Transportation Research Part C 125 (2021) 103047

Contents lists available at ScienceDirect

Transportation Research Part C

journal homepage: www.elsevier.com/locate/trc

OpenACC. An open database of car-following experiments to study

the properties of commercial ACC systems
Michail Makridis b, a, Konstantinos Mattas a, Aikaterini Anesiadou a, Biagio Ciuffo a, *
a
European Commission – Joint Research Centre (JRC), Ispra, VA, Italy
b
ETH Zürich, Institute for Transport Planning and Systems (IVT), Zürich, Switzerland

A R T I C L E I N F O A B S T R A C T

Keywords: Adaptive Cruise Control (ACC) systems are becoming increasingly available as a standard
Open data equipment in modern commercial vehicles. Their penetration rate in the fleet is constantly
Adaptive cruise control increasing, as well as their use, especially under freeway conditions. At the same time, limited
Empirical observations
information is openly available on how these systems actually operate and their differences
Car-following
Traffic flow
depending on the vehicle manufacturer or model. This represents an important gap because as the
Driver behavior number of ACC vehicles on the road increases, traffic dynamics on freeways may change
Vehicle dynamics accordingly, and new collective phenomena, which are only marginally known at present, could
Microsimulation emerge. Yet, as ACC systems are introduced as comfort options and their operation is entirely
under the responsibility of the driver, vehicle manufacturers do not have explicit requirements to
fulfill nor they have to provide any evidence about their performances. As a result, any safety
implication connected to their interactions with other road users escapes any monitoring and
opportunity of improvement.
This work presents a set of experimental car-following campaigns, providing an overview of the
behavior of commercial ACC systems under different driving conditions. Furthermore, the sug­
gestion of a unified data structure across the different tests facilitates comparison between the
different campaigns, vehicles, systems and specifications. The complete data is published as an
open-access database (OpenACC), available to the research community. As more test campaigns
will be carried out, OpenACC will evolve accordingly.
The activity is performed in the framework of the openData policy of the European Commission
Joint Research Centre with the objective to engage the whole scientific community towards a
better understanding of the properties of ACC vehicles in view of anticipating their possible
impacts on traffic flow and prevent possible problems connected to their widespread introduc­
tion. In this light, OpenACC, over time, also aims at becoming a reference point to study if and
how the parameters of such systems need to be regulated, how homogeneously they behave, how
new ACC car-following models should be designed for traffic microsimulation purposes and what
are the key differences between ACC systems and human drivers.

1. Introduction

Advanced in-vehicle technologies promise to disrupt road transportation as it is known today bringing benefits in several

* Corresponding author.
E-mail address: biagio.ciuffo@ec.europa.eu (B. Ciuffo).

https://doi.org/10.1016/j.trc.2021.103047
Received 11 April 2020; Received in revised form 14 February 2021; Accepted 14 February 2021
Available online 22 February 2021
0968-090X/© 2021 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

dimensions. More specifically, vehicle automation and connectivity promise to assist towards reducing traffic congestion, reducing
travel times, pollutant emissions and improving traffic safety and comfort. The automotive industry has made significant investments
on these technologies with the objective to compete on a potentially multi-billion market (European Commission, 2019). Even if full
vehicle automation still requires several years before mass deployment, partial levels of automation are already available in the
market. The Society of Automotive Engineers (SAE) has proposed a classification based on 5 vehicle automation levels (SAE Inter­
national, 2018). Many existing vehicles are classified as level 1 or 2 as they are able, under the constant supervision of the driver, to
automatically keep longitudinal and/or lateral control of the vehicle. The system in charge of maintaining the longitudinal vehicle
control is called Adaptive Cruise Control (ACC) and is one of the most widespread driving assistance systems available in the market.
More than 20 years have passed since the first introduction of ACC-equipped vehicles in the market (Xiao and Gao, 2010). Despite
the research efforts, the operational design of these systems is still almost a black box to the research community, as it is usually
protected by intellectual property rights and very few attempts have been made by researchers to reverse engineer it. A conceptual
review of the different ACC control strategies can be found in the work of He et al. (2019). The limited number of experimental test
campaigns involving ACC vehicles makes it also difficult to understand the level of similarity in the implementation logic used by
different manufacturers or available in different vehicle models.
Experimental campaigns are costly in terms of budget, organization and human resources, which poses difficulties in systematically
studying in-vehicle technological advances. Although an exhaustive research with empirical observations on all available vehicle
models and their ACC systems would be unfeasible, the development of a database collecting in a systematic and structured way the
results of experimental campaigns carried out over time by even different research teams can lead to important findings on how these
systems operate and how the control logic evolves over time. Such database can provide valuable insights on how automated driving
systems create new driving patterns, different from those of human drivers and what are the best ways for modeling and assess their
impact on road transport networks. The Joint Research Centre (JRC) of the European Commission, in line with its policy of providing
open access to its research data, aims at contributing towards this broader understanding idea.
In the literature, the properties of commercial ACC systems and their impact on various dimensions is discussed through simulation
studies (Hu et al., 2019; Li and Wagner, 2019; Liu et al., 2018; Makridis et al., 2020c; Mattas et al., 2018; Olia et al., 2018; Shladover
et al., 2012; Sun et al., 2018; Wang et al., 2018b; Ye and Yamamoto, 2018; Yu et al., 2018). However, simulation results are influenced
by a large number of known factors (assumptions to build the model, models’ design, precision and accuracy, calibration process,
limited or noisy input data etc.). Field tests and systematic study of the vehicle dynamics can certainly help towards more realistic
simulation studies and more reliable impact assessment conclusions.
On the other side, experimental campaigns are considered more reliable for drawing conclusions, but they demand more resources,
and variations in the commercially available systems leave small space for generalizations. However, progress on the accuracy of data
acquisition systems and computing provides also new opportunities for study and assessment. Recently, there is an increasing number
of publications with empirical observations in the literature, which shows the importance of having real world data in order to observe,
understand and model with improved precision traffic phenomena, advanced in-vehicle technologies and variation in the driving
behaviors. The experimental campaigns can be categorized in two main clusters; high-level campaigns that focus on the supervision of
a section or area in order to observe traffic and lower-level car-following campaigns that zoom into the individual vehicle dynamics.
In the first category, NGSIM (“NGSIM,” 2006), although developed more than a decade ago, is still probably the most well-known
database of traffic data and it has been extensively used for traffic simulation research. NGSIM dataset is a collection of real-world
trajectories, based on the use of cameras mounted on tall buildings and covering approximately 0.5–1 km long roadway sections
with a frequency of 10 Hz in several US locations. Although there are studies in the literature proposing post-processing methodologies
in order to cope with measurement inaccuracies and get the maximum out of this real-world tests (Punzo et al., 2011; Montanino and
Punzo, 2015; Coifman and Li, 2017), the superiority of modern, more precise data acquisition systems and the different patterns arising
from novel in-vehicle technologies, demand new, updated datasets and they leave little room for further exploitation of the NGSIM.
Recently, Barmpounakis and Geroliminis conducted the pNEUMA large-scale field experiment (Barmpounakis and Geroliminis, 2020)
aiming to record traffic streams in a multi-modal congested environment over an urban area using Unmanned Aerial Systems. The
dataset was generated by a swarm of 10 drones hovering over a traffic intensive area of 1.3 km2 in the city center of Athens, Greece,
covering more than 100 km-lanes of road network at 25 Hz. This initiative allows the deep investigation of critical traffic phenomena in
urban areas as never has been possible in the past. Similarly but on a smaller scale, Krajewski et al. (2018) created the highD (highway
drone dataset) with detailed vehicle trajectories over a road segment of around 420 m, from a drone hovering next to German highways
at 25 Hz. The same team produced two additional datasets one on an intersection and another on a roundabout. All these datasets can
be particularly useful for studying traffic related phenomena, but they do not provide technical specifications for the vehicles involved
in the tests, nor they can identify the advanced driving assistance and automation systems involved in the observations.
In the second category, there are several studies with small-scale car-following experiments providing insights on in-vehicle
technologies and the driving behaviors in the longitudinal direction. Milanés and Shladover (2014) carried out experimental tests
using up to four production Infiniti M56s provided by Nissan. The vehicles were equipped with a commercial ACC system that uses a
LIDAR for detecting the preceding vehicle. The aim of this work was to compare the performance of three different longitudinal
controllers, a production ACC, the IDM model and a Cooperative-ACC system. Furthermore, Knoop et al. (2019) conducted an
experiment with seven SAE level-2 vehicles driven in a car-platoon formation. The vehicles were driven on public roads in the
Netherlands for a trip of almost 500 km. The authors discuss the observed instability in the car-platoon, when all vehicles have ACC
activated. Severe variations in the speed were also observed, leading in cases to discomfort and even risks of rear-end collisions.
Furthermore, in Gunter et al. (2020), the authors assessed the string stability of seven 2018 model year ACC-equipped vehicles from
two makes. On another field test, Jiang et al. (2015) carried out car-following experiments with a 25-car-platoon on an open road

2
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

section, giving insights on the relation between a car’s speed and its spacing under various traffic conditions. Finally, there is a couple
of small scale car-following field tests, a campaign with 6 vehicles from Laval et al. (2014) and another with 5 vehicles from Punzo and
Simonelli (2005). Table 1 lists the most relevant datasets found in the literature and their characteristics, including the proposed
OpenACC database.
Limited publicly available empirical trajectory data is an important issue mentioned in bibliography hindering researchers from
evaluating and validating different methods and tools. Li et al. (2020), devoted a whole discussion section in their study regarding the
need for more empirical trajectory data in terms of quantity and diversity so that traffic flow phenomena and also other related topics
such as emissions estimation can be investigated. As other researchers argue, the limited amount of empirical data is an obstacle on
investigating traffic flow phenomena, and on calibrating or training traffic flow models (Li and Chen, 2017; Li et al., 2016; Wang et al.,
2018a; Wang et al., 2019a; Wang et al., 2017; Wang et al., 2019b, 2018c).
Aiming to contribute on the above research problems and in line with the open data policy of the European Commission, the Joint
Research Centre has decided to make publicly available OpenACC, the complete database resulting from the experimental campaigns
carried out in the past years and involving several commercial ACC systems. The objective of the present paper is to introduce the main
characteristics of OpenACC, including a description of the test-campaigns and the main challenges faced during their execution. In
addition, the paper proposes different types of analyses that researchers can perform with this type of data and provide preliminary
insights on the characteristics of commercially available ACC vehicles and the properties of a traffic stream composed by them. As new
test campaigns will be carried out and new data generated, OpenACC will be expanded accordingly aiming at becoming a reference
point for traffic research related to driving assistance and automation systems.
Before proceeding further with the description of the database, it is worth underlying that the present work does not aim to provide
an exhaustive comparison between the different commercially available Adaptive Cruise Control systems, vehicle models and/or
vehicle brands. It should be noted that the analysis of the response of a driving system subject to well-defined stimuli from the vehicle
in front is not necessarily comparable with the response of the same systems inside a platoon, as the control strategy of ACC vehicles
may act differently in the two conditions. The main objective of the paper is to capture the collective behavior of ACC systems when

Table 1
Summary of important experimental datasets currently available in the literature.
Dataset NGSIM pNEUMA highD inD rounD Milanes and Shladover

Year 2006 2020 2019 2020 2020 2014

Research focus Traffic Traffic Traffic Traffic Traffic Platoon
2 vehicles of the same
brand model
Network Highway Urban Highway Intersection Roundabout Highway
Data acquisition system Camera Drone Drone Drone Drone Differential GPS
Resolution 10hz 25hz 25hz 25hz 25hz 10hz
Area of observation 500 m 1.3 km2 area 400 m – – –
Measurements: Speed/ Yes Yes Yes Yes Yes –
Position/Altitude No Yes No No No
No No No No No
Vehicle information No Class of vehicles No No No Yes
ACC No No No No No Yes
Available online Yes No1 Yes2 Yes3 Yes4 No
Dataset Knoop et al. Gunter et al. Jiang Laval et al. Punzo and OpenACC
et al. Simonelli
Year 2019 2018 2015 2014 2005 2020
Research focus Platoon Platoon Platoon Platoon Platoon Platoon
7 vehicles, 4 brands, 7 vehicles, 2 brands, 25 6 vehicles 4 vehicles 28 vehicles, 18 brands,
4 models 7 models vehicles 27 models
Network Highway Highway Highway – – Highway/Urban
Data acquisition system CAN and GPS Ublox5 GPS GPS GPS Ublox6 and Differential
GPS
Resolution 10hz 10hz 10hz 5hz 10hz 10hz
Area of observation 465 km route 1200miles route 3.2 km – – >3000 km
Measurements: Speed/ – No Yes Yes Yes Yes
Position/Altitude Yes Yes No No Yes
No No Yes No Yes
Vehicle information Yes Yes No No Yes Yes
ACC Yes Yes No No No Yes
Available online No Yes6 No No No Yes7
1
Sample available at https://open-traffic.epfl.ch/. It will become fully available in the future.
2
Upon request on https://www.highd-dataset.com/#download.
3
Upon request on https://www.ind-dataset.com/#download.
4
Upon request on https://www.round-dataset.com/#download.
5
https://www.u-blox.com/en.
6
https://vanderbilt.box.com/v/accData.
7
https://data.jrc.ec.europa.eu/dataset/9702c950-c80f-4d2f-982f-44d06ea0009f.

3
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

combined in platoons and anticipate their impact on road traffic when their penetration in the fleet will increase. At the same time, the
data collected can also provide useful insights into the characteristic of commercial ACC systems.
The rest of the paper is organized as follows: Section 2 presents the characteristics of the OpenACC database. Section 3 provides a
brief description of each test campaign and the main specifications of the vehicles involved in the experiments. Furthermore, the data
acquisition and post-processing workflow is discussed along with a brief description of the OpenACC database structure. Section 4
briefly introduces the first findings after an initial processing of the dataset, some of which confirm findings already presented in the
current literature, but also others that provide new insights to the topic. The last section concludes the paper, discussing the plans for
further exploitation of this database in the future. A dictionary of abbreviations can be found in Appendix A.

2. The OpenACC database

The OpenACC database (Ciuffo et al., 2020), is an open-access database of different car-following experiments involving 28 ve­
hicles, 22 of which equipped with state-of-the-art commercial ACC systems. Experiments are carried out in the framework of four test
campaigns. The campaigns have been designed to study among others, vehicle dynamics in real world conditions, behavior of ACC
systems and car-following patterns. As already mentioned, as more test-campaigns will be carried out and the related data extracted,
OpenACC will also evolve and grow.
Fig. 1 presents a schematic overview for the characteristics of the OpenACC database that, combined, provide a unique repository.
The availability of data with different acquisition devices and thus different error levels helps in assessing the quality of the observed
data for a specific problem under study. Altitude information is important for instantaneous emissions and energy demand estimations
and poses an additional force on the vehicle, creating variation in the output of the ACC controller. Experiments with different engine
types are important for studying at least the transition period, when automated vehicles will have to co-exist with traditional onesss on
public roads.
Two out of the four test campaigns have been carried out on public roads, while the other two on protected test tracks. The first test
was organized with two and three vehicles in car-platoon formation. The personal experience of discomfort for the passengers of the
following vehicles when the ACC was enabled and reacting to the speed perturbations of the preceding vehicles, was the motivation of
more challenging field tests. The second real-world campaign was thus organized with five vehicles in car-platoon formation. Data
from the second experiment revealed very interesting findings on basic properties of the ACC controllers such as their relatively large
response time and time headway. Moreover, they question the systems’ ability to ensure string stability within the car-platoon. As in
the previous case, the campaign was carried out on a public motorway. The difficulty to maintain the platoon formation for long
periods of time due to the road configuration and the interaction with other road users, and the safety implications of open-road tests
suggested the need to perform additional campaigns in closed proving grounds. The third campaign was therefore conducted in the
AstaZero proving ground in Sweden, with a car-platoon of five vehicles in order to provide insights on the above issues in a more
systematic way and also with a more precise data acquisition system. Finally, the last experiment took place in two different test-tracks

Fig. 1. A schematic overview for the main characteristics of the OpenACC dataset.

4
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

at the ZalaZone proving ground in Hungary. In this campaign, car platoons of 10 commercially available ACC-equipped vehicles were
tested in different vehicle orders and settings of the ACC systems. A first analysis reveals important and novel observations concerning
the characteristics of commercial ACC systems and their differences with human driving patterns. Simultaneous observation of car-
platoons shows the interaction of multiple ACC systems on road and the impact that mass deployment of such vehicles can have on

Fig. 2. Path layout for the four test campaigns included in the first version of the OpenACC database.

5
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

public networks in the future. The current version of the OpenACC dataset is publicly available from the JRC Data Repository8.

3. Experimental campaigns

The aim of this extended database is to provide novel insights on the behavior of commercially available ACC systems, the dif­
ferences in the observed longitudinal vehicle dynamics between controllers and human drivers, the car-following interactions between
human drivers and ACC systems and the ability of highly automated vehicles to coexist and cooperate with other users on the road.
Although, it is not feasible to capture completely the stochasticity that exists on public roads, as new experiments will be organized,
more important characteristics on the above dimensions will be revealed. This first version of the OpenACC database includes indeed
the results of four distinct car-following campaigns and the authors’ aim is to populate it in the future with additional observations.
Fig. 2 illustrates the path layout of the four campaigns, whereas the basic specifications of the vehicles involved in the experiments are
given in Table 2. Vehicles with prefix “(L)” were used as leaders in the car-platoon tests. It should be noted that the Smart BME ADdv
vehicle, which was used as a leader vehicle in the last experimental campaign, is a programmable vehicle, without a driver, moving
according to a predefined path and therefore, no specifications are given here. Budapest University of Technology and Economics
(BME) developed the BME Automated Drive demonstrator vehicle (BME ADdv). This is a first-generation Smart ForTwo W450 and it
was developed for validation, testing and demonstration purposes of different automated functions. Finally, Table 3 presents general
information of the four campaigns.

• Experimental Campaign N.1 – Driving from Ispra (VA) to Cherasco (CO)

The first campaign was conducted in the last quarter of 2018 and involved two days of car-following testing, either with two or with
three vehicles in car-platoon formation on the public freeway roads from Ispra (VA) to Cherasco (CO) in northern Italy. The experi­
ments were scheduled to take place after peak morning hours and during lunch time hours, aiming to avoid the disturbances from other
road users.
The goal for the leader was to drive manually on free flow, without following other road users. The leader was instructed to drive
the car manually and to create frequent small perturbations in a random yet realistic way through deceleration and acceleration
around the desired speed. The follower was instructed to drive whenever possible with the ACC enabled.

Table 2
The main specifications of the vehicles involved in the experiments.
Vehicles Max power Drive-Fuel Engine Battery capacity Propulsion Top speed Model
(kW) displacement (cc) (kWh) type (km/h) year

(L) Fiat (500X) 103 diesel 1956 – ICE 190 2016

Volvo (XC40) 140 diesel 1969 – ICE 210 2018
(L) VW (Polo) 63 Gasoline and liquid 1390 – ICE 177 2010
propane gas
Hyundai (Ioniq hybrid) 104 gasoline 1580 1.56 HEV 185 2018
(L) Mitsubishi 59 gasoline 1193 – ICE 173 2018
(SpaceStar)
KIA (Niro) 77.2 gasoline 1580 8.9 PHEV 172 2019
Mitsubishi (Outlander 99 gasoline 2360 12 PHEV 170 2018
PHEV)
Peugeot (5008 GT Line) 130 diesel 1997 – ICE 208 2018
VW (Golf E) 100 electricity – 35.8 BEV 150 2018
Mini (Cooper) 100 gasoline 1499 – ICE 210 2018
Ford (S-Max) 110 diesel 1997 – ICE 196 2018
(L) Audi (A8) 210 diesel 2967 – ICE 250 2018
Tesla (Model 3) 150 electricity – 79 BEV 210 2019
BMW (X5) 195 diesel 2993 ICE 230 2018
Mercedes (A Class) 165 gasoline 1991 – ICE 250 2019
Audi (A6) 150 diesel 1968 – ICE 246 2018
(L) Smart (BME Addv) – – – – – – –
(L) Skoda (Octavia RS) 180 gasoline 1984 – ICE 250 2019
Tesla (model X) 386 electricity – 90 BEV 250 2016
Tesla (model 3) 250 electricity – 79 BEV 250 2019
Tesla (model S) 244 electricity – 75 BEV 225 2018
Mercedes-Benz (GLE 270 gasoline 2999 31.2 HEV 250 2019
450 4Matic)
Jaguar (I-Pace) 294 electricity – 90 BEV 200 2019
BMW (I3 s) 135 gasoline 647 33.2 HEV 160 2018
Audi (E-tron) 300 electricity – 83.6 BEV 200 2019
Toyota (Rav 4) 115 gasoline 2487 41.8 HEV 180 2019
Mazda (3) 96 gasoline 1998 – ICE 197 2019
Audi (A4 Avant) 140 gasoline 1984 0.69 HEV 238 2019

6
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

Table 3
Overview of the experimental campaigns included in the first version of the OpenACC database.
Campaign Ispra-Cherasco Ispra-Vicolungo (N.2) AstaZero (N.3) ZalaZone (N.4)
(N.1)

Leading vehicles Fiat 500X, Volvo Mitsubishi SpaceStar Audi A8 Smart BME ADdv, Octavia RS
XC40, VW Polo
Driving modes of Human Human ACC ACC
leading
vehicles
Vehicles involved Hyundai Ioniq KIA Niro, Mitsubishi Outlander Tesla Model 3, BMW Tesla Model X, Tesla Model 3, Tesla Model S,
in car- hybrid, Volvo PHEV, Ford S-Max, Peugeot 3008 GT X5, Mercedes A Class, Mercedes-Benz GLE 450 4Matic, Jaguar I-Pace, BMW
following XC40 Line, VW Golf E, Mini Cooper Audi A6 I3 s, Audi E-tron, Toyota Rav 4, Mazda 3, Audi A4
Avant
Driving modes of Human, ACC Human, ACC Human, ACC Human, ACC
following
vehicles
Time gap settings Short Short Short and long Short, long, medium and mixed (Short and long)

• Experimental Campaign N.2 – Driving from Ispra (VA) to Vicolungo (NO)

This campaign was conducted in the first quarter of 2019 and involves three days of car-following testing from Ispra (VA) to
Vicolungo (NO) and back, in northern Italy. The testing was performed with five vehicles of different brands and models driving on car-
platoon formation. As in the first campaign, the tests were scheduled for non-peak hours in order to minimize the disturbances from
other road users.
This campaign was a follow-up of the first one, when there were signs about unstable behavior of the ACC controller. The leader was
instructed to drive manually and perform occasional random deceleration and accelerations over a desired speed in a realistic way. The
followers, whenever possible were driving with ACC enabled.
Retrieving data from the first two test campaigns was challenging. As also reported by Knoοp et al. (Knoop et al., 2019), car-

Fig. 3. Challenges encountered during the experiment, Fig. a) is a cut-in situation due to an off-ramp, Fig. b) is car-platoon inconsistency due to
tolls, Fig. c) is a situation with low-quality GNSS signal due to a tunnel and Fig. d) is a lane change and overtaking situation due to a slow
truck ahead.

7
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

following experiments on public motorways are usually challenged by the behavior of the surrounding vehicles and the road geometry
and therefore, it is not always easy to preserve the platoon for very long distances. In addition, motorway driving usually limits the
speed range that can be analyzed. The shortest time headway setting for each vehicle driven by the ACC system was used to avoid cut-
in situations from other users. No matter of the short time gap setting, the random influences of the public road environment affected
the conduction of the experiments. Fig. 3 depicts some of the challenges encountered during the real-world experiments, showing four
indicative examples with physical and artificial obstacles. More specifically, Fig. 3 a) shows a cut-in situation when a road user needed
to exit the highway through an off-ramp, Fig. 3 b) is an example when the car-platoon consistency brakes on tolls, Fig. 3 c) is another
example inside a tunnel when the GNSS receiver provides position data of low quality and Fig. 3 d) shows a situation when a slow
moving truck was an obstacle for the experiment and overtaking had to be performed. Manual inspection of the dataset was performed
in order to remove problematic measurements and improve the quality of the car-following datasets. The implementation of a more
sophisticated filtering workflow, able to detect and isolate the above-mentioned challenges is currently an ongoing process.

• Experimental Campaign N.3 – AstaZero proving ground

This third campaign was conducted in the second quarter of 2019 and it involves two days of car-following testing on the Rural road
of the AstaZero test track in Sweden. The Rural road is approximately 5.7 km long. The campaign involves five high-end vehicles, from
four different makes, all different models.
Given the limitations of the on-road experiments and having significant insights from the first two campaigns about high response
times of the ACC controllers, large headway settings and instability, this campaign was organized in a way to become a research tool
for a more systematic study. The tests took place within a protected environment, in order to avoid risks and limitations that exist on
public roads. The experiments are organized in laps, where one lap corresponds to the path shown in Fig. 2. In all the tests, the leading
vehicle is the same and it is driven with the ACC enabled in order to avoid noisy fluctuations around the desired speed due to manual
maneuver. In general, two different car-following patterns are applied for the following vehicles, a) car-platoon with constant speed
and b) car-platoon with perturbations of the target speed (deceleration to a new desired speed) from an equilibrium point. For the
second pattern, a radio-based communication between the drivers of the first and last vehicles ensures that the speed of the last vehicle
is stable on the desired speed and therefore the car-platoon is close to an equilibrium state, prior to applying a new perturbation. Each
perturbation is triggered by the driver by setting the desired speed of the ACC system to a new lower desired speed value. Conse­
quently, the vehicle decelerates autonomously and when the new desired speed is reached, the driver resets the desired speed to the
previous setting. The duration of the perturbation is automatically adjusted based on the deceleration strategy applied by the
controller. This procedure is selected to perform the different perturbations in a controlled way, and it resembles the way that vehicles
with ACC enabled behave on road. For safety reasons in each lap the desired speeds were fixed to 13.9–16.7 m/s along the curves and
to 25–27.8 m/s on the straight parts.

• Experimental Campaign N.4 – ZalaZone proving ground

This last campaign was conducted in October 2019 and it involves two days of car-following testing on two test tracks, the Dynamic
Platform and the Handling Course of the ZalaZone ground in Hungary (see Fig. 2). The Dynamic Platform has a dimeter of 300 m which
is large enough to accomplish any kind of maneuver without any safety critical risk. The Handling Course contains curves and in­
clinations and has a length of 2.2 km, while the width of the asphalt lane throughout the entire module is 12 m. The campaign involves
eleven high-end vehicles, from 7 different makes, all different models. The programmable Smart BME ADdv vehicle was used as a
leader in the first day of the experiment. A wide range of different ACC systems was tested in low speeds under various vehicle orders
and ACC settings.
Tests in the Dynamic Platform were conducted in the circular plate of the road, where general ACC performance and related string
stability was investigated. The design of these tests was to drive with a stable speed and apply a perturbation in the form of a quick
deceleration to a new speed and a subsequent acceleration to restore the desired speed. The leader vehicle was always the Smart BME
ADdv and it was the one which applied all excitations. There was radio-based communication among all drivers, so as to ensure that the
speed of each vehicle is in the desired speed before the application of a new perturbation. All decelerations triggered by the leader were
executed at a target deceleration of − 3 m/s2, while all acceleration were performed with 1 m/s2. The automated leader vehicle assured
the exact reproduction of each single perturbation regardless of the equilibrium speed or the perturbation magnitude. Tests in the
Handling Course aimed at investigating the effect of the road geometry on the ACC performance. There, tests were organized in laps
and the speed of the vehicles was always set to a fixed desired speed. Each experiment involved 2 complete laps around the track. At the
end of the second lap, vehicles stopped at the straight part of the Handling Course using the ACC to reproduce stopping at traffic light

Table 4
Overview of the data related to the campaigns.
Type On road Proving ground

Campaign Campaign N.1 Campaign N.2 Campaign N.3 Campaign N.4

Driving mode Human ACC Human ACC Human ACC Human ACC
Duration (h) 5.70 5.28 4.31 5.69 3.12 15.59 5.32 50.88
Distance covered (km) 569 519 399 602 206 1029 188 1683

8
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

situations. Short, long, medium and mixed time gap settings were used.
Real world and test track experimental campaigns complement each other in the OpenACC database. The test on proving ground
ensure lower errors in data acquisition, repeatability of the same driving scenarios, improved safety and no distraction from other road
users. However, they are limited to specific road types and they introduce a bias, since the environment is controlled, there are safety
precautions and very high speeds cannot be reached and the drivers (or operators of the ACC systems) are aware of the perturbations.
Table 4 provides an overview for the OpenACC database with regard to different driving modes, the duration in hours and distance
travelled per test. Information on the duration and travel distance summed for all drivers categorized per experimental campaign and
driving mode are available. As can be seen the data connected to the test tracks are more extended due to the fact that the conditions
allow for more variable tests and without external destructions.

3.1. Data acquisition and post-processing

Position and speed vehicle data were acquired with different acquisition systems for the four experimental campaigns. For the first
two campaigns that took place on public roads, data acquisition was obtained in binary format from the U-blox M8 devices, one device
installed per vehicle. U-blox M8 devices are commercial systems equipped with accelerometers, gyroscopes and GNSS receivers. OBD
and CAN data were also available for some vehicles. The average horizontal accuracy reported by the receivers is less than 50 cm and
the sampling frequency is of about 3–5 Hz. Taking into account the difficulties of conducting experiments on public roads, manual
cleaning was applied to identify problematic trajectory parts and apply corrections in the OpenACC database. OBD data provide in­
formation of vehicle dynamics such as speed, acceleration, engine speed. OBD data are often incomplete, of low frequency and, in most
cases, they do not provide enough accuracy for trajectory/speed series reconstruction. However, they were useful in automatically
identifying trajectory parts, where the ACC was activated due to recorded acceleration pedal positions (OBD) or driving assistance
information (provided by the CAN). This information is available in the OpenACC database. Since the experiment was coordinated
among the vehicles’ drivers, detection of an enabled ACC system in the platoon was used as indication that all vehicles were per­
forming with their ACC systems (if they had this capability). At a final stage, visual inspection of speed series was performed in order to
detect noisy parts in the speeds and accelerations. These parts have been removed and are not included in the final dataset. For the first
two campaigns, the trajectories corresponding to urban conditions were removed as, for those parts, we were unable to detect per
vehicle when the ACC was enabled or not due to the technical problems with the data acquisition devices. Specific instances of cut-ins
or lane changes are not removed up to now from the database. An automated detection process is ongoing work for this task. Since
satellite reception capabilities are not uniform on public road experiments, at a final stage, splines data interpolation was applied to
ensure a constant sampling frequency of 10 Hz. A piecewise cubic polynomial being twice continuously differentiable on subintervals
of two points was used to interpolate the data. GNSS data contain significant noise levels especially when studying vehicle dynamics
with regard to vehicles’ acceleration or even jerk. Some works in the literature use linear interpolation in order to bring acceleration
values to acceptable levels, but we have found that this introduces step function pattern in the jerk domain. This makes the data
unsuitable for some topics, e.g. for studying the safety response of the vehicles. Therefore, in this work the piecewise cubic polynomial
was used. Inevitably, any post-processing introduces errors that counter-balance existing ones. In this case, for an insignificantly small
part of the data, the post-processed speed and acceleration values were found negatives (usually when the vehicles start from or arrive
at a stop). In those cases, the smoothed values were bounded to zero.
For the AstaZero test track experiment, data acquisition was performed with an inertial navigation system, the RT-Range S multiple
target ADAS measurements solution by Oxford Technical Solutions Company, with a differential GNSS accuracy, with precision of 2
cm/s in the speed and 2 cm in the positioning measurements with a sampling rate around 100 Hz. In the final database output, down
sampling to 10 Hz was applied as this frequency is enough for capturing the vehicle dynamics in the platoon. Since the test took place in
a protective environment, there is no interference from other users and output data ensure the absence of cut-ins or other types of
noise.

Table 5
An example of the columns contained in the .csv files of the database.
Column ID Description Units

Time Common time frame for all vehicles s

Speed Raw Speed (Doppler) m/s
Lat Latitude rad
Lon Longitude rad
Alt Altitude m
E East (x) coordinate in the local ENU plane (common center for all vehicles) m
N North (y) coordinate in the local ENU plane (common center for all vehicles) m
U Up (z) coordinate in the local ENU plane (common center for all vehicles) m
VE Speed in the East direction of the local ENU plane m/s
VN Speed in the North direction of the local ENU plane m/s
VU Speed in the Up direction of the local ENU plane m/s
IVS IVS computed from raw GNSS data after bumper to bumper correction. m

9
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

As for the ZalaZone proving ground experiment, fusion of different measurement systems was used. Three different systems were
used, a Race logic VBOX with a 2 cm (with RTK) accuracy on the position, 0.1 km/h speed error accuracy and 100 Hz sampling
frequency, Ublox 9 chipset with 0.3 m position accuracy and 0.5 km/h speed accuracy and up to 10 Hz frequency and finally, a tracker
app with 10 m position error and 1 km/h mean speed error, with frequency of 1 Hz. The measurements from the different systems were
synchronized based on UTC-time and in order to be able to compare and process those multisource data, special filters were applied to
achieve smooth data traces. Data have been fused and interpolated to 10 Hz using 2nd degree B-splines (Kitson, 1989). The final
dataset has been visually inspected in terms of positions, speed and acceleration values. In this process, the starting and end times of
each experiment have been noted. Data that correspond to the times between the end of one lap and the beginning of the second are
discarded from the final processed version. Moreover, instances for which a vehicle malfunctioned and had to leave the platoon have
also been recognized in the dataset, confirming the notes collected by the team during the experiments. Finally, the data has been
divided according to the platoon length, the steady state speed and the time-gap setting. For each different case, the specific exper­
iment settings are noted, together with the observed vehicle order.
Another dimension that directly affects the behavior of the ACC systems is the road elevation, which applies an additional force on
the vehicle. This is an important property that should be known to the researchers and therefore road elevation information is provided
in all four datasets per each time step under the variable name “Altitude” (see Table 5). Furthermore, the last campaign includes
experiments on a flat track. Fig. 4 illustrates sample elevation profiles, one per dataset. The two upper subfigures refer to the public
road experiments where the elevation profile present significant fluctuations. The two lower subfigures refer to the proving ground
tests where the variation in the elevation profile is lower and more controlled. Additionally, there are experiments in another track
included in the fourth dataset where the road is flat (apart from a minimum inclination to avoid the accumulation of water).

3.2. Database description

The final version of the OpenACC database, that is publicly available online8, contains four folders, one per experimental campaign.
Each folder has a list of .csv files that correspond car-platoon data series. Each .csv file contains a series of columns and an example of
such a file is shown in the Table 5. The folders, moreover, contain a file with the specifications of the vehicles and a brief description

Fig. 4. Elevation profiles for 4 tracks, each in a different experimental campaign.

10
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

with the experimental design. For illustrative purposes a sample file can be found in Appendix B.

4. A preliminary analysis of the OpenACC

This open science dataset creates unique opportunities for researchers in order to observe commercially available automation
controllers under real world conditions. An essential advantage according to the authors’ view is that researchers can use the diversity
of the OpenACC to draw novel conclusions regarding different topics:

– The study of human and ACC driving behavior for the same vehicle, or long car-platoons
– The ability to look at different levels of details, e.g. powertrain, vehicle, driver/ACC and platoon, with the opportunity to generalize
conclusions for impacts on traffic
– The presence of altitude information that is particularly useful for precise emissions/energy demand estimations
– Measurements of car-platoons under real world conditions but also repeatable driving scenarios in protected environments (test
tracks) that can assess the consistency in the controllers’ behavior
– Lower error levels due to modern data acquisition systems
– Measurements with a variety of modern highly automated commercial vehicles, giving important insights on the heterogeneity that
ACC systems can bring to future networks
– Data with different vehicle powertrains (i.e. electric, hybrid, diesel, gasoline)

The aim of this section is to present briefly the first findings after an initial processing of the dataset, some of which confirm results
already present in the literature but also others that provide new evidence. Since the main part of this work is the presentation of this
unique dataset, the authors here give only few but concrete examples on the above-mentioned scope.
The objective of the next sections is to provide the reader with general information about the behavior of these systems and to
illustrate with practical example the test carried out and the type of analyses that can be performed. The information provided here
should become a source of inspiration for the research community and especially the PhD students working in the field to imagine and
develop new strategies to use and analyze such data with the objectives to further understand how these systems will influence the
future traffic and to suggest ways to prevent the possible envisaged negative effects.
To the best of the authors’ knowledge, this is the first study that demonstrates the behavior of such heterogeneous and detailed car
dataset on car-following conditions and at the same time demonstrates the behavior of commercially available ACC systems regarding
some of their most important properties. The resulting OpenACC uses a simple data format that can be provided by most data
acquisition systems equipped with a GNSS and an antenna and it will be updated by the Joint Research Centre of the European
Commission, adding wherever possible new car-following tests. External parties willing to include their ACC experiments to the
OpenACC database are warmly invited to get in contact with the authors.

4.1. Lower speed variations but more outlier values for ACC drivers

Fig. 5 illustrates the instantaneous acceleration and deceleration distributions of each vehicle involved in the first campaign. The y-
axis shows the acceleration/deceleration values in m/s2 and the x-axis shows the different vehicles. The prefix ‘L’ before the name of

Fig. 5. Acceleration (m/s2) and deceleration distributions (m/s2) per vehicle and driving mode (raw measurements). Prefix “(L)” before the name of
a vehicle signifies that this vehicle was used as a leader of the car-platoon.

11
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

the vehicle in the x-axis signifies that this vehicle was used as a leading vehicle at some point during the campaign. Unfortunately, only
partial data of human-driven car-platoon are available in campaign N.1. Apart from the inference of realistic acceleration and
deceleration distributions based on normal non-critical car-following conditions, there is another interesting observation. Comparing
the human with the ACC distributions, the second has much higher peak and outlier values. This observation holds for both accel­
eration and deceleration distributions but for the second is much more obvious. Such measurements are derived probably due to
inability of the ACC system to monitor the conditions downstream and act proactively on imminent perturbations. Although these
observations do not constitute any proof that this is always valid, it is worth mentioning that this pattern (higher absolute peak values
for the ACC system) is observed across all campaigns. The above observations can be particularly helpful for calibration of car-
following models in microsimulation studies and the development of novel ones that can be parametrized based on the physical
interpretation of their parameters such as maximum acceleration as a function over speed, as well as maximum or comfortable
deceleration values.

4.2. Homogenous driving patters for ACC drivers

Fig. 6 shows two trajectory snapshots from campaign N.2. Each figure consists of four subfigures. The top-most one is the trajectory
plot of the vehicles. The second top-most refers to the oblique trajectories (Cassidy and Windover, 1995; Munoz and Daganzo, 2002)
that provide a better snapshot on the traffic dynamics. Oblique plots are a powerful alternative tool to visualize traffic conditions, as it
successfully tackles the scaling problem that appears in trajectory plots. For more information, we refer the reader to the above
publications. The third and fourth subfigures show the corresponding speed and time headway profiles of the vehicles.
In Fig. 6 a) all vehicles in the car-platoon drive with ACC on, while in Fig. 6 b) all vehicles have ACC off. Under stable car-following
conditions, the human drivers do not respect the ‘constant time headway’ policy that applies for the ACC controllers. It seems that the
distance from the leading vehicle is regulated by the driver’s perceived level of safety, which in turn, is based on the individual driver’s
characteristics and behavior. This is something that has been brought to attention in the literature (Jiang et al., 2015; Laval et al., 2014;
Ngoduy et al., 2019), and this is the reason why an increasing number of authors suggest the need for more accurate reproduction of
vehicle dynamics and driving behaviors in traffic modeling (Ciuffo et al., 2018; Fadhloun and Rakha, 2019; He et al., 2020; Makridis
et al., 2019a). There is also significantly higher variation in the speeds of the manually driven vehicle. ACC platoons behave much
more homogenously and although from different manufacturers, vehicles settle for very similar time headway values. Looking at the
oblique plots, it is clear that the ACC vehicles on constant speeds keep similar spacing and thus they have almost equal time headways.
The corresponding values for human-driven vehicles are quite random as they depend a lot on the driver’s behavior and style. Another
interesting observation is related to string (in)stability. Looking at the third subplot of Fig. 6 a), it is apparent that the perturbation
from the leader of the car-platoon is amplified upstream by the ACC controllers. The findings in Fig. 6 b) for trajectories with human-
driven vehicles are quite different. Manual drivers anticipate the changes in the speed of the car-platoon leader and absorb the
perturbation that occurs downstream. This confirms observations from other works in the literature such as Gunter et al. (Gunter et al.,
2020).

Fig. 6. Two snapshots from campaign’s N.2 trajectories. In Fig. a) the two followers are driving with ACC on, while in Fig. b) the following vehicle is
human driven.

12
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

4.3. Driver heterogeneity, time headway and response time

Driver heterogeneity on car-following has a direct impact on traffic flow (Makridis et al., 2020a) and it is reflected on time headway
measurements (Saifuzzaman and Zheng, 2014). Looking at the time headway distributions across the whole dataset of the third
campaign, Fig. 7 illustrates for the four following vehicles in the third campaign the headway distributions for ACC and human drivers.
For the ACC systems there are two distinctive peaks corresponding to the maximum and minimum settings of the controllers. The
estimated peak values found for each vehicle are 1.3 s and 2.7 s for the Tesla, 1.2 s and 2.5 s for the BMW, 1.2 s and 3.5 s for the Audi
and 1.2 s and 2.5 s for the Mercedes. The differences between the time headway distributions of the ACC controller and the human
driver are obvious, with the second one being much more uniform. It is worth noting that the ranges of observed time headway values
between human and ACC (minimum and maximum boundaries of most observations) are similar, roughly between 1 s and 4 s.
Furthermore, human drivers show different distributions with regard to their time headway strategy, reflecting their personal behavior
and driving style. The above observations have direct impact on the traffic flow. From one side, the commercial ACC controllers ensure
low speed variations in the longitudinal direction and under stable conditions, but it seems that they cannot efficiently respond to
(even small) perturbations that can occur on the road downstream. Human drivers on the other hand, limit the road capacity under
stable conditions but they are very efficient in anticipating traffic oscillations, as they have an overview of the traffic conditions
downstream.
Estimation of the response time of the ACC controllers is a hot topic for discussion in the literature (Makridis et al., 2019b, 2018;
Milanes et al., 2014; Milanés and Shladover, 2014; Shladover et al., 2015). Given the theoretical ability of the ACC systems to retrieve
instantly the leader’s speed, there is a number of works in the literature that expect instantaneous response time of the controller
(Green, 2000; Kesting et al., 2008; Tapani, 2012). This assumption has been adopted by traffic simulation studies. However, empirical
tests have shown that systems in commercial vehicles experience significant delays. Makridis et al. (2019c) proposed a method for
estimation of the response time of an ACC controller based on correlation of two signals, the acceleration of the follower and the speed
difference between the leader and the follower. This methodology was used in the present study as well. We refer the reader to the
corresponding publication for more details.
Fig. 8 illustrates the estimated response times for the ACC controllers of two vehicles based on partial trajectories. The top subplots
show the two signals that need to be correlated, the delta speed between the two vehicles and the acceleration of the follower. The
middle subplot shows the correlation function. The peak corresponds to the correlation coefficient. Values closer to one indicate better
confidence on the estimation. In both cases demonstrated here, the response time has been found to be around 2 s, which is much larger
than the one expected in most of the literature studies. The confidence for both cases is very high and in the third subplot, it is obvious
that the two signals become highly correlated for the estimated response times. It is important to note that these estimations refer to
normal, non-critical, driving conditions. For safety-related studies the corresponding times of the Automated Emergency Braking
ADAS functionality should be considered. However, response time under normal freeway conditions directly impacts traffic and string
stability.
For the sake of completeness, Fig. 9 illustrates the boundary estimated time headway values for human drivers and both the
headway and response time boundary values for the ACC systems under observation, per vehicle. Boundary time headways have
similar values for both human drivers (between 0.6 s and 3.8 s) and ACC systems (between 1.1 s and 4.1 s). Furthermore, the estimated
response times for ACC systems are quite high (between 0.6 s and 3.7 s), and comparable to the corresponding values of human drivers
reported in the literature. It has been observed that the behavior of each controller is not deterministic and varies depending on factors
such as road conditions (geometry, altitude), the car-platoon equilibrium speed, the time headway settings, the weather, the
magnitude and type of perturbation performed by the leading vehicles and others.

Fig. 7. Time headway distributions for all the vehicles involved in the campaign N.3.

13
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

Fig. 8. Indicative response time estimations for the ACC controllers of two vehicles from campaign’s N.2.

Fig. 9. Estimated minimum and maximum values for the time headways of human drivers as well as the time headway and response time of the
ACC controllers from all four campaigns.

14
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

4.4. String (in)stability of commercial ACC systems

Another interesting topic for researchers is whether commercial ACC systems are string stable or not. A verbal definition of the
string stability for a car-platoon assuming that it is under stable conditions and the leader performs a perturbation, can be given as
follows: a car-platoon can be considered weak string stable when the magnitude of the leading vehicle’s perturbation is greater than the
magnitude of the last following vehicle’s perturbation in the car-platoon (Monteil et al., 2019).
In the literature, there are several works arguing that commercially available ACC systems are string unstable. Such systems
currently deployed on the market have limitations due to their sensor in observing the traffic conditions downstream. This practically
means that they cannot react quickly on perturbations that are initiated 3 or 4 vehicles ahead. Consequently, they apply hard de­
celerations creating trajectory patterns like the ones discussed in Fig. 6. Such findings have been presented in the work of Gunter et al.
(Gunter et al., 2020), where the authors assessed the string stability of seven 2018 model year ACC-equipped vehicles from two makes.
Also, in the empirical study of Makridis et al. (2020b), it is discussed that even for slight perturbations derived by variability in the road
gradient, string instability can be observed.
However, as it is illustrated in Fig. 10, more research on this topic is needed. Of course, string stability should be studied across the
complete operational domain and under all possible conditions for a controller. Experimental observations from the fourth campaign
show that looking solely into isolated platoon trajectories, the platoon of 10 ACC-driven vehicles can demonstrate either a string stable
or string unstable behavior (using the weak string stability definition) depending on the time-gap desired setting applied by the drivers.
Fig. 10 shows examples of the effect of different perturbations on the platoon’s speed profile. When the vehicles’ ACC adopt a short
time-gap (Fig. 10 a and c), the magnitude of the perturbation increases over the platoon. On the contrary, with the long time-gap
setting (Fig. 10 b and d), the magnitude of the perturbation is attenuated over the platoon. This is consistently experienced inde­
pendently from the target speed and the vehicles’ order in the platoon. This observation implies a relation between the desired
headway setting and the string stability, where there should be a threshold time headway setting upon a platoon can be string stable
and below which the platoon can be string unstable. Of course, a more systematic research is needed on this topic, but this is considered
out of the present paper’s scope. Such results however show without doubt, the added value of acquiring as much as possible empirical
data in order to observe the behavior of such systems under as many as possible different scenarios and conditions.

Fig. 10. Speed profile of the platoon resulting from four different perturbations during the experiments on the fourth campaign. Charts a) and b)
refer to experiments with 30 km/h target speed, while charts c) and d) refer to 40 km/h target speed. Charts a) and c) refer to experiments with
vehicles adopting a short time-gap for their ACC, while charts b) and d) to long time-gap.

15
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

4.5. Traffic hysteresis and safety

The study of empirical data can contribute towards is a better understanding of the mechanism of traffic hysteresis and traffic
oscillations from the driver behavior perspective as described in the literature (Laval and Leclercq, 2010; Saifuzzaman et al., 2017). In
the third campaign dataset, the acceleration, deceleration and stable states are separated by experimental design. This characteristic
facilitates the study of traffic hysteresis and its magnitude as demonstrated in Fig. 11a. Here, the three main phases (i.e. deceleration,
acceleration, equilibrium) are colored differently, showing the differences in the density during the non-stable conditions, as well as
the different behavior of maximum and minimum headway settings of the controllers. For the minimum headway setting, during the
acceleration phase, the density is much smaller than during the deceleration phase, for similar speeds, so the produced flow is
decreased. On the other hand, there is no such phenomenon observable for the maximum headway setting. However, the flow is
already smaller than even the decreased flow produced by the minimum headway platoon during the acceleration phase. This is a
phenomenon discussed in the literature but not demonstrated with empirical observations. Again, the function that relates the
headway with the positive impact on traffic hysteresis and the negative impact on traffic flow is another open question in the literature.
Furthermore, analysis regarding traffic safety can be also performed looking at common Safety Surrogate Metrics such as Time-to-
Collision (TTC) as shown in Fig. 11b. Here the TTC instantaneous values are shown for two different headway settings in the vehicles of
the third campaign. The difference in the instantaneous TTC values for the vehicles with the maximum and minimum headway settings
is clear. It is worth noticing that with the minimum headway setting, the minimum TTC decreases moving backwards in the platoon of
vehicles. This is clearly related to the string instability and the need for vehicles further upstream to decelerate stronger than the
vehicles upstream. Looking at the particular case presented in Fig. 11b the last vehicle in the platoon has minimum TTC of 4 s cor­
responding to a no-risk situation (literature suggests a crisp threshold in the range of 1.5–3 s to identify a conflict). However, if the
platoon was longer it could be expected that vehicles further backwards would have achieved high collision risk even if the decel­
eration of the leader was mild.

5. Conclusions

This work summarizes the main features of the OpenACC, an open-access database of different car-following experiments involving
a total of 22 vehicles of different brands equipped with state-of-the-art commercial ACC systems and 6 more vehicles used as leaders in
the tests. The OpenACC database includes four experimental campaigns, two on public roads and two on test tracks. The resulted
dataset provides empirical evidence on various dimensions. It provides data with different modern acquisition devices and thus
different error levels, altitude information, vehicles of different ACC controllers and various powertrains, measurement on public roads
and protected test tracks and repeatable driving scenarios with and without ACC enabled for the vehicles in the platoon.
Considering the lack of empirical data in the literature, due to the high cost of organizing, financing and designing large experi­
mental campaigns, the development of a dataset in a systematic and structured way can lead to important findings on how these
systems operate, how they differentiate from human driving patterns and what are the best ways for modeling them and assess their
impact on future road transport networks.
By making the data available, the authors want to engage the whole traffic scientific community to use them, improve them, and
derive from them useful insights by means of which it is possible to avoid that new driving assistance and automation systems will
create additional problems to the road transport systems.
Preliminary results from a first analysis of all four experiments validate similar observations in the literature, confirm analytical
results and provide novel insights on a variety of topics:

– Lower speed variations are observed for the commercial ACC systems but their inability to monitor the traffic conditions down­
stream, leads to extreme acceleration (but mainly) deceleration peak values
– ACC controllers show homogenous headway distributions, contrary to the corresponding observations on human driven vehicles
– Boundary time headway values are similar for human drivers and ACC controllers
– The response time of commercial ACC systems is by no means instantaneous and similar to human’s reaction times as they are
reported in the literature
– String instability reported in the literature is confirmed for large time headway settings in the ACC systems, but novel findings show
the for large time headway setting, even large car-platoons can achieve weak string stability
– Large time headway settings have a negative impact on the capacity but positive impact on hysteresis
– Instability in car-platoon of ACC vehicles can lead to safety–critical situations according to the observed TTC values.

It is believed that this open science dataset creates unique opportunities for researchers in order to observe commercially available
automation controllers under real world conditions. Future plans involve data acquisition campaigns involving also lane changing
behaviors, other automated systems such as Automated Emergency Braking measurements, Automated Lane Keeping Systems, Traffic
Jam Chauffeur etc. and deeper investigation on the impact of automated vehicle functionalities on traffic flow, energy demand and
safety. Additional study is ongoing on the relation between the physical meaning of car-following parameters in traffic micro­
simulation modeling and the empirically observed values. Establishing a correlation between the two can be proved very helpful not
only for researchers but also for practitioners. Finally, heterogeneity in the driving behavior of human drivers is another interesting
topic, to what extent this applies for ACC systems and what are the most significant challenges that arrive with the mass deployment of
advanced automated technologies.

16
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

Fig. 11. Demonstration of traffic hysteresis based on the data from the campaign N.3.

CRediT authorship contribution statement

Michail Makridis: Conceptualization, Formal analysis, Writing - original draft, Writing - review & editing, Supervision, Visuali­
zation. Konstantinos Mattas: Formal analysis, Data curation, Visualization, Writing - original draft. Aikaterini Anesiadou: Data
curation, Visualization, Writing - original draft. Biagio Ciuffo: Conceptualization, Supervision, Writing - original draft, Writing -
review & editing.

Acknowledgments

The views expressed here are purely those of the authors and may not, under any circumstances, be regarded as an official position
of the European Commission. The authors are grateful to Eleftherios-Nektarios Grylonakis and Vincenzo Arcidiacono who contributed
to the data processing, as well as to Georgios Fontaras, Fabrizio Minarini, Norbert Brinkhoff-Button, Gianmarco Baldini, Daniele Borio,
Alessandro Tansini, Melania Susi, Lorenzo Maineri, Yinglong He, Akos Kriston, and Fabrizio Re for their support during the experi­
mental campaigns. Finally, the authors would like to thank Greger Rognelund and the staff in AstaZero proving ground in Sweden, and
Zsolt Szalay and the staff in ZalaZone proving ground in Hungary for their work during the experiments.

Appendix A. Indicative template of the information file

Each folder inside the OpenACC database contains a file with the specifications of the vehicle as well as a short description of the
experimental campaign. Below a sample is presented on how the information file is structured.
Description of the experiment: A main description of the campaign, and objectives. The number of trips included.
Date: DD/MM/YYYY.
Vehicle specs: Name of the vehicles and vehicles’ specifications.
Number of vehicles: Number of vehicles involved in each campaign.
Equipment: Data acquisition equipment.
Data processing: Technique used for the 10 Hz frequency achievement.
Trip comments: Some comments on the trips on the data of the trips.
Columns info: Description of the columns included in the csv files regarding the respective experimental campaign.

References

Barmpounakis, E., Geroliminis, N., 2020. On the new era of urban traffic monitoring with massive drone data: The pNEUMA large-scale field experiment. Transp. Res.
Part C Emerg. Technol. 111, 50–71. https://doi.org/10.1016/j.trc.2019.11.023.
Cassidy, M.J., Windover, J.R., 1995. Methodology for assessing dynamics of freeway traffic flow. Transp. Res. Rec.
Ciuffo, B., Makridis, M., Toledo, T., Fontaras, G., 2018. Capability of current car-following models to reproduce vehicle free-flow acceleration dynamics. IEEE Trans.
Intell. Transp. Syst. 19, 3594–3603. https://doi.org/10.1109/TITS.2018.2866271.
Ciuffo, B., Mattas, K., Anesiadou, A., Makridis, M., 2020. Open ACC Database Open ACC Database. European Commission, Joint Research Centre (JRC) [Dataset] PID:
http://data.europa.eu/89h/9702c950-c80f-4d2f-982f-44d06ea0009f.
Coifman, B., Li, L., 2017. A critical evaluation of the Next Generation Simulation (NGSIM) vehicle trajectory dataset. Transp. Res. Part B Methodol. 105, 362–377.
https://doi.org/10.1016/j.trb.2017.09.018.
European Commission, 2019. The future of road transport [WWW Document]. EU Sci. Hub - Eur. Comm. URL https://ec.europa.eu/jrc/en/publication/eur-scientific-
and-technical-research-reports/future-road-transport (accessed 8.23.19).

17
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

Fadhloun, K., Rakha, H., 2019. A novel vehicle dynamics and human behavior car-following model: model development and preliminary testing. Int. J. Transp. Sci.
Technol. https://doi.org/10.1016/j.ijtst.2019.05.004.
Green, M., 2000. “How long does it take to stop?” Methodological analysis of driver perception-brake times. Transp. Hum. Factors 2, 195–216. https://doi.org/
10.1207/STHF0203_1.
Gunter, G., Gloudemans, D., Stern, R.E., McQuade, S., Bhadani, R., Bunting, M., Monache, M.L.D., Lysecky, R., Seibold, B., Sprinkle, J., Piccoli, B., Work, D.B., 2020.
Are commercially implemented adaptive cruise control systems string stable? IEEE Trans. Intell. Transp. Syst. 1–12 https://doi.org/10.1109/TITS.2020.3000682.
He, Y., Ciuffo, B., Zhou, Q., Makridis, M., Mattas, K., Li, J., Li, Z., Yan, F., Xu, H., 2019. Adaptive cruise control strategies implemented on experimental vehicles: a
review. In: IFAC-Pap., 9th IFAC Symposium on Advances in Automotive Control AAC. https://doi.org/10.1016/j.ifacol.2019.09.004.
He, Y., Makridis, M., Mattas, K., Fontaras, G., Ciuffo, B., Xu, H., 2020. Introducing electrified vehicle dynamics in traffic simulation. Transp. Res. Rec. J. Transp. Res.
Board 036119812093184. https://doi.org/10.1177/0361198120931842.
Hu, S.G., Wen, H.Y., Xu, L., Fu, H., 2019. Stability of platoon of adaptive cruise control vehicles with time delay. Transp. Lett. 11, 506–515. https://doi.org/10.1080/
19427867.2017.1407488.
Jiang, R., Hu, M.-B., Zhang, H.M., Gao, Z.-Y., Jia, B., Wu, Q.-S., 2015. On some experimental features of car-following behavior and how to model them. Transp. Res.
Part B Methodol. 80, 338–354. https://doi.org/10.1016/j.trb.2015.08.003.
Kesting, A., Treiber, M., Schönhof, M., Helbing, D., 2008. Adaptive cruise control design for active congestion avoidance. Transp. Res. Part C Emerg. Technol. 16,
668–683. https://doi.org/10.1016/j.trc.2007.12.004.
Kitson, F.L., 1989. An algorithm for curve and surface fitting using B-splines. In: International Conference on Acoustics, Speech, and Signal Processing. Presented at
the International Conference on Acoustics, Speech, and Signal Processing, vol. 2, pp. 1207–1210. https://doi.org/10.1109/ICASSP.1989.266651.
Knoop, V.L., Wang, M., Wilmink, I., Hoedemaeker, D.M., Maaskant, M., Van der Meer, E.-J., 2019. Platoon of SAE level-2 automated vehicles on public roads: setup,
traffic interactions, and stability. Transp. Res. Rec. J. Transp. Res. Board 2673, 311–322. https://doi.org/10.1177/0361198119845885.
Krajewski, R., Bock, J., Kloeker, L., Eckstein, L., 2018. The highD dataset: a drone dataset of naturalistic vehicle trajectories on german highways for validation of
highly automated driving systems. In: 2018 21st International Conference on Intelligent Transportation Systems (ITSC). Presented at the 2018 21st International
Conference on Intelligent Transportation Systems (ITSC), pp. 2118–2125. https://doi.org/10.1109/ITSC.2018.8569552.
Laval, J.A., Leclercq, L., 2010. A mechanism to describe the formation and propagation of stop-and-go waves in congested freeway traffic. Philos. Transact. A Math.
Phys. Eng. Sci. 368, 4519–4541. https://doi.org/10.1098/rsta.2010.0138.
Laval, J.A., Toth, C.S., Zhou, Y., 2014. A parsimonious model for the formation of oscillations in car-following models. Transp. Res. Part B Methodol. 70, 228–238.
https://doi.org/10.1016/j.trb.2014.09.004.
Li, D., Wagner, P., 2019. Impacts of gradual automated vehicle penetration on motorway operation: a comprehensive evaluation. Eur. Transp. Res. Rev. 11, 36.
https://doi.org/10.1186/s12544-019-0375-3.
Li, L., Chen (Michael), X., 2017. Vehicle headway modeling and its inferences in macroscopic/microscopic traffic flow theory: a survey. Transp. Res. Part C Emerg.
Technol. 76, 170–188. https://doi.org/10.1016/j.trc.2017.01.007.
Li, L., Jiang, R., He, Z., Chen (Michael), X., Zhou, X., 2020. Trajectory data-based traffic flow studies: a revisit. Transp. Res. Part C Emerg. Technol. 114, 225–240.
https://doi.org/10.1016/j.trc.2020.02.016.
Li, M., Chen (Michael), X., Ni, W., 2016. An extended generalized filter algorithm for urban expressway traffic time estimation based on heterogeneous data. J. Intell.
Transp. Syst. 20, 474–484. https://doi.org/10.1080/15472450.2016.1153426.
Liu, H., Kan (David), X., Shladover, S.E., Lu, X.-Y., Ferlis, R.E., 2018. Modeling impacts of Cooperative Adaptive Cruise Control on mixed traffic flow in multi-lane
freeway facilities. Transp. Res. Part C Emerg. Technol. 95, 261–279. https://doi.org/10.1016/j.trc.2018.07.027.
Makridis, M., Fontaras, G., Ciuffo, B., Mattas, K., 2019a. MFC free-flow model: introducing vehicle dynamics in microsimulation. Transp. Res. Rec. 2673, 762–777.
https://doi.org/10.1177/0361198119838515.
Makridis, M., Leclercq, L., Mattas, K., Ciuffo, B., 2020a. The impact of driving homogeneity due to automation and cooperation of vehicles on uphill freeway sections.
Eur. Transp. Res. Rev. 12, 15. https://doi.org/10.1186/s12544-020-00407-9.
Makridis, M., Mattas, K., Borio, D., Ciuffo, B., 2019b. Estimating empirically the response time of commercially available ACC controllers under urban and freeway
conditions. In: 2019 6th International Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS). Presented at the 2019 6th
International Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS), pp. 1–7. https://doi.org/10.1109/MTITS.2019.8883341.
Makridis, M., Mattas, K., Borio, D., Giuliani, R., Ciuffo, B., 2018. Estimating reaction time in Adaptive Cruise Control System. In: Presented at the 2018 IEEE Intelligent
Vehicles Symposium (IV), pp. 1312–1317. https://doi.org/10.1109/IVS.2018.8500490.
Makridis, M., Mattas, K., Ciuffo, B., 2019c. Response time and time headway of an adaptive cruise control. An empirical characterization and potential impacts on
road capacity. IEEE Trans. Intell. Transp. Syst. 1–10 https://doi.org/10.1109/TITS.2019.2948646.
Makridis, M., Mattas, K., Ciuffo, B., Re, F., Kriston, A., Minarini, F., Rognelund, G., 2020b. Empirical study on the properties of adaptive cruise control systems and
their impact on traffic flow and string stability. Transp. Res. Rec. 2674, 471–484. https://doi.org/10.1177/0361198120911047.
Makridis, M., Mattas, K., Mogno, C., Ciuffo, B., Fontaras, G., 2020c. The impact of automation and connectivity on traffic flow and CO2 emissions. A detailed
microsimulation study. Atmos. Environ. 226, 117399 https://doi.org/10.1016/j.atmosenv.2020.117399.
Mattas, K., Makridis, M., Hallac, P., Raposo, M.A., Thiel, C., Toledo, T., Ciuffo, B., 2018. Simulating deployment of connectivity and automation on the Antwerp ring
road. IET Intell. Transp. Syst. 12, 1036–1044. https://doi.org/10.1049/iet-its.2018.5287.
Milanés, V., Shladover, S.E., 2014. Modeling cooperative and autonomous adaptive cruise control dynamic responses using experimental data. Transp. Res. Part C
Emerg. Technol. 48, 285–300. https://doi.org/10.1016/j.trc.2014.09.001.
Milanes, V., Shladover, S.E., Spring, J., Nowakowski, C., Kawazoe, H., Nakamura, M., 2014. Cooperative adaptive cruise control in real traffic situations. IEEE Trans.
Intell. Transp. Syst. 15, 296–305. https://doi.org/10.1109/TITS.2013.2278494.
Montanino, M., Punzo, V., 2015. Trajectory data reconstruction and simulation-based validation against macroscopic traffic patterns. Transp. Res. Part B Methodol.
80, 82–106. https://doi.org/10.1016/j.trb.2015.06.010.
Monteil, J., Bouroche, M., Leith, D.J., 2019. L2 and L-infinite stability analysis of heterogeneous traffic with application to parameter optimization for the control of
automated vehicles. IEEE Trans. Control Syst. Technol. 27, 934–949. https://doi.org/10.1109/TCST.2018.2808909.
Munoz, J.C., Daganzo, C.F., 2002. Fingerprinting traffic from static freeway sensors.
Ngoduy, D., Lee, S., Treiber, M., Keyvan-Ekbatani, M., Vu, H.L., 2019. Langevin method for a continuous stochastic car-following model and its stability conditions.
Transp. Res. Part C Emerg. Technol. 105, 599–610. https://doi.org/10.1016/j.trc.2019.06.005.
NGSIM [WWW Document], 2006. URL https://ops.fhwa.dot.gov/trafficanalysistools/ngsim.htm (accessed 12.11.19).
Olia, A., Razavi, S., Abdulhai, B., Abdelgawad, H., 2018. Traffic capacity implications of automated vehicles mixed with regular vehicles. J. Intell. Transp. Syst. 22,
244–262. https://doi.org/10.1080/15472450.2017.1404680.
Punzo, V., Borzacchiello, M.T., Ciuffo, B., 2011. On the assessment of vehicle trajectory data accuracy and application to the Next Generation SIMulation (NGSIM)
program data. Transp. Res. Part C Emerg. Technol. 19, 1243–1262. https://doi.org/10.1016/j.trc.2010.12.007.
Punzo, V., Simonelli, F., 2005. Analysis and comparison of microscopic traffic flow models with real traffic microscopic data. Transp. Res. Rec. J. Transp. Res. Board
1934, 53–63. https://doi.org/10.3141/1934-06.
SAE International, 2018. J3016B: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles - SAE International
[WWW Document]. URL https://www.sae.org/standards/content/j3016_201806/ (accessed 4.9.20).
Saifuzzaman, M., Zheng, Z., 2014. Incorporating human-factors in car-following models: a review of recent developments and research needs. Transp. Res. Part C
Emerg. Technol. 48, 379–403. https://doi.org/10.1016/j.trc.2014.09.008.
Saifuzzaman, M., Zheng, Z., Haque, Md.M., Washington, S., 2017. Understanding the mechanism of traffic hysteresis and traffic oscillations through the change in task
difficulty level. Transp. Res. Part B Methodol. 105, 523–538. https://doi.org/10.1016/j.trb.2017.09.023.

18
M. Makridis et al. Transportation Research Part C 125 (2021) 103047

Shladover, S.E., Nowakowski, C., Lu, X.-Y., Ferlis, R., 2015. Cooperative adaptive cruise control: definitions and operating concepts. Transp. Res. Rec. 2489, 145–152.
https://doi.org/10.3141/2489-17.
Shladover, S.E., Su, D., Lu, X.-Y., 2012. Impacts of cooperative adaptive cruise control on freeway traffic flow. Transp. Res. Rec. 2324, 63–70. https://doi.org/
10.3141/2324-08.
Sun, Jie, Zheng, Z., Sun, Jian, 2018. Stability analysis methods and their applicability to car-following models in conventional and connected environments. Transp.
Res. Part B Methodol. 109, 212–237. https://doi.org/10.1016/j.trb.2018.01.013.
Tapani, A., 2012. Vehicle trajectory effects of adaptive cruise control. J. Intell. Transp. Syst. 16, 36–44. https://doi.org/10.1080/15472450.2012.639641.
Wang, J., Huang, Y., Feng, Z., Jiang, C., Zhang, H., Leung, V.C.M., 2018a. Reliable traffic density estimation in vehicular network. IEEE Trans. Veh. Technol. 67,
6424–6437. https://doi.org/10.1109/TVT.2018.2803062.
Wang, M., Hoogendoorn, S.P., Daamen, W., van Arem, B., Shyrokau, B., Happee, R., 2018b. Delay-compensating strategy to enhance string stability of adaptive cruise
controlled vehicles. Transp. B Transp. Dyn. 6, 211–229. https://doi.org/10.1080/21680566.2016.1266973.
Wang, S., Li, L., Ma, W., Chen, X., 2019a. Trajectory analysis for on-demand services: a survey focusing on spatial-temporal demand and supply patterns. Transp. Res.
Part C Emerg. Technol. 108, 74–99. https://doi.org/10.1016/j.trc.2019.09.007.
Wang, W., Liu, C., Zhao, D., 2017. How much data is enough? A statistical approach with case study on longitudinal driving behavior. IEEE Trans. Intell. Veh. 1–1
https://doi.org/10.1109/TIV.2017.2720459.
Wang, X., Jiang, R., Li, L., Lin, Y., Zheng, X., Wang, F.-Y., 2018c. Capturing car-following behaviors by deep learning. IEEE Trans. Intell. Transp. Syst. 19, 910–920.
https://doi.org/10.1109/TITS.2017.2706963.
Wang, X., Jiang, R., Li, L., Lin, Y.-L., Wang, F.-Y., 2019b. Long memory is important: a test study on deep-learning based car-following model. Phys. Stat. Mech. Its
Appl. 514, 786–795. https://doi.org/10.1016/j.physa.2018.09.136.
Xiao, L., Gao, F., 2010. A comprehensive review of the development of adaptive cruise control systems. Veh. Syst. Dyn. 48, 1167–1192. https://doi.org/10.1080/
00423110903365910.
Ye, L., Yamamoto, T., 2018. Modeling connected and autonomous vehicles in heterogeneous traffic flow. Phys. Stat. Mech. Its Appl. 490, 269–277. https://doi.org/
10.1016/j.physa.2017.08.015.
Yu, S., Tang, J., Xin, Q., 2018. Relative velocity difference model for the car-following theory. Nonlinear Dyn. 91, 1415–1428. https://doi.org/10.1007/s11071-017-
3953-8.

19