Download as pdf or txt
Download as pdf or txt
You are on page 1of 36

Reviews of Human Factors

andhttp://rev.sagepub.com/
Ergonomics

Digital Human Modeling for Workspace Design


Don B. Chaffin
Reviews of Human Factors and Ergonomics 2008 4: 41
DOI: 10.1518/155723408X342844

The online version of this article can be found at:


http://rev.sagepub.com/content/4/1/41

Published by:

http://www.sagepublications.com

On behalf of:

Human Factors and Ergonomics Society

Additional services and information for Reviews of Human Factors and Ergonomics can
be found at:

Email Alerts: http://rev.sagepub.com/cgi/alerts

Subscriptions: http://rev.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://rev.sagepub.com/content/4/1/41.refs.html
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
>> Version of Record - Oct 1, 2008

What is This?

Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013


CHAPTER 2

Digital Human Modeling for Workspace Design


By Don B. Chaffin

Digital human modeling (DHM) technology offers human factors/ergonomics specialists


the promise of an efficient means to simulate a large variety of ergonomics issues early in the
design of products and manufacturing workstations. It rests on the premise that most prod-
ucts and manufacturing work settings are specified and designed by using sophisticated
computer-aided design (CAD) systems. By integrating a computer-rendered avatar (or homi-
noid) and the CAD-rendered graphics of a prospective workspace, one can simulate issues
regarding who can fit, reach, see, manipulate, and so on. In this chapter, I briefly describe
the development of various DHM methods to improve CAD systems. Past concerns about
early DHM methods are discussed, followed by a description of some of the recent major
developments that represent attempts by various groups to address the early concerns. In
this latter context, methods are described for using anthropometric databases to ensure that
population shape and size are well modeled. Efforts to integrate various biomechanical mod-
els into DHM systems also are described, followed by a section that outlines how human
motions are being modeled in different DHM systems. In a final section, I discuss recent work
to merge cognitive models of human performance with DHM models of manual tasks. Much
has been accomplished in recent years to make digital human models more useful and effec-
tive in resolving ergonomics issues during the design of products and manufacturing
processes, but much remains to be learned and applied in this rapidly evolving aspect of
ergonomics.

M any human factors/ergonomics specialists have long desired to have a robust, ana-
lytical model that would be capable of simulating the physical and cognitive performance
capabilities of specific, demographically defined groups of people. A 1990 report from
the U.S. National Research Council on human performance modeling (Baron, Kruser, &
Huey, 1990) highlighted the following benefits of such models:

a. Experts in ergonomics can simulate and test various underlying human behavior theories
with these models, thus better prioritizing areas of new research.
b. Experts can use the models to gain confidence about their own knowledge regarding peo-
ple’s performance under a variety of circumstances.
c. The models provide a means to better communicate human performance attributes and
capabilities to others who want to consider ergonomics in proposed designs.

Since the early 1960s, the quest to develop robust human performance models for
ergonomics has evolved into two distinct research directions. The first path, which is
DOI 10.1518/155723408X342844. Copyright 2008 by Human Factors and Ergonomics Society. All rights reserved.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
41
42 Reviews of Human Factors and Ergonomics, Volume 4

emphasized in this chapter, has concentrated on how various physical aspects of human
behavior can be modeled, such as predicting population reach capabilities, sitting and
standing eye locations, and space and clearance requirements. During the 1980s this type
of modeling of people’s physical functions became more sophisticated, mainly because of
the growing computational power of high-speed personal computers, which enabled the
precise measurement and modeling of human anthropometric and biomechanical attrib-
utes. Such computing power today also provides the means to render visually realistic-
looking graphic hominoids, or avatars of the human body, so that ergonomics specialists,
engineers, designers, and managers of firms can visualize how various people can be
positioned and function in different environments rendered by computer-aided design
(CAD) or virtual reality (VR).
In this chapter, I describe the development and use of some of the more popular phys-
ical human models, which are often referred to as digital human models, or DHMs. These
models focus on how human manual functions, such as grasping and moving objects and
tools, can be simulated in a CAD environment, thus enabling more accurate and faster
biomechanical analyses of the type described in a previous review by Marras and Radwin
(2006). Also discussed in this review are the many challenges that the research and devel-
opment communities have faced—and still face—when attempting to provide robust and
valid DHMs that can assist in solving complex human-workspace interface issues during
the early design of vehicle interiors and industrial workplaces.
What is not discussed much in this chapter is a second important type of human be-
havior modeling, which concentrates on predicting the perceptual-cognitive aspects of
human performance. Models of this type have been referred to as human performance
process models. The primary motivation for the development of these models has been to
understand and predict the time required for different people to perform a task without
errors, especially when the task has a high perceptual and/or cognitive load. A contempo-
rary example of this would be when a driver of a car is suddenly required to slow or steer
the car into a different lane to avoid a sudden change in perceived traffic flow or road con-
ditions. Accurately predicting under what specific driving conditions a person would be
able to maneuver the vehicle safely is of immense importance, especially when designing
or specifying various types of driver aids, such as heads-up visual displays, automatic col-
lision warning systems, and night vision systems.
An additional motivation for developing these types of perceptual-cognitive models
is to predict the amount of mental load or stress that results when a person is required to
make certain types of decisions, particularly when the time for such decisions is limited
and/or the consequences of an error in judgment are high. Human performance process
models to assist in solving these types of problems have been developed by several groups
over the past few decades. Some excellent reviews are provided in Polk and Seifert (2002);
Gluck and Pew (2005); Liu, Feyen, and Tsimhoni (2006); and Fisher, Schweickert, and
Drury (2006).
One issue that will be mentioned briefly toward the end of this chapter is the need to
integrate physical DHM simulation models with cognitive human performance process
models. Ianni (1999) proposed that such an integration was vital to the continuing growth
and use of both types of models. The work of one group to perform this type of integra-
tion is described near the end of the chapter.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 43

MOTIVATION AND EARLY DHM DEVELOPMENTS


During the 1970s and 1980s, several physical human simulation models were developed
to aid in the design of workspaces in which manual tasks were to be performed. Only
a few of the more widely used and referenced DHMs are discussed here. The reader is
referred to the following texts for a greater appreciation of the history of these models:
Automotive Ergonomics, edited by Peacock and Karwowski (1993); Computer Aided Ergo-
nomics, edited by Karwowski, Genaidy, Shihab, and Asfour (1990); Simulating Humans,
by Badler, Phillips, and Webber (1993); Digital Human Modeling for Vehicle and Workplace
Design, edited by Chaffin (2001); and the following review papers: “Simulating Humans:
Ergonomic Analysis in Digital Environments,” by Raschke, Schutte, and Chaffin (2001); and
“Digital Human Modeling for CAE Applications,” by Sundin and Örtengren (2006).

Early Seated Reach Models


From a historical perspective, one of the first attempts to develop a computerized human
performance model was in a doctoral dissertation by Kerry Kilpatrick (1970). He devel-
oped a 3-D human graphic model at the University of Michigan to depict how a seated
person would most likely be postured when performing reaches and moves constituting
a manual task. Simulations with his model required a user to provide a list of hand motion
requirements (e.g., “move Item A to Location 3”) and a file describing where objects were
located relative to the seat reference point of a simulated person. The latter file also con-
tained the weight of the objects and how they were oriented for grasping.
Kilpatrick’s model provided a means to scale the size of his 3-D avatar by using seg-
ment proportionality relationships from several existing anthropometric databases avail-
able at the time. For the posture predictions, an inverse kinematics structure was employed.
This structure provided a means to compute joint angles from any given position of
a linkage representation of a person. The representative postures for this linkage were pre-
dicted from a set of regression equations derived from 3-D photogrammetric data of seated
people reaching around their immediate environments (Chaffin, Kilpatrick, & Hancock,
1970; Kilpatrick, 1970; Snyder, Chaffin, & Schutz, 1972).
The output of the Kilpatrick model provided (a) predictions for population reach lim-
its; (b) a simple graphic avatar of various-sized people, as shown in Figure 2.1, which de-
picts postures of interest; and (c) the Methods-Time-Measurement™ predicted standard
performance times required to make various reaches and to move objects. An empirical
validation study showed that Kilpatrick’s postures were predicted with the joint center loca-
tions within 0.25 inches of the mean measured locations for 66% of the trials.
The Kilpatrick biokinematic model was one of the first such models to illustrate how
anthropometric data, when combined with geometric data describing a work environment
and a list of manual tasks, could aid a job designer in performing a physical ergonomics
assessment of a workspace. Its kinematics structure also became important during the early
development of the University of Michigan’s 3D Static Strength Prediction Program™,
described later in this section.
Perhaps the most noted original development of a digital human model to assist in
providing better ergonomics within a vehicle was provided by Ryan and Springer (1969)
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
44 Reviews of Human Factors and Ergonomics, Volume 4

Figure 2.1. Computer graphic representation of a predicted posture from the University of
Michigan’s first biokinematics model developed by Kilpatrick (1970).

for the Boeing aircraft company. This effort was called the First Man program, and later
was retitled simply Boeman. It was supported by a U.S. Naval Air Development Center
contract with Boeing. The goal was to provide an analytical tool for simulating the reach
postures and capabilities of pilots while seated with a full-torso restraint in various air-
craft. The motivation was to reduce the expense and time to build and test various pro-
totype crew compartments in advanced aircraft.
The Boeman model used a mathematical optimization scheme for selecting postures
for the three links composing the upper extremities (arms, forearms, and hands). Ryan
and Springer chose an optimization procedure that minimized the deviation in joint
angles from a so-called preferred posture. Although it could take up to 20 min to gener-
ate a sequence of simulated reach postures, the model did provide a tool for designers to
begin to visualize how their suggested placement of controls and displays would affect a
future pilot’s capabilities to reach to selected points in the cockpit when the torso was re-
strained by a four-point safety harness.
The basic structure of Boeman was adopted in the early 1970s by the U.S. Air Force
Aerospace Medical Research Laboratory (AMRL). The AMRL Crew System Interface Divi-
sion simplified the posture prediction model to gain computational speed and then added
the ability to simulate a variety of male and female anthropometric dimensions of pilots
seated in different types of aircraft. The data and algorithms used to refine the reach as-
sessment predictions were developed within a program referred to as the Crew Assess-
ment of Reach (CAR) program at AMRL. The resulting DHM model became known as
COMBIMAN. AMRL provided the program to various federal agencies and contractors
during the 1980s.

Early Biomechanical Models


Figure 2.2 depicts COMBIMAN being used to simulate sight lines from the predicted eye
position to displays of concern in a proposed aircraft. As can be seen, the program uses a
simple 3-D graphic depiction of the pilot without solid forms or shading enhancements.
During the early 1980s, AMRL personnel also recognized a need to simulate aircraft
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 45

Figure 2.2. COMBIMAN being used to simulate sight lines for a helicopter crewstation.
Adapted from McDaniel (1998).

manual maintenance tasks. As a result, COMBIMAN was reconfigured to stand, stoop,


kneel, and bend at the direction of the user while not only reaching about the immediate
environment but also lifting, pulling, and pushing on various tools and objects placed in
the hands. This new model was referred to as Crewchief. The reach and sight-line analy-
ses provided within Crewchief were a derivative of the algorithms existing in COMBIMAN,
but the strength prediction functions required totally new population data. These were
acquired in a series of empirical studies conducted by the AMRL staff during the 1980s
and resulted in an extensive catalog of force predictions that could be accessed through
Crewchief.
An example of Crewchief rotating a wrench is shown in Figure 2.3. Both interference
in the rotation of the ratchet handle and the capability of the population’s strength when
attempting to turn the wrench with a given torque could be simulated.
About the same time, I and my colleagues in the Center for Ergonomics at the Univer-
sity of Michigan developed a more general biomechanical model for population strength
prediction (Chaffin, 1969). Its 3-D biomechanical logic was completely described later in
Chaffin and Andersson (1984). During the early 1970s, a major initial application of the
model was to predict the static strength capabilities of astronauts when lifting, pushing,
or pulling on objects during various extravehicular activities in space and when on the
lunar surface, because it was very difficult to physically evaluate reduced-gravity effects
on manual exertions. The logic of the human strength prediction model was based on the
concept that when a person performs an exertion, the hand force requirements combine
with body segment weights to produce load moments at all the major body joints. When
a user of the program inputs (a) the amount of hand force required to perform a task,
(b) the anthropometry of the person, and (c) the posture used by the person, the model
predicts these joint load moments.
Based on the earlier Kilpatrick (1970) biokinematics model, as well as additional pho-
togrammetric studies in the early 1970s depicting people’s functional reach postures, an
inverse kinematics algorithm was developed to assist a user in providing posture predic-
tions. The population strength data, which are used as limits to the joint load moments,
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
46 Reviews of Human Factors and Ergonomics, Volume 4

Figure 2.3. Crewchief is shown rotating a ratchet wrench wherein an obstruction exists on
the left but not on the right. Adapted from McDaniel (1998).

were based on measurements of more than 2,000 workers who were tested by the Michigan
group during the 1970s. The model also predicted the compressive forces acting on the
lumbar spine, which were later compared in the model with cadaver vertebral body frac-
ture limits established by the U.S. National Institute of Occupational Safety and Health in
1981. Static body balance and static foot slip potential also were predicted by the model.
Taken together, during the 1980s, these output indices enabled users to quickly predict
the capability of a given population during different high-exertion, static, manual tasks
of interest.
The resulting software continues to be supported by the Michigan Center for Ergonom-
ics and is referred to as the 3D Static Strength Prediction Program, or simply 3DSSPP.
A demo version of the program can be accessed at http://www.umichergo.com. A screen
shot of this software is shown in Figure 2.4, which depicts the simulation of a large stock
reel being lifted from the floor.

European Model Development


During the same period of the 1970s and 1980s in which both the Boeman-Crewchief and
3DSSPP programs were being developed in the United States, Case, Porter, and Bonney at
Nottingham and Loughborough Universities in the United Kingdom developed SAMMIE
(System for Aiding Man-Machine Interaction Evaluation). SAMMIE was conceived in the
late 1960s as a very general human kinematics model for assessing various reach, inter-
ference, and sight-line issues posed by a designer. It uses a sophisticated statistical method
to assemble the population anthropometric data needed to predict the percentile size and
shape of given somatotype subgroups of interest.
Recently, the SAMMIE avatar has become a solid-segmented hominoid and contin-
ues to be fully integrated into a CAD program that enables users to insert the avatar
into a graphic environment of interest and then visualize the potential effects of various
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 47

Figure 2.4. The University of Michigan 3DSSPP™ model shown simulating the lifting of a
200-Newton stock reel from the floor. For various postures, hand forces, and demographic
groups specified, the model outputs population static strength capabilities, balance and foot
friction requirements, and low back compression forces relative to NIOSH norms. Reprinted
with permission of the University of Michigan Regents (1989).

workspace and vehicle designs (see Figure 2.5). Alternative postures can be selected from
menus or by direct joint manipulations. Hidden-line algorithms allow the avatar to appear
within a 3-D space, and sight lines with mirrors are provided as an additional feature for
vehicle designers. The SAMMIE program has been formulated to represent various spe-
cific individuals, which enhances the use of individual anthropometric attributes in a pop-
ulation. This development is referred to as the HADRIAN project and is discussed in the

Figure 2.5. A development version of the SAMMIE system on a Windows NT platform.


From Porter, Marshall, Freer, and Case (2004).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
48 Reviews of Human Factors and Ergonomics, Volume 4

next section. More information on SAMMIE can be found by contacting the developers
at http://www.lboro.ac.uk/departments/cd/research/groups/erg/sammie/.
For vehicle interior package designs, the German model RAMSIS (Realistic Anthro-
pological Mathematical System for Interior Comfort Simulation) deserves special note.
It was developed by a consortium of German automobile manufacturers that supported
a cooperative arrangement between a software company, TECMATH, and the Lehrstuhl
für Ergonomie at the Technical University of Munich beginning in 1987. Empirical stud-
ies of drivers were undertaken to determine both the postures they chose to use and their
psychophysical discomfort when seated for different periods in a variety of laboratory seat
bucks and vehicles. These data were then combined with an anthropometrically scalable
avatar to allow a designer to visualize people of different sizes and shapes as they drive
vehicles having various interior and seat configurations.
RAMSIS includes a sophisticated statistical method for representing different popu-
lation subgroups and employs an optimization method for matching the avatar’s pos-
ture with a large database of empirical data of people sitting in and driving various cars.
The kinematics optimization method also provides a means for a designer to move the
RAMSIS avatar within a vehicle of interest to assess the location of controls and displays.
The current avatar in RAMSIS uses a fully enfleshed, deformable graphic with hidden
lines and shadowing to provide a realistic-looking person.
Figure 2.6 depicts the RAMSIS avatar in a contemporary digital mockup of a vehicle.
A new aspect of the RAMSIS development includes further improvements in the realism
of the avatar when depicting a variety of people. This is referred to as the ANTHROPOS
project and is discussed in the next section. More information on RAMSIS can be found
at http://www.human-solutions.com/automotive_industry/ramsis_en.php.

Contemporary Models
During the late 1980s, another, more general-purpose, model was developed at the Ecole
Politechnique in Montreal, Canada. Now known as SAFEWORK, it was the first to

Figure 2.6. RAMSIS model used for analysis of control and pedal locations in an aircraft
cockpit depicted on the right. From Seidl (2004).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 49

incorporate a special statistical model that considers the multivariate correlation of anthro-
pometric dimensions that define human population size and shape. The model also fea-
tures an inverse kinematics method for assisting designers in selecting postures of interest
and simulating simple motions. The SAFEWORK avatar is a fully enfleshed and realistic-
looking human graphic embedded in a 3-D CAD system to render very complicated scenes.
Sight lines and physical interference with objects in the virtual environment are provided,
along with other ergonomics analyses.
Figure 2.7 depicts the SAFEWORK avatar used for assessing sight lines and control
locations. The present version of SAFEWORK is supported as an application within the
DELMIA software suite provided by Dassault Systemes and runs in a CATIA CAD envi-
ronment. More information can be accessed at http://www.safework.com/delmia/delmia_
sw.html.
Another general-purpose physical human simulation model, JACK, began as a NASA-
supported effort within the Department of Computer and Information Science at the
University of Pennsylvania during the mid-1980s (Badler et al., 1993). It was referred to
originally as the TEMPUS model for simulating astronauts assembling the International
Space Station. Questions of how an astronaut could best reach and fasten items, as well
as see objects while in a spacesuit, posed major issues to be considered in the early simu-
lations. Within a period of a few years, researchers realized that these same issues existed
for the operation and maintenance of military aircraft and other ground-based vehicles.
By the early 1990s, JACK was made available to the public with some of the following
features: (a) a means to use different published anthropometric data sets to produce a scal-
able linkage and avatar, (b) a flexible spine and multisegmental limbs that can be articu-
lated and positioned through an inverse kinematics model, (c) a method for creating a
solid-form environment in which the JACK hominoid can be positioned for reach and

Figure 2.7. Dassault Systemes digital human modeling products used to perform access
validation in an automotive factory. Courtesy of DELMIA World News #13.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
50 Reviews of Human Factors and Ergonomics, Volume 4

visual interference analysis, and (d) a strength-guided posture and motion prediction al-
gorithm. A suite of ergonomics analysis tools can be accessed through the JACK software.
Several CAD software companies took over the development of JACK during the
1990s. It is now available from the Siemens Corporation as part of its Product Life-Cycle
Management PLM software. More information can be found at http://www.plm.
automation.siemens.com/en_us/products/tecnomatix/. Figure 2.8 depicts a contemporary
use of JACK for simulating a workplace layout.
In the early 1990s, researchers at Aalborg University in Denmark proposed a new 3-D
biomechanical model that was based on the theory that the mechanical structure of the
musculoskeletal system shows a mathematical resemblance to a solid articulated struc-
ture in which the load-carrying elements (bones, muscles, and ligaments) are used to their
full potential to stabilize the configuration of the structure. They started to experiment
with algorithms for muscle recruitment in inverse dynamics that transform simple motions
into estimates of joint forces and moments needed to produce such motions. Using this
approach, the researchers were able to simulate well how the muscles of the lower extrem-
ities propel a bicycle. By 1997, they formed the AnyBody Research Group within the uni-
versity, proceeded to build a more complex whole-body model, and developed ergonomics
optimization in design as an engineering discipline within mechanical engineering at their
university.
In 2002, the AnyBody Research Group spun off a company, AnyBody Technology, and
was able to attract funding that made it possible to develop the AnyBody Modeling System.
This allowed researchers to license their model to others for testing and evaluation pur-
poses. The fundamental principle behind their software is different from typical digital
human models described earlier: It is more similar to finite-element systems, in the sense
that it enables the user to build a musculoskeletal structure from the bottom up. Users
can select from libraries that assist in determining how muscles, bones, and ligaments
should be located and function, rather than a fixed-linkage model (Damsgaard, Rasmussen,
Christensen, Surma, & de Zee, 2006).
The Aalborg University research project continues to develop. It has a detailed repos-
itory of various human musculoskeletal models, many of which were developed by users.

Figure 2.8. JACK being used to simulate a typical assembly line task. Image courtesy of
UGS Corporation.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 51

More information is available at http://www.anybody.aau.dk/. It should be noted that in


the AnyBody framework, the complexity of a given model can be composed of more than
500 individual muscle units. A typical example of the AnyBody structure is depicted in
Figure 2.9.
One of the most recent human body models under development is SANTOS, from the
University of Iowa. Although not yet released to the public for independent testing and
validation as of this writing, it promises future users that it can assist in performing sev-
eral types of ergonomics assessments. The structure, fully described by Abdel-Malek et al.
(2006), uses an avatar that is a visually realistic man or woman with deformable skin and
muscles to display the effects of different postures, as shown in Figure 2.10 (page 52). Opti-
mization methods are used within an inverse kinematics structure, so the user can move
the avatar without having to use direct joint manipulation methods. Users can specify
several optimization criteria to simulate manual tasks. For instance, a user may wish to
simulate a person attempting to minimize potential energy levels, self-collision, and end-
effector (hand) path finding.
An internal muscle force prediction model is being developed to predict joint stresses
and provide guidance on postures and exertions to minimize potentially harmful tissue
stresses. Empirical validation studies of the various posture prediction methods have be-
gun. More information about the SANTOS project can be found at http://www.digital-
humans.org.

PAST DEFICIENCIES IN DIGITAL HUMAN MODELING

The human simulation methods just described certainly indicate the emergence of a ma-
jor new design technology that is being rapidly developed and disseminated by several

Figure 2.9. A depiction of the AnyBody avatar and typical musculoskeletal structure perform-
ing an egress maneuver from a proposed vehicle. From J. Rasmussen at AnyBody Tech-
nology, Aalborg University.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
52 Reviews of Human Factors and Ergonomics, Volume 4

Figure 2.10. The SANTOS avatar in a predicted posture for lifting a large box. With permis-
sion of K. Abdel-Malek, University of Iowa.

different sophisticated and well-organized groups. This new technology has the potential
to drastically change and improve the process by which most designers decide on the ap-
propriate human physical attributes to consider in a workspace when attempting to meet
ergonomics and production goals.
In anticipation of the growing use of DHM technology for design, the U.S. National
Research Council (NRC) released a report titled “Ergonomic Models of Anthropometry,
Human Biomechanics and Operator-Equipment Interfaces” (Kroemer, Snook, Meadows,
& Deutsch, 1988), which was based on a 1985 workshop. This NRC report provided the
following recommendations to enhance the usability and effectiveness of the technology:

a. Improve the integration of anthropometric and biomechanical databases in DHMs.


b. Develop models that predict coordinated body segment motions required for simulating
a large variety of manual tasks.
c. Develop better biomechanical models to analyze and predict musculoskeletal stresses dur-
ing manual work simulations.
d. Develop robust and generic methods to improve the interface between existing DHMs and
workstation CAD programs.

More than a decade later, in 1995, an attempt was made to assess which specific features
would be most desirable in future human simulation software. The Society of Automotive
Engineers (SAE) G-13 Committee on Human Modeling Technology and Standards car-
ried out this evaluation and commissioned a survey (conducted in 1996) of designers re-
garding their potential uses for DHM methods. The survey contained a list of almost 500
potential attributes that could be included in future human simulation software. This list
was sent to 250 designers around the world with a request that they indicate their prefer-
ence for each attribute by using a scale ranging from 1 (no use) to 5 (highly useful).
About 20% responded, most of whom were involved in aerospace applications. These
designers showed a strong preference (a score of 4.0 or better) for features such as these:

a. being able to include different anthropometric data sets and population demographic sub-
groups;
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 53

b. including a variety of clothing, gloves, and helmets;


c. predicting the strength and endurance of different populations in a task;
d. being able to simulate realistic motions and postures with minimum task input descriptions
and in both physically constrained and unconstrained conditions;
e. providing hand grip, strength, and visual sight lines with and without mirrors and obstruc-
tions;
f. providing task timeline analyses;
g. performing reach and fit analyses for a variety of conditions; and
h. seamlessly accepting Input and Output commands and data and/or executing them within
various CAD systems that are commonly used in rendering and specifying products, tools,
and workstations.

It should be noted that the 1985 NRC workshop and the 1996 SAE survey participants
were disparate, more than 10 years apart in time; the former had only a few end users
involved in the workshop, whereas the later survey included only end users. Nevertheless,
there appeared to be considerable overlap in their assessments of DHM technologies and
future needs.
What follows are descriptions of how some of the major DHM developers and research
groups have been addressing the issues identified in both the NRC workshop and the
SAE survey.

RECENT DEVELOPMENTS IN DIGITAL HUMAN MODELING


FOR ERGONOMICS PROBLEM SOLVING

Incorporation of Anthropometric Databases Into


DHM Simulation
Perhaps no other single functionality in DHM development has received more attention
since about 2000 than making sure existing anthropometric data were referenced easily
and accurately to determine the size and shape of an avatar to be used in a particular sim-
ulation. Most often this has been accomplished by using percentile designations for the size,
weight, and inertial properties of a set of body segments of interest, and then referencing
a recognized anthropometric data set for the parameter estimates. In the latter regard, most
current DHM programs contain several popular anthropometric databases for quick ref-
erence, or they allow the user to manually enter the appropriate data.
It may sound trivial to scale an avatar to represent a designated proportion of a given
population once a good anthropometric database is made available, but it is not as sim-
ple as it sounds. The complexity begins when one realizes that each particular chosen pos-
ture affects how the size of multiple segments combines to provide a functional prediction
of interest. A second complexity arises in that the correlation of one body segment’s at-
tributes with another is not consistent between people or demographic groups. In statis-
tical terms, this multiple colinearity means that you cannot simply add segment dimensions
to create a human form that is representative of a particular percentile of the population
of interest.
Fortunately, some of the DHM models available today take these complexities into
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
54 Reviews of Human Factors and Ergonomics, Volume 4

consideration when providing a simulation to address the following typical workplace de-
sign question: “What percentage of the population could reach to a specific location?” To
provide the answer to such a practical question, researchers commonly use three statisti-
cal methods: the single-parameter method, the multiple-parameter method, and the bound-
ary manikin method. The single-parameter method uses a set of linear regressions to relate
segment parameter values to a single population attribute, such as population stature. The
resulting regressions capture the interdependencies of one segment’s size and shape with
those of another segment. The multiple-parameter approach expands the regressions to in-
clude several population attributes as the basis for the predictions (e.g., stature, gender, age,
and body mass are typically used). The boundary manikin method runs repeated Markov-
type simulations with the avatar in the posture of interest, using many different estimates
of segment size data, until a combination of the data is found that meets the percentile
accommodation goal set by the user. Dai, Teng, and Oriet (2003) and Greil and Jurgens
(2000) further discussed these methods and their implications in DHM simulations.
One additional approach to the problem of simulating people of different sizes and
shapes begins with providing DHM users with a highly deformable and visually appealing
avatar that can be configured to approximate a diverse set of anthropometric dimensions.
This approach is provided by the ANTHROPOS project in Germany (Seidl, 2004). The
avatar the researchers developed contains 3,200 skin surface points and as many as 40,000
reference points. Multiple regression methods are used to relate the location of each skin
surface point to identifiable skeletal structures, allowing the user to choose from a highly
diverse population database and to visualize a chosen avatar’s form in varied postures. An
inverse kinematics structure facilitates posturing of the manikin and moving it about in
a CAD or VR environment, as depicted in Figure 2.11.
Finally, an anthropometric scaling method—advocated by Porter, Case, Marshall, Gyi,
and Sims (2004) in the United Kingdom—uses a set of individual anthropometric data

Figure 2.11. Model structure of ANTHROPOS project, which provides a flexible and deform-
able avatar for realistic visualization of varied demographic groups (Seidl, 2004).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 55

from a diverse group of people. These researchers developed a program called HADRIAN
(Human Anthropometric Data Requirements Investigation and Analysis). It relies on an
efficient search method for finding individuals in their database whose anthropometric
characteristics could be accommodated by a specific design requirement. The proportion
of people who fit in the specified CAD environment indicates the level of accommodation.
In conclusion, there is no doubt that since the 1985 NRC workshop identified the need
to improve the anthropometric scaling aspects of DHMs, there have been major break-
throughs in this aspect of the technology. With newer 3-D body-scanning methods, com-
bined with new geometric statistical methods, this aspect of DHM will continue to improve
greatly in the near future.
One last point in this regard: With the increased degree of visual realism of the avatars
that are being developed today, a separate use for DHM technology is arising in ergonom-
ics: training people to operate and maintain complex products. By inserting realistic
human-looking avatars into special virtual reality video programs, and combining these
with motion prediction methods (to be discussed later in this chapter), one can create in-
teractive and self-paced training sessions that are specific to a person’s anthropometry.

Integration of Biomechanics in DHM Methods for Ergonomics


All the DHM software described earlier has the ability to access biomechanical models.
Some do this better than others, and most use what many would refer to as a backward
or inverse dynamics method. This method begins with the user specifying the kinematics
(segment displacements, velocities, and accelerations) of a posture or a motion. New-
tonian principles are then used, along with estimates of the inertial properties of body
segments and any external forces, to provide predictions of the resultant extrinsic forces
and load moments at the various skeletal joints of interest. Once such extrinsic joint forces
and moments are provided, separate muscle activation models are accessed to predict the
muscle-tendon-ligament internal reactionary forces required to stabilize and move
the body segments.
To provide this latter result, researchers first use geometric data from various anatom-
ical databases to describe the joint structures. Estimates of internal tissue forces needed
to balance the extrinsic joint moments then are used to predict the stresses and resulting
strains on the tissues that are most expected to fail or limit a person when performing the
exertion being simulated. As described in an earlier review by Marras and Radwin (2006),
this is not a simple matter because of the mechanical redundancy of muscles and liga-
ments required to stabilize and move segments when subjected to external forces and
moments. Figure 2.12 depicts this inverse dynamics logic being used to predict muscle reac-
tion forces for given motion kinematics (Zajac, 1993).
The inverse dynamics approach has many qualities that are very attractive in DHM sim-
ulations. Perhaps the most appealing is that the user simply has to concentrate on mak-
ing sure that the posture or motion data specified as inputs in a simulation are correct.
Unfortunately, this is also one of the major problems of this approach, in that small devi-
ations in postures and motions can result in very large errors in the external moments
and forces, which then result in errors in the prediction of who can perform a particular
task safely (Chaffin & Erig, 1989). Other major sources of errors are in the assumptions
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
56 Reviews of Human Factors and Ergonomics, Volume 4

Figure 2.12. An illustration of the typical method used to estimate muscle forces and joint
stresses and strains in most DHM models (modified from Zajac, 1993). The muscle actions
needed to create the joint torques Tn are estimated by optimization models of some form (see
text for more on this subject). Once joint muscle forces are known, they are combined with
the joint external forces and ligament forces to produce joint stress and strain estimates.

needed to resolve the indeterminacy of the highly redundant muscle and ligament sys-
tems that stabilize and move the skeleton.
Most of the muscle models use a variety of mathematical techniques for solving the
latter “tissue redundancy” problem. A review by Erdemir, McLean, Herzog, and van den
Bogert (2007) of 49 muscle force prediction models raises questions about the precision
of the predicted forces because many of the models have been validated only by using indi-
rect measures (primarily electromyography [EMG] magnitudes during controlled exer-
tions) to indicate individual muscle responses. Another issue is that the redundancy of the
muscle systems permits individual muscle activation patterns to vary greatly among indi-
viduals, even when they are performing simple movements.
Despite these reservations, various muscle force prediction models have provided a
means to begin to understand how different postures, loads, and motions combine with
a particular set of anthropometric attributes to cause tissue failures. Perhaps the most
popular use of this approach, beginning in the late 1960s, has been in the development of
torso muscle prediction models. The primary goal of these torso models is to predict the
vertebral tissue injury risk associated with lifting and moving heavy objects in industry.
Chaffin, Andersson, and Martin (2006) described the evolution of these types of models
and how they are used to estimate potentially injurious compression and shear forces in
the lumbar motion segments. Hughes (2000) reviewed how the prediction of internal
torso muscle forces is affected by different muscle optimization methods.
Similar models have been developed to predict lower-extremity joint forces incurred
during walking and running. Shoulder and upper-extremity models also have been devel-
oped for predicting specific joint tissue stresses during various sporting activities. As Hatze
(2002) commented:

Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013


Digital Human Modeling for Workspace Design 57

Although these biomechanical models are very important in helping to learn about
the complexity of human tissue loading, the lack of structural complexity in many
muscle and joint tissue models, the widely differing assumptions about muscle acti-
vation patterns, and the inability to accurately validate the predictions for a large range
of human endeavors means that much more fundamental research is needed to
improve the prediction accuracy of models that are based on an inverse dynamics
method. (p.109)

The AnyBody modeling approach described earlier is an enlightened attempt to address


these concerns. It allows the user to modify easily both the anatomical assumptions and
individual muscle and ligament structures and their functions as needed, or as is justified
from the literature, for a specific simulation. Thus, the user can explore how sensitive a
particular outcome is to the many assumptions needed when using a particular musculo-
skeletal model.
Because of the aforementioned concerns, some researchers have taken a different ap-
proach for ergonomics workplace assessments, correlating the predicted joint torques with
psychophysically determined limits, either discomfort or muscle strengths. Some DHM
models exploit this approach, such as the 3D Static Strength Prediction Program described
earlier, which compares joint moments with a population database of psychophysically
determined joint muscle strengths. Kim, Martin, and Chaffin (2004) and Dickerson, Martin,
and Chaffin (2007) took a similar approach, using joint torques to predict shoulder and
torso psychophysical effort perception levels.
In contrast to this inverse dynamics modeling approach, which begins with posture or
motion data and predicts joint tissue stresses, is a forward dynamics approach. This begins
by assuming that a person will contract muscles in a certain pattern to create a posture
or motion of interest. To use this method, one needs a highly sophisticated biomechan-
ical model, coupled with a motor control or neural control model of the muscles. Once
again, Zajac (1993) described this general logic, as shown in Figure 2.13.

Figure 2.13. A forward dynamics model for predicting body motions based on knowledge of
muscle activation and joint biomechanics. Modified from Zajac (1993).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
58 Reviews of Human Factors and Ergonomics, Volume 4

As one can imagine, it is possible to develop optimal muscle control models given
enough assumptions about how the various redundant muscle systems behave. Much
work has been done in this area for simple motions, such as bicycle riding or unimpeded
cyclical walking (Kuo, 2002). Unfortunately, the forward dynamics approach, though
useful in studying rhythmic gait motions with a limited number of links involved, may
not be generalized easily to the wide range of discrete movement and posture requirements
that are necessary for ergonomics assessments.
Dysart and Woldstad (1996) attempted to use a forward dynamics modeling approach
for simulating the motions incurred during the lifting of a load. They studied sagittal plane
lifting motions with a five-link model, assuming three objective functions: minimization
of total joint torques, minimization of the percentage of strength required at each joint,
and maximization of balance. They found that the minimization of total joint torques was
the most statistically significant predictor of the movements, though considerable vari-
ance within the group data existed.
Perez (2005) statistically compared these and other objective functions using a large
set of lifting motion data, and concluded that each formulation yielded different joint
motion trajectories. These trajectories were dependent on (a) the type of movement be-
ing attempted, (b) the load being lifted, and (c) the final postures required by the task.
Researchers engaged in current work at the University of Iowa on the SANTOS project
described earlier do not assume that one objective function can be used to predict all types
of discrete, complex motions; rather, the user can choose among several optimization
schemes when simulating a particular motion (Abdel-Malek et al., 2006).
Because of limited knowledge about what exactly a specific person is attempting to
optimize when performing a particular motion or exertion, a modification to a strict for-
ward dynamics approach has developed for ergonomics assessments of complex manual
tasks. As an example, to compensate for not knowing how muscles are most often acti-
vated, researchers have developed a method referred to as a biologically driven muscle mod-
eling approach. This method requires that one take EMGs of a group of people to estimate
average muscle recruitment patterns used during a variety of exertions. Marras and Granata
(1995) and McGill (1992) pioneered the use of this approach to predict lumbar spinal
forces, for instance.
By combining imaging studies of the joint structures, anthropometric data, and torso
motion data, along with the EMG-based estimates of muscle forces, researchers can pro-
duce good predictions of vertebral tissue stresses. Furthermore, to provide the predic-
tion of these vertebral forces directly from only the lumbar moments, McGill, Norman,
and Cholewicki (1996) developed linear regressions of the vertebral compression forces
that were extracted from the much more complex, biologically driven EMG models used
to analyze people while lifting loads with a variety of motions and end postures.
It would appear that in the future, a person performing an ergonomics analysis with
a digital human model will be able to simulate many different types of tissue and even
cellular responses that are expected from groups of people when performing a particular
physical activity. Nuzzo (2007) presented what this future development might look like
in Figure 2.14, with an inverse dynamics feedback loop on the top superimposed over
a forward dynamics structure. Although a great deal of research is needed to provide the
necessary biomechanical, tissue, and neural-motor models, it is only a matter of time and
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 59

Figure 2.14. A schematic showing the integration of biomechanical models across multiple
scales. With permission from Nuzzo (2007).

effort before this integrated approach will become a reality. When it does, DHM simula-
tions will become even more powerful tools for proactive ergonomics and thus will real-
ize the goal stated by 1985 NRC workshop participants.

Lifelike Posture and Motion Modeling in DHM Methods


It should be clear that predicting human postures and motions when attempting to
predict human reach capability, sight lines, and physical accommodation is technically chal-
lenging. Many approaches have been taken within traditional inverse kinematics frame-
works to resolve the inherent behavioral variability associated with the wonderfully
complex, multiple-linkage system that comprises the human body. It is beyond the scope
of this chapter to review these methods in detail, but the reader is referred to several
excellent reference sources, including Allard, Cappozzo, Lundberg, and Vaughan (1998);
Gleicher (1997); Jagacinski and Flach (2003); Nigg, MacIntosh, and Mester (2000); and
Zatsiorsky (1998).
Given that many approaches to human motion prediction are being developed and
promoted by various research groups for special applications, the following criteria should
be considered when evaluating them for use in DHMs that are designated for proactive
ergonomics:

a. Simulated motions must be based on real human motion data to provide internal construct
validity and overall empirical validity.
b. Models of motions should be capable of representing motions that are not in an existing
database, that have extrapolation capability, and that retain the essential lifelike motion
behaviors contained in the motion database.
c. Models should be computationally fast and portable for real-time simulations performed
in commercial CAD-DHM software products.
d. Models should be adaptable so that they can assimilate new motion data and algorithms,
thereby becoming more robust in predicting novel motion situations of interest to a
designer.

Empirical posture and motion modeling. Researchers have used motion capture sys-
tems and statistical methods since the early 1970s in an attempt to document the vari-
ability that is so obvious when people select various postures and motions to perform a
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
60 Reviews of Human Factors and Ergonomics, Volume 4

manual task. One of the earlier photogrammetric studies of reaching postures by Chaffin,
Schutz, and Snyder (1972) documented seated postures of a group of 35 young men. In
this early attempt, statistical regressions were used to determine the normative 3-D spa-
tial coordinates of body landmarks as a function of where the person was attempting to
reach, and individual anthropometry. The resulting multiple-regression equations later
were used as part of the posture prediction algorithm in the 3DSSPP model (along with
other data on standing and manual lifting postures) to assist a user in selecting among al-
ternative postures when performing a simulation of a manual task.
With the advent of much-improved, digitally based 3-D posture and motion capture
techniques since the early 1990s (Chaffin et al., 2006), it is now possible to capture accu-
rately not just static postures but 3-D whole-body motions. In 1998, the Human Motion
and Simulation Laboratory (HUMOSIM) was formed at the University of Michigan to
exploit this technology for DHM development. To date, more than 300 people of both
genders, ranging in age from 18 to 81 years, have participated in a series of motion cap-
ture studies. These studies have resulted in almost 120,000 motion data sets, which have
been made available to 12 research institutions for DHM development purposes. For more
information on accessing these data, as well as a description of the laboratory studies,
see http://www.HUMOSIM.org. What follows is a review of how these types of data are
being modeled for DHM use.

Empirically based motion modeling. Faraway (1997) developed a functional regression


model to predict body linkage motions. It is based on transforming the 3-D body marker
locations, obtained from a set of motion capture experiments, into a set of body joint
angles or spatial coordinates that vary over time. Once these joint angle–time patterns
are documented, a time-dependent, least-squares error regression method is applied to
the following form:
θ (t) = βo(t) + Cχβχ (t) + C y βy(t) + Czβz (t)
+ CχC y βxy (t) + C y Czβ yz(t) + CzCxβzx (t)+ Cx2βx2 (t) + C y 2β y 2 (t)
+Cz2βz2 (t) + D
where θ (t) are the predicted joint angles over time; Cχ, C y, and Cz are target coordinates;
β(t) are parametric functions to be estimated by the regression process; and D are select
demographic variables (e.g., age, stature, and gender) that could modify the predictions.
Faraway (1997) found that this quadratic regression model accounted for approxi-
mately 80% of the joint angle deviations measured in a large set of reach data. The method
provides a set of algebraic equations that are computationally fast when simulating a vari-
ety of motions. Because this method provides statistical estimates of both the means and
standard errors in a motion trajectory, Chaffin, Faraway, Zhang, and Woolley (2000) used
this method to describe statistically the dominant role that stature plays in predicting
seated reaching motions compared with gender and age effects. As powerful as this empir-
ical method is for identifying the different sources that account for the large variation in
normal human motions, the resulting prediction equations are susceptible to error prop-
agation throughout the multiple segments comprising the kinematics chain in a whole-
body simulation, thus affecting end-link motion accuracies (i.e., a predicted hand motion
may not end where the user intended).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 61

To correct for the endpoint prediction error when the functional regression method
was used for whole-body motion prediction, Faraway (1997) developed a second formu-
lation for motion prediction, referred to as the stretch pivot method. The stretch pivot mo-
tion prediction method combines the former statistical functional regression model of
joint angle and location predictions with an estimate of the hand coordinates predicted
by a regression of the trajectory and orientation of the hand as it is moved from the ori-
gin to its destination in a task. This is done in such a manner that the hand is guaranteed
to land where it is supposed to land in any given reach task simulation.
Once a hand trajectory is predicted, the multisegmented body is decomposed into two
or three link groups, which can both stretch and pivot as they are moved as groups within
an inverse kinematics computational structure. Functional regressions of critical angles
and/or joint coordinates of past data are used to resolve any kinematics indeterminacies.
Testing of this method showed that it worked well for many types of hand-reaching
motions.
Faraway, Zhang, and Chaffin (1999) developed a complementary method that com-
bined the functional regression equations with the use of a pseudo-inverse Jacobian com-
posed of the differences between the predicted joint angles and the joint angles needed
to ensure that the endpoint on the linkage followed the intended or predicted hand tra-
jectory. This method was shown to greatly improve the accuracy of the predictions of link-
age motions required to reach specific locations.
Zhang (1997) developed another use of the pseudo-inverse Jacobian method in this
context. He showed that he could fit, by regression, a set of weighting factors within the
Jacobian function for the velocities of the joint angles during motions of a three-link sys-
tem composed of the torso, upper arm, and lower arm. The weighting factors in this case
represented how much each link’s motions contribute to the kinetic energy involved in a
particular segment motion. An evaluation by Zhang and Chaffin (2000) showed that this
method was robust and accurate (the median joint angle error was 4.7 deg) for moder-
ately complex, seated reaching motions, but it did not perform as well for extreme reaches,
such as when trying to open the door on the opposite side of a large vehicle or when the
wrist angle had to deviate from a near-neutral posture to grasp an object.
Jung, Kee, and Chung (1995) developed a four-link seated reach model that used a
Jacobian formulation to transform the Cartesian coordinates of the joints to joint angle
velocities. They then employed a search method to minimize a weighted deviation of the
joint angles from a set of neutral posture angles, normalized for the range of motion for
each joint. They reported correlations of the mean joint coordinates with r2 values over
0.95 compared with data from 5 participants reaching to targets that did not require ex-
treme motions. The challenge for these different empirical methods appears to be how
to predict normal motions that involve more than three or four links, especially when it
is known that the motion variability between and within participants is high.
An artificial neural network (ANN) method is another empirical approach to motion
prediction. In the ANN method, data that have been observed to have a great deal of vari-
ation are systematically compared with predictions from an artificial neural network,
wherein several input conditions are chosen to drive the network response. Figure 2.15
shows a network configuration used by Perez (2005) for studying standing load-lifting
motions. In this particular configuration, the input conditions are the 3-D coordinates
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
62 Reviews of Human Factors and Ergonomics, Volume 4

Sigmoid Transfer Output Linear Transfer


Function Function
Units
Left Ankle Alpha
Input Hidden Units Left Ankle Beta
Variable bias
Layer (× 100) Right Ankle Alpha
End Effector x Position Right Ankle Beta
Variable bias Left Knee Alpha
Left Knee Beta
End Effector y Position Right Knee Alpha
Right Knee Beta
Left Hip Alpha
End Effector z Position Left Hip Beta
Right Hip Alpha
Right Hip Beta
Strength Coefficient Left Lower Torso Alpha
Left Lower Torso Beta
Right Lower Torso Alpha
Anthropometry Coefficient 1 Right Lower Torso Beta
Upper Torso Alpha
Upper Torso Beta
Anthropometry Coefficient 2 Left Clavicle Alpha
Left Clavicle Beta
Right Clavicle Alpha
Anthropometry Coefficient 3 Right Clavicle Beta
Left Shoulder Alpha
Left Shoulder Beta
Anthropometry Coefficient 4 Right Shoulder Alpha
Right Shoulder Beta
Left Elbow Alpha
Left Elbow Beta
Right Elbow Alpha
Right Elbow Beta
Left Wrist Alpha
Left Wrist Beta
Right Wrist Alpha
Right Wrist Beta
Neck Alpha
Neck Beta

Figure 2.15. One of several ANN structures used by Perez (2005) to predict postures and
motions involved in lifting moderate loads.

of the hand at the end of a lift, as well as anthropometric and strength data describing
the subjects. The outputs from the network are major joint angles or coordinates relative
to a global reference system.
In developing a particular ANN, researchers must use a recursive search method to find
a combination of weights that associate particular input values with the output values.
At least one hidden set of units is used in an ANN between the input and output units to
allow complex interactions to be included, which some consider joint coordination effects.
The general rule always is to attempt to use as few hidden units as needed to fit the data.
Perez (2005) used a large set of motion data from the HUMOSIM laboratory to “train”
his model of lifting motions—that is, to determine the set of weighting factors that, in
combination, provide the joint coordinates that best match the measured joint coordinates
(or joint angles). As is required in the ANN development process, another set of motion
data is then used to validate the model under conditions other than those used to train
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 63

the model, thus providing a means to determine how robust the outputs are in general.
Perez found very high correlations between the predicted and actual joint coordinates,
with mean average errors in the predicted joint locations of about 4 to 7 cm, depending
on the location of the reach targets.
Using data from simple upper-extremity, planar motions, Wada and Kawato (1993)
showed how ANN models could be used to solve forward dynamic optimization formu-
lations using one of the following objectives: minimize joint torque changes, minimize
muscle tension changes, or minimize motor neural command changes. They concluded
that for highly practiced upper-extremity motions, the ANN results indicated that such
motions were more consistent with minimizing motor command changes than with min-
imizing joint torques or muscle tension changes. They speculated that the latter may occur
while learning new motions.

Motion engineering algorithm development. Park, Chaffin, and Martin (2004) pro-
posed a motion engineering system approach consisting of three components: a motion
database (memory of motions completed), a motion search and comparison method
(motion retrieval), and a motion modification algorithm (generalization method). Theo-
retically, an organized computerized motion database can be considered a surrogate for
one’s memory of past motions. This is important in that several motion memory experts
have proposed that we retain an imprint or templates of our past motions, and we then
use them at a subconscious level to assist us when attempting to perform a new motion.
This is referred to as the general motor program (GMP) theory (Schmidt & Lee, 1999).
The first part of the method by Park et al. (2004) structures a motion memory file
as a set of joint angle–time trajectories derived from motions that participants performed
in various experiments. Using the algorithm by Park et al., one makes a new motion sim-
ulation request by defining the ending and beginning conditions of the motion, as well
as some anthropometric specifications of the people to be simulated. The algorithm then
searches the database for motion conditions that are similar (i.e., none or many could be
found, depending on how the user sets the criteria for “similar” motions). Assuming that
at least one stored motion set meets the user’s criteria, the algorithm by Park et al. sys-
tematically applies a gradient search algorithm to modify the joint angle–time trajecto-
ries that were retrieved from the database to satisfy the newly designated reach beginning
and end-point conditions.
Statistical comparisons of joint locations with standing reach data obtained from 20
participants revealed that the predicted joint locations were within about 3 cm of the
actual mean locations, provided that the hand end point of a chosen motion was within
15 cm of the one stored in the database. If the end-point location being simulated was
30 cm away, then the joint locations averaged about 4 cm from the actual mean values.
Park, Singh, and Martin (2006) demonstrated how such a memory-based model can be
used to search for motions that would prevent a collision between a body segment and
an obstacle.
Park, Martin, Choe, Chaffin, and Reed (2005) developed a joint angle contribution
vector (JCV), which is an index of how much each joint angle change contributes to a
particular complex motion. The JCV allows the user to understand how much two
motions differ and which joint motion contributes most to the difference. Knowing the
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
64 Reviews of Human Factors and Ergonomics, Volume 4

variability in motions for a particular task could influence the choice of which one to
use in subsequent ergonomics evaluations of a task, or how some people’s joint motion
patterns differ from others.
In summary, the memory-based motion engineering algorithm that Park et al. (2004)
developed can be considered a motion-editing system that contains a motion library, a
search engine to find motions stored in the library that are close to a new motion of inter-
est, and a systematic means to modify them to fit new scenarios. In this context, the
method meets the criterion of being capable of accommodating new motions easily by
simply adding these to the library. It also meets the criterion of being capable of modify-
ing existing data and thus extrapolating past motions to meet new motion requirements.

Optimization-assisted motion prediction methods. As mentioned earlier, various in-


vestigators use optimization methods to predict body segment motions for subsequent
ergonomics analyses. A couple of optimization modeling studies that also involve com-
parisons with actual motion data are described in the following section.
One attempt at using optimization methods described earlier was by Dysart and
Woldstad (1994), who employed a five-link biomechanical model to simulate sagittal
plane lifting. They formulated several different objective functions that minimized vari-
ous biomechanical indices, such as total joint torques, proportion of joint torque relative
to strength required, and loss of balance.
Lin, Ayoub, and Bernard (1999) developed a similar model of sagittal plane lifting and
included a fully dynamic analysis of the joint forces and moments. They minimized the
sum of the squared ratio of joint torques divided by strength moments over the total time
allowed for a lift. Joint range of motion and body balance limits were used to prevent
extreme postures in both of these efforts. Joint trajectories were assumed to be regular in
time and were fit with a two-parameter polynomial with zero velocity at the beginning
and end of a motion of the hands. A nonlinear optimization method was used to solve
the system of equations. Lin et al.’s comparison with motion data acquired from meas-
urements of 10 participants who were lifting weights from floor level to a shelf at various
heights indicated that the average joint motion trajectories were captured well with about
a 5-deg mean error, but individual joint trajectories varied by over 25 deg, depending
on the task specifications.
Wang (1999) developed a more heuristic, rule-based algorithm for predicting upper-
extremity seated reaching motions. The algorithm was implemented by providing data
about both the hand motion trajectory and the orientation of the hand. It was based on
the assumption that during a reach, the torso would not move except when the range of
motion of the shoulder joint was approached. Geometric rules also were provided that
used the inputted hand locations and orientations to direct forearm rotations and elbow
angles. An optimal control algorithm that minimized select joint velocities was used to
predict the joint motion trajectories.
Wang showed that the algorithm was more computationally efficient and was as accu-
rate as other optimization algorithms that are used for predicting body segment locations
during different types of seated reaches. His mean elbow error predictions were on average
within about 3.4 cm of the mean locations measured during a set of seated reaches. What
is interesting is that his approach combines knowledge of the anatomically determined
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 65

limits to certain motions with optimization methods to select values for segment motion
kinematics variables.
From the preceding, it should be obvious that to use optimization methods effectively
to predict how people move, one requires knowledge about musculoskeletal physiology
and the motor behavior of people so that computational challenges that accompany non-
linear and quadratic programming methods can be reduced. The other issue that plagues
this approach is that different people use different varieties of motions to accomplish the
same goal, depending on their experiences and perceptual-motor capabilities. In other
words, various objective functions must be considered to model even simple motions, and
choosing which one to use for a particular group of people and situations presents a large
challenge to most users of these models.

Optimal control models of motions. Because it is well known from various neurolog-
ical studies of human motions that neural feedback exists during a motion, it must be
assumed that it can play an important part in assisting a person to perform novel or dif-
ficult motions. People are known to receive proprioceptive, kinesethic, and tactile feed-
back, as well as visual and auditory feedback, and all this informs the central nervous
system about the state of a particular motion. Therefore, it is reasonable for motion mod-
elers to attempt to use optimal control methods to study and simulate movements. Zajac
and Winters (1990) described the form that such models follow conceptually (see Fig-
ure 2.16). Models of this type attempt to recognize that multiple feedback loops affect
neural control (activation dynamics), muscle tendon forces (contractile dynamics), and
joint torque production (joint geometry), all of which are moderated by the body segment
motion dynamics.
Inspection of Figure 2.16 should remind the reader of how difficult it is to develop an
optimal control model that is robust in predicting how various people move in a variety
of disparate situations. The fact is that we have limited physiological knowledge about
many of the concepts contained in Figure 2.16. Even assuming we know the input com-
mands to muscles (which are often approximated by muscle EMG levels), the muscle ten-
don forces must be modeled as a function of the length and velocity of shortening or
lengthening of each muscle, often using only crude (Hill-type) muscle models. Joint geom-
etries have complex shapes that alter the muscle force-to-torque relationships, further
complicating the modeling process. And it is known that previously performed motions

Figure 2.16. A conceptual diagram of a control system showing multiple feedback loops that
are known to affect motion control. From Zajac & Winters (1990).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
66 Reviews of Human Factors and Ergonomics, Volume 4

can affect new motion dynamics through our motion memory, not to mention physical
fatigue and/or past neuromuscular pain experiences.
At best, optimal control models have been used to simulate simple, planar motions of
small linkage systems, such as when pedaling a bicycle. Although the method does pro-
vide an important tool in studying motor coordination, as Zajac (1993) stated, there are
several important reasons that this method has seen limited use in predicting more com-
plicated motions:
• Considerable effort and expertise are needed to formulate a multiple muscle dynamics model
of even simple planar motions.
• Computational time is very long—days or weeks for a single simulation on a desktop com-
puter (Menegaldo, Fleury, & Weber, 2003).
• Muscle coordination is poorly understood, even when EMG patterns are used as inputs to
the models.
• Dynamic optimization algorithms to find acceptable or even feasible solutions are (a) few,
(b) not robust, and (c) computationally time intensive.

Top-down motion engineering methods. It should be clear that one particular motion
modeling method is not best for all situations—that is, it does not meet the four criteria
stated at the beginning of this section. One approach that has been taken by Reed, Faraway,
Chaffin, and Martin (2006) is to use a motion framework that integrates several different
modeling concepts to allow robust simulation of a large variety of motions.
The guiding principles used by Reed and colleagues (2006) in the development of the
top-down motion framework are as follows:
Modularity. This means that a set of well-documented, rational models of particular
types of motions could be accessed by the framework as required by a midlevel task plan-
ner. For instance, if the task planner calls for simulating a walk to a bin of parts, the frame-
work organizes the geometric and other task data needed to permit the use of a foot
placement algorithm, such as the one from Wagner, Reed, and Chaffin (2005). Similar
modules regarding eye-hand coordination (Kim, 2005), spinal curvatures (Reed, Parkinson,
& Chaffin, 2003), and body balance (Parkinson, Chaffin, & Reed, 2006) are accessed as
specific tasks are simulated. The importance of this particular principle is that as new mo-
tion modules are developed, they can be added to the framework to enable it to provide
more robust motion simulations.
Algorithmic. Each module used in the framework needs to be documented completely
in a published paper so that motion simulation software developers can integrate them
into various commercial motion simulation applications.
Behavior based. This requirement simply recognizes the wonderful complexity and
variations that exist in human motions. Until we have a much deeper understanding of
all the psychological and physiological factors that affect our individual movements, we
will have to accept a certain amount of empirical representation, with prediction models
and algorithms derived from carefully constructed human motion studies. Luckily, since
the mid-1990s, large motion databases have begun to be assembled and used as the basis
for many of the motion modules required in the framework. Its reliance on these data-
bases gives the framework its intrinsic validity.

Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013


Digital Human Modeling for Workspace Design 67

Coordinated actions. A fundamental requirement of the framework is to coordinate


the input and output of various complex motion modules. At a minimum, this requires
an inverse kinematics type of structure that ensures that the body segments do not dis-
assemble during a motion when competing motion modules predict different movement
scenarios. The framework uses empirical guidelines that give priority to the sequence of
segment motions needed to simulate a required whole-body motion. In essence, the frame-
work acts like a high-level motion controller, not unlike the human cognitive central
nervous system, which is believed to choose and coordinate simple stored motions from
memory.
In the context of the motion framework, simple segment motions are represented as
a set of low-level motion modules, shown at the bottom of Figure 2.17. The resource
scheduler in the center of Figure 2.17 determines which modules will be given priority
to produce a motion, and others are used to check the feasibility of the motion. An exam-
ple would be a task that first required walking to a bin and reaching for a part in the bin.
The foot placement module would be accessed along with the lower-extremity modules
to simulate walking to the bin. The upper-extremity reach prediction module would then
be used to reach to the part. If the reach is extreme for a designated population or there
is a biomechanical strength limit, or if balance limits are exceeded, the torso module may
be used to adjust the shoulder locations, and the upper-extremity reach module is ap-
plied again. Similarly, the visual gaze module may be used to ensure that the head is posi-
tioned to provide a line of sight if required to see the part while reaching to it.
Robustness. The framework is an attempt to provide a systematic means to extrapolate
specific motion studies and data to provide novel and complex motion simulations. This
implies that some motions will not be as well validated as others. If the framework is
extrapolating beyond certain limits, then it warns the user. Of course, as a larger array of
motion modules is provided within the framework and more independent users test the
framework in unusual situations, validity should be less problematic.
Integrated analyses. Reed et al. (2006) recognized that motion simulation is rarely the
goal in ergonomics but is a necessary requirement. Most often, the user wishes to know
how much time it will take to perform a new task or how difficult and possibly risky such
an endeavor might be for a specified population. The motion framework he and his col-
leagues have developed provides a structure for motion time prediction and biomechan-
ical analyses to be performed as an inherent part of a motion simulation. For instance,
the user may set a criterion that in all the motion simulations to be performed, 95% of the
specified population should have the necessary muscular strength. This command to
the framework would give priority to a biomechanical strength prediction module that
would then be activated during each motion iteration, ensuring that various muscle groups
are not overly stressed when the final motion simulation is produced.
The status of the motion framework by Reed et al. (2006) is that it is a relatively new
approach. Although based on a great deal of empirical modeling at the University of
Michigan’s Human Motion Simulation Laboratory, it still remains to be evaluated thor-
oughly by independent users. Parts of the framework have been integrated into JACK
and SAFEWORK prototype systems for testing using special and limited problem sets.
The framework approach is in some ways an extension of Wang’s (1999) behavioral-based
IK for upper-extremity reaching motions. Wang argued that optimization-based methods
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
68 Reviews of Human Factors and Ergonomics, Volume 4

Figure 2.17. A top-down motion framework developed by Reed, Faraway, Chaffin, and
Martin (2006).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 69

alone become too computationally inefficient and do not take advantage of the empiri-
cal knowledge that exists regarding various ways people perform complex motions.
Perez (2005) also concluded, after comparing his empirical ANN model results for
various whole-body motions with optimization-based methods, that a top-down frame-
work that would combine empirically based modules of simple motions into a heuristic
motion prediction system would be useful. The motion framework approach by Reed
et al. (2006) is a highly formal attempt to provide such a method.

INTEGRATION OF COGNITIVE AND PHYSICAL HUMAN


PREDICTION MODELS
As stated at the beginning of this chapter, it is not my intention to discuss the state of cog-
nitive models, sometimes referred to as human performance process models or HPPMs.
A set of references for this purpose is given near the beginning of the chapter. What is im-
portant is to acknowledge that the next big advances in human performance modeling
will require the integration of these HPPMs with the digital physical human models de-
scribed in this chapter. This integration is so important that it has become one of the
motivating factors in the establishment of a new technical group in the Human Factors
and Ergonomics Society, the Human Performance Modeling Technical Group. This group
recognizes that the designed environment affects both the cognitive and physical states
of a person and that to understand this phenomenon, one must use models that draw on
both types of knowledge.
Such integration is in its infancy, but an example from Tsimhoni and Reed (2007) is
provided to indicate the potential benefits from such work. As was described earlier, the
motion framework by Reed et al. (2006) combines various motion modules to simulate
how people might respond to well-specified manual task requirements (e.g., reach to an
object at Location A or look at the driver’s side mirror on a vehicle). Clearly, the person
using the framework must direct or choreograph task sequences such as these, and this
may require more knowledge about human behaviors than most engineers and designers
possess (Chaffin, 2007). For instance, instructing an avatar to look at the driver’s side mir-
ror while driving is simple enough, but how often should such a head-eye motion be
made, and what would be the reaction of the avatar if an approaching car in the next lane
suddenly appeared as the driver was attempting to merge into traffic? Such a scenario is
exactly what Tsimhoni and Reed have in mind as they combine the motion framework
with the queuing network–model human processor (QN-MHP) cognitive model devel-
oped by Liu et al. (2006). The logic used in the QN-MHP model is shown in Figure 2.18.
The QN-MHP model treats the human mind as a network of information servers, each
having a prescribed distribution of the service times required to process various streams
of information packets flowing through the network. Structuring human information
processing in this fashion, Liu et al. (2006) have shown that such a model is capable of
predicting the time required to make complex decisions involving multiple mental tasks,
as well as predicting the expected mental workload to process various amounts of in-
formation in a given time. By combining the QN-MHP model with the motion frame-
work, Tsimhoni and Reed (2007) have been able to predict how drivers would react to
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
70 Reviews of Human Factors and Ergonomics, Volume 4

Figure 2.18. The queuing network–model human processor (QN-MHP) developed by Liu,
Feyen, and Tsimhoni (2006) to predict people’s multitasking behaviors.

high-information-loading tasks and have demonstrated the expected physical behaviors


(e.g., steering the vehicle, reaching to adjust a control surface, gazing at the mirror).
In the future, such an advanced performance model will be capable of not just demon-
strating the physical consequences of high-information-loading tasks but also aiding an
analyst concerned with the cognitive consequences that might be associated with a phys-
ical feature in a vehicle or industrial process (e.g., larger displays, magnifying mirrors,
heads-up night vision displays, stalk- or wheel-mounted controls, audio warning systems,
tactile warning systems). Clearly, the integration of cognitive and physical human
models is a major undertaking. Much remains to be studied and developed in this inte-
grative work, but the potential benefits from these types of human simulation models
truly are unbounded.

SUMMARY

It has been more than 20 years since the 1985 U.S. National Research Council workshop
attendees suggested that some basic advances in human modeling would be needed to
provide effective ergonomics assistance in the design of various workspaces. In this chap-
ter, I have attempted to illustrate some of the advances that have been accomplished.
Specifically, anthropometric data from many sources can be accessed and used easily with
sophisticated statistical methods to ensure that a specified percentile of the population
is accommodated in human reach analyses, sight-line determinations, and human fit and
clearance simulations. The performance effects of restrictive clothing, however, are still
being studied empirically (Uppu, Aghazadeh, & Nabatilan, 2006).
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 71

Biomechanical models of exertions have provided the basis for some predictions of
human static strength and low back stresses. These models are often integrated with high-
fidelity graphic representations of human forms, or avatars, to assist designers who are
concerned with the adverse effects of high-exertion tasks. Yet much remains to be done
to improve the fundamental biomechanical models that are being used in DHM simu-
lations, especially those related to dynamic task simulations (Marras & Radwin, 2006).
Where biomechanical models are limited, psychophysical discomfort ratings have been
used in some DHMs to provide population limits to specific manual tasks.
Predicting lifelike human motions for a large variety of tasks has been a major goal
of many DHM developers. Sophisticated motion optimization methods, control system
models, and functional regression methods are being used, along with very large motion
databases, to derive the best means of accomplishing this goal. Major progress has been
made in predicting realistic movement characteristics of healthy individuals. Yet the mo-
tions associated with tasks that require persons to brace themselves, reach or look into
confined areas, or exert unusual forces are not well modeled. More specifically, the com-
pensatory type of motions often used by people with back or other musculoskeletal im-
pairments or by people working close to their physical capacities are not predicted well
(Chaffin, Woolley, Dickerson, & Parkinson, 2004; Larivière, Gagnon, & Loisel, 2000;
Lindbeck & Kjellberg, 2001).
One of the most difficult aspects of human motion prediction is to incorporate a
means to predict the elegant ways in which people carefully coordinate the movements of
various body members when performing a specific task. Perhaps no better example of this
is when attempting to simulate the human hand as it grasps objects of various sizes,
shapes, and textures. Each of these object attributes, plus the intended goal of the grasp-
ing motion, affects the positioning of individual fingers. Armstrong (under review) de-
scribes various attempts to model grasping postures.
The good news is that several research and software development groups around the
world are fully engaged in providing digital human models that will enable users to sim-
ulate a large variety of human endeavors. Indeed, it is a very exciting time for those in-
volved in this evolving technology. As shown by the seven case studies comprising the text
Digital Human Modeling for Vehicle and Workplace Design (Chaffin, 2001), this technol-
ogy is already providing answers to difficult questions about human capabilities that are
posed by engineers and designers of aircraft, automobiles, spacecraft, ships, and work-
places. It is hoped that this review will help many readers realize the potential benefits and
challenges in developing and using DHM programs, and encourage additional use and
contributions to the technology in the near future.

ACKNOWLEDGMENTS

I thank the sponsors of the HUMOSIM Laboratory at the University of Michigan, par-
ticularly the U.S. Army–TARDEC group, Ford Motor Corporation, General Motors
Corporation, International Truck Corporation, UGS Corporation, and the U.S. Postal
Service, which for 9 years have supported some of the work discussed in this chapter.
Finally, I thank Patricia Terrell for assisting in the preparation of this manuscript.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
72 Reviews of Human Factors and Ergonomics, Volume 4

REFERENCES
Abdel-Malek, K., Yang, J., Marler, T., Beck, S., Mathai, A., Zhou, X., et al. (2006). Towards a new generation of
virtual humans. International Journal of Human Factors Modelling and Simulation, 1(1), 1–39.
Allard, P., Cappozzo, A., Lundberg, A., & Vaughan, C. (Eds.). (1998). Three-dimensional analysis of human loco-
motion. West Sussex, UK: Wiley.
Armstrong, T. J. (under review). Development of hand models for ergonomic applications. In V. Duffy (Ed.),
Handbook of digital human models. London: Taylor & Francis.
Badler, N. I., Phillips, C. B., & Webber, B. L. (1993). Simulating humans: Computer graphics animation and con-
trol. New York: Oxford University Press.
Baron, S., Kruser, D. S., & Huey, B. M. (1990). Quantitative modeling of human performance in complex, dynamic
systems. Washington, DC: National Academy Press.
Chaffin, D. B. (1969). A computerized biomechanical model: Development and use in studying gross body
actions. Journal of Biomechanics, 2, 429–441.
Chaffin, D. B. (Ed.). (2001). Digital human modeling for vehicle and workplace design. Warrendale, PA: Society of
Automotive Engineers.
Chaffin, C. (2007). Human motion simulation for vehicle and workplace design. Human Factors and Ergo-
nomics in Manufacturing, 17, 475–484.
Chaffin, D. B., & Andersson, G. B. J. (1984). Occupational biomechanics. New York: Wiley.
Chaffin, D. B., Andersson, G. B. J., & Martin, B. J. (2006). Occupational biomechanics (4th ed.). Hoboken, NJ:
Wiley.
Chaffin, D. B., & Erig, M. (1989). 3-Dimensional biomechanical static strength prediction model sensitivity to
postural and anthropometric inaccuracies. IIE Transactions, 23, 215–227.
Chaffin, D. B., Faraway, J. J., Zhang, X., & Woolley, C. (2000). Stature, age, and gender effects on reach motion
postures. Human Factors, 42, 408–420.
Chaffin, D. B., Kilpatrick, K. E., & Hancock, W. M. (1970). A computer-assisted manual work design model. IIE
Transactions, 2, 347–354.
Chaffin, D. B., Schutz, R. K., & Snyder, R. G. (1972). A prediction model of human volitional mobility (SAE Paper
720002). Detroit, MI: Society of Automotive Engineering.
Chaffin, D. B., Woolley, C., Dickerson, C., & Parkinson, M. (2004). Modeling of object movement capability in
the spinal cord injured population. International Journal of Industrial Ergonomics, 33, 229–236.
Dai, J., Teng, Y., & Oriet, L. (2003). Application of multiparameter and boundary mannequin techniques in auto-
motive assembly process. In Proceedings of the Digital Human Modeling Conference (CD-ROM). Warrendale,
PA: Society of Automotive Engineers.
Damsgaard, M., Rasmussen, J., Christensen, S. T., Surma, E., & de Zee, M. (2006). Analysis of musculoskeletal
systems in the AnyBody modeling system. Simulation Modelling Practice and Theory, 14, 1100–1111.
Dickerson, C. R., Martin, B. J., & Chaffin, D. B. (2007). Predictors of perceived effort in the shoulder during load
transfer tasks. Ergonomics, 50, 1004–1016.
Dysart, M. J., & Woldstad, J. C. (1996). Posture prediction for static sagittal-plane lifting. Journal of Biome-
chanics, 29, 1393–1397.
Erdemir, A., McLean, S., Herzog, W., & van den Bogert, A. (2007). Model-based estimation of muscle forces
exerted during movements. Clinical Biomechanics, 22, 131–154.
Faraway, J. J. (1997). Regression analysis for a functional response. Technometrics, 39, 254–261.
Faraway, J. J., Zhang, X., & Chaffin, D. B. (1999). Rectifying postures reconstructed from joint angles to meet
constraints. Journal of Biomechanics, 32, 733–736.
Fisher, D., Schweickert, R., & Drury, C. (2006). Mathematical models in engineering psychology: Optimizing
performance. In G. Salvendy (Ed.), Handbook of human factors and ergonomics (3rd ed., pp. 997–1024).
New York: Wiley.
Gleicher, M. (1997). Motion editing with spacetime constraints. In Proceedings of 1997 Symposium on Interactive
3D Graphics. New York: ACM Press.
Gluck, K. A., & Pew, R. W. (Eds.). (2005). Modeling human behavior with integrated cognitive architectures:
Comparison, evaluation. Mahwah, NJ: Erlbaum.
Greil, H., & Jurgens, H. W. (2000). Variability of dimensions and proportions in adults or how to use classic
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
Digital Human Modeling for Workspace Design 73

anthropometry in man modeling. In K. Landau (Ed.), Ergonomic software tools in product and production
design: A review of recent developments in human modeling and other design aids (pp. 309–327). Stuttgart,
Germany: Ergon Verlag.
Hatze, H. (2002). The fundamental problem of myoskeletal inverse dynamics and its implications. Journal of
Biomechanics, 35, 109–115.
Hughes, R. E. (2000). Effect of optimization criterion on spinal force estimation during asymmetric lifting.
Journal of Biomechanics, 33, 225–229.
Ianni, J. D. (1999). A specification for human action representation. In Proceedings of the Digital Human
Modeling Design and Engineering International Conference and Exposition (paper no. 1893). Warrendale,
PA: Society of Automotive Engineers.
Jagacinski, R. J., & Flach, J. M. (Eds.). (2003). Control theory for humans: Quantitative approaches to modeling
performance. Mahwah, NJ: Erlbaum.
Jung, E. S., Kee, D., & Chung, M. K. (1995). Upper body reach posture prediction for ergonomic evaluation
models. International Journal of Industrial Ergonomics, 16, 95–107.
Karwowski, W., Genaidy, A. M., Shihab, S., & Asfour, S. S. (Eds.). (1990). Computer-aided ergonomics. London:
Taylor & Francis.
Kilpatrick, K. E. (1970). Computer aided workplace design. Journal of Methods–Time Measurement, 14(4), 24–33.
Kim, K. H. (2005). Modeling of head and hand coordination in unconstrained three-dimensional movements.
Unpublished doctoral dissertation, University of Michigan, Ann Arbor.
Kim, K. H., Martin, B. J., & Chaffin, D. B. (2004). Modeling of shoulder and torso perception of effort in man-
ual transfer tasks. Ergonomics, 47, 927–944.
Kroemer, K. H. E., Snook, S. H., Meadows, S. K., & Deutsch, S. (Eds.). (1988). Ergonomic models of anthropom-
etry, human biomechanics and operator-equipment interfaces. Washington, DC: National Academy of Sciences.
Kuo, A. D. (2002). The relative roles of feedforward and feedback in the control of rhythmic movements. Motor
Control, 6(2), 129–145.
Larivière, C., Gagnon, D., & Loisel, P. (2000). Effect of load on the coordination of the trunk for subjects with
and without chronic low back pain during flexion-extension and lateral bending tasks. Clinical Biome-
chanics, 15, 407–416.
Lin, C. J., Ayoub, M. M., & Bernard, T. M. (1999). Computer motion simulation for sagittal plane lifting activ-
ities. International Journal of Industrial Ergonomics, 24, 141–155.
Lindbeck, L., & Kjellberg, K. (2001). Gender differences in lifting technique. Ergonomics, 44, 202–214.
Liu, Y., Feyen, R., & Tsimhoni, O. (2006). Queueing network-model human processor (QN-MHP): A compu-
tational architecture for multitask performance in human-machine systems. ACM Transactions on Computer-
Human Interaction, 13, 37–70.
Marras, W. S., & Granata, K. P. (1995). A biomechanical assessment and model of axial twisting in the thora-
columbar spine. Spine, 20, 1440–1451.
Marras, W. S., & Radwin, R. G. (2006). Biomechanical modeling. In R. S. Nickerson (Ed.), Reviews of human
factors and ergonomics (Vol. 1, pp. 1–88). Santa Monica, CA: Human Factors and Ergonomics Society.
McDaniel, J. W. (1998, April). Human modeling: Yesterday, today, and tomorrow. Keynote address at the SAE
International Conference, Digital Human Modeling for Design and Engineering, Dayton, OH.
McGill, S. M. (1992). A myoelectrically based dynamic three-dimensional model to predict loads on lumbar
spine tissues during lateral bending. Journal of Biomechanics, 25, 395–414.
McGill, S. M., Norman, R. W., & Cholewicki, J. (1996). A simple polynomial that predicts low-back compres-
sion during complex 3-D tasks. Ergonomics, 39, 1107–1118.
Menegaldo, L. L., Fleury, A., T. & Weber, H. I. (2003). Biomechanical modeling and optimal control of human
posture. Journal of Biomechanics, 36, 1701–1712.
Nigg, B. M., MacIntosh, B. R., & Mester, J. (Eds.). (2000). Biomechanics and biology of movement. Champaign,
IL: Human Kinetics.
Nuzzo, R. (2007, Winter). Computational biomechanics. Biomedical Computation Review, pp. 10–19.
Park, W., Chaffin, D. B., & Martin, B. J. (2004). Toward memory-based human motion simulation: Development
and validation of a motion modification algorithm. IEEE Transactions on Systems, Man, and Cybernetics—
Part A: Systems and Humans, 34, 376–386.
Park, W., Martin, B. J., Choe, S., Chaffin, D. B., & Reed, M. P. (2005). Representing and identifying alternative
movement techniques for goal-directed manual tasks. Journal of Biomechanics, 38, 519–527.
Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013
74 Reviews of Human Factors and Ergonomics, Volume 4

Park, W., Singh, D., & Martin, B. J. (2006). A memory-based model for planning target reach postures in the
presence of obstructions. Ergonomics, 49, 1565–1580.
Parkinson, M. B., Chaffin, D. B., & Reed, M. P. (2006). Center of pressure excursion capability in performance
of seated lateral-reaching tasks. Clinical Biomechanics, 21, 26–32.
Peacock, B., & Karwowski, W. (Eds.). (1993). Automotive ergonomics. Bristol, PA: Taylor & Francis.
Perez, M. A. (2005). Prediction of whole-body lifting kinematics using artificial neural networks. Unpublished
doctoral dissertation, Virginia Polytechnic Institute and State University, Blacksburg.
Polk, T. A., & Seifert C. M. (Eds.). (2002). Cognitive modeling. Cambridge: MIT Press.
Porter, J. M., Case, K., Marshall, R., Gyi, D., & Sims, R. (2004). “Beyond Jack and Jill”: Designing for individu-
als using HADRIAN. International Journal of Industrial Ergonomics, 33, 249–264.
Porter, J. M., Marshall, R., Freer, M., & Case, K. (2004). SAMMIE: A computer-aided ergonomics design tool.
In N. Delleman, C. Haslegrave, & D. Chaffin (Eds.), Working postures and movements: Tools for evaluation
and engineering (pp. 454–462). Boca Raton, FL: CRC Press.
Raschke, U., Schutte, L., & Chaffin, D. B. (2001). Simulating humans: Ergonomic analysis in digital environments.
In G. Salvendy (Ed.), Handbook of industrial engineering (pp. 1110–1130). New York: Wiley.
Reed, M. P., Faraway, J., Chaffin, D. B., & Martin, B. J. (2006). The HUMOSIM ergonomics framework: A new
approach to digital human simulation for ergonomics analysis (SAE Technical Paper Series 2006-01-2365).
Warrendale, PA: Society of Automotive Engineers.
Reed, M. P., Parkinson, M. B., & Chaffin, D. B. (2003). A new approach to modeling driver reach. SAE Transac-
tions: Journal of Passenger Cars—Mechanical Systems, 112, 709–718.
Ryan, M. P., & Springer, W. E. (1969). Cockpit geometry evaluation final report (JANAIR Rep. 69105, Vol. V).
Washington, DC: Office of Naval Research.
Schmidt, R. A., & Lee, T. D. (1999). Motor control and learning: A behavioral emphasis. Champaign, IL: Human
Kinetics.
Seidl, A. (2004). The RAMSIS and ANTHROPOS human simulation tools. In N. Delleman, C. Haslegrave, &
D. Chaffin (Eds.), Working postures and movements: Tools for evaluation and engineering (pp. 445–454). Boca
Raton, FL: CRC Press.
Snyder, R. G., Chaffin, D. B., & Schutz, R. K. (1972). Link system of the human torso (HSRI Rep. 71-112). Ann
Arbor: Highway Safety Research Institute, University of Michigan.
Sundin, A., & Örtengren, R. (2006). Digital human modeling for CAE applications. In G. Salvendy (Ed.), Hand-
book of human factors and ergonomics (3rd ed., pp. 997–1024). New York: Wiley.
Tsimhoni, O., & Reed, M. P. (2007). The virtual driver: Integrating task planning and cognitive simulation with
human movement models (SAE Technical Paper Series 2007-01-1766). Warrendale, PA: Society of Auto-
motive Engineers.
Uppu, N., Aghazadeh, F., & Nabatilan, L. (2006). Effect of pressure suit on functional reach. Occupational
Ergonomics, 6, 129–142.
Wada, Y., & Kawato, M. (1993). A neural network model for arm trajectory formation using forward and inverse
dynamics models. Neural Networks, 6, 919–932.
Wagner, D. W., Reed, M. P., & Chaffin, D. B. (2005, June). Predicting foot positions for manual materials handling
tasks. Paper presented at the SAE Digital Human Modeling for Design and Engineering Conference, Iowa
City, IA.
Wang, X. (1999). A behavior-based inverse kinematics algorithm to predict arm prehension postures for
computer-aided ergonomics evaluation. Journal of Biomechanics, 32, 453–460.
Zajac, F. E. (1993). Muscle coordination of movement: A perspective. Journal of Biomechanics, 26(Suppl. 1),
109–124.
Zajac, F. E., & Winters, J. M. (1990). Modeling musculoskeletal movement systems: Joint and body segmental
dynamics, musculoskeletal actuation, and neuromuscular control. In J. M. Winters & S. L.-Y. Woo (Eds.),
Multiple muscle systems: Biomechanics and movement organization (pp. 121–148). New York: Springer-Verlag.
Zatsiorsky, V. M. (1998). Kinematics of human motion. Champaign, IL: Human Kinetics.
Zhang, X. (1997). Development of a three-dimensional dynamic posture prediction model for seated operator motion
simulation. Unpublished doctoral dissertation, University of Michigan, Ann Arbor.
Zhang, X., & Chaffin, D. (2000). A three-dimensional dynamic posture prediction model for simulating in-
vehicle seated reaching movements: Development and validation. Ergonomics, 43, 1314–1330.

Downloaded from rev.sagepub.com at UNIV OF MICHIGAN on June 24, 2013

You might also like