occurs immediately after passing the
blindingly bright headlights of oncoming vehicles and
having to adapt to the comparatively dark roadway
ahead. These examples illustrate that adaptation of
the HVS to instantly changing stimuli is the most
critical issue in the automotive environment.
However, the models describing human day and
night vision refer to steady states of global adaptation
to defined ranges of light levels and characterize the
responses of the two types of retinal sensors, rods and
cones, to these light levels. Three distinct adaptation
states of the human eye have been identified: photopic,
mesopic, and scotopic.
● Photopic vision in bright light is dominated by cone
receptors of relatively low sensitivity that provide
full color at high acuity, which is highest at a central
foveal field of about 2
. The spectral sensitivity
function for photopic vision is measured as the
combined response of the long-, medium-, and
short-wavelength-sensitive cones and standardized
as the photopic luminous efficiency function V(l)
with peak sensitivity at 555 nm (yellowish green).
● Scotopic vision, at very low light levels, relies on
highly sensitive rods that provide only
a monochrome image with poor acuity at the center
of the field of vision but increased acuity toward the
periphery. The peak of the scotopic luminous
efficiency function V 0 (l) is at 510 nm, shifted
toward bluish green, the so-called Purkinje-shift.
● Mesopic vision is a transition phase between
photopic and scotopic vision that is invoked at
luminance levels below 10 cd/m2 and above
0.01 cd/m2. The mesopic luminous efficiency
function has not yet been standardized. Proposals
range from a simple weighted average of the
photopic and scotopic sensitivity functions to
more complex chromatic models that weigh the
contributions from the individual cones and yield
sensitivity functions with two or three peaks [6].
Mesopic vision thus represents a balance between
photopic and scotopic vision where the response of
monochrome and color receptors will depend
not only on overall light level but also on the
scene-dependent local distribution of brightness
on the retina. For example, the response within
those areas of the retina that are stimulated by
bright colored traffic lights is still cone-dominated
while the response of retinal areas that are stimu-
lated by dark background objects is rod-dominated.
Due to the presence of artificial lights in the auto-
motive environment at night, the eye’s adaptation
never quite reaches the scotopic state but remains
between or in the photopic and mesopic states. Color
vision in very dark scenes is restricted to bright colored
lights (traffic, tail, and brake lights), retroreflective
signage, and surfaces brightly illuminated by headlights
and streetlights. As these are relatively small and iso-
lated by a dark background, they are referred to as
“unrelated colors” [7]. Thus, relevant traffic lights are
perceived as colored, but most dark objects (foliage,
pedestrian’s clothes) are perceived as monochrome.
Although the simultaneous dynamic range of the
human eye, also known as the instantaneous sensitivity
window, is limited to about 10,000:1, adaptation
 True Color Night Vision Video Systems in Intelligent Vehicles 963

enables the HVS to adapt over time to a much wider True Color Night Vision Video Systems in Intelligent
range of luminance levels (Fig. 5). The rate of adapta- Vehicles. Table 1 Visual impact of aging on human vision
tion of the HVS to changing light levels is critical to performance
traffic safety. Adaptation works with two time
Parameter Effect of aging Visual impact
constants: fast (transient) adaptation occurs within
Sensitivity Decreases Less light
seconds whereas slow adaptation happens on a much
reaches retina
longer timescale, ranging from several minutes up to
three quarters of an hour. Fast adaptation is the instant Color perception Deteriorates Loss of blue
sensitivity
reaction to a sudden change in brightness, for example
tunnel entry or exit, and vision is impaired until adap- Rate of dark Slows “Black holes”
adaptation
tation is achieved, usually after 1–2 s. Therefore, fast
adaptation is of utmost importance in the performance Glare Increases Loss of
of the HVS during driving. The rate of slow adaptation dynamic range
depends on the direction (dark takes longer than bright Contrast sensitivity Decreases Loss of detail
adaptation) and the light levels before and after Peripheral vision Decreases Restricted
adaptation. For example, the brighter the initial light
level and the darker the final light level, the more time is
required for slow dark adaptation. Headlight glare
from oncoming vehicles constantly interrupts the phenomenon of “black holes,” a temporary inability
process of dark adaptation. to perceive dark background detail after blinding.
Although all drivers experience black holes, older
drivers take longer to recover vision after the experi-
Challenges to the Human Visual System
ence. Many societies have an increasing proportion of
Driving under night conditions is very stressful to the older drivers, so the implications of this issue are ever
HVS. First, the range of luminance levels is extremely greater.
large, ranging from blinding oncoming headlights to The simulation in Fig. 1 illustrates the loss of
dark objects beyond the cone of one’s own headlights. sensitivity and shift in color vision due to yellowing
The constant change in peak brightness and position of with age.
the highlights puts the eye under the constant strain
of global and local adaptation so that the retina is
Conclusion: Producing Natural Images Under
constantly registering either photopic or mesopic
Mesopic Conditions
brightness levels. Color vision is restricted to bright
lights and brightly illuminated objects whose position Night vision images displayed for the driver have to be
in the field of vision changes rapidly and constantly. rendered “naturally” so that traffic-relevant objects can
The rate of change in stimuli will almost always exceed be quickly seen and recognized. This requires that
the rate of transient (fast) adaptation. Visual night illuminated dark details such as pedestrians are
driving performance can be evaluated by the responses rendered as a sufficiently bright monochrome image
to three visual subtasks that are defined by questions while “unrelated color” objects such as traffic lights are
and evaluated by corresponding parameters: “Can it be rendered with their correct hue and sufficient satura-
seen?” (contrast threshold), “How quickly?” (reaction tion (Fig. 2). In other words, the best compromise is to
time), and “What is it?” (recognition threshold) [6]. reconstruct a good luminance image then add those
The ability of the HVS to fulfill these tasks decreases unrelated colors that are relevant to the traffic scene.
with age (Table 1). The most serious aspect of age- This way, the automotive vision system will mimic the
related visual deterioration is the slowing of transient mesopic state of the HVS, but with two key improve-
adaptation, making it more likely that the driver is ments. First, as adaptation to sudden changes in light
“blind” immediately after sudden changes of bright- levels is almost instantaneous, the driver is provided
ness. At night, headlight glare can lead to the with visual information by the safety system in
964 True Color Night Vision Video Systems in Intelligent Vehicles

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 1

Simulated change in night color vision with age (a) 20, (b) 60, and (c) 70 [42]

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 2

Same scene showing (a) related color at daytime and (b) unrelated color at nighttime [8]

situations where the HVS adapts slowly. Second, the background. Although color cameras do not experi-
displayed luminance image is brighter and more ence a shift in spectral sensitivity akin to the Purkinje
detailed at a wider field of view in comparison to the shift, image processing algorithms (such as white
corresponding rod-based HVS perception of the dark balancing) still have to be applied to correct color shifts
 True Color Night Vision Video Systems in Intelligent Vehicles 965

in illuminated objects caused by altered illuminant
color temperature, for example moving between areas
lit by mercury and sodium vapor lamps.
The Automotive Imaging Environment
A possible methodology for estimating the dynamic
range of the automotive vision environment is based
on photometric and spectral measurements together
with the statistical photospace model. In addition,
knowledge of the spectral characteristics of colored
light sources is important when designing VIS–NIR
color night vision systems.
Scene Photometry
To determine dynamic range requirements, a quanti-
tative model of the image capture environment is
needed. First, a measure of the brightness of light
sources and illuminated objects in automotive scenes
is required. The intensity of light is measured in
radiometric power units, which are then converted
to photometric units that represent the perceived
brightness. This is done by weighting the radiant
power at each wavelength by a spectral sensitivity
function that models human brightness sensitivity,
for example the photopic luminous efficiency func-
tion V(l) for the bright-adapted eye, or the scotopic
luminosity function V 0 (l) for the dark-adapted eye.
The amount of visible light reflected by or emitted from
a surface area is measured as luminance with units of
candela per square meter [cd/m2]. Luminance is a good
indicator of how bright a light source or illuminated
object will appear. It measures the luminous power
perceived at the surface of a light source or an illumi-
nated object. It is independent of viewing distance and
not changed by optical systems, e.g., lenses. Luminance
measurements of objects in automotive scenes require
spot luminance meters.
In the case of color and multispectral imaging,
radiance must be measured [W/cm2/sr] at specific
wavelengths to characterize the night vision environ-
ment. This is done by a spot spectroradiometer that
can measure not only the visible but also the NIR
range.
Scene analysis relies on measuring the luminance of
individual scene elements rather than the scene average
[9]. Scene photometry in combination with scene
analysis has been used to build statistical distributions
of light levels within a scene, ranging from the brightest
lights to the darkest illuminated surfaces. Figure 3
shows an automotive vision example with luminance
ranges for light sources and reflective surfaces typical of
a night traffic scene.
Photospace in Automotive Vision
Photospace describes the environment as a statistical
distribution that is populated by measuring average

Headlights

Traffic signals

Taillights

Traffic sign (retro-reflective)

Pavement markings (retro-reflective)

Traffic signs (non retro-reflective)

Pedestrians

Road surface

Intra-scene dynamic range at night 120dB

0.1 10 1000 100000

Luminance [cd / m2]

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 3

Luminance ranges for light sources and reflective surfaces in typical night traffic scenes [8]
966 True Color Night Vision Video Systems in Intelligent Vehicles
scene luminance levels for a great number of
application-typical scenes [10]. By analyzing at least
ten daytime and ten night traffic scenes, extended
photospace distributions can be built based not on
scene averages but rather on the probability that
a scene element of a given luminance will occur
(Fig. 4) [4, 11]. As more image data are added,
such distributions will ever more closely represent
application conditions. These extended photospace
distributions for automotive vision by day and night
show clusters representing light sources at high
luminance levels (reflections of sun, vehicle lights,
street lights, and traffic lights), and illuminated sur-
faces at lower luminance levels (road surface, traffic
signs, and traffic participants). The luminance ratio
within night scenes (106) is anticipated to be higher
than that of daylight scenes (104–105) since poorly
illuminated objects at night have much lower lumi-
nance. The highest luminance ratios in daylight
scenes occur when the roadway enters or exits deep
shade (bridges, tunnels). For the camera to detect
image detail at night simultaneously in both the
brightest (differentiation of headlights) and darkest
scene areas (pedestrian detection), a dynamic range of
50%
40%
Probability Distribution
30%
20%
10%
0%
0.1
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 4
Extended photospace distribution showing clusters for light sources and illuminated surfaces [5]
106:1 is required. However, a wide dynamic range is
not the only requirement for an automotive camera
to work in a night vision environment. The other
necessity is that the camera is sensitive to the lowest
luminance levels of the automotive photospace. The
noise-limited sensitivity of the camera at the video
frame rate should be such that its dynamic range covers
the entire automotive photospace, including most of
the mesopic range plus as much of the photopic range
as possible (Fig. 5).
For a monochrome system, it is sufficient to ren-
der the brightest lights near the maximum signal level
of the pixel in order to minimize halos and flare, thus
maximizing differentiation between various lights,
for example, headlights. However, in the case of
color capture, this may be insufficient; a colored
light captured near the maximum signal level will be
too desaturated to be recognizable. As a rule of
thumb, if a pixel’s set of RGB signals is to be perceived
as “green” (10% color saturation between the hue
angles of 85
and 155
), each of the R and B signals
has to be at least 10% lower on average than the
G signal. However, if the green light increases in
brightness, the G signal will reach then exceed
Night Daylight
Light
Sources
Illuminated
Surfaces
1 10 100 1,000 10,000 100,000
Scene Luminance (cd/m2)
Luminance ratio day = 104
Luminance ratio night = 106

Range darkest night sky - solar disk
Automotive photospace: day
night
Human vision: photopic
mesopic
scotopic
Human simultaneous range
1E–06 0.0001
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 5
Luminance levels and adaptation of the HVS [42]
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 6
Typical urban night traffic color scene where limited
dynamic range results in clipped highlight colors
appearing white (centers of green traffic lights, red tail
lights) [35]
saturation level before R and B, thus decreasing the
G/R and G/B ratios so that the green color is barely
distinguishable from “white” (see Fig. 6). Therefore,
the dynamic range of a night vision color system must
be wide enough to prevent any of the RGB color signals
from being clipped.
True Color Night Vision Video Systems in Intelligent Vehicles 967
0.01 1 100 10000 1E+06 1E+08 1E+10
Luminance [cd/m2]
Spectral Distribution of Light Sources
The diversity of spectral characteristics of colored light
sources poses a challenge to automotive vision systems
that combine NIR sensitivity with color differentiation.
For example, a green traffic light might be an LED array
or an incandescent lamp behind a colored filter.
The former has a narrow spectral distribution peaking
in the 500–600 nm range. The latter has a more
complicated spectral distribution. Although in the
VIS range, it has a peak in the green region of the visible
spectrum, since most commonly used color filter
materials are almost transparent in the NIR range and
the spectral emission of incandescent lamps (color
temperature 3,200 K) peaks in the NIR at about
800 nm, the strong NIR contribution might exceed
that from the VIS range (400–700 nm), see Fig. 7.
Spectral measurements of objects in automotive scenes
require spot spectroradiometers.
This high NIR content of colored lights does not
pose a problem for human vision and common RGB
color cameras because the eye is insensitive to IR, and
the latter has an IR-cut filter.
It follows that a VIS–NIR color night vision system
has to combine high sensitivity in both spectral areas
with good reproduction of those colors relevant to
night traffic scenes. This requires the imager to have
WDR to avoid signal saturation in colored lights, plus
968 True Color Night Vision Video Systems in Intelligent Vehicles
Spectral emission
[arbitrary units]
incandescent lamp
lamp with green filter
green LED
400 500 600 700 800 900 1000 1100
Wavelength [nm]
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 7
Spectral VIS–NIR emission curves of an incandescent lamp
(3,200 K), an incandescent lamp with a green glass filter,
and a green LED [35]
sufficient separation of NIR and color signals so that
colors can be reproduced independently of the light
source’s IR content.
Conclusion: Automotive WDR Environment
Intra-scene dynamic range is determined by the range
of intensities between bright lights (some of them
carrying color information) and dark, dimly illumi-
nated objects. The automotive vision environment
can be measured by scene photometry to quantify the
luminance ranges of traffic-relevant light sources and
illuminated objects, and to populate statistical
photospace distributions that show the probability
that scene elements of a given brightness (luminance)
will occur at night or day. Automotive scenes typically
cover a WDR, highest at night due to lower background
luminance levels. Challenges in daylight conditions are
deep shadows (parked vehicle under bridge, tunnel
entry) and blinding (solar glare, exit from tunnels or
bridges). Night vision requires both high sensitivity (to
distinguish shadow detail) and WDR (to distinguish
highlight color detail). Preserving the color informa-
tion of traffic-relevant lights requires an even wider
dynamic range than for monochrome vision.
Extending the Dynamic Range
A possible methodology to extend the dynamic range
of image sensors without compromising the transfer
of scene information to image information can be
developed from an information-theoretical model.
Definition of Dynamic Range
Dynamic range generally relates to an imager’s ability
to capture information about both bright and dark
objects in the same scene at the same exposure settings.
There are several definitions of dynamic range. In its
most general sense, dynamic range describes the ratio
between the largest and smallest possible signal values
of a changeable quantity such as light [4]. Dynamic
range can be either input-referred (ratio of real-world
light levels) or output-referred (ratio of the response
signals delivered by a light detector). Since the output-
referred definition can lead to dynamic range numbers
built solely on technical specifications (such as the bit
depth of an A/D converter) rather than on actual mea-
surements, it has limited value in predicting dynamic
range [13]. Therefore, dynamic range should refer strictly
to input signals, for example luminance L in photopic
units of cd/m2. The relationship between input and
output signals is established from the imager response
curve.
In silver halide photography, the luminance ratio of
the highlights to the shadows is called “object ratio”
and the logarithm of this ratio “object range” [4]. The
equivalent of “object range” in digital imaging is the
input-referred dynamic range, which can be expressed
in units of decibels [dB] by calculating 10 log of the
luminance ratio. However, technical specifications of
digital and video image sensors traditionally referred to
ratios of output voltages or digitized light levels so that
decibels represent 20 log of the ratio (even when the
represented input luminance is directly proportional to
the output voltage or level) not to its square [14]. In
order to avoid disadvantaging input-referred data, the
20 log rule is also used to calculate the input-referred
dynamic range from luminance ratios:


Lmax
DR ¼ 20  log10 : ð1Þ
Lmin
Reliability Criterion
Ascertaining dynamic range requires determining the
maximum and minimum signal levels. Detection of the
smallest possible signal is limited by the noise floor. To
become detectable, a signal must exceed the noise floor; in
other words, the signal-to-noise ratio (SNR) must be
greater than 1. The largest possible detected signal is
 True Color Night Vision Video Systems in Intelligent Vehicles 969

limited by saturation. Any input signal above the satura- Image

signal D
tion level can only reproduce a saturation level output
signal and thus remains indistinguishable from the
saturation signal. This introduces the concept of signal
variation as the carrier of information: object detail is
D +ΔD
carried by small variations in light level. Scene informa-

tion can only be gathered reliably if the image capture
system can detect and reproduce the small signal varia-
tions carrying that information. Thus, dynamic range
needs a more precise specification to ensure detection
reliability: it is the ratio between the maximum
and minimum luminance signals where a camera can
simultaneously capture detail carried by small signal
variations.
A distinction has to be made concerning the spatial
dimension of object details. Below a certain size, the
details in an image are no longer formed indepen-
dently; the signal at a given image point depends not
only on the luminance at the corresponding object
point but also on that of its surroundings [4]. This
effect can be accounted for by the modulation transfer
function (MTF). However, this estimation of dynamic
range refers only to the reproduction of details large
enough that the loss of detail due to MTF can be
ignored. It is desirable to deduce dynamic range from
objective measurements. This methodology is laid
down in a progression of ISO/TC42 imaging perfor-
mance standards [15–17].
Figure 8 shows the transfer of large-area detail
(small luminance variation) by the response curve in
the presence of noise [18]. The response curve is the
relationship D(L) between input luminance signal L
and output digital image signal D. ISO/TC42 refers to
the response curve as opto-electronic conversion func-
tion (OECF). The slope of the OECF, the change dD in
image signal due to a change dL in the luminance
signal, is referred to as incremental gain g(L):
dD
gðLÞ ¼ : ð2Þ
dL
Detection error is given by noise. Since dynamic
range refers to the input luminance signal L, the
noise sL has to be defined as the luminance-equivalent
of the noise sD that can be observed and measured
in the digital image [19]. At each luminance level
L, the equivalent luminance variation sL can be
26D
D
Noise σD
Luminance signal L
L L+ΔL
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 8
Transfer of large-area detail (small luminance variation DL)
by the response curve D(L) in the presence of noise sD [5]
calculated from the image noise sD via the incremental
gain:
sD ðLÞ
sL ðLÞ ¼ : ð3Þ
gðLÞ
The criterion for detection reliability is the SNR, the
ratio of the input luminance signal L and the
luminance-equivalent noise sL. Detection reliability
requires the SNR to stay above a threshold of one to
fulfill the following condition:
L gðLÞ  L
¼ > 1: ð4Þ
sL ðLÞ sD ðLÞ
The expression in (4) is referred to as “incremental
SNR” (iSNR, also S/Nx and SNRinc); the incremental
signal is defined as iS(L) = g(L)∙L. An equivalent
concept for determining detection reliability in silver
halide photography is known as noise-equivalent
contrast (NEC) [20].
The threshold condition iSNR > 1 means that at
any given luminance level L, detail carried by a small
variation in luminance can only be reliably detected if
the corresponding incremental image signal is high
enough and the noise low enough. Fulfilling the iSNR
970 True Color Night Vision Video Systems in Intelligent Vehicles
criterion will be first explored for imagers with a linear
response. At very low luminance levels, noise keeps
iSNR below its threshold. The minimum reliable lumi-
nance level Lmin is reached only when iSNR rises above
the threshold. At very high luminance levels, the onset
of saturation will clip further signal increases and thus
cause the incremental gain to drop sharply. The
maximum reliable luminance level Lmax is reached
when iSNR once more drops below the threshold.
It follows that dynamic range for linear imagers is
bounded by the two extreme luminance levels Lmin
and Lmax at which the iSNR condition of (4) still holds.
The iSNR condition is now applied to imagers with
extended dynamic range. This requires shifting satura-
tion to higher input signal levels by a nonlinear
response curve: at a so-called breakpoint, the response
curve’s slope (incremental gain) is suddenly lowered to
delay saturation to higher luminance levels than for
a linear imager (Fig. 9). However, lowering incremental
gain will, in turn, decrease iSNR, pointing to a potential
iSNR trade-off when increasing dynamic range. Rather
than just using iSNR to determine the boundary
luminance levels Lmin and Lmax, the minimum iSNR
criterion has to be maintained at every luminance level
in between. Verifying this requires iSNR be measured as
a function of luminance L. An equivalent strategy was
used for silver halide photography where NEC curves
were measured as function of exposure [20].
Therefore, nonlinear imagers require a more
general definition beyond the recommendations of
ISO/TC42; dynamic range is the input luminance
range within which the threshold criterion for
Saturation
Image signal D
linear
non-linear
Noise
Breakpoint
Noise floor
Luminance signal L
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 9
Extending dynamic range with a nonlinear response curve [5]
iSNR(L) is met at all levels Lmin  L  Lmax.
This suggests a strategy for maximizing the dynamic
range of imagers with a nonlinear response curve:
maintain the necessary minimum of iSNR(L) over the
largest possible range of luminances L. Thus, iSNR
becomes a metric describing the reliability of detecting
scene detail. The theoretical detection limit is 1, but the
reliability requirements of applications may require
higher thresholds. For example, a minimum of 10 is
suggested for “acceptable” photographic image quality,
and between 1.5 and 4 are estimated for machine-
vision applications [18].
Technologies to Extend Dynamic Range
The following section reviews a selection of WDR
technologies and explores the iSNR trade-offs of
nonlinear dynamic range extension. A WDR video
camera for automotive applications should combine
good sensitivity with low noise at a given frame rate
(10–60 fps) plus a WDR extension that can be adapted
quickly to match typical automotive photospace
conditions (80–120 dB). The first group of technolo-
gies for WDR extension are those that directly change
the CMOS pixel response function (such as logarithmic
response [21], linear-log combination [22], in-pixel
light-to-frequency conversion [23], pixel-level A/D
conversion [24], or pixels with overflow capacitors
[25]), while a second group changes pixel response
indirectly by altering the timing (multiple exposure,
multiple partial reset [26]).
The first group, directly changing the response func-
tion, involves extensive pixel circuitry resulting in
low fill-factor and thus poor low-light sensitivity.
For example, in-pixel light-to-frequency conversion pro-
vides 115–130 dB WDR, but the required circuitry
decreases the fill factor to 25%, and reduces low-light
sensitivity. Similar sensitivity degradation currently
limits the usefulness for automotive video of both
pixel-level A/D conversion and pixels with overflow
capacitors. Although pixels with direct logarithmic com-
pression can provide WDR above 170 dB at low cost, the
WDR extension is not adaptive, and at present, high
fixed-pattern noise and dark current decrease low-light
SNR. The combined linear-log CMOS pixel does avoid
these drawbacks, but the required seven transistors per
pixel limit the fill factor.

WDR can be covered with a linear image sensor by
multiple exposure, but application to automotive video
requires either fast readout (high frame rate with
corresponding loss in low-light sensitivity), or local
frame storage and processing.
Among WDR technologies that alter the timing,
CMOS image sensors with multiple partial reset
(multiple-slope) technology offer good imaging
performance at low cost and are thus well suited to
automotive video capture [4]. The example of
a multiple-slope imager explains how to extend
dynamic range while maintaining detection reliability.
At the heart of the CMOS pixel is a light-sensitive
photodiode with a parasitic capacitance that is charged
by applying a voltage pulse V0 at frame begin t0, then
discharged as exposure generates photoelectron-hole
pairs. Figure 10 shows the charge–time diagram of
a CMOS pixel. The rate of discharge depends on the
input luminance; a darker pixel is discharged at a lower
rate, a brighter at a higher one. After integration time
tint, the remaining charge is read out and converted into
the image signal D. This is proportional to the degree of
discharge between t0 and tint; no discharge corresponds
to D = 0, and full discharge to saturation level (Fig. 10).
A darker pixel will only be partially discharged at tint,
proportional to the input luminance; its response
will therefore resemble that of a linear pixel. A very
bright pixel will be fully discharged before tint , making
it impossible to register any further increase in
luminance; it has reached the saturation level of its
t int
Compression ~ Pixel
t int − t 1
V0 Output:
V1 black
Charge
dark pixel
bright pixel
white
Frame Partial Pixel
begin reset readout
t0 Exposure time t1 t int
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 10
Charge–time diagram for a CMOS WDR pixel with a single
partial reset [5]
True Color Night Vision Video Systems in Intelligent Vehicles 971
response curve. However, this bright pixel could
resume registering photons if a reset voltage pulse V1
is applied at the time t1 < tint to replenish the charge of
each capacitor. The reset is only partial; the reset
voltage V1 is chosen to be lower than the initial V0 so
as to discriminate between darker pixels that would not
be saturated at tint and brighter ones that would. The
partial reset does not affect those darker pixels that are
still above the reset level V1. It only recharges
the brighter pixels that are already discharged
below V1. For them, the reset overwrites any memory
of previous exposure, and they resume registering
photons in the shorter exposure time tint  t1. The
corresponding pixel response curve will be piecewise
linear, but with a lower incremental gain after
the breakpoint that corresponds to the reset at t1.
The nearer t1 is to tint, the lower the incremental gain
after reset, and the higher the saturation luminance
Lmax becomes. The reset allows a higher luminance
range to be “compressed” into the same limited output
image signal range; thus, the ratio of the total exposure
time tint to the remaining exposure time tint  t1 is
referred to as “compression.” To increase compression
further, additional partial resets of successively decreas-
ing voltage can be applied within the remaining time
between t1 and tint, see Fig. 11a, thus creating a piece-
wise linear response curve where the incremental gain
is further decreased after each of the breakpoints, see
Fig. 11b. Theoretically, the number of additional resets
is only limited by the smallest increments of timing and
recharge voltage control, but in reality, it is limited by
the requirement of detection reliability.
Extending Dynamic Range Reliably
The partial resets of a multiple-slope WDR imager require
precise control of the timing and recharge voltage so that it
can adapt optimally to the changing dynamic range of
a natural scene. Experimental observations and model
calculations show how the iSNR criterion can be used to
maximize dynamic range without compromising detec-
tion of scene detail. The methodology for measuring iSNR
has been described in detail [5].
Experimental Observations One important goal
of dynamic range extension is minimizing the number
of saturated pixels and thus maximizing the capture of
972 True Color Night Vision Video Systems in Intelligent Vehicles
dark pixel
bright
V0 brighter
V1 brightest
Charge
V2
t0 Partial resets: t1 t2
a Exposure time
3
Image Signal
2 highlight detail more visible on the trunk but at the cost
brightest
1 brighter
bright
Breakpoints dark pixel
b has lost all texture detail. There is a trade-off; the higher
Luminance
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 11
(a) Charge–time diagram and (b) response curve for
a CMOS WDR pixel with multiple partial resets [5]
highlight detail. With the single-reset system shown in
Fig. 10, high compression can be achieved by moving
the reset time t1 as close as possible to the frame end tint.
The highest compression is limited only by the
increments of reset timing, given by one row time in
common multiple-slope imagers and by one pixel time
for the most advanced multiple-slope imagers. This can
be explained in the example of a VGA imager with 480
rows of 640 pixels. The voltage reset can be placed
as close as one row time before tint (i.e., after 479
rows), achieving a maximum compression of c = 480/
(480–479) = 480. This means that the ratio of exposure
times in compressed and linear modes is 1:480 and that
the pixel is 480 less sensitive in compressed mode.
From exposure time ratios, dynamic range can be
estimated to be extendable by 54 dB. In an advanced
VGA imager where single pixel times are addressable,
the dynamic range extension could be even
more impressive. Each row contains 640 pixels. The
shortest exposure time would be one pixel time before
tint = 480640 pixels so that a maximum compression
of c = (640480)/(640–639) = 307,200 could be
reached, extending the dynamic range by 110 dB.
Experiments like those shown in Fig. 12 and the
V3
corresponding imager response model of section
“Imager Response Model” show that although the
practical implementation of high compression with
t3 tint just one partial reset can effectively prevent highlight
saturation, it does come at the cost of usefulness, seen
as severe loss of midtone image contrast. Figure 12a
shows an image taken in linear mode where saturation
Saturation prevents visibility of detail on the trunk of the car.
Progressing from Fig. 12b to d, increased compression
decreases the proportion of saturated pixels, making
of lowered contrast of midtone detail (lane markings
versus road surface). Figure 12d shows the limit of
a single-reset system. There are still a fair number of
saturated pixels, but the lane markings have become
almost indistinguishable from the road surface, which
the compression, the lower the midtone contrast. At
levels of compression still far from theoretical limits,
portions of midtone detail almost disappear, and
the image becomes unnatural and less useful.
The underlying mechanism of this is explained by the
following imager response model.
Imager Response Model Response curve and noise
models show that the loss of midtone detail as observed
in Fig. 12 can be explained by a drop of iSNR that
occurs at the reset point. Figure 13 shows modeled
curves of OECF and iSNR as functions of log
luminance. The model assumes a constant dark noise
floor without considering photonic shot noise.
After the partial reset breakpoint, the slope of
the response curve (and thus the incremental signal)
is reduced so that saturation is delayed to
a correspondingly higher luminance level. With
increased compression, the drop in iSNR at the
breakpoint becomes deeper until it drops below
the reliability threshold as shown in Fig. 13c, thus
creating an undesirable zone of extinguished local con-
trast, a so-called iSNR hole. The luminance range
 True Color Night Vision Video Systems in Intelligent Vehicles 973

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 12

Natural scene captured with (a) linear camera and with a single-reset WDR camera at (b) low, (c) medium, and (d) high
compression, demonstrating the trade-off between increased dynamic range and decreased midtone contrast [5]

within and near the “iSNR hole” corresponds to the
luminance levels of those details such as road texture
and lane markings that became almost indistinguish-
able in Fig. 12d.
Along with saturated pixels or those below the noise
floor, “iSNR holes” make the detection of scene detail
impossible. In linear imagers, the zones of iSNR failure
occur only at very dark pixels (below noise floor) and
very bright pixels (above saturation level). However, in
multiple-slope WDR imagers, zones of iSNR failure
can also occur at luminance levels corresponding to
reset times. These levels cannot be predicted without
exact knowledge of the response curve for a given set of
reset times. If the resets are automatically adapted to
the intra-scene dynamic range, then the luminance
levels corresponding to the reset times would depend
on scene content and thus be variable, making iSNR
holes potentially interruptive to the visibility and auto-
matic detection of scene detail. It follows that resets
have to be controlled to prevent iSNR holes occurring
in the first place. Dynamic range can only be extended
as long as the incremental signal iS can be kept above
the noise floor at every luminance level. Thus, iSNR is
not just a criterion for determining the upper and lower
dynamic range limits but becomes critical when con-
trolling the reset times within the dynamic range.
Adaptive WDR extension requires separate control
loops for integration time and compression that both
adapt the imager’s response curve to include the
darkest and brightest scene details. The control
algorithms for timing the multiple partial resets in
adaptive WDR extension must be designed specifically
to avoid iSNR holes.
Reliably Extending Dynamic Range The example of
a multiple-slope WDR camera demonstrates how the
iSNR criterion can be used to extend dynamic range
without sacrificing detection reliability.
974 True Color Night Vision Video Systems in Intelligent Vehicles

Compression: c
iSNR

Image Signal D(L)

iSNR OECF(linear)
drop OECF(low c)
100 iSNR(low c)
Dynamic Range
10 where iSNR>1

0.1
−3 −2 −1 0 1 2 3
a log Luminance L [log cd/m2]

Compression: medium c
iSNR
Image Signal D(L)

iSNR OECF(linear)
drop
OECF(medium c)
100 iSNR(medium c)
Dynamic Range
10 where iSNR>1

0.1
−3 −2 −1 0 1 2 3
b log Luminance L [log cd/m2]

Compression: high c
iSNR
later
saturation
Image Signal D(L)

iSNR
OECF(linear)
“Hole”
OECF(high c)
100
iSNR(high c)
10 Dynamic Range
where iSNR>1
1

0.1
−3 −2 −1 0 1 2 3
c log Luminance L [log cd/m2]

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 13

Log-linear response curves (OECF) and incremental SNR of a single-reset model at (a) low compression, (b) medium, and
(c) high compression [5]

Image Signal D(L)
−3 −2 −1 0 1
log Luminance
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 14
Lower pixel-level noise increases dynamic range in a single-reset imager [5]
If the number of resets is limited to one or two, then
noise reduction is the most effective way of increasing
dynamic range. Figure 14 compares two single-reset
model imagers, one where altered pixel design (i.e.,
larger size, correlated double sampling, etc.) has cut
noise in half (6 dB). This leads to a twofold increase in
dynamic range. Firstly, the 6 dB lower noise floor leads
to a 6 dB lower noise-equivalent luminance signal Lmin.
In addition, the lower noise floor allows – without
creating an iSNR hole – the compression to be
increased by a factor of 2, halving the slope after reset
and doubling the saturation luminance Lmax, thus
increasing the dynamic range by another 6 dB, a total
of 12 dB. Assuming a minimum iSNR of 1, dynamic
range is increased from 88 to 100 dB or twice that of
a linear imager of 50 dB.
Where pixel-level noise reduction is not effective
(small pixels, low fill factor, high operating tempera-
tures, inexpensive pixel design), then increasing the
number of partial resets is a better strategy. The drop
in iSNR at a reset point depends on the ratio of incre-
mental gains after and before the reset, k = g/g0. The
more g(L) is reduced by a reset, the deeper iSNR will
drop. After each partial reset n, the sensitivity decreases
geometrically by the factor 1/kn. When deploying mul-
tiple partial resets, their timing and charge (voltage)
can be chosen so that the linear output signal range is
equally distributed between the resets [5]. This ensures
that iSNR at each reset point drops to the same mini-
mum level (see Fig. 15). Having more resets has the
added advantages: the decrease in incremental gain can
True Color Night Vision Video Systems in Intelligent Vehicles 975
Noise
reduction
iSNR
linear
reference
-6dB noise reduction
100
iSNR reference
10
iSNR (-6dB NR)
1 DR (-6dB NR)
0.1 DR reference
2 3 4
be approached more gradually, and the iSNR threshold
can be raised from one to a more practical level.
Figure 15 shows a dynamic range model for an increas-
ing number of resets. Assuming an increased iSNR
threshold of 5, the linear imager will have a dynamic
range of only 36 dB. Compression with a single reset
will extend the dynamic range to 62 dB; two resets will
reach 98 dB, and six resets 138 dB.
Conclusion
Although the concept of iSNR-based reliability has
been modeled and tested only for multiple-slope
CMOS imagers, it can be deduced that the concept
applies to any other WDR extension technology that
involves sudden changes in the incremental gain of
the pixel response curve. In contrast, sensors with log-
arithmic response are in no danger of “iSNR holes” as
their incremental gain decreases continuously during
exposure time.
For the example of multiple-slope CMOS imagers,
it has been shown that iSNR is a critical criterion
governing the number, timing, and recharge voltage
of partial resets. Dynamic range can be increased by
reducing noise at the pixel level, by increasing the
number of partial resets, and by distributing them
equally to maintain a minimum iSNR necessary for
reliable detail detection throughout the entire dynamic
range. As technologies for extending the dynamic range
evolve, new solutions will be added to the repertoire
and tested for their suitability for intelligent vehicle
976 True Color Night Vision Video Systems in Intelligent Vehicles
Image Signal D(L)
−3 −2 −1 0 1
log Luminance
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 15
More resets increase dynamic range while allowing a higher minimum iSNR threshold of 5 [5]
technology. However, the methodology described here
will provide a guideline for optimizing the pixel
response for maximum dynamic range at the highest
possible iSNR.
Extending Spectral Sensitivity into NIR
In order to capture dark scene detail illuminated by
ambient light, the camera has to be very sensitive with
high SNR. Exposure time is limited by the video frame
rate and the need to limit motion blur. Sensitivity can
be increased by either lowering noise (increasing SNR)
and/or extending the range of spectral energy sensed.
IR Sensing
The spectral sensitivity of silicon-based image sensors
extends beyond the long-wavelength edge of the visible
spectrum (730 nm) into the infrared (IR). As longer
wavelengths penetrate further into silicon, CMOS
imagers can be made more sensitive to IR by collecting
photo charge carriers from greater depth and by increas-
ing substrate thickness. The IR cut-off in CMOS imagers
can thus be shifted as high as 1,100 nm. This spectral
range coincides with the shorter wavelengths of the IR-A
or near-infrared (NIR) range (700–1,400 nm). Imagers
for longer wavelengths beyond the NIR rely on semi-
conductors with a narrower band gap, for example
InGaAs (950–2,600 nm). Such far-infrared (FIR) cam-
eras can sense the black body radiation that surfaces emit
as a function of their temperature, and obtain a thermal
# of resets
iSNR
OECF: linear
1 barrier
100
3 barriers
10 6 barriers
1 iSNR > 5 (6 barriers)
Dynamic Range
0.1
2 3 4 5 6 7
image of the environment. They do not require any
other source of radiation and can operate in complete
darkness. Compared to NIR systems, FIR night vision
systems are more costly, cannot be placed behind
glass, and deliver unnatural images that are difficult to
interpret [27, 28].
Passive and Active NIR Vision
The detection of objects by a NIR-sensitive camera
requires them to be irradiated by a NIR source so that
the camera can detect radiance image from reflected
NIR. Passive sensing relies on ambient NIR. Ambient
NIR at night comes mostly from artificial incandescent
light sources such as vehicle headlights; vapor street
lighting and natural light sources (moon, stars) emit
comparatively little NIR. Daylight especially direct sun-
light contains extremely high background NIR that is
strongly reflected by vegetation; cameras may have to
operate in WDR mode to avoid signal saturation.
Active sensing requires sources that are mounted to
the same vehicle as the NIR camera. Conventional
metal halide (tungsten) headlamps emit more than
60% of their spectral energy in the NIR. Dedicated
NIR-only headlamps, either tungsten lights with
a VIS-cut filter or NIR LEDs at 850 nm, have the
advantage that they are invisible to humans. As they
cannot blind oncoming drivers, they can be operated
on high-beam, thus extending the range of vision
beyond that of dipped headlights [29]. One of the

Spectral Responsivity
1 photopic (daytime)
0.8
R(λ) [normalized]
0.6
0.4
0.2
0
400 500
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 16
Comparison of photopic and scotopic luminous sensitivity with the spectral sensitivity of a monochrome automotive
CMOS image sensor [42]
first active systems implemented was the Mercedes-
Benz Night View Assist (2005).
Comparing the spectral sensitivity of a mono-
chrome CMOS image sensor with the photopic and
scotopic luminous sensitivity functions shows the
range of spectral energy that the image sensor can
utilize. For example, when the scene is illuminated by
metal halide headlamps whose emission spectra peak in
the NIR at about 900 nm, sensitivity stretches to the
long-wavelength cut-off of the image sensor (Fig. 16).
The example in Fig. 17 shows an urban traffic scene
taken without and with ambient NIR, showing the main
benefit of extended NIR sensitivity: better visibility of the
roadway where it is illuminated by dipped headlights.
The range of visibility can be extended even further,
without blinding oncoming drivers, if dedicated NIR-
only high-beam headlights are used. These will, of
course, “blind” any NIR-sensitive night vision cameras
of other vehicles, unless they are capable of performing
in a WDR environment.
Adding Color Differentiation to VIS + NIR-
Sensitive Imager
Using NIR to increase the sensitivity of color imagers
requires separation of the NIR and RGB color signals
because the color filter materials used in standard
True Color Night Vision Video Systems in Intelligent Vehicles 977
CMOS pixel
scotopic (night) vision
600 700 800 900 1000
Wavelength [nm]
CMOS and CCD technology are highly transparent to
NIR so that each color pixel also detects NIR (Fig. 18).
Mixing color and NIR signals can result in extreme
color de-saturation if the illumination contains high
amounts of IR. This is the case when incandescent
lamps are used with color temperatures typically
about 3,200 K, where spectral power distribution
peaks at about 800 nm (Fig. 7). Here, the signal con-
tributions to RGB from the NIR are even higher than
those from the actual colors so that all colors are ren-
dered as off-gray, for example shades of pink (Fig. 19).
The example demonstrates that the differentiation of
color requires the NIR signal to be separated from the
color signals.
Increasing the Spectral Sensitivity of Color Imagers
Color image sensors simply suppress the NIR infor-
mation by an IR-cut filter. The sensitivity of such
imagers can be improved by adding filterless “white”
pixels to the RGB filter pattern [30, 31] (Fig. 20).
Compared to the RGB color pixels, these “white” pixels
have about three times higher spectral sensitivity.
As scene illuminance decreases, information from
color pixels is increasingly mixed with luminance infor-
mation from the W-pixels by pixel binning. This
requires high pixel counts plus special (nonstandard)
978 True Color Night Vision Video Systems in Intelligent Vehicles
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 17
Monochrome image of urban traffic scene (a) without NIR and (b) with NIR
filter patterns, and the NIR remains unutilized for
night vision.
Extending the Spectral Sensitivity of Color Image
Sensors into the NIR
Extending the light sensitivity of color imager sensors
into the NIR requires a separation of the NIR and the
color signals. This could be achieved by using a beam
splitter and two image sensors, one monochrome for
NIR, and one color for VIS. However, cost and form
factor do not recommend this solution for automotive
vision applications.
Single-imager solutions are based on the addition
of a fourth pixel type to the RGB-pixels (Fig. 20c).
This fourth pixel, which, for example, replaces the
second G-pixel in the Bayer pattern, is either NIR-
sensitive or “panchromatic,” sensitive to both visible
and NIR light. A variety of methods have been
proposed to separate NIR signals from color signals:
● Patterned optical NIR-cut filters can be added to
mosaic filter arrays, for example RGBW or CymG
(cyan-yellow-magenta-green) [32], to protect all or
a portion of color pixels from the effects of NIR
radiation. The filters can be of the absorptive or
interference type. Interference filters are preferred
for their sharper absorption edge toward the NIR,
but they do require patterned deposition of more
than ten dielectric layers [36], which is both costly
compared to dye-based filters and uses dielectric
materials that are incompatible with standard Si
processes. The effectiveness of such filters is limited
by optical crosstalk of NIR into the color pixels after
passing the NIR cut filter, especially at low incident
angles.
● Nano-optical filters, currently still in the research
stage, can be designed as short-pass edge filters or
band-pass filters. They can either be integrated as
metal layers into the pixel stack [37] or deposited
 True Color Night Vision Video Systems in Intelligent Vehicles 979

Color Filter Transmission

1

0.8

0.6
T(λ) B
0.4 G
R
0.2

0
1

0.8
R(λ) [normalized]

0.6

0.4

0.2

0
400 500 600 700 800 900 1000
Wavelength [nm]

NIR

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 18

VIS–NIR spectral transmission curves of typical RGB color filters used in CMOS imagers [35]

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 19

Raw images of a Macbeth color checker chart illuminated by incandescent light (3,200 K) taken with a CMOS color camera
(a) with IR-cut filter and (b) without IR-cut filter [35]

on top of the imager [38]. Although nanostruc-
tured metal layers within the pixel stack have to
follow the Si design rules for structure size (0.15,
0.13, 0.11, 0.09 mm) which limits the degrees of
freedom for spectral tuning, they are practically
free from optical crosstalk. Filters on top of the
imager can be created by nanoimprint technolo-
gies which allow the creation of an array of
plasmonic band-pass filters. However, the peak
transmission of band-pass filters is relatively
980 True Color Night Vision Video Systems in Intelligent Vehicles

low (<0.4). Peak filter transmission decreases,
and halfwidth increases with wavelength, making
nano-optical filters less effective in the NIR
(Fig. 21) [38].
● Virtual NIR filters separate NIR and color signals by
image signal processing. One possible implementa-
tion adds filterless pixels to the RGB filter pattern,
but in contrast to RGBW, no optical NIR-cut filter is
used [33–35]. The filterless pixel has two functions:
Firstly, it has a “panchromatic” spectral response
R G1 W B W G R′
B W G W
G2 B P′
a c
W G W R
that includes the NIR and is equivalent to that of
a monochrome imager. The panchromatic P0-pixel
is more than three times as sensitive as a color
imager, capable of delivering a much brighter lumi-
nance image. Secondly, the P0-pixel delivers the
reference signal necessary to estimate the NIR signal
contribution to the R0G0 B0 pixel signals. The raw
color pixel responses are R 0 = R + I, G 0 = G + I, and
B 0 = B + I (Fig. 22). The panchromatic pixel
response P 0 = R + G + B + I provides the reference
signal for estimating the NIR portion in the inci-
dent light, which is then used in the virtual NIR
filter matrix to remove as much as possible of the
G′ NIR contributions to the R0 G0 B 0 signals. The
response of the P0-pixel can also be used to simply
B′ retrieve a panchromatic VIS–NIR luminance
channel LVIS–NIR.

G W R W An example of raw spectral pixel responses (R 0 (l),

b
G (l), B 0 (l), P 0 (l)) shows the high NIR signal contri-
0

True Color Night Vision Video Systems in Intelligent bution to the color signals as well as high amounts of
Vehicles. Figure 20 crosstalk between the four pixel types. Since both dete-
CFA layouts: (a) standard Bayer, (b) RGBW, and (c) RGBP [35] riorate colors, they have to be minimized by determining

nano-optical filter set

0.3 1

B′
0.25 G′
0.8
R′
Filter T * Pixel response

0.2 IR
0.6
Imager
CPX DU

0.15

0.4
0.1

0.2
0.05

0 0
400 500 600 700 800 900 1000 1100
Wavelength [nm]

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 21

Simulated response curves for CMOS imager with plasmonic nano-optical filters [38]
 True Color Night Vision Video Systems in Intelligent Vehicles 981

B G R IR the most relevant colors are not reflective but rather

emissive colors whose brightness might lie far out-
side the set, and whose illuminant, and thus, rela-
tive NIR content varies widely.
● The spectral calibration method optimizes the
matrix coefficients bmn so that the filtered spectral
B′ G′ R′ P′ response functions (R(l), G(l), B(l), I(l)) fulfill
application-specific conditions.
2 3 2 3 2 0 3
R ð lÞ b11 b12 b13 b14 R ð lÞ
6 G ðlÞ 7 6 b21 b22 b23 b24 7 6 G 0 ðlÞ 7
6 7 6 7 6 7
4 B ðlÞ 5 ¼ 4 b31 b32 b33 b34 5  4 B 0 ðlÞ 5
I ð lÞ b41 b42 b43 b44 P 0 ðlÞ
B+IR G+IR R+IR BGR+IR ð6Þ

For example, in a machine-vision application, the

Virtual NIR filter
B G R IR
True Color Night Vision Video Systems in Intelligent
Vehicles. Figure 22
Light, pixel responses of the R0 G0 B0 P0 imager and image
signals after virtual NIR filtering [35]
the virtual NIR filter coefficients, for which two calibra-
tion methods are commonly used, global and spectral.
● The global method relies on a set of known refer-
ence colors, for example those of a Macbeth color
checker chart, that have to be reproduced as faith-
fully as possible. The color match can be done in
a standardized color space, for example CIE 1931
XYZ, and the matrix MG optimized so that the
transformed colors match the reference colors
[XYZ] as closely as possible,
2 03
2 3 R
X nation conditions with high amounts of NIR, for
6 G0 7
4 Y 5  MG  6 0 7: ð5Þ
4B 5
Z 0
P
Global calibration works well for colors within and
close to the reference set and under illumination
conditions similar to those used in the calibration,
but it is poorly suited for color night vision where
coefficients are optimized to remove from the
transformed spectral responses (R(l), G(l), B(l), I(l))
as much of the NIR and color crosstalk signals
as possible. Instead of retrieving an IR channel,
the P0- response can be mapped into a panchromatic
luminance channel LVIS–NIR. In human vision applica-
tions, a colorimetric calibration is better suited for
achieving natural-looking colors. Colorimetric calibra-
tion as shown in Fig. 23c tries to optimize the matrix
MS so that the filtered spectral response curves of the
imager approximate the spectral color matching
functions (b(l), g(l), r(l)) of the human visual system
as specified in ISO 17321 [39]:
2 0 3
2 3 R ð lÞ
r ð lÞ 6 0 7
4 g ðlÞ 5  MS  6 G0 ðlÞ 7: ð7Þ
4 B ð lÞ 5
b ð lÞ
P 0 ð lÞ
Spectral calibrations have the advantage of not rely-
ing on a particular set of reference colors, and also,
they do not rely on a specific illuminant. The virtual
NIR filter with colorimetric spectral calibration reduces
the NIR crosstalk into color signals even under illumi-
example incandescent light at 3,200 K color tempera-
ture. The example in Fig. 24 shows images before (a),
(b), and after virtual NIR filtering (c) with increasing
amounts of NIR. Increasing NIR reduces color satura-
tion of the raw R0 G0 B0 color channels, and increases
brightness of the panchromatic P0-channel. The colors
reconstructed by the virtual NIR filters become darker
982 True Color Night Vision Video Systems in Intelligent Vehicles

1 B′
0.8 G′

Pixel response
0.6 R′

0.4 P′

0.2

0
a 400 500 600 700 800 900 1000
Wavelength [nm]

2
Channel response

1.5 B
G
1
R
0.5
IR
0

–0.5
400 500 600 700 800 900 1000
b
Wavelength [nm]

2 B (transformed)
G (transformed)
1.5 R (transformed)
Channel response

b (aim)
1
g (aim)
0.5 r (aim)

–0.5

–1
c 400 500 600 700 800 900 1000
Wavelength [nm]

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 23

Spectral pixel responses of (a) the R0 , G0 , B0 , and P0 pixels and spectral channels after (b) separating NIR from the RGB color
channels and (c) colorimetric calibration [42]

with increasing levels of NIR because of decreasing Extending the Dynamic Range of Color Imagers with
color-to-NIR signal ratios in the raw images. NIR Sensitivity
However, residual spectral leakage limits the per-
Signal saturation in color imagers leads to loss in color
formance of the virtual NIR filter, especially in WDR
information; as the signals in the R, G, and B channels
mode where initial signal ratios are reduced. Careful
approach saturation levels, colorfulness decreases. In a
management of NIR can further reduce the effects of
RGB color imager, signal clipping in one or more color
NIR crosstalk and has to include pixel and microlens
channels leads to a whitening of colors; clipping in
design to minimize optical and electrical crosstalk,
a four-channel system such as R0G0 B0 P0 can also lead
high SNR to counter the noise-amplifying effect
to false colors. WDR extension is therefore essential to
of high filter coefficients, and NIR-only headlights to
capture color detail carried by bright lights. However,
minimize the proportion of NIR radiation with the
corrections have to be made for the effects of WDR
highest leakage into the color signals.
 True Color Night Vision Video Systems in Intelligent Vehicles 983

Short-pass filtered with cut-off wavelengths at:

650nm 700nm 750nm 850nm 1100nm

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 24

Macbeth color checker chart illuminated by incandescent light (3,200 K) and imaged with short-pass filters allowing
increasing amounts of NIR: (a) raw R0 G0 B0 image, (b) raw P0 image, and (c) RGB image after virtual NIR filtering [42]. The false
colors in the brightest patch are caused by clipping of the P0 -signal

extension on the color signals. In the example of mul-
tiple exposures with varying exposure time, each expo-
sure maps a different range of input radiance signals
into the same range of output image signals. Within the
sequence of images to be combined into a single WDR
image, the same output signal may be produced by
different combinations of input signals so that the
relationship between input and output signals is not
unique. Therefore, the combination of exposures into
the WDR image has to be preceded by linearization,
which uses the camera response function in each color
channel to create a unique map between the digital
signal levels in the WDR image and the input radiance
levels. The response function can be derived by tech-
niques such as those proposed by Debevec and Malik
[40], or Mitsunga and Nayar [41].
Direct WDR methods that extend the dynamic
range in a single exposure have the great advantage of
providing a unique map between input radiance and
output image signal levels (Fig. 9). However, if the
response function is nonlinear, for example piecewise
linear as shown in Fig. 11, the signal ratios between the
color channels will decrease with decreasing slope of
the response curve. Increasing compression will thus
decrease the input signal ratios between the raw
R0 G0 B0 P0 channels, thus decreasing chromaticity and
iSNR of the output colors after virtual NIR filtering.
To recover the color components and remove the NIR
component from the highly compressed signals of
bright colored lights, the raw R0G0 B0 P0 pixel signals
have to be linearized with the inverse response curve
of the imager so that the image signals can be referred
back to the corresponding scene radiances (inverse
gamma correction). The effect of linearization is
shown in Fig. 25 where it restores the colors in the
highly compressed images of bright colored lights
below the Macbeth color checker chart [42].
Visual Representation of the VIS + NIR Color
Information
Section “Conclusion: Producing Natural Images Under
Mesopic Conditions” outlined the requirements for
rendering a natural-looking night vision image. In the
example of a WDR imager with panchromatic and
color pixels, NIR sensitivity increases the range of
984 True Color Night Vision Video Systems in Intelligent Vehicles

Pixel response WDR extension Dynamic Range

130dB
118dB

110dB
Image digit

94dB

78dB

48dB

–3 –2 –1 0 1 2 3 4
a
log Luminance
Linear: Increasing WDR compression:
48dB 78dB 94dB 110dB 118dB 130dB

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 25

WDR color scene consisting of dark reflective colors (top), bright transmissive colors with (center) and without NIR (bottom),
taken with (a) imager response of increased compression and displayed (b) without and (c) with linearization [42]

visibility, and the panchromatic image information can
be displayed in the luminance channel. The virtual
NIR-cut filter can recover traffic-relevant color infor-
mation from backlit or retroreflective signage, traffic
lights, and taillights. Virtual NIR filtering shows rea-
sonably good performance in reconstructing colors
even under illumination conditions with high amounts
of NIR, for example incandescent light at a color tem-
perature of 3,200 K having 20% signal in the visual and
80% in the NIR range (Fig. 24). Since the filter array
relies on standard RGB filter materials and fabrication
processes, it can be produced at comparably low costs
as conventional RGB imagers, without the extra cost of
a patterned optical NIR-cut filter.
The VIS and NIR image information has to be
represented in a way that the human observer viewing
video on a dashboard screen can easily interpret. The
VIS + NIR color camera with virtual NIR filter can
provide multispectral image information, for example,
in four channels, RGB and NIR. Multichannel image data
could simply be mapped into a false-color representa-
tion, but this creates unnatural images that do not agree
well with visual expectations and can even be confusing
rather than useful. Natural image data representation
that agrees with human expectations can be achieved by
following certain design guidelines [43]: (1) orthogonal
encoding to represent information with as little redun-
dancy as possible; (2) anthropometric optimality to max-
imize usefulness to the human observer; (3) ecological
invariance so ensure similarity of image interpretation
over a variety of environmental parameters such as time
of day, illuminants, and weather conditions; (4) compu-
tational simplicity to allow video implementation. For
example, the achromatic channel of the HVS carries over

95% of the scene information and is much more sensitive
to spatial information than the opponent color channels
which process color information. Therefore, true-color
image fusion is best suited for this application because it
uses the extra information from the NIR to enhance the
spatial and luminance information in the achromatic
channel while mapping the color information from
the visible range into the chrominance channels
[32, 44, 45]. First, the RGB image obtained by
virtual NIR filtering (6) must be transformed into
a luminance–chrominance color space; then, its lumi-
nance component LVIS is replaced by the luminance
LVIS+NIR estimated from the P0-pixels. Additional
processing of LVIS+NIR ensures that the fusion
increases the brightness of the dark scene background
but not so much that of colored lights, thus preventing
the loss of color saturation in colored lights after
fusion. For example, tone mapping using global,
local, or Retinex algorithms significantly improves
the visibility of grayscale and color detail in WDR
images [46].
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 26
Night scene with tungsten headlights as only light source, taken without image fusion (a) and with image fusion (b)
True Color Night Vision Video Systems in Intelligent Vehicles 985
The night scene in Fig. 26, illuminated only by the
car’s headlights, shows how image fusion effectively
extends the field of vision, increases visibility of detail,
and retains color in the retroreflective sign.
A previously described WDR scene simulator [5]
composed of a digital display and a projector has been
used for creating test scenes with controlled luminance
levels, covering a wide dynamic range of 100 dB.
Individual scene elements such as the road surface, lane
markings, traffic signs, pedestrians, and headlights
shown in Fig. 27 were calibrated so that their absolute
luminance levels were within the ranges of typical traf-
fic scenes as shown in Fig. 3. The example demonstrates
that even if little or no IR is present in the scene, image
fusion will still increase the brightness of dark scene detail
because the filterless P0-pixels are more sensitive than any
of the color pixels. Fusion utilizes the brighter luminance
image from the P0-pixels, making the pedestrians in the
dark background much apparent. The retroreflective sign
and traffic lights are also brightened, but only to the
extent that color saturation is not adversely affected.
986 True Color Night Vision Video Systems in Intelligent Vehicles
True Color Night Vision Video Systems in Intelligent Vehicles. Figure 27
Simulated 100 dB WDR traffic color scene (a) without and (b) with image fusion [35]
In real-world night vision scenes, the panchromatic
pixel can utilize the large NIR component of incandes-
cent (tungsten) headlights that is invisible to the driver
(Fig. 7). Furthermore, high-beam NIR-only headlights
can be used to increase visibility beyond the distance
range of headlights at low beam.
Example: Render Traffic-Relevant Color Information
in Natural Images
Figure 28 demonstrates the image rendering of a
VIS–NIR true-color WDR night vision camera for
a typical urban night traffic scene.
The monochrome image in Fig. 28a shows that the
wide spectral sensitivity of the panchromatic P0-
pixels delivers a sufficiently bright monochrome
image even at a frame rate of 30 fps. The color
image in Fig. 28b is comparatively dark, consisting
mainly of the “unrelated colors” of traffic lights and
vehicle taillights against an almost black background.
The spectral energy in the visible band is small
compared to the total energy within the imager’s
sensitivity range so that the luminance component
of the filtered colors is much lower than the lumi-
nance from the P0-pixels. WDR compression and
subsequent linearization ensure that the bright high-
light colors are preserved. The fused image in Fig. 28c
then combines the bright monochrome background
with the unrelated colors, leading to a more natural
rendering of the scene that contains all traffic-relevant
color information from traffic lights and car signal
lights.
Conclusion
True-color night vision systems that utilize the NIR
sensitivity and WDR capability of CMOS image
sensors can aid the driver in dangerous situations
where human vision is inadequate or challenged
(Table 2). NIR sensitivity extends the range of vision
when the safe braking distance exceeds the range of
headlights. Cameras with adaptive WDR compres-
sion provide natural images of scenes where the
intra-scene dynamic range exceeds that of the
human eye, for instance when blinded by headlights
of oncoming vehicles. Such cameras are able to
 True Color Night Vision Video Systems in Intelligent Vehicles 987

True Color Night Vision Video Systems in Intelligent Vehicles. Figure 28

Urban night traffic scene: (a) panchromatic P0 image, (b) RGB color image after virtual NIR filtering, and (c) final image after
image fusion [8]

adapt to sudden changes in scene brightness much Future Directions

faster than the HVS so that image information is
Vision-based DAS are one of a wide range of intelligent
available while the eye is still in the process of
vehicle technologies that can be implemented either as
adapting, for example to the “black hole” after
stand-alone or cooperative systems. Some are already
bright headlights of a passing car have left the
in use (antilock braking system, electronic stability
field of vision.
control) while others are still in the research stage,
988 True Color Night Vision Video Systems in Intelligent Vehicles

True Color Night Vision Video Systems in Intelligent signals. Furthermore, they have to perform under
Vehicles. Table 2 Advantages of camera-based night adverse weather conditions that severely tax human
vision systems vision, such as rain, drizzle, fog, snow, and salt spray.
The performance and reliability of future night
True-color
Human VIS–NIR vision systems might be improved in many ways,
visual camera Advantages of perhaps including:
system with WDR camera
● Improved NIR management by integrating
NIR No Up to Extended range adaptive NIR-only headlights into the night vision
sensitivity 1,100 nm of vision
system [49]
Adaptation Transient 3–6 frames Image ● Increased SNR of the image sensor to increase
to brightness 1–2 s (0.2–0.1 s) information
dark sensitivity and counter the increased
and dynamic available at WDR
range and during visibility of color noise after virtual NIR
sudden changes filtering
in brightness ● Reduced NIR crosstalk into the color signal through
Color vision Mesopic Color Consistent color CFA and color pixel design, for example, nano-
shift constancy information optical color filters [37, 38]
● Reduced NIR leakage in virtual NIR filtering by
using algorithms that further reduce NIR leakage,
for example, neural networks [50]
under development, or in the process of being intro- ● Improved WDR color processing, for example,
duced into the market [47, 48]. Retinex-based tone mapping [46]
Vision-based driver assistance systems aim to
In addition, head-up displays such as windscreen
enhance the visual perception of the road ahead in
projection will be able to communicate visual informa-
situations where the human visual system is challenged
tion to the driver without shifting attention away from
or even fails to perceive dangerous situations. The scene
the road as necessitated by a head-down display outside
information captured by the camera is processed into
the field of vision, for example, the dashboard displays
a natural-looking image and displayed in the driver’s
of early night vision systems [28].
field of view. Since more than 90% of driving decisions
The main shortcoming of any display-based vision
and actions are based on vision, the fastest and most
system is that it only improves traffic safety if the driver
efficient human–machine interface is a visual display.
pays attention to the display at the critical moment. For
Their combination of relative simplicity with high
example, it is of no use if the driver does not check the
capacity and the speed of the visual information
display as an obstacle beyond the field of vision is
channel recommended display-based vision systems to
displayed by the NIR-sensitive system before the driver
early proponents of intelligent vehicle technology.
can see it through the windshield. Recent developments
Active NIR-sensitive night vision systems such as the
in intelligent vehicle technology work toward addressing
Toyota Night View (2002) and Mercedes-Benz Night
this shortcoming and move beyond the early display-
View Assist (2005) were pioneered in top-of-the-range
based vision systems. The two main strategies are intel-
cars. Despite the relative simplicity of camera-to-display
ligent stand-alone systems to enhance vehicle safety, and
night vision systems versus machine-vision-based
cooperative systems to improve road traffic safety [48].
automatic driver assistance systems (such as lane keep-
ing, obstacle detection, etc.), the technical challenges ● Stand-alone, in-vehicle systems use sensors to
and reliability requirements are so high that their detect possible dangers and try to avoid or mitigate
penetration into the market is still low. The night vision accidents by following a three tier strategy:
system must deliver useful natural images under an (1) inform the driver as early as possible, (2) issue
extremely wide range of lighting conditions while warnings if there is no driver reaction to the infor-
experiencing changing proportions of color and NIR mation, and (3) actively assist or ultimately

intervene in the vehicle control [48]. In-vehicle
systems will rely on multiple vision sensors and
combine different sensor technologies such as
radar and LIDAR (light detection and ranging)
alongside vision sensors to improve overall
detection reliability. Display-based vision systems
can be expected to remain an important integral
part of on-board safety systems because a visual
display more immediately informs the driver of
the nature of the danger than an acoustic warning
would.
● Cooperative systems increase the capacity for gath-
ering information by using communications from
infrastructure or other sensor-equipped vehicles to
make an individual vehicle less dependent on its
own sensors and local environmental conditions.
For example, if a car drives into a bank of fog, it
can be guided by infrastructure and information
from other vehicles outside the danger zone.
Forward-looking VIS–NIR cameras with color dis-
crimination and WDR will continue to produce useful
image information for both human and machine
vision systems. Even if camera-based night vision
systems with visual human–machine interfaces turn
out to be only a transition stage toward future intelli-
gent vehicle solutions, the technical experience
gleaned by the implementation and application of
pioneering vision systems may still prove
indispensible for the future development of intelligent
vehicle technology.
Bibliography
Primary Literature
1. Traffic Accident Causation in Europe (TRACE) (2007) Review of
crash effectiveness of intelligent transport systems, Project
No. 027763, Deliverable D4.1.1–D6.2, 55
2. Jones WD (2006) Safer driving in the dead of night – infrared
vision systems are set to become standard in high-end cars.
IEEE Spectrum Green Technology. http://spectrum.ieee.org/
green-tech/advanced-cars/safer-driving-in-the-dead-of-night
3. Yendrikhovskij S (1999) Image quality: between science and
fiction. In: Proceedings of the IS&T PICS, Savannah, pp 173–178
4. Hertel D, Betts A, Hicks R, ten Brinke M (2008) An adaptive
multiple-reset CMOS wide dynamic range imager for automo-
tive vision applications. In: Proceedings of the 2008 IEEE intel-
ligent vehicles symposium, IEEE Press, Piscataway, pp 614–619
True Color Night Vision Video Systems in Intelligent Vehicles 989
5. Hertel D (2010) Extended use of incremental signal-to-noise
ratio as reliability criterion for multiple-slope wide-dynamic-
range image capture. J Electron Imaging 19(1):011007
6. Marjukka E, Liisa H (2005) MOVE – performance based model
for mesopic photometry. Helsinki University of Technology
Report No. 35, ISBN 951-22-7566-X, Espoo, Finland
7. Fu C, Li C, Luo MR, Hunt RWG, Pointer MR (2007) Quantifying
colour appearance for unrelated colour under photopic and
mesopic vision. In: Proceedings of the IS&T 15th color imaging
conference, Albuquerque, pp 319–324
8. Hertel D, De Locht C (2010) Utilizing near-infrared and
colour information in an automotive wide-dynamic-range
night vision system for driver assistance. In: Meyer G,
Valldorf J (eds) Advanced microsystems for automotive
applications 2010 – smart systems for green cars and
safe mobility. Springer, Dordrecht/London/New York,
pp 145–154
9. Carlson PJ, Urbanik T (2004) Validation of photometric model-
ing techniques for retroreflective traffic signs. Transp Res Rec
1862:109–118
10. Keelan BW (2002) Handbook of image quality. Dekker,
New York, pp 329–399
11. Hertel D, Chang E (2007) OECF characterization of a non-linear
HDR color camera for automotive applications. In: Proceed-
ings of the IS&T 15th color imaging conference, Albuquerque,
pp 244–248
12. Dierickx B (2008) Wide dynamic range: what should I care
about? IEEE international solid-state circuits conference
(ISSCC ‘08), imager design forum, wide dynamic range
imaging, San Francisco
13. Williams D (2003) Debunking specsmanship: progress on ISO/
TC42 standards for digital capture imaging performance. In:
Proceedings of the IS&T PICS, Rochester, pp 77–81
14. Sangwine SJ, Horne REN (1998) The colour image processing
handbook. Springer, New York, p 127
15. International Standard ISO 12232 (1997) Photography –
electronic still picture cameras – determination of ISO
speed
16. International Standard ISO 14524:1999(E) (1999) Photography –
electronic still-picture cameras – methods for measuring opto-
electronic conversion functions (OECFs)
17. International Standard ISO 15739:2003(E) (2003) Photography
– electronic still-picture imaging – noise measurements
18. Görgens E (1987) Objective evaluation of image quality. J Inf
Rec Mater 15:305–321
19. Kerr DA (2008) The ISO definition of the dynamic range of
a digital still camera. http://dougkerr.net/Pumpkin/articles/
ISO_Dynamic_range.pdf
20. Dainty JC, Shaw R (1974) Image science. Academic Press,
London, pp 152–171
21. Chamberlain SG, Lee JPY (1984) A novel wide dynamic range
silicon photodetector and linear imaging array. IEEE J Solid-
State Circuits SC-19(2):41–48
22. Storm G, Henderson R, Hurwitz JED, Renshaw D, Findlater K,
Purcell M (2006) Extended dynamic range from a combined
990 True Color Night Vision Video Systems in Intelligent Vehicles
linear-logarithmic CMOS image sensor. IEEE J Solid-State
Circuits 41(9):2095–2106
23. Wang X, Wong W, Hornsey R (2006) A HDR CMOS image
sensor with inpixel light-to-frequency conversion. IEEE Trans
Electron Devices 53(12):2988–2992
24. Yang D, El Gamal A, Fowler B, Tian H (1999) A 640  512 CMOS
image sensor with ultrawide dynamic range floating-point pixel-
level ADC. IEEE J Solid-State Circuits SC-34(12):1821–1834
25. Kawahito S (2006) A sensitivity and linearity improvement on
a 100 dB dynamic range image sensor using a lateral overflow
integration capacitor. IEEE J Solid-State Circuits 41(4):
851–858
26. Decker S, McGrath RD, Brehmer K, Sodini CG (1998) A 256 
256 CMOS imaging array with wide dynamic range pixels and
column-parallel digital output. IEEE J Solid-State Circuits SC-
33(12):2081–2091
27. Knoll PM (2002) A NIR based system for night vision improve-
ment. SAE Convergence 2002, Detroit
28. Kallhammer J-E (2006) Night vision: requirements and possible
roadmap for FIR and NIR systems. SPIE 6198, 61980 F
29. Hörter MH, Stiller C, Koelen C (2009) Hardware and software
framework for automotive intelligent lighting. In: Proceedings of
the 2009 IEEE intelligent vehicles symposium, Xi’an, China, p 299
30. Hamilton JF, Compton JT (2007) Processing color and pan-
chromatic pixels. US Patent Application 20070024879
31. Honda H, Iida Y, Itoh G, Egawa Y, Seki H (2007) A novel Bayer-
like WRGB color filter array for CMOS image sensors. In: Pro-
ceedings of the SPIE-IS&T electronic imaging, SPIE, vol 6492,
pp 64921 J.1–64921 J.10
32. Kriesel M, Gat N (2008) True-color night vision cameras. In:
Proceedings of the SPIE, vol 6540, pp 65400D.1–65400D.10
33. Frame WW (2006) One chip, low light level color camera. US
7,012,643 B2
34. Kidono K, Yoshiki N (2007) Visibility estimation under night-
time conditions using a multiband camera. In: Proceedings of
the 2007 IEEE intelligent vehicles conference, Istanbul, Turkey,
pp 1013–1018
35. Hertel D, Marechal H, Tefera DA, Fan W, Hicks R (2009) A low-
cost VIS-NIR true color night vision video system based on
a wide dynamic range CMOS imager. In: Proceedings of the
2009 IEEE intelligent vehicles symposium, Xi’an, China, p 273
36. Koyama S et al (2009) A day and night MOS imager spectrally
adjusted for a wide range of color temperatures. SPIE 7249,
72490 S
37. Catrysse PB, Wandell BA (2007) Integrated color pixel. U.S.
Patent No. 7,248,297 (Filed 2001 – Issued 2007)
38. Choi B (2007) Nano-technology for color: plasmonic nano-
optic filter technology for color capturing, expression, and
management. In: Proceedings of the IS&T CIC15, Albuquerque
39. International Standard ISO 17321:1998 (1998) Graphic technol-
ogy and photography – colour target and procedures for
the colour characterization of digital still cameras (DSCs). ISO,
Geneva, Switzerland
40. Debevec PE, Malik J (1997) Recovering high dynamic
range radiance maps from photographs. In: Proceedings
of the SIGGRAPH 97 annual conference, Los Angeles,
pp 369–378
41. Mitsunga T, Nayar SK (1999) Radiometric self calibration. In:
Proceeding of the IEEE computer vision and pattern recogni-
tion, Fort Collins
42. Hertel D (2010) Using spectral response measurements to
optimise the colour processing of a wide dynamic range
colour camera with near-infrared sensitivity. The royal photo-
graphic society, symposium on colour imaging 2010, http://
www.rps-isg.org/SCI2010_programme.php
43. Scott Tyo J (2007) Displaying advanced optical imagery infor-
mation for human observers. In: Proceedings of the IS&T
CIC15, Albuquerque, pp 174–177
44. Ngo H, Tao L, Zhang M, Livingston A, Asari V (2005) A visibility
improvement system for low vision drivers by nonlinear
enhancement of fused visible and infrared video. IEEE com-
puter society conference on computer vision and pattern
recognition (CVPRW’05) – Workshops, p 25
45. Fredembach C, Süsstrunk S (2008) Colouring the near-infrared.
In: Proceedings of the IS&T 16th color imaging conference,
Portland, pp 176–182
46. Meylan L, Süsstrunk S (2006) High dynamic range image ren-
dering using a retinex-based adaptive filter. IEEE Trans Image
Process 15(9):2820–2830
47. Bertozzi M, Broggi A, Fascioli A (2000) Vision-based intelligent
vehicles: state of the art and perspectives. Robot Auton Syst
32:1–16
48. European Commission, Information Society and Media:
i2010 intelligent car initiative. http://ec.europa.eu/information_
society/activities/intelligentcar/docs/icar/intelligent_car.pdf
49. Carley L (2009) News technologies for safer driving: adaptive
lighting, night vision, backup cameras & blind spot
detection. Import car, http://www.import-car.com/Article/
60749/News_Techn.aspx
50. Kang HR, Anderson PG (1992) Neural network applications to
the color scanner and printer calibrations. J Electron Imaging
1:125–134
Books and Reviews
Frieser H (1975) Photographische Informationsaufzeichnung.
Focal Press, New York
Hunt RWG (2004) The reproduction of colour. Fountain Press,
Tolworth, England
Proceedings of IEEE intelligent vehicles symposium 2007, Istanbul,
Turkey, 13–15 June 2007, ISBN 1-4244-1068-1
Proceedings of IEEE intelligent vehicles symposium 2008, Eindho-
ven, The Netherlands, 4–6 June 2008
Proceedings of IEEE 2009 IEEE intelligent vehicles symposium,
Xi’an, China, 3–5 June 2009, ISBN 978-1-4244-3502-9
Reinhard E, Ward G, Pattanaik S, Debevec P (2006) High dynamic
range imaging – acquisition, display, and image-based
lighting. Morgan Kaufmann Publishers, San Francisco
Sharma G (2003) Digital color imaging. CRC Press, Boca Raton
Wyszecki G, Stiles WS (2000) Color science. Wiley, New York
 Unscented Kalman Filter in Intelligent Vehicles 991

Unscented Kalman Filter in the dynamical responses of a vehicle in various

in-motion situations.
Intelligent Vehicles Vertical (normal) tire forces They are responsible to
MOUSTAPHA DOUMIATI1, DANIEL LECHNER2 support the weight of the vehicle.
1
B2i Automotive Engineering Company, Massy, France
2
Department of Accident Mechanism Analysis, Definition of the Subject
IFSTTAR-MA Laboratory, Salon de Provence, France
The principal concerns in driving safety with standard
vehicles are understanding and preventing risky situa-
tions. A close examination of accident data reveals that
Article Outline
losing the vehicle control is the main reason for most
Glossary car accidents. To help the driver to prevent such acci-
Definition of the Subject dents, vehicle control systems may be used. For their
Introduction optimal operation, these control systems require cer-
Estimation Process Description tain input data concerning vehicle dynamic parameters
Four-Wheel Vehicle Model and vehicle–road interaction. Unfortunately, some fun-
Tire-Road Interface damental parameters like the tire-road forces and the
Observers Design sideslip angle are difficult to measure in a car, for both
Experimental Results technical and economic reasons. To face this problem,
Conclusion and Future Perspectives this study presents a dynamic modeling and observation
Acknowledgments method to estimate these variables. The ability to accu-
Bibliography rately estimate lateral tire forces and sideslip angle is
a critical determinant in the performances of many
vehicle control systems. To address nonlinearities and
Glossary unmodeled vehicle dynamics, an observer derived from
unscented Kalman filtering technique is proposed. The
Lateral tire forces They are responsible to hold on
estimation process method is based on the dynamic
the vehicle during a turn.
response of a vehicle instrumented with easily available
Longitudinal tire forces They are responsible to
and potentially integrable sensors. Performances are
accelerate/brake the vehicle.
tested using an experimental car in real driving situa-
Observer or estimator It models a real system in
tions. Experimental results show the potential of the
order to provide an estimate of its internal state,
proposed estimation method.
given measurements of the input and output of the
real system.
Introduction
Sideslip angle It is the angle between the velocity
heading and the true heading of the vehicle. Vehicle dynamics and stability have been of consider-
Tire forces The developed forces (longitudinal and able interest to automotive engineers, automobile
lateral) are function of tire properties (material, manufacturers, government, public safety groups, and
tread pattern, tread depth, profile, etc.), the normal general public for a number of years. The obvious
load on the tire, and the velocities experienced by dilemma is that people naturally desire to drive faster
the tire. and faster on the roads and highways, yet they expect
Vehicle control systems They provide commands their vehicles to be stable and safe during all normal
and instructions to control the movements of the and emergency maneuvers. For the most part, people
vehicle in order to maintain stability and enhance pay little attention to the limited handling potential of
passengers security and comfort. their vehicles until some unusual behavior is observed
Vehicle dynamics It includes analytical and experi- that often results in fatality. According to statistics,
mental technology used to study and understand worldwide, an estimated 1.2 million people are killed

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3
992 Unscented Kalman Filter in Intelligent Vehicles
in road crashes each year and as many as 50 million are
injured [1]. Preventing car accidents requires to know
what determines vehicle dynamics during motion [2].
Today, automotive electronic technologies are
developing for safe and comfortable traveling of drivers
and passengers. Nowadays, there are a lot of vehicle
control system such as the Anti-lock Braking System
(ABS) that prevents wheel lock during braking [3], and
the Electronic Stability Control (ESC) that enhances
lateral vehicle stability [4, 5]. These control systems
installation rate is increasing all around the world.
Vehicle control algorithms have made great strides
toward improving the handling and safety of vehicles.
For example, experts estimate that ESC prevents 27%
of loss-of-control accidents by intervening when
emergency situations are detected [6]. While nowadays
vehicle control algorithms are undoubtedly a lifesaving
technology, they are limited by the available vehicle
state information.
Vehicle control systems currently available on pro-
duction cars rely on available inexpensive measure-
ments such as longitudinal velocity, accelerations, and
the vehicle yaw rate. Sideslip rate can be evaluated
using the yaw rate, lateral acceleration, and vehicle
velocity [7]. Calculating the sideslip angle is possible
from the sideslip rate integration. However, it is prone
to uncertainty and errors from sensor bias. Besides,
these control systems use unsophisticated, inaccurate
tire models to evaluate lateral tire dynamics. In fact,
measuring tire forces and sideslip angles is very difficult
for technical and economic reasons. Therefore, these
important data must be observed or estimated. If
control systems were in possession of the complete set
of lateral tire characteristics, namely lateral forces, side-
slip angle, and the tire-road friction coefficient, they
could greatly enhance vehicle handling and increase
passenger safety.
As the motion of a vehicle is governed by the forces
generated between the tires and the road, knowledge of
the tire forces is crucial when predicting vehicle
motion. For example, a vehicle can turn because of
the applied lateral tire forces. In fact, what happens is
that when the front wheels of a vehicle are steered, a slip
angle is created, which gives rise to a lateral force. This
lateral force turns or yaws the vehicle. Under normal
driving situations (low slip angle), a vehicle responds
predictably to the driver’s inputs. As the vehicle
approaches the handling limits, for example, during
an evasive emergency maneuver, or when a vehicle
undergoes high accelerations, high slip angle occurs,
and the vehicle’s dynamic becomes highly nonlinear
and its response becomes less predictable and poten-
tially very dangerous.
Accurate data about tire forces lead to a better eval-
uation of the vehicle possible trajectories, to a better
vehicle control and rollover prevention. Moreover, it
enables the development of a diagnostic tool for eval-
uating the potential risks of accidents related to poor
adherence or dangerous maneuvers.
In the literature, many studies have looked at the
vehicle dynamic states estimation. Several ones have
been conducted regarding the estimation of tire/road
forces [8–20]. For example, in [8], a study of
a 14-degrees-of-freedom (DOF) vehicle model is pro-
posed where the dynamics of the roll center are used to
calculate vertical tire forces. In [9], the authors propose
an estimation method in order to estimate vertical and
lateral forces per axle. The authors in [10–13] estimate
the vehicle vertical forces and other dynamic states for
a four-wheel vehicle model (FWVM) comprising four
DOF. Consequently, lateral tire forces at each tire are
calculated based on the estimated states and using
a quasi-static tire model. In [14], Ray estimates the
vehicle dynamic states and lateral tire forces per axle
for a nine-DOF vehicle model. The author uses mea-
surements of the applied torques as inputs to his
model. We note that the torque is difficult to get in
practice; it requires expensive sensors. More recently,
authors in [15, 16] propose observers to estimate lateral
forces per axle without using torque measures. In [17],
the authors propose an estimation process based on
a three-DOF vehicle model, as a lateral tire force esti-
mator. In [15–17], lateral forces are modeled with
a derivative equal to random noise. The authors in
[17] remark that such modeling leads to a noticeable
inaccuracy when estimating individual lateral tire
forces, but not in axle lateral forces. This phenomenon
is due to the non-representation of the lateral load
transfer when modeling. Studies in [21–24] focus on
the tire–road friction estimation.
The main goal of this study is to present an estima-
tion method that uses simple vehicle-road models and
a certain number of valid measurements in order to
estimate in real time and in accurate way the sideslip
 Unscented Kalman Filter in Intelligent Vehicles 993

angle and the lateral force at each individual tire/road
contact point. This study presents two significant
particularities (Fig. 1):
1. First, the estimation process does not use the
measurements of wheel torques which are very
expensive.
2. Second, the estimation process uses accurate nor-
mal tire forces, in contrast to many existing
approaches that assume constant vertical forces.
This approach is more realistic since during
cornering, accelerating, and braking, the load dis-
tribution varies significantly in a car, thus cornering
stiffness and lateral forces evaluation are directly
affected.
The developed estimation process is model-based
and built using Kalman filter technique. The Kalman
filter is known as the most commonly used real-
time estimator for linear and nonlinear systems. In
order to show the effectiveness of the estimation
method, some real-time validation tests were carried
out on an instrumented vehicle in realistic driving
situations.
Estimation Process Description
The estimation process is shown in its entirety by the
block diagram in Fig. 3, where ax and aym are, respec-
tively, the longitudinal and lateral accelerations; c_ is the
yaw rate, y_ is the roll rate, Dij (the index i represents the
front (1) or the rear (2) and the index j represents
the left (1) or the right (2)) is the suspension deflection;
wij is the wheel velocity; Fzij and Fyij are, respectively,
the normal and lateral tire-road forces; b is the sideslip
angle at the center of gravity (cog). The whole estima-
tion process consists of two blocks, and its role is to
estimate sideslip angle at the cog, normal, and lateral
forces at each tire/road contact point, and consequently
evaluate the used lateral friction coefficient. The
following measurements are needed:
● Yaw and roll rates measured by gyrometers
● Longitudinal and lateral accelerations measured by
accelerometers
● Suspension deflections using suspension deflec-
tions sensors
● Steering angle measured by an optical sensor
● Rotational velocity for each wheel given by
magnetic sensors

The first block aims to provide the vehicle’s mass,

lateral load transfer, normal tire forces, and the
corrected lateral acceleration ay (by canceling the grav-
Suspension
Δij itational acceleration component that distorts the
deflection
sensors Block 1 accelerometer signal aym). It contains observers based
on vehicle’s roll dynamics and model that couples lon-
. Roll plane model +
θ gitudinal and lateral accelerations. Authors have looked
coupling of the
Inertial ax, aym longitudinal / lateral at the first block in some previous studies [10, 11]. This
sensors dynamics work focuses only on the second block, whose main
. role is to estimate individual lateral tire force and
ψ
Fzij ax, ay sideslip angle. The second block makes use of the
estimations provided by the first block. In fact, as will
Block 2 be shown in the sections, the impact of including
Magnetic wij accurate normal forces in the calculation of lateral
sensors forces is fundamental.
Four wheel vehicle
model + Dugoff tire One specificity of this estimation process is the
δij model use of blocks in series. By using cascaded observers,
Optical
sensor the observability problems entailed by an inappropri-
ate use of the complete modeling equations are
Fyij β
avoided, enabling the estimation process to be carried
Unscented Kalman Filter in Intelligent Vehicles. Figure 1 out in a simple and practical way. In the following, the
Process estimation diagram model-based observer of the second block will be
994 Unscented Kalman Filter in Intelligent Vehicles

explained in details. Therefore, the vehicle-road model braking or acceleration. While considering their effect
and the estimation method will be illustrated, is certainly important, its inclusion makes solving the
respectively. lateral estimation problem considerably more com-
plex. Thus, it may be desirable to solve the lateral
Four-Wheel Vehicle Model estimation problem in the absence of longitudinal
forces first and include them in later studies. This can
The Four-Wheel Vehicle model (FWVM) is chosen
be done by focusing on solving the estimation problem
for this study because it is simple and corresponds
when the vehicle is driven at constant speeds [20].
sufficiently to our objectives. The FWVM is widely
This study extended the hypothesis of moving in
used to describe transversal vehicle dynamic behavior
a constant speed and addresses the case of a front-wheel
[12–14, 17, 18].
drive, where rear longitudinal forces are neglected
Figure 2 shows a simple diagram of the FWVM
relative to the front longitudinal forces. Longitudinal
model in the longitudinal and lateral planes.
front axle forces are considered by assuming that:
In order to simplify the lateral and longitudinal
dynamics, rolling resistance is neglected. Additionally, Fx1 ¼ Fx11 þ Fx12 ð1Þ
the front and rear track widths (E) are assumed to be
The longitudinal force evolution is modeled with
equal. L1 and L2 represent the distance from the vehi-
a random walk model, where its derivative is equal to
cle’s center of gravity to the front and rear axles, respec-
random noise (F_ x1 ¼ 0). This is due to the lack of
tively. The sideslip at the vehicle center of gravity (b) is
knowledge on the longitudinal slip and the effective
the difference between the velocity heading (Vg) and
_ is radius of the tire [25].
the true heading of the vehicle (c). The yaw rate (c)
The lateral dynamics of the vehicle can be obtained
the angular velocity of the vehicle about the center
by summing the forces and moments about the vehi-
of gravity. The forward and lateral velocities are,
cle’s center of gravity. Consequently, the simplified
respectively, V and U. The longitudinal and lateral
FWVM is formulated as the following dynamic
forces (Fx,y,i,j) are shown for front and rear tires of
relationships [19]:
the vehicle.
8 2 3
Longitudinal forces should be taken into account to > Fx1 cosðb  dÞ þ Fy11 sinðb  dÞ
>
>
enable accurate lateral forces estimation during vehicle >
> 16 7
>
> V_ g ¼ m 6
>
> 4 þ Fy12 sinðb  dÞ þ ðFy21 þ Fy22 Þ sin b 5
7
>
>
>
>
>
>
>
> 2 3
>
> L1 ½Fy11 cos d þ Fy12 cos d þ Fx1 sin d
δ11 >
>
>
> 6 7
>
α11 >
>
> €¼16
c 6
L2 ½Fy21 þ Fy22  7
7
Fy11 < Iz 4 5
Fx11 E
y E > þ ½F sin d  F sin d
>
>
y11 y12
x >
> "2 #
z >
>  F sinðb  dÞ þ Fy11 cosðb  dÞ
> _ 1 x1
δ12 >
>
> b ¼  c_
Vg α12 >
> mVg þ Fy12 cosðb  dÞ þ ðFy21 þ Fy22 Þ cos b
>
>
Fy21 Fx21 U >
>
Fx12 >
> 1
α21 β >
> ay ¼ ½Fy11 cos d þ Fy12 cos d þ ðFy21 þ Fy22 Þ þ Fx1 sin d
V >
> m
.
ψ Fy12 >
>
>
> 1
: ax ¼ ½Fy11 sin d  Fy12 sin d þ Fx1 cos d
m
ð2Þ
Fy22
L1
Fx22 α22 where m is the vehicle mass, and Iz is the yaw moment
of inertia.
L2 The tire slip angle (aij), as shown in Fig. 2, is the
difference between the tire’s longitudinal axis and the
Unscented Kalman Filter in Intelligent Vehicles. Figure 2 tire’s velocity vector. The tire velocity vector can be
Four-wheel vehicle model obtained from the vehicle’s velocity (at the cog) and
 Unscented Kalman Filter in Intelligent Vehicles 995

the yaw rate. Assuming that rear steering angles are longitudinal forces, the simplified nonlinear lateral
approximately null, the direction or heading of the tire forces are given by:
rear tires is the same as that of the vehicle. The heading
Fyij ¼ Cyij tanaij f ðlÞ ð4Þ
of the front tires includes the steering angle (d).
The front steering angles are assumed to be equal where Cyij is the cornering stiffness, aij is the slip angle,
(d11 = d12 = d). The forward velocity V, steering angle and f(l) is given by:
d, yaw rate c,_ and the vehicle body slip angle b are then 
ð2  lÞl; if l < 1
used to calculate the tire slip angles aij, where: f ðlÞ ¼ ð5Þ
1; if l  1
8 " #
>
> V b þ L1 c_ mFzij
>
> a11 ¼ d  arctan l¼ ð6Þ
>
> V  E c=2_ 2Cyij jtanaij j
>
>
>
> " #
>
>
>
> V b þ L1 c_ In the above formulation, m is the friction coeffi-
>
> a ¼ d  arctan
< 12 V þ E c=2_ cient and Fzij is the vertical load on the tire. This
" # ð3Þ
>
>  _ simplified tire model assumes pure slip conditions
> a ¼  arctan
>
V b L 2 c
>
> 21
_
with negligible longitudinal slip, a uniform pressure
>
> V  E c=2
>
> " # distribution, a rigid tire carcass, and a constant
>
>
>
>
> V b  L2 c_ friction coefficient for sliding rubber. The original
>
: a22 ¼  arctan _
V þ E c=2 Dugoff tire model has a constant stiffness in
respect to weight transfer. It is worthy to note that,
Tire–Road Interface according to [28], load transfer affects the cornering
stiffness Cyij . It can be represented by a second-order
As the motion of a vehicle is governed by the forces
polynomial with respect to the normal force as
generated between the tires and the road, knowledge of
shown in (7):
the tire forces is crucial in order to predict the vehicle’s
motion. This section presents the tire–road interaction Cyij ðFz Þ ¼ ðaFzij  bFzij2 Þ ð7Þ
phenomenon, especially the lateral tire forces. Since the
quality of the observer largely depends on the accuracy where a and b are, respectively, the first and
of the tire model, the underlying model must be pre- second-order coefficient in the cornering stiffness poly-
cise. Taking real-time calculation requirement, the tire nomial. They are identified once using a set of experi-
model should also be simple. mental data treated offline, where the tires remain
in their linear operation zone. Therefore, (7) is com-
pared to the calculated cornering stiffness obtained
Dugoff Tire Model
from the ratio of the measured lateral force to the
Many different tire models, based on the physical measured tire’s slip angle. Hence, a and b are calculated,
nature of the tire and/or on empirical formulations in such a way that the (7) fits the calculated cornering
deriving from experimental data, can be found in the stiffness well.
literature. These models include the Burckhardt, This study proposes a modified Dugoff tire model,
Dugoff, and Pacejka models [19, 26, 27]. One of the where the cornering stiffness varies with respect to load.
most commonly used model is the Pacejka’s “Magic As shown in (4), vertical forces and the tire slip
Formula.” It is an effective method for predicting real angles can be used to find the lateral force on
tire behavior. However, it requires a large number of each tire. Figure 3, based on the modified Dugoff ’s
tire-specific parameters that are usually unknown. tire model, is a graph of the lateral force versus
Another commonly used model is the Dugoff tire tire slip angle. It will be noted that as the load increases,
model. It synthesizes all the tire property parameters the peak lateral force occurs at somehow higher
into two constants Cx and Cy, referred to as the longi- slip angle.
tudinal and cornering stiffness of the tire. Dugoff ’s It is clear that for small slip angles, the force profile
model is the one used in this study. Neglecting can be defined by a linear region. When operating in
996 Unscented Kalman Filter in Intelligent Vehicles

9000
Load: Fz = 10000 N.
8000
Load: Fz = 8000 N.
Lateral force: - Fy (N) 7000

6000 Load: Fz = 6000 N.

5000
Load: Fz = 4000 N.
4000

3000

2000

1000

0
0 5 10 15 20
Sideslip angle (°)

Unscented Kalman Filter in Intelligent Vehicles. Figure 3

Generic tire curve: lateral force versus slip angle

this region, a vehicle responds predictably to the 

_
XðtÞ ¼ f ðXÞðtÞ; U ðtÞÞ þ wðtÞ
driver’s inputs. ð8Þ
Y ðtÞ ¼ hðXðtÞ; U ðtÞÞ þ uðtÞ
As the slip angle continues to grow, the tire begins
to saturate and reaches a peak value; this area is com- The input vector, U, comprises the steering angle
monly called the nonlinear region of the tire curve. and the normal forces considered estimated by the first
It represents the tire limits, and it is rarely reached block (see Fig. 3):
under normal driving conditions. If the front tires
saturate first, the vehicle is said to display understeer U ¼ ½d; Fz11 ;Fz12 ;Fz21 ;Fz22 T ¼ ½u1 ; u2 ;u3 ;u4 ;u5 T
and may plow out of a bend. If the rear tires saturate ð9Þ
first, the vehicle limit oversteers and may spin out. The measure vector, Y, comprises yaw rate, vehicle
Because most drivers are not accustomed to operate velocity (approximated by the mean of the rear-wheel
in the nonlinear handling regime, both of these velocities calculated from wheel-encoder data), longi-
responses are potentially very dangerous. Noting that tudinal and lateral accelerations:
oversteer situation is much more difficult to be con-
trolled than the understeer [29]. _ g ;ax ; ay T ¼ ½y1 ;y2 ;y3 ;y4 T
Y ¼ ½c;V ð10Þ

The state vector, X, comprises yaw rate, vehicle

Observers Design
This section presents a description of the observer
devoted to lateral tire forces and sideslip angle. The
state-space formulation, the observability analysis,
and the estimation method will be presented.
Stochastic State-Space Representation
The nonlinear stochastic state-space representation of
the system described in previous section is given as:
velocity, sideslip angle at the cog, lateral forces, and
the sum of the front longitudinal tire forces:
_ g ; b;Fy11 ;Fy12 ;Fy21 ;Fy22 ;Fx1 T
X ¼ ½c;V
ð11Þ
¼ ½x1 ;x2 ;x3 ; x4 ;x5 ;x6 ;x7 ;x8 T
Authors would like to emphasize that the consider-
ation of the lateral forces as states allows:
1. A better evaluation of the tire forces. In fact, what-
ever the complexity of the tire models, there are
 Unscented Kalman Filter in Intelligent Vehicles 997

several reasons why such models do not match the The observation function h(.) is:
actual tire forces perfectly [30]. From these reasons, 8
>
> h ¼ x1
we can cite especially the changes in the tire’s pres- > 1
>
>
> h ¼ x2
sure and temperature and the changes in the road >
< 2
characteristics. Therefore, authors believe that 1
> h3 ¼ ½x4 sin u1  x5 sin u1 þ x8 cos u1 
>
> m
according to the closed loop observer theory, the >
>
>
> 1
integration of the tire forces in the state vector may : h4 ¼ ½x4 cos u1 þ x5 cos u1 þ ðx6 þ x7 Þ þ x6 sin u1 
lead to better results than just using an open loop m
tire model. ð13Þ
2. A better understanding of the tire behaviors using The state vector X(t) will be estimated by applying
the relaxation-length formulation, especially in the unscented Kalman filter technique.
transient maneuvers [31].
3. The forces reconstruction to be done robustly with Observability
respect to some parameter variations. In fact, it is
well known that the Kalman filters have proven to Observability is a measure of how well the internal
be robust to parameter changes. states of a system can be inferred from knowledge of
its inputs and external outputs. This property is often
Taking these observations in mind, one can infer presented as a rank condition on the observability
the contribution of this study with respect to others matrix. Using the nonlinear state-space formulation
existing studies in the literature like [12, 13], which of the system presented in section “Stochastic State-
estimate dynamic variables of the vehicle, and then Space Representation,” the observability definition is
assess the tire forces using a properly adjusted tire local and uses the Lie derivative [32]. An observability
model. analysis of this system was undertaken in [33]. It was
The process and measurements noise vectors, shown that the system is observable except when:
respectively, w and u, are assumed to be white, zero
mean, and uncorrelated. ● Steering angles are null.
Consequently, the particular nonlinear function ● The vehicle is at rest (Vg = 0).
f(.) of the state equations is given by: For these situations, we assume that lateral forces
8 2 3 and sideslip angle are null, which approximately corre-
>
> L1 ½x4 cos u1 þ x5 cos u1 þ x8 sin u1 sponds to the real cases.
>
>
>
>
> 1 6 6  L2 ½x6 þ x7 
7
7
>
> f ¼ 6 7
>
>
1
I 4 5
>
>
z
E
>
> þ ½x sin u  x sin u  Estimation Method: Kalman Filter Algorithms
>
> 4 1 5 1
>
> " 2 #
>
> x8 cosðx3  u1 Þ þ x4 sinðx3  u1 Þ
>
> 1 The aim of an observer or a virtual sensor is to estimate
>
> f2 ¼
>
> m þ x5 sinðx3  u1 Þ þ ðx6 þ x7 Þsinðx3 Þ a particular unmeasurable variable from available
>
>
>
> " #
>
>  x8 sinðx3  u1 Þ þ x4 cosðx3  u1 Þ measurements and a system model in a closed loop
>
<f ¼ 1
3  x1 observation scheme, as illustrated in Fig. 4. Because of
mx2 þ x5 cosðx3  u1 Þ þ ðx6 þ x7 Þcos x3
>
> the vehicle system-model mismatches (unmodeled
>
> x2
>
> f4 ¼ ðx4 þ Fy11 ða11 ;u2 ÞÞ dynamics, parameter variations, etc.) and the presence
>
> s1
>
>
>
> x2 of unknown and unmeasurable disturbances, the cal-
>
> f5 ¼ ðx5 þ Fy12 ða12 ;u3 ÞÞ
>
> s culation obtained from vehicle’s model would deviate
>
> 1
>
> x2 from the actual values over time. In order to reduce the
>
> f6 ¼ s ðx6 þ Fy21 ða21 ;u4 ÞÞ
>
>
>
> 2 estimation error, at least some of the measured outputs
>
> x2
>
> f7 ¼ ðx7 þ Fy22 ða22 ;u5 ÞÞ are compared to the same variables estimated by the
>
>
>
>
s2
observer. The difference is fed back into the observer
:
f8 ¼ 0
after being multiplied by a gain matrix K, and so we
ð12Þ have a closed loop observer (see Fig. 4). All developed
998 Unscented Kalman Filter in Intelligent Vehicles

Measurements

Inputs

X=f(X,U) h(X, U) +
sensors –
evolution
++
K
X
correction Kalman Gain
observer

Unscented Kalman Filter in Intelligent Vehicles. Figure 4

Process estimation diagram

observers are implemented in a first-order Euler The UKF can be formulated as follows [37–40]:
approximation discrete form. At each iteration, the Initialization
state vector is first calculated according to the evolution
(
equation and then corrected online with the measure-  0 ¼ E½X0 
X
ment errors (innovation) and filter gain K in a recursive ð15Þ
 0 Þ ðX0  X
P0 ¼ E½ðX0  X  0 ÞT 
prediction–correction mechanism. In this study, the
gain is calculated using the Kalman filter method 0 and P0 are, respectively, the initial state and
where X
which is a set of mathematical equations and is widely
the initial covariance.
represented in [34–36]. Sigma points calculation and time update
The EKF (Extended Kalman Filter) is probably the
most commonly used estimator for nonlinear systems. 8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
> w ¼ ½X k1 ; X
 k1  ðn þ lÞPk1 
However, in this study, the UKF (Unscented Kalman >
>
k1
>
> wkjk1 ¼ f ðwk1 ; Uk1 Þ
>
>
Filter) [38–40] is chosen for the following fundamental >
> X2n
>
>
reasons: >
> xkjk1 ¼ w m wi;kjk1
>
> i¼0 i
>
> X2n
● The high nonlinearities of the vehicle-road model. >
< Pkjk1 ¼ w c ðwi;kjk1  X  kjk1 Þ
i¼0 i
● The calculation complexity of the Jacobian matrices ð16Þ
>
>  ðgi;kjk1  X kjk1 ÞT þ Q
which causes implementation difficulties. >
> qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
>
>
> ¼ ½  ;   ðn þ lÞPkjk1 
>
> w kjk1 X k1 X k1
>
>
>
>
UKF Algorithm In this subsection, the principle of > gkjk1 ¼ hðwkjk1 Þ
>
>
the UKF is summarized. Consider the general discrete >
> X2n
: Y
kjk1 ¼ wim gi;kjk1
nonlinear system: i¼0
(
Xkþ1 ¼ f ðXk ; Uk Þ þ wk where
ð14Þ 8
Yk ¼ hðXk Þ þ vk l
>
> w0m ¼
>
> þ
where Xk 2 Rn is the state vector, Uk 2 Rr is the known >
> n l
>
< c l
input vector, Yk 2 Rm is the output vector at time k, wk, w0 ¼ þ ðn  a2 þ bÞ ð17Þ
> nþl
and vk are, respectively, the disturbance and sensor- >
> m
>
> w ¼ wic ¼ 2ðnþl1
Þ i ¼ 1; . . . ; 2n
noise vector, which are assumed to be Gaussian white >
> i
:
noise with zero mean and uncorrelated. l ¼ nða  1Þ
2
 Unscented Kalman Filter in Intelligent Vehicles 999

Measurement update Experimental Results

In this section, the experimental car used to evaluate
8 X2n the observer performances is presented. Moreover, the
>
>
> PY k Y k ¼ w c ðg
i¼0 i i;kjk1
 Y kjk1 Þ
>
> test conditions and the results of the previously devel-
>
>
>
>  ðgi;kjk1  Y kjk1 ÞT þ R oped observers are discussed and analyzed.
>
> X2n
>
>  kjk1 Þ
>
< PX k Y k ¼ w c ðwi;kjk1  X
i¼0 i Experimental Car
ð18Þ
>
>  ðgi;kjk1  Y kjk1 ÞT
>
> The experimental vehicle shown in Fig. 4 is the
>
> Kk ¼ PX k Y k PY1
>
>  k Y k INRETS-MA (Institut National de Recherche sur les
>
>
>
> P ¼ Pkjk1  Kk PY k Y k KkT Transports et leur Sécurité – Département Mécanismes
>
> k
: k ¼ X  kjk1 þ Kk ðYk  Y kjk1 Þ d’Accidents) Laboratory’s test vehicle [7]. It is
X
a Peugeot 307 equipped with:
● Gyrometers and accelerometers that measure,
where the variables are defined as follows: Xk and Ykjk1
respectively, the rotations (roll, pich, and yaw
are the estimations, respectively, of the state and of the
rates) and accelerations (longitudinal, lateral, and
real measurement at each instant k, wi is a set of scalar
vertical) of the car body.
weights, and n is the state dimension; the parameter a
● Suspension sensors that measure the distances
determines the spread of the sigma points around X 
between the wheels and the car body.
and is usually set to 1e – 4 < a < 1. The constant b is
● Three Correvit noncontact optical sensors:
used to incorporate part of the prior knowledge of the
distribution of X, and for Gaussian distributions, b = 2 1. One is located in chassis rear overhanging posi-
is optimal. Q and R are, respectively, the disturbance tion, and it measures longitudinal and lateral
and sensor-noise covariance: R takes into account the vehicle speeds.
uncertainty in the measured data and Q is tuned 2. The other two are installed on the front right
depending on the model quality. Remember that the and rear right tires, and they measure front- and
computation of the Kalman gain is a subtle mix rear-tires longitudinal/lateral velocities and
between process and observation noises. The less sideslip angles.
noise in the operation compared to the uncertainty in ● Dynamometric wheels fitted on all four tires, which
the model, the more the variables will be adapted to are able to measure tire forces and wheel torques in
follow measurements. Since the lateral forces are and around all three dimensions.
modeled using a relaxation model based on reliable ● Steering-rack displacement sensor that is used to
tire models, the uncertainty affected to them is not determine the steering angle.
too high. However, the longitudinal force per front ● Magnetic sensors that measure rotational velocity
axle is not modeled at all; hence, it is represented by for each wheel.
a high noise level. The other states (yaw rate, longitu-
It is important to note that the Correvit and the
dinal and lateral vehicle velocity) are modeled using the
wheel-force transducers (see Fig. 5) are very expensive
vehicle’s equations. Therefore, they are said to have an
sensors. They are used in this study as a reference for
average noise. On the other hand, since the embedded
validating the estimation process. The sampling
sensors have good accuracy, the noises on the measure-
frequency of the different sensors is 100 Hz.
ments are quite small. In order to reduce the complex-
ity of the problem, both measurement covariance
Test Conditions
matrix, R, and the process covariance matrix, Q, are
assumed to be constant and diagonal. The off-diagonal Test data from nominal as well as adverse driving con-
elements are set to 0. That means that both the mea- ditions were used to assess the performance of the
surement noises and the process noises are supposed observer presented in section “Observers Design,” in
uncorrelated. realistic driving situations. We report a lane-change
1000 Unscented Kalman Filter in Intelligent Vehicles
Wheel force transducer
Unscented Kalman Filter in Intelligent Vehicles. Figure 5
Wheel-force transducer and sideslip sensor installed at the tire level
maneuver where the dynamic contributions play an
important role. Figure 6 presents the Peugeot’s trajec-
tory (on a dry road), its speed, steering angle, and “g-g”
acceleration diagram during the course of the test. The
acceleration diagram, that determines the maneuvering
area utilized by the driver/vehicle, shows that large
lateral accelerations were obtained (absolute value up
to 0.6 g). This means that the experimental vehicle was
put in a critical driving situation.
The estimation process algorithm was written in
C++ and has been integrated into the laboratory car
as a DLL (Dynamic Link Library) that functions
according to the software acquisition system.
Validation of Observers
The observer results are presented in two forms: as
tables of normalized errors, and as figures comparing
Correvit
the measurements and the estimations. The normal-
ized error for an estimation z is defined as:
k zobs  zmeasured k
Ez ¼ 100  ð19Þ
max ðk zmeasured kÞ
where zobs is the variable calculated by the observer,
zmeasured is the measured variable, and max(kzmeasuredk)
is the absolute maximum value of the measured
variable during the test maneuver.
Figures 7 and 8 show lateral forces on the front and
rear wheels. According to these plots, the observers are
relatively good with respect to measurements. Some
small differences during the trajectory are to be noted.
These might be explained by neglected geometrical
parameters, especially the camber angles, which also
produce a lateral forces component [41]. It is
also shown that the lateral forces on the right-hand
tires exceed those on the left-hand tires. This result is
 Unscented Kalman Filter in Intelligent Vehicles 1001

200
110

Speed (km.h1)
Y position (m)
100
100

0 90

80
−100
0 500 1000 0 20 40
X position (m) Time (s)

Longitudinal acceleration (g)

0.04
0.1
Steering angle (rad)

0.05
0.02
0 0
−0.02 −0.05

−0.1
−0.04
0 20 40 −0.4 −0.2 0 0.2 0.4 0.6
Time(s) Lateral acceleration (g)

Unscented Kalman Filter in Intelligent Vehicles. Figure 6

Experimental test: vehicle trajectory, speed, steering angle, and acceleration diagrams for the lane-change test

Front left lateral force Fy11 (N)

2000
measurement
1000 Estimation
0

−1000

−2000
0 10 20 30 40 50

Front right lateral force Fy12 (N)

4000

2000

−2000
0 10 20 30 40 50
Time (s)

Unscented Kalman Filter in Intelligent Vehicles. Figure 7

Estimation of front lateral tire forces
1002 Unscented Kalman Filter in Intelligent Vehicles
Rear left lateral force Fy21 (N)
1000 measurement
Estimation
−1000
0 10 20
Rear right lateral force Fy22 (N)
3000
2000
1000
−1000
0 10 20
Unscented Kalman Filter in Intelligent Vehicles. Figure 8
Estimation of rear lateral tire forces
clearly a consequence of the load transfer produced
during cornering from the left- to the right-hand
side of the vehicle. In fact, lateral force increases as
normal force increases. Figure 9 shows how sideslip
angle changes during the test. Reported results are
relatively good.
Table 1 presents maximum absolute values, nor-
malized mean errors, and normalized std (standard
deviations) for lateral tire forces and sideslip angles.
Despite the simplicity of the model, we can deduce that
for this test, the performance of the observer is satis-
factory, with normalized error globally less than 8%.
Given the vertical and lateral tire forces at each tire-
road contact level, the estimation process is able to
evaluate the used lateral friction coefficient m. This is
defined as the ratio of friction force to normal force,
and is given by [29, 41]:
Fyij
mij ¼
Fzij
The lateral friction coefficients in Figure 10 show
that the estimated mij are close to the measured values.
A closer investigation reveals that the used lateral fric-
tion coefficients m12 and m22 corresponding to the
overloaded tires during cornering are lower than m11
30 40 50
30 40 50
Time (s)
and m21. This phenomenon is due to the tire load
sensitivity effect: the lateral friction coefficient is nor-
mally higher for the lighter loads, or conversely, falls off
as the load increases [29, 41].
This test also demonstrates that m11 and m21 are
high, especially for lateral accelerations up to 0.6, and
that they attain the limit for the dry road friction
coefficient. In fact, dry road surfaces show a high fric-
tion coefficient in the range 0.9–1.2 (implying that
driving on these surfaces is safe), which means that
for this test the limits of handling were reached.
Conclusion and Future Perspectives
This paper presents an interesting method for estimat-
ing lateral tire forces and sideslip angle, that is to say,
two of the most important parameters affecting vehicle
stability and the risk of leaving the road. Consequently,
ð20Þ the developed observer could feed control systems with
fundamental vehicle-dynamics data in order to
enhance vehicle safety.
The proposed observer is derived from a simplified
four-wheel vehicle model and is based on unscented
Kalman filtering technique. Tire–road interaction is
represented by the Dugoff model. A comparison with
 Unscented Kalman Filter in Intelligent Vehicles 1003

Sideslip angle at the cog (rad) μ11 μ12

0.01 1
measurement 0.5
Estimation
0.005 0.5
0
0 0
−0.5
−0.005
0 20 40 0 20 40
−0.01 Time (s) Time (s)

−0.015 μ21 μ22

1 0.5
−0.02
0.5
0
−0.025
10 20 30 40 50 60 0
−0.5
Time (s)
0 20 40 0 20 40
Unscented Kalman Filter in Intelligent Vehicles. Figure 9 Time (s) Time (s)
Estimation of the sideslip angle at the cog
Unscented Kalman Filter in Intelligent Vehicles.
Figure 10
Used lateral friction coefficients developed by the tires

Unscented Kalman Filter in Intelligent Vehicles.

Table 1 Maximum absolute values, normalized mean
errors, and normalized Std
major importance to study the effect of coupling lon-
Max kk Mean% Std% gitudinal/lateral dynamics on lateral tire behavior.
Fy11 2180 (N) 8.23 4.80
Fy12 4070 (N) 3.70 3.74 Acknowledgments

Fy21 1441 (N) 7.51 3.52 This work was done at Heudiasyc Laboratory UMR
Fy22 2889 (N) 1.91 1.77 CNRS 6599, UTC University (Compiègne, France) in
collaboration with Alessandro Victorino and Ali
b 0.027 (rad) 8.32 6.41
Charara.

Bibliography

real experimental data demonstrates the potential of Primary Literature

the estimation process. It is shown that it may be
possible to replace expensive correvit and dynamomet-
ric hub sensors by real-time software observers. This is
one of the important results of this report. Another
important result concerns the estimation of individual
lateral forces acting on each tire. This can be seen as an
advance with respect to the current vehicle-dynamics
literature.
Future studies will improve the vehicle-road model
in order to widen validity domains for the observer
(take into account road irregularities and road bank
angle), and make it adaptive with the road conditions
(especially the road friction). Moreover, it will be of
1. FARS (2007) Fatality analysis reporting system (FARS). Techni-
cal report, National Highway Traffic Safety Administration,
National Center for Statistics and Analysis
2. Aparicio F, Paez P, Moreno F, Jimenez F, Lopez A (2005)
Discussion of a new adaptive speed control system incorpo-
rating the geometric characteristics of the road. Int J Veh
Auton Syst 3(1):47–64
3. Bosch R (2004) Bosch automotive handbook, 6th edn. SAE
Society of Automotive Engineers, Bosch company, New York
4. ESC (2003) Electronic stability control fact sheet. Technical
report, ESC Coalition and DEKRA Automotive research
5. Van Zanten A (2002) Evolution of electronic control systems
for improving the vehicle dynamic behavior. In: Proceedings
of the 6th international symposium on advanced vehicle con-
trol (AVEC), Hiroshima, pp 7–15
1004 Unscented Kalman Filter in Intelligent Vehicles
6. Hsu Y, Gerdes JC (2005) Stabilization of a steerby-wire vehicle
at the limits of handling using feedback linearization. In:
Procedings of IMECE2005, AMSE international mechanical
engineering congress and exposition, Florida
7. Lechner D (2008) Embedded laboratory for vehicle dynamic
measurements. In: International symposium on advanced
vehicle control, Kobe
8. Shim T, Ghike G (2007) Understanding the limitations of dif-
ferent vehicle models for roll dynamics studies. Veh Syst Dyn
45:191–216
9. Doumiati M, Baffet G, Lechner D, Victorino A, Charara A (2008)
Embedded estimation of the tire/road forces and validation in
a laboratory vehicle. In: International symposium on
Advanced Vehicle Control, Kobe
10. Doumiati M, Victorino A, Charara A, Lechner D, Baffet G (2008)
An estimation process for vehicle wheelground contact nor-
mal forces. In: IFAC World Congress, Seoul Korea
11. Doumiati M, Victorino A, Charara A, Lechner D (2009) Lateral
load transfer and normal forces estimation for vehicle safety
experimental evaluation. Veh Syst Dyn 47(12):1511–1533
12. Wenzel TA, Burnham KJ, Blundell MV, Williams RA (2006)
Estimation of the nonlinear suspension tyre cornering
forces from experimental road test data. Veh Syst Dyn
44:153–171
13. Dakhlallah J, Glaser S, Mammar S, Sebsadji Y (2008) Tire-road
forces estimation using extended Kalman filter and sideslip
angle evaluation. In: American control conference, Seattle
14. Ray LR (1997) Nonlinear tire force estimation and road friction
identification: simulation and experiments. Automatica
33(10):1819–1833
15. Baffet G, Charara A, Dherbomez G (2007) An observer of tire
road forces and friction for active-security vehicle systems.
IEEE/ASME Trans Mechatron 12:651–661
16. Baffet G, Charara A, Lechner D, Thomas D (2008) Experimental
evaluation of observers for tire-road forces, sideslip angle and
wheel cornering stiffness. Veh Syst Dyn 45:191–216
17. Wilkin MA, Manning WJ, Crolla DA, Levesley MC (2006) Use of
an extended Kalman filter as a robust tyre force estimator. Veh
Syst Dyn 44:50–59
18. Venhovens PT, Naab K (1999) Vehicle dynamics estimation
using Kalman filters. Veh Syst Dyn 32:171–184
19. Kiencke U, Daiss A (1997) Observation of lateral vehicle
dynamics. Control Eng Pract 5(8):1145–1150
20. Hsu YHJ (2009) Estimation and control of lateral tire forces
using steering torque. PhD, Department of mechanical engi-
neering, Stanford University
21. Haffner L, Kozek M, Shi J, Jorgel HP (2008) Estimation of the
maximum friction coefficient for a passenger vehicle using the
instantaneous cornering stiffness. In: American control confer-
ence, Seattle
22. Gustafsson F (1997) Slip-based tire-road friction estimation.
Automatica 33(6):1087–1099
23. Hahn JO, Rajamani R, Alexander L (2002) GPS-Based real time
identification of tire-road friction coefficient. IEEE Trans Con-
trol Syst Technol 10(3):331–343
24. Uchanski MR (2001) Road friction estimation for automobiles
using digital signal processing methods. PhD thesis, University
of California, Berkley
25. Koo SL, Tan HS (2007) Tire dynamic deflection and its impact
on vehicle longitudinal dynamics and control. IEEE/ASME
Trans Mechatron 12(6):623–631
26. Dugoff J, Fanches P, Segel L (1970) An analysis of tire proper-
ties and their influence on vehicle dynamic performance.
Society of Automotive Engineers Paper 700377
27. Pacejka HB (2002) Tyre and vehicle dynamics. Elsevier,
Amsterdam
28. Gillespie TD (1992) Fundamental of vehicle dynamics. Society
of Automotive Engineers, Warrendale
29. Rajamani R (2005) Vehicle dynamics and control. Springer,
New York
30. Lidner L (1993) Experience with the magic formula tyre model.
In: Proceedings of the 1st international colloquium on tyre
models for vehicle dynamic analysis, Amsterdam
31. Svendenius J (2007) Tire modeling and friction estimation.
PhD thesis, Lund University, Sweden
32. Nijmeijer H, Van der Schaft AJ (1991) Nonlinear dynamical
control systems. Springer, New York
33. Doumiati M, Victorino A, Charara A, Lechner D (2009)
Unscented Kalman filter for real-time vehicle lateral tire forces
and sideslip angle estimation. In: IEEE intelligent vehicles sym-
posium, Xi’an, Shaanxi
34. Kalman RE (1960) A new approach to linear filtering and
prediction problems. Trans ASME J Basic Eng 82(D):35–45
35. Welch G, Bishop G (2001) An introduction to the Kalman filter,
Course 8. University of North Carolina, Departement of com-
puter science
36. Durrant-Whyte H (2001) Multi sensor data fusion. Technical
report, Australian center for field robotics, University of
Sydney
37. Mohinder GS, Angus PA (1993) Kalman filtering theory and
practice. Prentice Hall, Englewood Cliffs
38. Julier SJ, Uhlman JK (1997) A new extension of the Kalman
filter to nonlinear systems. In: International symposium
aerospace/defense sensing, simulation and controls, Orlando
39. Julier SJ, Uhlman JK, Durrant-Whyte HF (1995) A new exten-
sion approach for filtering nonlinear systems. In: American
control conference, Seattle
40. Wan EA, Merwe RVD (2000) The unscented Kalman filter for
nonlinear estimation. In: Proceedings of the adaptive systems
for signal processing, communications, and control sympo-
sium, Atlanta, pp 153–158
41. Milliken W, Milliken DL (1995) Race car vehicle dynamics.
Society of Automotive Engineers, Warrendale
Books and Reviews
Aga M, Okada A (2003) Analysis of vehicle stability control (VSC)’s
effectiveness from accident data. In: Proceedings of the 18th
international technical conference on the enhanced safety of
vehicles, Nagoya

Aguirre LA, Letellier C (2005) Observability of multivariate differ-
ential embeddings. J Phys A Math Gen 38:6311–6326
Bakker E, PAcejka HB, Lidner L (1987) A new tire model with
application in vehicle dynamic systems, SAE Paper 890087
Besanon G (2007) Nonlinear observers and applications. Springer,
Berlin
Bevly DM, Gerdes JC, Wilson C, Zheng G (2000) The use of
GPS based velocity measurements for improved vehicle
state estimation: proceedings of the American Control
Conference, Chicago. American Automatic Control Council,
New York
Brown T, Hac A, Martens J (2004) Detection of vehicle rollover. SAE
World Congress, Detroit
Unscented Kalman Filter in Intelligent Vehicles 1005
Doumiati M, Sename O, Martinez J, Poussot C, Dugard L (2010)
Gain scheduled steering and braking control for improvement
of vehicle handling and stability. In: IEEE conference on deci-
sion and control, Georgia
Doumiati M, Sename O, Martinez J, Dugard L, Gaspar P, Szabo Z,
Bokor J (2011) Vehicle yaw control via coordinated use of
steering/braking systems. IFAC World Congress, Milano
Heydinger G, Howe J (2000) Analysis of vehicle response data
measured during severe maneuvers, SAE Paper 2000-01-1644
Jazar RN (2008) Vehicle dynamics, theory and application.
Springer, New York
Smith ND (2004) Understanding parameters influencing tire
modeling. Colorado State University, Formula SAE Platform
1006 Vehicle Biofuels

Vehicle Biofuels the most valuable because they are easily stored, have
a high energy density, and are readily metered into
MARK HOLTZAPPLE engines. Biofuels are infinitely renewable provided
Department of Chemical Engineering, Texas A&M appropriate agricultural practices are followed.
University, College Station, TX, USA

Introduction
Article Outline
Historically, biological feedstocks have been used
Glossary to ‘power transportation. For example, early steam-
Definition of the Subject powered trains were fueled by wood, Rudolph Diesel
Introduction designed his engine to run on peanut oil, and Henry
Vehicle Fuels Ford designed the Model T to run on ethanol.
Biomass From 1880 to 1973, the price of crude oil was
Biomass Processing $15–30/bbl (current dollars), so it was very difficult
Conversion Processes for biofuels to compete. As a consequence, during this
Future Directions time period, very few vehicles were powered by
Bibliography biofuels. However, during crisis situations, biofuels
were employed. For example, during World War II,
Glossary Europeans had difficulty obtaining gasoline, so
Biodiesel Methyl or ethyl ester of fatty acids. millions of vehicles were adapted to run on gasified
Biomass Biological material from living or recently wood. The Arab Oil Embargo (1973) and Iranian Hos-
living organisms. tage Crisis (1979) caused the price of crude oil to rise to
Bio-oil The liquid resulting from biomass pyrolysis. about $90/bbl (current dollars), which precipitated
Equivalence ratio The actual oxygen fed to the reactor interest in biofuels. In 1978, the US Energy Tax Act
divided by the stoichiometric oxygen needed for provided tax credits for ethanol ($0.54/gal ethanol),
complete combustion. which promoted the conversion of corn into biofuel.
Gasohol A mixture of gasoline and ethanol, typically This act was designed to reduce dependence on foreign
90% gasoline and 10% ethanol. oil (36% imports in 1978) and to create a new market
Lignocellulose Biomass composed of lignin, cellulose, for corn, which had depressed prices at the time.
and hemicellulose. In 2005, the US government provided a tax subsidy
Oleaginous microorganisms Microorganisms that for biodiesel ($1.00/gal). In 2008, the Farm Bill
accumulate triacylglycerol (TAG) within their cells. included a subsidy for cellulosic ethanol ($1.01/gal).
Saccharification Hydrolysis of polysaccharides to Through these subsidies, the US government is
produce sugar. promoting the use of biofuels for the following
Synthesis gas Mixture of carbon monoxide and reasons:
hydrogen. ● Energy security – In 2009, the United States imported
Triacylglycerol (TAG) A natural product containing 59% of its net crude oil consumption, which makes
three fatty acids linked to glycerol via ester bonds. its economy vulnerable to supply disruptions.
● Balance of trade – When crude oil sells for about
$90/bbl, the United States spends about $1 billion
Definition of the Subject
per day on imports. If retained within the
Vehicle biofuels are solid, liquid, and gaseous fuels United States, this money would help develop
derived from biomass (e.g., corn, sugarcane, grasses, domestic jobs.
and wood) used for transportation (e.g., automobiles, ● Rural economic development – The sale of biofuels
trucks, planes, and trains). Among biofuels, liquids are will bring additional revenue to rural economies.

M. Ehsani et al. (eds.), Transportation Technologies for Sustainability, DOI 10.1007/978-1-4614-5844-9,

# Springer Science+Business Media New York 2013

Originally published in
Robert A. Meyers (ed.) Encyclopedia of Sustainability Science and Technology, # 2012, DOI 10.1007/978-1-4419-0851-3

● Environment – The production of biofuels from
polluting biomass feedstocks (e.g., manure, munic-
ipal solid waste) removes these hazards from the
environment.
● Global warming – The combustion of biofuels
does not contribute net carbon dioxide to the
atmosphere; any CO2 released from the combustion
of biofuels is fixed via photosynthesis which recycles
it as plant matter.
Vehicle Fuels
For transportation, liquid fuels are preferred because of
their high energy density, ease of transport, and
controllability; therefore, they command a premium
price as described below:
● Liquid (e.g., gasoline, jet fuel, diesel) – $15 to $25/GJ.
● Gaseous (e.g., methane) – $2 to $5/GJ.
● Solid (e.g., coal) – $1 to $3/GJ.
Conventional Fuels
Gasoline is used primarily in spark-ignited piston
engines (Otto cycle). It is composed of hydrocarbons
containing 4–12 carbon atoms (average about 7–8).
Gasoline grades are determined by the octane number,
which characterizes the tendency of the fuel to knock,
i.e., prematurely detonate. Gasoline-powered engines
have a compression ratio (maximum: minimum
gas volume) of about 7:1–12:1. During the compres-
sion of the air–fuel mixture, the temperature increases.
Fragile molecules (e.g., large linear hydrocarbons)
readily detonate whereas stable molecules
(e.g., branched hydrocarbons, aromatics, oxygenates)
resist detonation. High-compression engines are more
efficient and powerful, and require high-octane fuels to
prevent knocking. Because knocking pressurizes the gas
at the wrong point in the engine cycle, it can rob
efficiency and potentially cause damage. (Note: A fuel
with an octane number of 100 has knocking character-
istics identical to an idealized fuel containing 100%
iso-octane and 0% n-heptane. Similarly, a fuel with
an octane number of 0 has knocking characteristics
identical to an idealized fuel containing 0% iso-octane
and 100% n-heptane. Other fuel octane numbers
correspond to other idealized mixtures.)
Vehicle Biofuels 1007
Jet fuel is used primarily in jet engines (Brayton
cycle). It is composed of hydrocarbons containing
8–16 carbon atoms (average of about 12), which is
similar to kerosene. Jet fuel formulations are adjusted
to ensure the fuel does not freeze at high altitudes,
where the temperature is cold. Also, the composition
is adjusted to reduce smoke formation.
Diesel fuel is used primarily in diesel engines. It is
composed of hydrocarbons containing 10–22 carbon
atoms (average of about 16). Diesel engines have a high
compression ratio (typically 14:1–24:1), which greatly
increases the air temperature during the compression
stroke. To ignite the fuel, they do not employ a spark;
rather, they rely on the high temperature of the
compressed air. When atomized diesel fuel is injected
into the hot compressed air, it decomposes and ignites.
The burn characteristics of diesel fuel are determined
by the cetane number, which quantifies how quickly the
fuel starts to auto-ignite in a diesel engine. Analogous
to octane rating, the cetane rating assigns a value of 100
to a fuel that behaves like cetane (n-hexadecane)
whereas a value of 0 is assigned to a fuel that behaves
like 1-methyl napthalene.
Biofuels
Biofuels come in many forms, as described in Table 1.
Table 2 provides properties of both biofuels and
conventional fuels.
In the United States, ethanol is the dominant biofuel.
Figure 1 shows the US ethanol capacity from 1980 to 2010.
In 2010, the US ethanol capacity was 13.2 billion gallons
per year and US gasoline consumption was 138.5 gal per
year (Fig. 2); thus, about 8.7% of the gasoline–ethanol
pool was ethanol. In 2010, approximately 40% of
the US corn crop was converted to fuel ethanol. Table 3
shows global ethanol production in 2009.
In 2009, the US biodiesel capacity was 2.0 billion
gallons per year [4] and US diesel consumption
was 55.7 billion gallons per year [5]; thus, about 3.5%
of the diesel pool was biodiesel.
Biofuel Blends
Biofuels are often blended with other components.
Common designations are described below.
1008 Vehicle Biofuels
Vehicle Biofuels. Table 1 Examples of biofuels
Fuel type Generic
Primary alcohols R–OH
Secondary alcohols
OH
R –C– R ’
Ethers
R –O – R’
Ketones
O
R –C– R ’
Esters
O O
R –CO – R ’
E10 E10 is a blend of 10% ethanol and 90% gasoline,
and is commonly described as gasohol. It is widely
used in the United States; nearly the entire gasoline
pool contains ethanol. Modern gasoline engines are
designed to be compatible with E10 without the need
for modifications.
Addition of ethanol to gasoline has some benefits,
such as (1) reducing gasoline imports, (2) improving
the fuel octane rating, and (3) improving combustion.
Regarding the latter point, the 1990 US Clean Air Act
required the addition of oxygenates to reformulated
gasoline in cities that could not attain carbon monoxide
or ozone standards. In those cities, fuels were required
to contain 2.7% oxygen, which could be met by adding
oxygenates such as methyl tertiary butyl ether (MTBE)
or ethanol. When MTBE was found to contaminate
groundwater, its use was banned so the demand for
ethanol increased dramatically.
Addition of ethanol to gasoline also has some
detriments, such as (1) reducing fuel economy because
of its lower energy content (
3%); (2) raising the
fuel vapor pressure, which reduces the amount of
low-cost, high-octane butanes that can be added to
fuel; (3) increasing production of formaldehyde and
acetaldehyde in tail-pipe emissions; and (4) absorbing
water into the fuel. Regarding the latter point,
ethanol is very polar compared to gasoline and
Examples
H3C–OH methanol
H3CH2C–OH ethanol
OH
isopropanol
H3C – C – CH3
H3C–O–CH3 dimethyl ether
H3CH2C–O–CH2CH3 diethyl ether
O
Acetone
H3C – C – CH3
Biodiesel
H34C18 – CO– CH3
attracts water, which is also polar. Common-carrier
pipelines that transport hydrocarbon fuels often
contain water. If ethanol-containing fuel were
transported through these pipelines, it would
absorb water, which adversely affects fuel quality.
To avoid this problem, anhydrous ethanol is
transported via rail or trucks and is “splash
blended” at the fuel terminal.
E15 E15 is a blend of 15% ethanol and 85% gasoline.
As ethanol is increasingly added to the gasoline
pool, the United States will soon reach a “blend wall”
for E10. The US Environmental Protection Agency
has authorized the use of E15 in certain vehicles
(e.g., automobiles, SUVs, light-duty trucks) after
model year 2001, which potentially would allow the
percentage of ethanol to increase by 50%. The ruling
has spurred controversy regarding potential damage to
engines, emission systems, and air quality. Further, E15
would need to be sold in a manner to prevent it from
being used in unauthorized vehicles (e.g., motorcycles,
heavy-duty vehicles, off-road vehicles, and vehicles
older than model year 2000).
E85 E85 is a blend of 85% ethanol and 15%
gasoline. Because ethanol has lower energy content
than gasoline, the fuel mileage of E85 decreases by
 Vehicle Biofuels 1009

Vehicle Biofuels. Table 2 Properties of fuelsa

Heat of combustionb Density Heat of combustionb
Fuel (MJ/kg) (kg/L) (MJ/L) Octane ratingc
Hydrocarbons
Diesel 44.8 0.83 37.2 15–25
Jet 46.6 0.81 37.7
Gasoline 47.3 0.74 35.0 91–99
Esters
Biodiesel 39.9 0.88 35.1
Alcohols
Methanol 19.9 0.79 15.7 106
Ethanol 28.9 0.79 22.8 108
n-Propanol 30.7 0.80 24.6
Isopropanol 30.5 0.80 24.4
n-Butanol 33.1 0.81 26.8 96
Isobutanol 33.0 0.80 26.4
Tertbutanol 32.6 0.78 25.4 103
n-Pentanol 34.7 0.81 28.1
Ethers
Dimethyl ether 28.7 0.67 19.2
Diethyl ether 33.9 0.71 24.1
Dipropyl ether 36.4 0.74 26.9
Dibutyl ether 37.8 0.77 29.1
Ketones
Acetone 28.6 0.78 22.3
Methyl ethyl ketone 31.5 0.80 25.2
Diethyl ketone 35.7 0.82 29.3
a
Data derived from Wikipedia
b
Higher heating value (i.e., water product is a liquid)
c
Research Octane Number (RON)

25–30%. Compared to gasoline, ethanol has a higher
latent heat of vaporization and lower vapor pressure;
thus, it is potentially difficult to start the engine during
cold weather. To overcome this problem, gasoline is
a component of E85. During very cold weather, fuel
retailers will increase the percentage of gasoline to 30%.
Conventional spark-ignited engines are not designed to
combust E85; however, flexible-fuel vehicles (FFVs)
can accept any ratio of E0 to E85. In FFVs, the metals,
polymers, and elastomers in the fuel system and
engine components must be compatible with the
range of fuels. Further, engine sensors detect the fuel
composition and automatically adjust the fuel injector
and spark timing. For the automobile manufacturer,
the marginal cost of producing an FFV is about $150
per vehicle.
1010 Vehicle Biofuels

14

12

Ethanol Capacity (billion gallons per year)

10

0
1980 1985 1990 1995 2000 2005 2010
Year

Vehicle Biofuels. Figure 1

US ethanol capacity [1]

160
Gasoline Consumption (billion gallons per year)

140

120

100

80

60

40

20

0
1980 1985 1990 1995 2000 2005 2010
Year

Vehicle Biofuels. Figure 2

US gasoline consumption [2]

Vehicle Biofuels. Table 3 Global fuel ethanol production
in 2009 [3]
Country Annual Production (million gallons)
United States 10,600.00
Brazil 6,577.89
European Union 1,039.52
China 541.55
Thailand 435.20
Canada 290.59
India 91.67
Colombia 83.21
Australia 56.80
Other 247.27
Total 19,963.70
E100 E100 is 100% ethanol and 0% gasoline.
The previously discussed fuels (E10, E15, E85) contain
ethanol blended with gasoline. In these fuels, only
anhydrous ethanol can be used; otherwise, water will
phase out when the gasoline is added. However, E100 is
a neat fuel (i.e., it contains no additives), so it can
contain water. In the production of ethanol, it must
be distilled from water. The top of the distillation
column reaches the azeotrope, where the composition
of the liquid and vapor are identical (95.63% ethanol
and 4.37% water, by weight). To break the azeotrope
requires additional technology (e.g., the use of molec-
ular sieves to remove water from the solution), which
adds cost. In Brazil, many vehicles operate on
azeotropic ethanol; therefore, the cost of the final
dehydration step is eliminated. To overcome potential
cold-start problems, it is common practice to start the
engine on gasoline and then switch to E100 once
the engine is warmed. Direct injection of ethanol
into the engine shows an effective octane rating of
130. With appropriate engine controls, it is possible
to burn neat ethanol in engines with a compression
ratio up to 19.5:1, which has a thermal efficiency
similar to diesel engines. The higher efficiency of the
high-compression engine compensates for the lower
energy content of ethanol, so the fuel mileage is similar
to gasoline.
Vehicle Biofuels 1011
Biodiesel Biodiesel is produced from vegetable oil
or animal fats. It has a carbon number similar to
petroleum-derived diesel fuel, so it can be combusted
in conventional diesel engines. Common blends
include B2, B5, B10, and B100, which contain
2%, 5%, 10%, and 100% biodiesel, respectively.
Disadvantages of biodiesel include the following:
(1) potential cold-start problems (not unlike
conventional diesel), which can be overcome using
additives that prevent clogging of filters; (2) deposits
and clogging from low-quality or oxidized fuel;
(3) slightly higher NOx emissions; and (4) lower energy
content. Regarding the latter point, because biodiesel
contains oxygen, B100 has 3–5% less energy content
than petroleum-derived diesel. Advantages of biodiesel
include the following: (1) better lubricity (short-term
studies with biodiesel show that it has less wear
than conventional diesel) and (2) lower emissions.
Regarding the latter point, because of its oxygen
content, biodiesel burns more completely. Further, it
has no sulfur or aromatic emissions.
Although biodiesel is more commonly used in die-
sel engines, it can be used in aircraft engines as well.
Usually, it is blended with conventional jet fuel;
however, there are some reports of aircraft operating
with B100.
E-Diesel Because of its hydroxyl group, ethanol has
hydrophilic properties and is completely soluble in
water in any ratio. Because of its ethyl group, ethanol
has oleophilic properties and is completely soluble in
diesel fuel in any ratio. However, if a small amount of
water (0.2%) is present in the fuel, two phases form:
one ethanol-rich and the other diesel-rich [6]. To
prevent phase separation, additives (e.g., surfactants,
emulsifiers, and co-solvents) can be included in the
fuel. The addition of biodiesel to petroleum-derived
diesel allows significant quantities (5–15%) of ethanol
to be added to the fuel [7]. Rather than adding ethanol
directly to diesel fuel, ethanol can be stored in
a separate tank and introduced as a fog or vapor into
the air intake of the engine.
Biomass
Biomass is biological material from living or recently
living organisms. In the context of biofuels, usually
1012 Vehicle Biofuels

biomass is used to describe plant matter such as wood, Cellulose

grass, agricultural residues, energy crops, and algae.
The dominant components of biomass are described
below.
Free Sugars
Free sugars are found in fruits and plant juices. Figure 3
shows examples including glucose, fructose, and
sucrose (a disaccharide of glucose and fructose).
As shown in Fig. 5, cellulose is an unbranched polymer
of glucose joined by b bonds. Because of extensive
internal hydrogen bonds (Fig. 6), cellulose is crystalline
and rigid; therefore, it is used as a structural
component of roots, stems, and leaves. Because
these components are so common in plants, cellulose
is the most abundant biological material produced
on earth.

Starch
As shown in Fig. 4, starch is a polymer of glucose
joined by a bonds. It occurs in two forms:
amylose (unbranched) and amylopectin (branched).
In plants, starch is used primarily as an energy-
storage compound.
Hemicellulose
As shown in Fig. 7, hemicellulose is a polymer with
a backbone of xylose (a five-carbon sugar) joined
by b bonds. In addition to xylose, hemicellulose
contains glucose, mannose, galactose, rhamnose,
arabinose, mannuronic acid, and galacturonic acid.

CH2OH CH2OH CH2OH CH2OH

OH CH2OH
O O OH O O O
OH OH HO OH HO
OH OH CH2OH OH O CH2OH
OH
OH OH OH OH OH

α-D-Glucose β-D-Glucose Fructose Sucrose

Vehicle Biofuels. Figure 3

Free sugars. (Note: For simplicity, the hydrogens bonded to carbon are not shown)

OH

O O
CH2OH CH2OH CH2OH HO
OH HO
O O O O O
O
OH OH OH HO
OH O O OH HO O
O OH
OH OH OH HO
300–600
HO O
O
HO
HO
O
Amylose Amylopectin

Vehicle Biofuels. Figure 4

Two forms of starch. (Note: For simplicity, the hydrogens bonded to carbon are not shown)
 Vehicle Biofuels 1013

CH2OH CH2OH CH2OH

H C O H C O H C O
H H H
C OH H C O C OH H C O C C O
OH H
C C H C C C C
H H
H OH H OH H OH

Cellulose

Vehicle Biofuels. Figure 5

Cellulose structure

H H

O
O H
H H H CH2
CH2 H O H
H O H
H H O
H H O O O
O O H
H H
HO H O O OH
O O H
H
O H H H H O H
H O H CH2 H
H CH2 H H H
O
O O
H O H H
H H CH2
H CH2 H O
H O H
H O
H O H H
H O O H
O O H
O H O OH
HO H O O H H
O H H H H
H O
H O H CH2 H
H CH2 H H
H O
O
H O
H O H
H H CH2
H CH2 H O
H O H H
H O H H O
H O O O H
O H
HO H O OH
H O O
O H H H
O H H H
H O
H O H CH2 H
H CH2 H O
O
H H

Vehicle Biofuels. Figure 6

Cellulose has extensive hydrogen bonds, shown as dotted lines

Hemicellulose is randomly acetylated, which helps it Lignin

resist degradation by sterically hindering hemicellu-
Lignin is a complex polymer composed of three mono-
lose enzymes.
mers (Fig. 8), which are highly cross-linked (Fig. 9).
1014 Vehicle Biofuels

OOC
H3CO O
HO H3C O
OH
O O
O
HO O HO O
O O O O O
OH n
O
O OH
H3CO
O
OH
HO

Hemicellulose

Vehicle Biofuels. Figure 7

Hemicellulose structure

CH2OH CH2OH CH2OH Proteins

CH CH CH
Figure 11 shows the chemical structure of an amino
CH CH CH
acid, which contains an amine group (NH2) and
a carboxylic acid (COOH). In naturally occurring
OCH3 CH3O OCH3 amino acids, there are 20 standard R groups, which
OH OH OH are coded by DNA. By eliminating water, the amine and
coniferyl alcohol sinapyl alcohol p-coumaryl alcohol carboxylic acid groups form a peptide bond, thus
creating a protein polymer. In nature, proteins have
Vehicle Biofuels. Figure 8
many functions, including catalyzing reactions
Lignin monomers
(enzymes) and providing mechanical structure.

Lignin is the “glue” that holds biomass together. Biomass Processing

By analogy to fiberglass composites, lignin functions
like epoxy resin whereas cellulose functions like glass
fibers. Lignin is hydrophobic and reduces evaporation
of water from plant vessels. Further, lignin resists
biological attack by insects and microorganisms.
Triacylglycerol
Triacylglycerol (TAG) is the main component of vege-
table oil and animal fats and is composed of fatty acids
bonded to glycerol via ether linkages (Fig. 10). In
Fig. 10, the value of n typically ranges from 14 (palmitic
acid) to 16 (oleic and stearic acids). Although Fig. 10
shows the hydrocarbon chain is fully saturated with
hydrogen, naturally occurring fatty acids often have
some unsaturated bonds as well.
Figure 12 shows an overview of processes that convert
biomass into fuels, which is described in more detail
below.
Raw Materials
Lignocellulose includes wood, grass, agricultural
residues, and aquatic plants. Lignocellulose is the struc-
tural component of biomass and is composed of
cellulose (30–50%), hemicellulose (20–40%), and
lignin (20–30%). By far, lignocellulose is the dominant
form of biomass produced on earth. Table 4 catalogs
the land available in the United States. Table 5
summarizes the potential lignocellulose resources in
the United States. Tables 6 and 7 focus on waste
 Vehicle Biofuels 1015

H2COH

H2COH CH2

CH2 H2COH
CH2
OH
H2COH HO CH
OCH3 CH2

CH3 CH HCO
CH3O CH3O
CH O CH
HCOH CH3O
HOC CH CH2OH HC (Carbohydrate) H2COH
OH OCH3
HC O OCH3
O H2COH HC O
CH3O
H2COH
CH3O OCH3 O CH
HC OH
CH3O
O CH

HC O HC O C2H H2COH

OCH3 HC O
HC HC CH H2COH
CH3O
O CH HCOH
HOCH2 OH H2C O CH H2COH

HOCH OCH3 HOCH HC O

HC H2COH CHO
CH3O CHO OCH3

H2C O CH O CH
O CH CH3O
HOC CH CH2OH CH
O CH HC O

CH3O O HOCH H2COH

HOCH2
CH3O O CH HC
OCH3
OCH3 CH3O
HOCH2 HC O
CH OH
H2COH CO
O CH
CH3O
CH2O HC CH HC O CH2
HOCH2 OCH3
HC CH2 HO CH HC O
O
H2COH HCOH
HCOH

HCOH CH3O
O OCH3
H2CO
OH

Vehicle Biofuels. Figure 9

Lignin structure

biomass whereas Table 8 describes the productivity of
energy crops.
Sugar crops include sugarcane, sweet sorghum, and
sugar beets. The dominant sugar is sucrose with minor
amounts of glucose and fructose. Table 9 shows the
productivity of sugar crops.
Starch crops include grains (e.g., corn, grain
sorghum) and tubers (e.g., potatoes, cassava). Table 10
shows the productivity and abundance of US grain
crops. Figure 13 shows the historical increase in US
corn productivity.
Oil crops include palm, soybeans, rape, and Chinese
tallow. Also, animal fats and waste frying oil can be
used as raw material. Table 11 shows the productivity of
oil crops.
Algae include primarily microalgae (e.g., cyan-
obacteria, diatoms) but also macroalgae (e.g., kelp, sea-
weed). Table 12 shows the range of lipid contents and
1016 Vehicle Biofuels

biomass productivities of microalgae grown under contains a substantial amount of oxygen and is
laboratory conditions. Table 13 shows lipid contents, water soluble. A typical composition includes sac-
biomass productivities, and lipid productivities for charides (2.4–3.3%), anhydrosugars (6.5–6.8%),
microalgae grown in outdoor ponds. aldehydes (10.1–14.0%), furans (0.35%), ketones
(1.24–1.4%), alcohols (1.2–2.0%), carboxylic acids
Intermediates (8.5–11.0%), and pyrolytic lignin (16.2–20.6%).
Because of the acids, the pH is low (1.5–3.8) [27].
Figure 12 describes the following intermediates:
Synthesis gas is the common name for a mixture of
Bio-oil is the liquid resulting from biomass pyrolysis. carbon monoxide and hydrogen, so named because
Visually, it looks like crude oil; however, the com- it can be catalytically converted to a wide variety
position is very different. Unlike crude oil, bio-oil of products.
Acid degradation products result when biomass is
heated with strong mineral acids. Components
O
include acetaldehyde, formic acid, 2-furfural,
H3C (CH2)n C O CH2 glucose, 5-hydroxymethyl-furfural, 5-methyl-2-
O
furfural, levulinic acid, saccharinic acid, and
2-furoic acid [28, 29].
H3C (CH2)n C O CH Sugar intermediates include glucose, xylose, arabinose,
O galactose, mannose, rhamnose, and fructose, which
are derived from sucrose, starch, cellulose, and
H3C (CH2)n C O CH2
hemicellulose.
Triacylglycerol Carboxylates are salts of carboxylic acids and include
salts of volatile fatty acids (e.g., acetate, propionate,
Vehicle Biofuels. Figure 10 butyrate, valerate, caproate, heptanoate) and fatty
Structure of triacylglycerol acids (e.g., palmitate, stearate, oleate).

H R O H R O H R O H R O
H—N—C—C—OH H—N—C—C—N—C—C—N—C—C—OH
H H H H
n

Amino acid Protein

Vehicle Biofuels. Figure 11

Structure of amino acids and proteins

Raw Material Intermediate Fuel

(1) Lignocellulose (a) Bio-oil (A) Hydrocarbon

(2) Sugar crops (b) Synthesis gas (B) Alcohol

(3) Starch crops (c) Acid degradation products (C) Ether

(4) Oil crops (d) Sugar (D) Ester

(5) Algae (e) Carboxylate (E) Ketone

Vehicle Biofuels. Figure 12

Overview of processes that convert biomass to fuels
 Vehicle Biofuels 1017

Vehicle Biofuels. Table 4 US land categories [8] Vehicle Biofuels. Table 6 Residues that are sustainably
recovered under moderate crop yield increases without
Land Land
(million (million land use changes [8]
Land categories Percentage (%) acre) hectare) Potential
Forest 33 747 302 Yield Availability gasolinea
Biomass (dry ton/ (million dry (billion gal/
Grassland 26 588 238
resource (acre·year)) ton/year) year)
pasture and
range Crop residues
Crop 20 453 183 Corn 4.1 169.7 11.88
Special use (e.g., 8 181 73 Sorghum, 1.7 1.3 0.09
public facilities) grain
Misc. (e.g., urban, 13 294 119 Barley 2.2 2.8 0.20
swamp, desert) Oats 1.9 0.7 0.05
Total 100 2,263 916 Wheat, 2.3 27.4 1.92
winter
Wheat, 1.4 7.4 0.52
Vehicle Biofuels. Table 5 Summary of potential lignocel- spring
lulose feedstocks [8] Soybeans 1.8 0.0 0.00

Potential Rice 5.1 10.3 0.72

Availability gasolinea Cotton lint 1.1 5.5 0.39
(million dry ton/ (billion gal/
Biomass resource year) year) Other crops 1.2 20.8 1.46

Woody Wastes

Logging and other 64 4.5 Manure – 43.5 3.05

residues Fats and – 2.0 0.14
Fuel treatments 60 4.2 grease

Urban wood residues 47 3.3 Municipal – 29.4 2.06

solid waste
Fuelwood 52 3.6
a
Assumed yield = 70 gal/t
Sub-total 223 15.6
Herbaceous
Perennial crops 377 26.4 Conversion Processes
Crop residues 446 31.2 Biomass is converted to fuels using either thermo-
Process residues 87 6.1 chemical or biological processes [30].
Sub-total 910 63.7
Thermochemical
Grand total 1,133 79.3
a
Assumed yield = 70 gal/t At elevated temperatures, thermochemical conversion
processes react biomass with air or oxygen to form
solid (char), liquid, and gaseous products. Except for
minor amounts of nitrogen and sulfur, the elemental
Products
composition of biomass is dominated by carbon,
Figure 12 describes the following products: hydrocar- hydrogen, and oxygen (Table 14). Free of ash, nitrogen,
bons, alcohols, ethers, esters, and ketones. All of these or sulfur, a typical formula for woody biomass is
products have been described previously. CH1.4O0.59 [31].
1018 Vehicle Biofuels

Vehicle Biofuels. Table 7 Maximum potential of municipal solid waste (2009) [9]
Wet weight Assumed Dry weight Gasoline potential (bill
Percentage (%) (mill ton/year) moisture (%) (mill ton/year) gal/year)
Organic
Wood 6.5 15.8 7 14.7 1.03
Yard trimmings 13.7 33.3 70 10.0 0.70
Food scraps 14.1 34.3 80 6.9 0.48
Paper & paperboard 28.2 68.5 6 63.7 4.46
Inorganics
Glass 4.8 11.7 0 11.7 0
Metal 8.6 20.9 0 20.9 0
Plastic 12.3 29.9 0 29.9 0
Rubber/leather/textiles 8.3 20.2 0 20.2 0
Other 3.5 8.5 0 8.5 0
Total 100.0 243.0 186.5 6.67

Vehicle Biofuels. Table 8 High-yield lignocellulosic crops

Yield Potential gasolinea Potential gasolinea

Crop (dry ton/(acre·year)) (gal/(acre·year)) (L/(ha·year))
Mixed prairie grass [10] 1.64–2.67 115–187 1,070–1,740
Switchgrass [11] 2.3–4.9 161–343 1,500–3,200
Poplar [12] 4.5–6.7 315–469 2,940–4,380
Willow [12] 4.5–6.7 315–469 2,940–4,380
Miscanthus [12] 5.3–13.4 371–938 3,460–8,760
Photoperiod-sensitive sorghum [13] 8–17 560–1,190 5,230–11,100
Conventional sugarcane [14] 17 1,190 11,100
Giant cane (Arundo donax) [15] 25 1,720 16,300
Energy cane [16] 30 2,100 19,600
Elephant grass (Pennisetum purpurcum) [17] 37–47 2,590–3,290 24,200–30,700
Water hyacinth [18] 111 7,770 72,500
Water hyacinth with enriched CO2 [18] 146 10,220 95,400
a
Assumed yield = 70 gal/t

The equivalence ratio f is defined as the actual the stoichiometric amount of oxygen is calculated
oxygen fed to the reactor divided by the stoichio- as follows:
metric oxygen needed for complete combustion.
Using a typical formula for woody biomass, CH1:4 O0:59 þ 1:055O2 ! CO2 þ 0:7H2 O:
 Vehicle Biofuels 1019

Vehicle Biofuels. Table 9 US sugar crops (2010)

Land Yield Sugar content Yield Potential ethanol
harvested (wet ton/ (ton sugar/t wet (ton sugar/ yield
Crop (acre) (acre·year)) biomass) (acre·year)) (gal/(acre·year))
Sugarcane [8] 1,766,400 31.8 0.12 3.8 536
Energy cane [16] Negligible 100 0.09 9 1,270
Sugar beets [8, 19] 1,155,700 27.6 0.15 4.1 578
Sorghum, sweet negligible 13–20 0.06 0.78–1.2 110–169
[20, 21]
Ton = 2,000 lb
Assumed yield = 141 gal per ton of sucrose

Vehicle Biofuel