Content uploaded by Ramin Arvin
Author content
All content in this area was uploaded by Ramin Arvin on Dec 03, 2020
Content may be subject to copyright.
Safety Evaluation of Connected and Automated Vehicles in Mixed Traffic
with Conventional Vehicles at Intersections
Ramin Arvin a, Asad J Khattak a,
*
, Mohsen Kamrani a, Jackeline Rios-Torres b
a Department of Civil and Environmental Engineering, The University of Tennessee
b Energy & Transportation Science Division, Oak Ridge National Laboratory
ABSTRACT
Connected and Automated Vehicles (CAVs) can potentially improve the performance
of the transportation system by reducing human errors. This paper investigates the safety
impact of CAVs in a mixed traffic with conventional vehicles at intersections.
Analyzing real-world AV crashes in California revealed that rear-end crashes at
intersections is the dominant crash type. Therefore, to enhance our understanding of the
future interactions between human-driven vehicles with CAVs at intersections, a
simulation framework was developed to model the mixed traffic environment of
Automated Vehicles (AV), cooperative AVs, and conventional human-driven vehicles.
In order to model AVs driving behavior, Adaptive Cruise Control (ACC) and
cooperative ACC (CACC) models are utilized. Particularly, this study explores system
improvements due to automation and connectivity across varying CAV market
penetration scenarios. ACC and CACC car following models are used to mimic the
behavior of AVs and cooperative AVs. Real-world connected vehicle data are utilized
to modify and tune the acceleration/deceleration regimes of the Wiedemann model.
Next, the driving volatility concept capturing variability in vehicle speeds was utilized
to calibrate the simulation to represent the safety performance of a real-world
environment. Two surrogate safety measures are used to evaluate the safety performance
of a representative intersection under different market penetration rate of CAVs: the
number of longitudinal conflicts and driving volatility. At low levels of ACC market
penetration, the safety improvements were found to be marginal, but safety improved
substantially with more than 40% ACC penetration. Additional safety improvements
can be achieved more quickly through the addition of cooperation and connectivity
through CACC. Furthermore, ACC/CACC vehicles were found to improve mobility
performance in terms of average speed and travel time at intersections.
Keywords: Connected and Automated Vehicles; Intersection Safety; Driving Volatility; Time to
Collision; Adaptive Cruise Control; Cooperative Adaptive Cruise Control; Simulation; Mobility
*
Corresponding author: Email: akhattak@utk.edu
Arvin, Khattak, Kamrani, Rios-Torres
2
INTRODUCTION
According to the National Highway Traffic Safety Administration, there were more than 7.27
million vehicle crashes across the United States that caused 2.17 million injuries and 37,914
fatalities in 2016 (Anon 2018). Given that human error is contributing to more than 90 percent of
crashes (Singh 2015), Automated Vehicles (AV) aim to improve the current state of transportation
through eliminating human drivers from the driving task. A system of AVs offers significant
potential to reduce crashes and related losses as well as reduce traffic congestion. The development
and deployment of AVs creates a world of possibilities and questions. As prototype AVs are
already commuting in the transportation network, widespread use of AVs is anticipated in the near
future. The interaction between AVs and non-automated vehicles raises many concerns as the
driving styles and driving responses to the environment vary from driver to driver, and that can
cause a disturbance in the traffic network. Referring to the AVs, Adaptive Cruise Control (ACC)
is introduced as one of the driving assistant systems performing longitudinal control of AVs by
automatically controlling vehicle movements and responding to the preceding vehicle. On the
other hand, Connected and Automated Vehicles (CAVs) integrate connectivity and automation
using Cooperative ACC (CACC), which is extending ACC by utilizing vehicle-to-vehicle (V2V)
communication which allows cooperation among AVs. While the impact of AVs on transportation
safety remains unknown due to lack of real-world data, microsimulation study can be the most
cost-effective approach for evaluation of potential benefits/drawbacks of mass market adoption of
AVs. This study presents a simulation framework to assess the impact of varying levels of AVs
market penetration in a mixed traffic with conventional vehicles which can provide initial safety
impact understanding on the implementation of AVs. While a few studies have explored the safety
impact of AVs, the main goal of this paper is to explore the safety implications of AVs interacting
with conventional vehicles at a signalized intersection.
This study analyzes the real-world AV-involved crashes in California in order to
understand the current crash mechanism. Next, the interaction of AVs with conventional human-
driven vehicles at a signalized intersection is simulated. The ACC and CACC models are utilized
to simulate longitudinal control of AVs with and without cooperation. Furthermore, real-world
Connected Vehicle (CV) data are used to revise the acceleration/deceleration regimes of the
Wiedemann car-following model to accurately represent longitudinal vehicle control and
following behavior of human drivers. Furthermore, the CV data was harnessed to calibrate the
simulation to truly represent the actual traffic in the baseline scenario. Finally, using a developed
behavioral framework, multiple scenarios are defined and simulated to study the mixed traffic of
conventional vehicles, ACC, and CACC vehicles.
LITERATURE REVIEW
An analysis of traffic crashes in the United States revealed that the main contributing factor in 93
percent of crashes is human error (Anon 2008). In an attempt to minimize human errors, new
technologies and advanced driving assistant systems (ADAS) are being developed to minimize
such errors by providing additional, crucial information to drivers. AV systems offering driver
assistance functions can transfer the driving tasks from the human to the Automated Driving
System (ADS) to minimize human error. The future of AVs has been studied from various
viewpoints, such as policy, society, public opinion, adoption, ethical issues, planning, and so forth
(Loeb et al. 2018, Ebnali et al. 2019, Nodjomian and Kockelman 2019, Motamedi et al. 2020).
Milakis et al (Milakis et al. 2017) reviewed AV-related studies and classified them into a three-
fold order. The first order of impacts focuses on issues related to travel time, travel cost, travel
Arvin, Khattak, Kamrani, Rios-Torres
3
choices, and road capacity. The second order of impacts focuses on vehicle ownership, vehicle
sharing, infrastructure, and land use. Finally, the third order of impacts is concerned with safety,
energy and fuel efficiency, public health, etc. Literature suggests that AVs would reduce vehicle
ownership (Fagnant and Kockelman 2014), travel time (Kesting et al. 2008, Li et al. 2013), parking
lots (Fagnant and Kockelman 2014), and emissions (Choi and Bae 2013, Fagnant and Kockelman
2014). It is also suggested that AVs would increase road capacity (Van Arem et al. 2006, Shladover
et al. 2012, Mahdavian et al. 2019), traffic flow stability (Talebpour and Mahmassani 2016, Parsa
et al. 2020), vehicle miles traveled (Fagnant and Kockelman 2014, de Almeida Correia and van
Arem 2016), fuel efficiency (Asadi and Vahidi 2011, Rios-Torres and Malikopoulos 2017b, a, Liu
et al. 2020), and safety (Arvin et al. 2018, Virdi et al. 2019). However, these benefits are
unachievable unless AV penetration reaches a certain threshold. Moreover, this study concludes
that the impact of AVs on some aspects such as value of time, long-term energy consumption and
air pollution, equity, economy, and public health are still unclear. Furthermore, various factors are
contributing to the pace of adoption, such as public opinion, willingness to pay, cyber-security and
safety, and policy barriers.
A few studies have attempted to quantify potential safety impacts of AVs on the
transportation system. Microscopic simulation can help researchers evaluate the impact of varying
levels of AVs market penetration in a mixed traffic with conventional vehicles. Tibljaš et al.
(Deluka Tibljaš et al. 2018) attempted to quantify safety effects of CAVs at roundabouts using
VISSIM microsimulation and changing the Wiedemann 99 car-following model parameters to
emulate the following behavior of CAVs. Their analysis revealed that there might be a slight
increase in conflicts with introduction of CAVs. It is worth noting that the Wiedemann model that
they used for simulating CAV longitudinal control does not truly represent real-life CAV behavior.
Similarly, Stanek et al. (Stanek et al. 2018) simulated CAV behavior by adopting Wiedemann 99
model parameters from the literature and studied potential impacts. In another study, Rahman and
Abdel-Aty (Rahman and Abdel-Aty 2018) studied the impact of platooning AVs on expressways
using VISSIM software. Surrogate safety measures are used to quantify crash risk, and the results
reveal that forming platoons can reduce time to collision between 19% to 28%. Papadoulis et al.
(Papadoulis et al. 2019) also studied the safety impact of CAVs on motorways by analyzing time
to collision using their developed algorithm, however, the model is not calibrated using real-world
CAV data. Li et al (Li et al. 2017) studied the impact of CACC vehicles on rear-end collision risk
with different market penetrations at freeways. By reviewing the literature, it can be inferred that
the limitation of the aforementioned studies is that they are focusing on roadway segments, while
the interaction of CAVs in a mixed traffic environment at intersections remains unknown.
In the literature, different car-following models are developed to simulate the longitudinal
and lateral movements of AVs. The ACC car-following model is one of the most well-known
models used in the literature to emulate longitudinal following behavior of AVs. The ACC vehicles
are able to perform longitudinal control of the vehicle by maintaining the desired speed and gap
without any input from the human driver. By combining automated driving (ACC system) and
wireless communication, CACC models further enhance AV’s performance by simultaneously
responding to the speed and acceleration changes of the leader (Shladover et al. 2015, Mahdinia
et al. 2020). Literature has introduced several ACC and CACC frameworks to model AVs behavior
(Shladover et al. 2012, Milanés and Shladover 2014, Amoozadeh et al. 2015, Xiao et al. 2017),
and their widely explored potential benefits in the mixed traffic of conventional vehicles on the
mobility in terms of capacity (Shladover et al. 2012, Olia et al. 2018), traffic flow stability
(Milanés and Shladover 2014, Talebpour and Mahmassani 2016), and desired headway
(Nowakowski et al. 2011). Furthermore, additional potential mobility impacts of CAVs via
Arvin, Khattak, Kamrani, Rios-Torres
4
platooning (Amoozadeh et al. 2015, Gong et al. 2019), merging areas (Rios-Torres and
Malikopoulos 2017a, Liu et al. 2018b), and intersections and traffic signals (Ala et al. 2016, Sun
et al. 2017, Almannaa et al. 2019, Liu et al. 2019) are discussed in the literature.
While the potential mobility benefits of AVs are widely investigated in the literature, the
safety impacts of AVs in the mixed traffic of conventional vehicles remain unknown, specifically
at intersections. This study attempts to develop an understanding of the interaction of AVs and
conventional vehicles in a mixed traffic environment at intersections, with and without
communication between AVs. Notably, the ACC and CACC models developed by (Milanés and
Shladover 2014, 2016, Xiao et al. 2017, Liu et al. 2018a) are utilized to simulate driving behavior
of AVs. It worth noting that the model is validated in a real-world testbed and can truly represent
the driving behavior of AVs in the simulation. The main goal of this study is to build the behavioral
framework to simulate the interaction of conventional, ACC, and CACC vehicles, and to explore
the impact of AVs on the safety performance at intersections by defining and simulating multiple
scenarios. In this study, new and unique CV data obtained from the Safety Pilot Model
Development (SPMD) collected in Ann Arbor, MI, is used and integrated into the simulation to
accurately simulate conventional vehicles driving behavior. The safety performance of an
intersection is analyzed using different safety metrics considering various penetration rates of AVs
in mixed traffic.
METHODOLOGY
Traffic micro-simulation models are widely used to assess the impacts of new technologies on
traffic flow, road system, and society in general (Esfahani and Song 2019, Sheng et al. 2019, Wu
et al. 2019, Abdulsattar et al. 2020). This study investigates the safety impact of automated
vehicles in a mixed traffic environment involving conventional vehicles. In order to include a
better representation of the driver behavior of conventional vehicles, the real-world connected
vehicle data from the SPMD in Ann Arbor, MI, was used to modify the Wiedemann car-following
model and calibrate the simulation. In order to calibrate the simulation with conventional vehicles,
the cautiousness and situational awareness coefficients are tuned to reflect the observed driving
volatilities at the intersection in the real-world condition. To model the AVs, the ACC and CACC
models developed by (Milanés and Shladover 2014, 2016, Xiao et al. 2017, Liu et al. 2018a) were
used. Next, to evaluate the safety performance of the intersection under mixed traffic, different
scenarios were defined in the simulation considering different market penetrations of AVs. Two
surrogate safety measures are used to evaluate the safety performance of a representative
intersection under different market penetration rate of CAVs: the number of longitudinal conflicts
and driving volatility. Furthermore, mobility impact of AVs in a mixed traffic is explored.
Evidence in AV crashes
In September of 2014, California state allowed permit-holding companies to test their AVs on the
public roadways, including freeways, highways, and local streets. As of May 2020, more than 60
permit-holders are testing their automated vehicles in public roads of California within a mixed
traffic environment containing human-driven vehicles (Boggs et al. 2020). Although AVs are
currently in the prototype stage, analyzing the mechanism of the crashes will provide insight to the
potential issues of AVs in a mixed traffic environment. According to the California Vehicle Code
Section 38750, manufacturers are required to report all AV crashes to the California Department
of Motor Vehicles (DMV 2017). From September 2014 to May 2020, there were 243 crash reports
involving an AV, and these are used as a database for analysis in this section. It is worth noting
that significant effort was taken to manually extract this information from crash report files and
narratives. Descriptive statistics for the crash reports are illustrated in Table 1. Among these 243
Arvin, Khattak, Kamrani, Rios-Torres
5
crashes, 72.8 percent (177 crashes) occurred at the intersections, and 27.1 percent at other locations
(66 crashes). It can be inferred that intersections are the most dangerous location due to interrupted
traffic flow, and AVs have higher chance of being involved in a crash at intersections. Referring
to the type of crashes at intersections, 63.8 percent (155 crashes) were rear-end, 7.8 percent (19
crashes) were angle crashes, 21.0 percent (51 crashes) were sideswipe, and 7.4 percent (18 crashes)
are other types of crashes. It is worth noting that rear-end crashes are the dominant crash type in
AV-involved crashes. Further details can be found in Table 1. While the results show that
longitudinal conflicts at intersections are the dominant type of conflicts between AVs and
conventional vehicles, it should be noted that these AVs are prototypes and their behavior is timid.
Further evaluation is required in the future when ADS has matured, and the AVs are fully
commercialized.
Table 1 Summary of California State crash reports from September 2014 to May 2020
Variable
Category
Freq.
Percent
Location
Intersection
177
72.8%
Other
66
27.1%
Crash type
Rear-end
155
63.8%
Angle
19
7.8%
Sideswipe
51
21.0%
Other
18
7.4%
Severity
Property-damage-only
199
81.9%
Minor injuries
44
18.1%
Number of involved vehicles
1
18
7.4%
2
221
90.9%
3
4
1.6%
Time of day
Day
175
72.0%
Night
60
24.7%
Unknown
8
3.3%
Car-Following Model
Automated Vehicles
In this paper, ACC and CACC models are utilized to simulate the following behavior of AVs. In
the following, details of each model are provided.
ACC Car-Following Model
The currently available ACC systems enable vehicles to follow a front vehicle by transferring the
control task of throttle and brake to the ADS. Specifically, range sensors (i.e. radar, camera, and
lidar) enable the ACC vehicle to measure the relative distance and rate of change of the distance
of a front vehicle. The ACC system enables the vehicle to adjust its speed to the preceding vehicle
considering the desired headway. It should be noted that in the absence of leader, the ACC system
will follow a desired speed that is specified by the user.
Arvin, Khattak, Kamrani, Rios-Torres
6
This study utilizes the ACC car-following model developed and validated with real-world
AV data (Milanés and Shladover 2014, 2016, Xiao et al. 2017, Liu et al. 2018a). The developed
control algorithm in this model relies on four modes depending on the motion purposes: 1- speed
control; 2- gap control; 3- gap-closing control; and 4- collision avoidance mode. The latter model
is developed by the Transition Areas for Infrastructure-Assisted Driving (TransAID) team (Mintsis
2018), which tries to avoid rear-end collisions between vehicles. Details on each control mode are
discussed.
Speed control: This mode is activated when there is not a leader in front of the subject vehicle, or
when the leading vehicle is outside of the detection range of the radar, which is considered 120 m
(Xiao et al. 2017, Liu et al. 2018a). This mode maintains the vehicle’s speed close to the desired
speed. The speed control mode recommends acceleration of vehicle i at time k+1:
where is the desired speed, is the current speed of the vehicle i at time k; and is the
control gain, which determines the rate of speed deviation for acceleration. This value is suggested
to be 0.3-0.4 by (Xiao et al. 2017). In this study, a value of 0.4 is selected.
Gap control: The goal of the gap control mode is to keep the desired gap between the ACC and
front vehicle (Shladover et al. 2012). When this mode is activated, the acceleration at time k+1
can be written as a second-order function based on the speed and gap error with the leading vehicle:
where is the gap error at time k; and are the control gain parameters for positioning and
speed deviations. According to Xiao et al. (Xiao et al. 2017), the proposed values for these
parameters can be: and . Furthermore, according to Milanés and
Shladover (Milanés and Shladover 2014, 2016), the gap error can be written as:
where and denote the position of the following and leader vehicles at the current time
k, and is the desired time gap for the ACC system.
Gap-closing control: The study by Milanes and Shladover (Milanés and Shladover 2016)
developed the gap-closing mode to model the following behavior of AVs when distance with the
front vehicle is lower than 100 m. However, this mode was not fully modeled in this study and
Xiao et al (Xiao et al. 2017) tuned the gap-closing control mode by setting parameters of equation
(2) as and . It is worth noting that if the gap between the vehicles lies
between 100 m and 120 m, the control mode returns to the previous strategy to guarantee a smooth
transition between two control modes.
Collision avoidance control: This control mode is introduced by the TransAID team and added to
the ACC car-following model to avoid rear-end collisions. The nature of the collision avoidance
control is similar to the gap controller, but is activated when the gap error with the leader is
negative and the speed deviation is smaller than 0.1 m/s. In this manner, values of
and are suggested to guarantee a hard break for safe operation of ACC vehicles in
simulation even in critical events (Mintsis 2018).
Arvin, Khattak, Kamrani, Rios-Torres
7
CACC Car-Following Model
CACC systems extend the functionality of ACC systems by exploiting information generated by
surrounding vehicles via V2V communication. Therefore, it is expected that such an enhanced
system further improves transportation performance in terms of safety and mobility. In this study,
the CACC model is based on the algorithm that is verified in a real-world condition (Milanés and
Shladover 2014, 2016, Xiao et al. 2017, Liu et al. 2018a). Similar to the ACC model, it has four
control phases including: 1- speed control; 2- gap control; 3- gap-closing control; and 4- collision
avoidance mode.
The speed control is similar to the ACC model, as the control aims to match the vehicle’s
speed to the desired speed. It should be noted that transmitted information does not affect the
vehicle’s cruising mode. The acceleration in the speed control mode can be obtained as:
where is the speed control gain and is equal to 0.4 (Milanés and Shladover 2014).
The gap control of the CACC model is a first-order transfer function, which can be written
as:
where is the gap error and is the first derivative of the gap error. The values of and
are suggested as 0.45 s-2 and 0.0125 s-1, respectively (Liu et al. 2018a). Furthermore, the derivative
of the gap error can be defined as:
where is the acceleration at time k, and is the CACC controller defined desired time gap.
The collision avoidance mode of the CACC system is designed to avoid rear-end collisions
in the simulation. The controller’s logic is similar to the gap control mode, and the difference is
the gain values which are and (Mintsis 2018).
Human-driven Vehicles
The Wiedemann car-following model was used to model human driving behavior for conventional
vehicles. This psychophysical model introduced by Rainer Wiedemann in 1974 has been widely
used by researchers to model car-following (Wiedemann 1974). To model the driving behavior of
the following vehicle, Wiedemann considered four regimes of driving:
1- Free-driving regime: This regime illustrates the condition in which the behavior of the
vehicle is not affected by the lead vehicle since the clearance gap with it is more than 150
meters. In this regime, the driver just reaches and maintains the desired speed.
2- Closing-in regime: In this regime, the distance with the leader vehicle is less than 150
meters and the driver perceives that he/she is approaching the leader due to higher speed
and attempts to adjust it to maintain a safety gap.
3- Following regime: In this regime, the subject vehicle attempts to maintain a desired
headway with the front vehicle by adjusting the acceleration/deceleration.
4- Emergency regime: In this regime, the clearance gap between the leader and following
vehicle is smaller than the desired distance gap and the following vehicle decelerates with
his/her maximum capability to avoid crash with the leading vehicle.
Arvin, Khattak, Kamrani, Rios-Torres
8
Figure 1 illustrates the regimes and thresholds of the Wiedemann model, which are defined
and utilized according to the relative speed and distance between the following and leading vehicle
(Wiedemann 1974). The thresholds can be defined as:
SDV: The threshold at which the driver realizes speed difference
CLDV: The threshold where the driver realizes positive speed differences where the gap is
decreasing.
OPDV: The threshold at which the driver realizes negative speed differences where the gap
is increasing.
AX: The desired distance gap between stationary vehicles
ABX: The desired minimum following distance at low relative speeds
SDX: The maximum following distance.
Figure 1 Thresholds and driving regimes in the Wiedemann model
Various formulations of the Wiedemann model are presented by researchers. In this paper,
we used the formulation suggested by Werner (Werner 2010), which defines the thresholds as:
Arvin, Khattak, Kamrani, Rios-Torres
9
where:
AX: desired minimum distance between two stationary vehicles
ABX: desired minimum following distance
SDX: maximum desired following distance
SDV: perception threshold
CLDV: threshold at which the driver starts to react
OPDV: threshold at which the driver perceives speed difference
L: vehicle length
α: Driver’s cautiousness
β: Driver’s situational awareness
rand: random number with mean 0.5 and standard deviation 0.15
Safety Surrogate Performance Measures
In the microsimulation, the environment is collision free because the vehicles follow a pre-defined
behavior set known as a car-following model. However, it is still possible to obtain insightful
safety information regarding the relative changes in the system, especially when trying to predict
future states. Therefore, in this paper, the safety impact of AVs is evaluated utilizing the concepts
of number of conflicts and driving volatility, which has been shown to be highly correlated with
the crash frequency at intersections (Kamrani et al. 2018b, Arvin et al. 2019b).
Number of Conflicts
A common measure to predict collisions is the number of observed conflicts. Amundsen and
Hyden (Amundsen and Hydén 1977) stated that crashes can be considered as a subset of conflicts.
They have shown that the number of conflicts is highly correlated with crash frequency at
intersections (Dijkstra et al. 2010). In the literature, one of the most common approaches for
quantifying the number of conflicts is time to collision (TTC), which can be written as
(Saccomanno et al. 2008):
where , and are the positions and speed of the leader, and are the speed and speed
of the following vehicles, is the vehicle length, and t refers to the time. In this study, a TTC
lower than 0.5 seconds was considered as a serious conflict.
Driving Volatility
Recently, the concept of location-based driving volatility was introduced and utilized to quantify
instantaneous variations in driving behavior in a particular geographic area (Kamrani et al. 2018b,
Arvin et al. 2019b), or an event (Kamrani et al. 2018a, Wali et al. 2019, Wali and Khattak 2020).
Various measures of volatility have previously been introduced (Kamrani et al. 2018b) and it was
found that speed volatility is highly associated with the crash frequency at intersections (Kamrani
et al. 2018b, Arvin et al. 2019b). In this paper, we used this measure as a surrogate safety index.
This study utilized the speed volatility which intends to identify the times that the drivers showed
volatile driving behavior in speed by defining a threshold and counting the number of outliers.
Arvin, Khattak, Kamrani, Rios-Torres
10
Therefore, we can write:
Where k is the number of times that observed speeds that lie beyond the defined threshold, and
n is the total number of observations. The threshold is defined as:
where is the average speed of the vehicles passing the intersection, and is the standard
deviation of the observed speeds (Kamrani et al. 2018b, Hoseinzadeh et al. 2020). Notably, the
simulated speed samples during the times that vehicles stopped at the intersection (zero speeds)
are removed for the volatility calculation. Further details can be found in a related study (Kamrani
et al. 2018b).
Simulation setup and calibration
This study utilized SUMO (Simulation of Urban Mobility) open-source microsimulation software
(Behrisch et al. 2011). In order to evaluate the pre-defined scenarios involving multiple degrees of
automation, the intersection of Huron parkway and Washtenaw Avenue in Ann Arbor, Michigan
was selected (Figure 2, right). We have simulated 600 ft in each direction of the intersection, and
intersection territory is defined as 250 ft from the intersection center (Figure 2, left), which is
recommended by the Highway Safety Manual for the intersection safety analysis (2010). Since
intersection characteristics are highly correlated with the safety outcome (Kamrani et al. 2018b,
Wali et al. 2018, Arvin et al. 2019b), this information is retrieved from the Metropolitan Planning
Organization Website (http://semcog.org/). In particular, the speed limit, average annual daily
traffic (AADT), and number of lanes were used in the simulation.
Figure 2 - Intersection of Washtenaw Ave. and Huron Parkway in Ann Arbor, MI. Study
area shown by yellow boundaries
Model Calibration
In order to calibrate the micro-simulation to generate reasonable approximation of real-world
traffic conditions at the selected intersection, the following steps were taken:
Arvin, Khattak, Kamrani, Rios-Torres
11
1- The Wiedemann car-following model was modified to generate more realistic acceleration
and speed patterns.
2- The simulation is calibrated by incorporating safety measures for the conventional vehicles
(base scenario).
In the following, the details for each step will be discussed.
Modifying the Wiedemann Car-following model
Realistic acceleration limits were extracted from a plot of real-world acceleration versus speed to
obtain accurate behavior trends in the simulation. The emergence of big data from connected
vehicles provides enriched data which enabled us to compare the real-world data with simulation
data. The data for this study was retrieved from the Intelligent Transportation System DataHub
website (https://www.its.dot.gov/data/), collected in Ann Arbor, Michigan, utilizing more than
2800 connected vehicles operating under the real-world condition. The vehicles transmitted
instantaneous vehicle information including their speed, acceleration, and coordinates at the
frequency of 10 Hz via Basic Safety Messages (BSM). Using the coordinates of the intersection,
along with the vehicle speed and acceleration, observations from more than 2300 vehicles passing
through the intersection were extracted (n = 6,911,987 BSMs). It should be mentioned that the
extracted data is also used for the model calibration (see following subsection).
Figure 3 (top) provides the plot of acceleration versus speed for the real-world CV data
passing the selected intersection. As discussed in the literature (Arvin et al. 2018, Kamrani et al.
2018b), it can be observed that with an increase in the vehicle speed, the acceleration and
deceleration capability of the vehicles decline. We compared this result with the simulation output
of SUMO for the conventional vehicles following the default Wiedemann model in the software
and the results are provided in Figure 3 (left). Comparing the results, it can be observed that the
plot of speed versus acceleration for the default Wiedemann model does not represent the real-
world behavior. The acceleration regime of vehicles showing a constant value and the deceleration
regime does not reflect the real-world trend. This is due to the coding of the acceleration regime
of the Wiedemann car-following model in SUMO that does not consider the fact that vehicles at
high speeds are not able to accelerate and decelerate at their maximum capability. In order to
overcome this issue, the acceleration/deceleration regimes of the Wiedemann model is modified
by revising the boundaries of maximum acceleration/deceleration of the vehicles. Two
polynomials were fitted to the real-world data and implemented in the software (as shown in Figure
3). The fitted equations are formulated as:
(16)
(17)
The preceding two equations were used in the Wiedemann model in SUMO and after
running a baseline simulation, the speed versus acceleration plot was generated again (Figure 3
right). Comparing the new results with the real-world data, it can be observed that the modified
Wiedemann model significantly improved the representation of the speed versus acceleration
pattern. In order to compare the following behavior of vehicles before and after modification, a
sample driving profile for three following vehicles is provided in Figure 4. The top two figures
illustrate the speed and acceleration trajectories of vehicles following default model, and the two
Arvin, Khattak, Kamrani, Rios-Torres
12
bottom figures are the revised model. The acceleration/deceleration rates of vehicles are
unrealistically high in the original Wiedemann model, it can be inferred that the original model
does not truly simulate real-life behavior. The modified model is a better representation of real-
world following behavior.
Figure 3 Vehicle speed vs. Acceleration plot of real-world connected vehicle observations
(top, N = 6,911,987); default Wiedemann car-following model used in SUMO (bottom, left),
and modified Wiedemann car-following model (bottom, right)
Arvin, Khattak, Kamrani, Rios-Torres
13
Note: Blue, green, and red vehicles represent trajectories of three preceding vehicles passing through the study area
Figure 4 Speed and acceleration profiles of vehicles passing through the default and modified
Wiedemann models in SUMO
To ensure the generated speeds from the simulation accurately represent real-world data,
the speed density of the simulation results with the modified Wiedemann and the real-data are
plotted and shown in Figure 5. It should be mentioned that due to the difference in signal timing
of real-data and simulation, the zero speeds are removed from our analysis. In addition, the
descriptive statistics of the simulation output and real-data is provided in Table 2. Comparison of
the density plots and the descriptive statistics show that the speed generated by the simulation is
representative of the real-world traffic condition.
Table 2 Descriptive statistics of vehicle speeds passing through intersections for real
CV data and in SUMO simulation
Number of
observations
Mean
Standard
deviation
Min*
Max
Real-Life CV
data
6,911,987
7.68
4.7
0.1
35.2
Simulation
(SUMO)
2,700,491
7.50
4.2
0.1
24.6
Difference
-
-2.34%
10.6%
-
29.8%
*Note: Since the signal timing for connected vehicle real-world data and the SUMO simulation
data are different, speeds lower than 0.1 m/s are removed from the analysis.
Arvin, Khattak, Kamrani, Rios-Torres
14
Calibrating the simulation for the conventional vehicles (Base case study)
In the previous steps, the model was modified and verified based on the speed and acceleration
output of the simulation, in this step, the simulation was calibrated using the driving volatility as
the surrogate safety performance of the intersection. As discussed before, there are two parameters
in the Wiedemann model which need to be calibrated in SUMO (their default value is 0.5). The α
and β parameters were calibrated based on the aforementioned driving volatility measures at the
intersection. The calibration process leads to α value of 0.08 and β value of 0.4. The speed volatility
measure for the connected vehicles passing the study area is 7.11, and simulated vehicles at the
intersection shows volatility of 6.88, which shows 3.23% difference and it is acceptable.
Previous studies (Ahmed 1999, Kamrani et al. 2014) presented an equation to estimate the
minimum number of simulation runs required to ensure that simulation results meet the desired
confidence interval:
where is the number of simulation runs, is the standard deviation of the simulation output,
is the t value with m-1 degree of freedom and
two-tail confidence interval,
is the mean of the simulation output, and is the percentage error. In this study, a 95 percent
confidence interval is accepted. The calibration procedure leads to the value of α to equal 0.08,
and the value of β equal to 0.4. Based on the equation (18), is lower than the number of required
simulation runs, indicating that with 95 percent confidence interval, the generated volatilities in
the SUMO simulation is similar to the real-world connected vehicle conditions.
RESULTS
Definition of Scenarios
In order to explore the safety impact of AVs in mixed traffic with conventional vehicles, this study
Figure 5 Comparison of density plot of the speed for the SUMO simulation and real-world CV data
Speed (m/s)
Speed (m/s)
Density
Density
Real Data
Simulation
Arvin, Khattak, Kamrani, Rios-Torres
15
considered two set of scenarios. The first set of scenarios focuses on different market penetration
rates of AVs with no coordination, following ACC car-following model. The penetration rate of
ACC vehicles ranges from zero to 100% in 10% increments. The second scenario considers
coordination of AVs assuming that there is a V2V communication between AVs and the vehicles
are following CACC car-following model. If the front vehicle is conventional vehicle or the
distance is higher than 120 m, the CACC model downgrades to the ACC mode. It should be noted
that the vehicle platooning is not considered in this study due to a limited study area that does not
allow for the creation of vehicle platoons.
Prior to running the simulations, the intersection was empty and a one hour warm up is
considered. The simulation was run for the length of one hour, with a step-length of 0.1 second.
We considered homogeneous traffic, i.e., all vehicles are assumed to have the same physical
characteristics. To model the lane-changing behavior of all vehicle types, we used the default
model in SUMO (Erdmann 2015). For each scenario, the simulation was run 20 times, and the
safety performance of scenarios in terms of driving volatility and number of conflicts was assessed.
Simulation results
Safety performance measures
Number of conflicts
This section discusses the number of conflicts in the study area to quantify the safety performance
of the system under different penetration rates of AVs, with and without cooperation of AVs. For
each simulation, 20 runs with random seeds were performed, and each conflict is considered the
time that the TTC of two vehicles reaches a 0.5 second threshold. It should be noticed that a conflict
is measured as a unique incident in which TTC between two vehicles was below 0.5 seconds,
regardless of its duration. In other words, we are counting unique conflicts between consecutive
vehicles. The simulation results for 10% increments in AVs market penetration are provided in
Figure 6. Referring to the figure, it can be observed at current state with zero penetration of AVs,
number of potential conflicts is 9.94 for one hour of simulation.
In scenario 1, where AVs have no coordination with each other, a slight increase in number
of conflicts is observed at 10% market penetration compared to the baseline. However, with the
higher increase in AVs’ penetration rate, this reduction is more substantial and significant. Finally,
at 100% penetration of AVs, the number of conflicts reached zero. It can be inferred that a uniform
AV response can reduce the relative speed of following vehicles, which leads to an increase in
TTC, which then decreases the number of conflicts. Furthermore, by decreasing the penetration
rate of conventional vehicles, the frequency of interaction between conventional vehicles will be
reduced, which will naturally decrease the likelihood of longitudinal conflicts between human-
driven vehicles.
On the other hand, by adding coordination to AVs (i.e. CACC vehicles), AVs are capable
of adjusting their behavior in terms of speed with the front vehicle by receiving trajectory
information via V2V communication. Based on the results, this coordination capability can further
improve safety by decreasing the number of conflicts at the study area. By comparing the ACC
and CACC at each penetration rate, it can be observed that there is a substantial reduction in
number of conflicts for CACC compared to ACC, which indicates the importance of cooperation
for additional improvement in safety performance.
Overall, the results indicate that holding the traffic demand constant, the increase in the
market penetration of AVs will decrease the number of conflicts. However, this reduction is not
linear and is not substantial. The results revealed that at lower market penetration, the
improvements are not as significant as reductions in high penetration of AVs (i.e. +20 percent).
Arvin, Khattak, Kamrani, Rios-Torres
16
The results are consistent with the findings of Virdi et. al (Virdi et al. 2019) which has shown that
the number of CAV conflicts at intersections might not be substantial at low penetrations of CAVs,
while an increase in CAV market penetration, will cause more improvements in term of conflicts.
Figure 6 – Number of conflicts observed at simulated intersection and confidence intervals
in the study area, in 10% increments of AV market penetration
Driving volatility
Previous studies has shown that driving volatility is highly associated with risk and severity of
crashes (Arvin et al. 2019a, Kamrani et al. 2019). As discussed in the calibration section, the
baseline scenario (zero market penetration of AVs) is calibrated for the driving volatility in order
to truly represent safety performance of the vehicles in the simulation. Figure 7 provides the
driving volatility for the different penetration rates of ACC and CACC vehicles ranging from zero
to 100%. The red line illustrates the ACC, and the blue line indicates CACC vehicles. Based on
the results, with 100% penetration of ACC and CACC, a significant drop in driving volatility can
be observed, and for the mixed traffic scenarios, this reduction is nonlinear and different for ACC
and CACC vehicles.
An increase in the market penetration (up to 40%) for ACC vehicles in mixed traffic is
associated with a slight increase in speed volatility. This increase in volatility could be associated
with the instability of the ACC model when multiple ACC vehicles are driving consecutively
(Seiler et al. 2004, Milanés et al. 2013). Increasing the penetration rate of ACC from 40% to 60%
is associated with a substantial decrease in speed volatility. However, moving from 60% to 80%
market penetration contributes to a slight increase in speed volatility. Finally, there is a significant
reduction in volatility with more than 80% ACC penetration, and the speed volatility reaches zero
Arvin, Khattak, Kamrani, Rios-Torres
17
when all vehicles follow the ACC model. Referring to the CACC vehicles, it can be observed that
by adding communication to AVs, the speed volatility drops substantially relative to the baseline,
and this reduction is much higher compared to the ACC vehicles. The speed volatility nonlinearly
dropped with the increase in market penetration of CACC vehicles. By comparing the speed
volatility and number of conflicts for the ACC and CACC vehicles, it can be inferred that speed
volatility is showing a similar pattern.
Figure 7 Speed volatility at simulated intersection for ACC vehicles and CACC vehicles
Overall, the safety performance measures indicate that the safety performance of the
intersection could substantially improve with a high market penetration of AVs that excludes
human drivers from the driving task. Both the number of conflicts and the driving volatility
substantially decreased, which might be due to uniform AV response which can reduce the relative
speed of the lead and following vehicles, that could improve safety by decreasing abrupt vehicular
movements. Furthermore, conventional vehicle human drivers would change their driving
behavior with the passage of time and increase in AVs penetration by learning the AVs response.
An analysis of AV involved crashes in California revealed that in almost all of the crashes, AVs
were hit by conventional vehicles, since the human driver did not expect the AV behavior (Arvin
et al. 2018). This learning process by humans could lead to further improvement in the safety
outcome of intersections by reducing the conflicts frequency and driving volatility. However, it is
also possible that drivers will game the limitations of AVs and behave more aggressively in their
vicinity, which might reduce the improvements.
Mobility Analysis
In order to evaluate AV performance in terms of mobility, average travel time and speed of
Arvin, Khattak, Kamrani, Rios-Torres
18
vehicles passing the intersection are considered. The average travel time of vehicles passing the
study area is calculated as:
where is the average travel time, n is the number of vehicles, and is the travel time of vehicle
i passing the study area. The average traveling speed of vehicles is obtained by averaging the mean
travel speed of vehicles passing the study area:
where is the average traveling speed of vehicles passing the intersection, and is the average
speed of vehicle i passing the intersection influence area. It is worth noting that for the mobility
analysis, all the simulation time intervals during which the vehicles were passing the intersection
was considered, including stoppage times at the intersection.
The results for the scenarios are summarized in Table 3. The results indicate that the
increase in the penetration rate of ACC and CACC vehicles significantly improved the mobility in
terms of travel time and average speed. The travel time decreased by 24.1% and 39% with ACC
market penetration of 50% and 100%, respectively. The cooperation between AVs can further
enhance the mobility performance. Compared to the baseline, the average travel time of passing
vehicles decreased by 27.1% and 43.4% for market penetration rates of CACC vehicles 50% and
100%, respectively.
Referring to the traveling speed, it can be inferred that the average speed of vehicles passing
through the intersection significantly increased relative to the baseline. With no cooperation
between AVs, 50% and 100% penetration rate of ACC vehicles can increase average speed by
48.3% and 71.7%, respectively. Enabling communication between AVs improved the average
speed of passing vehicles by 55.2% and 82.8% with 50% and 100% market penetration of CACC
vehicles compared to the baseline.
Arvin, Khattak, Kamrani, Rios-Torres
19
Table 3 – Average travel time and traveling speed of ACC and CACC vehicles in different
market penetration scenarios at simulated intersection
Pen.
Rate
Travel Time (sec)
Traveling Speed (m/s)
ACC
CACC
Diff of CACC
& ACC (%)***
ACC
CACC
Diff of CACC
& ACC (%)
0
Base*
base
-
base
base
-
10
88.56 (37.36)**
88.32 (38.94)
0.22
5.03 (2.79)
5.13 (2.98)
1.99
20
83.73 (35.07)
81.84 (32.81)
3.65
5.71 (2.96)
5.91 (2.98)
3.50
30
79.45 (32.02)
77.51 (31.77)
2.39
5.92 (3.06)
6.17 (3.11)
2.53
40
75.2 (31.54)
73.02 (30.92)
3.28
6.18 (3.35)
6.46 (3.52)
4.53
50
72.31 (30.93)
69.41 (30.88)
2.75
6.66 (3.52)
6.97 (2.94)
4.65
60
69.94 (30.46)
67.27 (30.34)
2.85
7.04 (3.79)
7.34 (4.02)
4.26
70
67.23 (30.50)
64.36 (30.65)
4.14
7.31 (4.03)
7.78 (4.21)
4.96
80
64.37 (30.18)
61.27 (30.34)
4.99
7.52 (4.15)
7.92 (4.46)
6.27
90
62.1 (29.91)
58.36 (30.18)
5.31
7.6 (4.37)
8.12 (4.57)
4.70
100
58.11 (38.48)
53.9 (29.9)
6.16
7.71 (4.48)
8.21 (4.74)
5.82
Note:
* At zero percent market penetration, the mean travel time and traveling speed for conventional vehicles
is 95.28 (SD=42.07) and 4.49 (SD=2.91), respectively.
** Numbers in parentheses show the standard deviations.
*** Percentage of improvements in travel time and speed gained by CACC vehicles compared to ACC
vehicles.
LIMITATIONS
This study investigates the safety impact of automated vehicles in mixed traffic, through
development of a simulation framework and integration of real-world connected vehicle data.
Though efforts have been made to integrate realistic measures and accurate calibration procedures,
the study still has some limitations. As discussed, this study considers identical ACC and CACC
models for all AVs, while real-life scenarios will contain multiple algorithms developed and
implemented by different manufacturers, and this variability is not considered in this paper. In
addition, the presented results are based on simulations which have well-known limitations, given
their hypothetical nature. The results are limited to the selected intersection and might not be
generalizable widely to other intersections with varying geometries and different traffic conditions.
CONCLUSIONS
The study aims to predict the future impact of connected and automated vehicles in mixed traffic
with conventional vehicles at an intersection, by integrating real-world connected vehicle data into
simulation. In order to simulate the longitudinal control of AVs, ACC model is utilized, and
cooperation and connectivity between AVs is considered using CACC model to represent CAV
vehicles. A modified Wiedemann car-following model was used to model the driving behavior of
conventional vehicles. Connected Vehicle (CV) data was obtained to calibrate the simulation from
a real-world testbed in Ann Arbor, Michigan. First, to reflect the speed-acceleration relationship
of real-world CV data, the Wiedemann car-following model was modified. Then, the simulation
was calibrated using driving volatility, as a proxy for the safety performance of the intersection,
and speed profile of vehicles passing the intersection. The safety outcome of the simulation is
quantified using two performance measures: 1- number of longitudinal conflicts, 2- driving
volatility. Numerous studies have used the number of conflicts to describe the safety outcome. On
Arvin, Khattak, Kamrani, Rios-Torres
20
the other hand, recent studies have revealed that driving volatility at intersections is highly
correlated with the number of crashes, which can be used as a surrogate safety performance
measure. In terms of mobility performance, average travel time and speed are utilized to study AV
impacts under different market penetration scenarios.
The interaction of conventional vehicles with ACC and CACC vehicles was studied by
investigating several scenarios considering different market penetration of AVs. The simulation
results revealed that at early stages of ACC vehicle adoption, the safety improvement might not
be substantial, while improvements are significant with +40 percent ACC vehicle market
penetration. The results indicate a nonlinear safety improvement, where at low market penetration,
there is not a substantial safety improvement, while with higher penetration rates of ACC vehicles
the number of conflicts and driving volatility drop substantially. On the other hand, by adding
cooperation between AVs, a significant safety improvement, in terms of reduction in number of
conflicts and driving volatility, is observed. This improvement might be due to faster response of
AVs and smoother driving behavior that could improve safety by decreasing abrupt vehicular
movements. Based on the results, ACC/CACC vehicles can improve mobility statistics in terms of
average speed and travel time at intersections.
On the other hand, AVs can respond to the leader vehicles faster due to vehicle automation,
and the uniform driving behavior could decrease the number of longitudinal conflicts and driving
volatility. In this study, it is assumed that the behavior of human-driven vehicles remains the same.
However, it is possible that human drivers might alter their behavior with the passage of time and
their exposure to the AVs. One possible scenario is with the familiarization of human drivers with
Automated Driving Systems (ADS), conflicts will further reduce. However, it is also possible that
human-drivers game the cautiousness and limitations of AVs and behave more aggressively in
their vicinity. Therefore, a longitudinal study can incorporate the learning process and behavior
adjustments of human drivers into the analysis. Furthermore, although different manufacturers will
use different ACC and CACC systems in practice, it is assumed that all the AVs are following the
same algorithm for longitudinal vehicle control and their performance is identical, similar to the
state of the art (Talebpour and Mahmassani 2016, Jeong et al. 2017, Papadoulis et al. 2019,
Rahman et al. 2019). Therefore, this variation in ADSs might increase the conflicts/volatility.
ACKNOWLEDGEMENT
This paper is based upon work supported by the Collaborative Sciences Center for Road Safety, a
U.S. DOT funded National Transportation Center, under the Project R-27, the Tennessee
Department of Transportation, and in part by the Laboratory Directed Research, Development
Program of the Oak Ridge National Laboratory, Oak Ridge, TN 37831 USA, managed by UT-
Battelle, LLC, for the DOE. Authors also benefited from discussions with Mr. Mani Amoozadeh
and his input is greatly appreciated. The authors thank Ms. Kinzee Clark and Mr. Zachary Jerome
for editing the manuscript.
Arvin, Khattak, Kamrani, Rios-Torres
21
REFERENCES
Aashto, 2010. Highway safety manual. Washington DC.
Abdulsattar, H., Mostafizi, A., Siam, M.R., Wang, H., 2020. Measuring the impacts of connected vehicles
on travel time reliability in a work zone environment: An agent-based approach. Journal of
Intelligent Transportation Systems 24 (5), 421-436.
Ahmed, K.I., 1999. Modeling drivers' acceleration and lane changing behavior. PhD Thesis,
Massachusetts Institute of Technology.
Ala, M.V., Yang, H., Rakha, H., 2016. Modeling evaluation of eco–cooperative adaptive cruise control in
vicinity of signalized intersections. Transportation Research Record: Journal of the
Transportation Research Board (2559), 108-119.
Almannaa, M.H., Chen, H., Rakha, H.A., Loulizi, A., El-Shawarby, I., 2019. Field implementation and
testing of an automated eco-cooperative adaptive cruise control system in the vicinity of
signalized intersections. Transportation research part D: transport environment 67, 244-262.
Amoozadeh, M., Deng, H., Chuah, C.-N., Zhang, H.M., Ghosal, D., 2015. Platoon management with
cooperative adaptive cruise control enabled by vanet. Vehicular communications 2 (2), 110-123.
Amundsen, F., Hydén, C., 1977. Proceedings of first workshop on traffic conflicts. Oslo, TTI, Oslo,
Norway and LTH Lund, Sweden.
Anon, 2008. National highway traffic safety administration. National motor vehicle crash causation
survey: Report to congress. National Highway Traffic Safety Administration Technical Report
DOT HS 811, 059.
Anon, 2018. Federal highway administration. Highway Statistics, 2016.
Arvin, R., Kamrani, M., Khattak, A., 2019a. The role of pre-crash driving instability in contributing to
crash intensity using naturalistic driving data. Accident Analysis & Prevention 132.
Arvin, R., Kamrani, M., Khattak, A.J., 2019b. How instantaneous driving behavior contributes to crashes
at intersections: Extracting useful information from connected vehicle message data. Accident
Analysis & Prevention 127, 118-133.
Arvin, R., Kamrani, M., Khattak, A.J., Rios-Torres, J., 2018. Safety impacts of automated vehicles in
mixed traffic. Transportation Research Board 97th Annual Meeting Washington DC.
Asadi, B., Vahidi, A., 2011. Predictive cruise control: Utilizing upcoming traffic signal information for
improving fuel economy and reducing trip time. IEEE transactions on control systems technology
19 (3), 707-714.
Behrisch, M., Bieker, L., Erdmann, J., Krajzewicz, D., Year. Sumo–simulation of urban mobility: An
overview. In: Proceedings of the Proceedings of SIMUL 2011, The Third International
Conference on Advances in System Simulation.
Boggs, A.M., Arvin, R., Khattak, A.J., 2020. Exploring the who, what, when, where, and why of
automated vehicle disengagements. Accident Analysis & Prevention 136, 105406.
Choi, J.-E., Bae, S.-H., 2013. Development of a methodology to demonstrate the environmental impact of
connected vehicles under lane-changing conditions. Simulation 89 (8), 964-976.
De Almeida Correia, G.H., Van Arem, B., 2016. Solving the user optimum privately owned automated
vehicles assignment problem (uo-poavap): A model to explore the impacts of self-driving
vehicles on urban mobility. Transportation Research Part B: Methodological 87, 64-88.
Deluka Tibljaš, A., Giuffrè, T., Surdonja, S., Trubia, S., 2018. Introduction of autonomous vehicles:
Roundabouts design and safety performance evaluation. Sustainability 10 (4), 1060.
Dijkstra, A., Marchesini, P., Bijleveld, F., Kars, V., Drolenga, H., Van Maarseveen, M., 2010. Do
calculated conflicts in microsimulation model predict number of crashes? Transportation
Research Record: Journal of the Transportation Research Board (2147), 105-112.
Dmv, 2017. Report of traffic accident involving an autonomous vehicle (ol 316). California DMV.
Ebnali, M., Hulme, K., Ebnali-Heidari, A., Mazloumi, A., 2019. How does training effect users’ attitudes
and skills needed for highly automated driving? Transportation research part F: traffic psychology
behaviour 66, 184-195.
Erdmann, J., 2015. Sumo’s lane-changing model. Modeling mobility with open data. Springer, pp. 105-
123.
Arvin, Khattak, Kamrani, Rios-Torres
22
Esfahani, H.N., Song, Z., 2019. A new method for microsimulation model calibration: A case study of i-
710. Civil and Environmental Engineering Faculty Publications.
Fagnant, D.J., Kockelman, K.M., 2014. The travel and environmental implications of shared autonomous
vehicles, using agent-based model scenarios. Transportation Research Part C: Emerging
Technologies 40, 1-13.
Gong, S., Zhou, A., Peeta, S., 2019. Cooperative adaptive cruise control for a platoon of connected and
autonomous vehicles considering dynamic information flow topology. Transportation research
record 2673 (10), 185-198.
Hoseinzadeh, N., Arvin, R., Khattak, A.J., Han, L.D., 2020. Integrating safety and mobility for
pathfinding using big data generated by connected vehicles. Journal of Intelligent Transportation
Systems, 1-17.
Jeong, E., Oh, C., Lee, S., 2017. Is vehicle automation enough to prevent crashes? Role of traffic
operations in automated driving environments for traffic safety. Accident Analysis & Prevention
104, 115-124.
Kamrani, M., Abadi, S.M.H.E., Golroudbary, S.R., 2014. Traffic simulation of two adjacent unsignalized
t-junctions during rush hours using arena software. Simulation Modelling Practice and Theory 49,
167-179.
Kamrani, M., Arvin, R., Khattak, A.J., 2018a. Analyzing highly volatile driving trips taken by alternative
fuel vehicles. Transportation Research Board 97th Annual Meeting Washington DC, United
States.
Kamrani, M., Arvin, R., Khattak, A.J., 2018b. Extracting useful information from basic safety message
data: An empirical study of driving volatility measures and crash frequency at intersections.
Transportation Research Record 2672 (38), 290-301.
Kamrani, M., Arvin, R., Khattak, A.J., 2019. The role of aggressive driving and speeding in road safety:
Insights from shrp2 naturalistic driving study data. Transportation Research Board 98th Annual
Meeting. Washington DC.
Kesting, A., Treiber, M., Schönhof, M., Helbing, D., 2008. Adaptive cruise control design for active
congestion avoidance. Transportation Research Part C: Emerging Technologies 16 (6), 668-683.
Li, Y., Wang, H., Wang, W., Xing, L., Liu, S., Wei, X., 2017. Evaluation of the impacts of cooperative
adaptive cruise control on reducing rear-end collision risks on freeways. Accident Analysis &
Prevention 98, 87-95.
Li, Z., Chitturi, M., Zheng, D., Bill, A., Noyce, D., 2013. Modeling reservation-based autonomous
intersection control in vissim. Transportation Research Record: Journal of the Transportation
Research Board (2381), 81-90.
Liu, H., Kan, D.K., Shladover, S., Lu, X.-Y., Wang, M., Schakel, W., Van Aren, B., 2018a. Using
cooperative adaptive cruise control (cacc) to form high-performance vehicle streams California
PATH Progam, Institute of Transportation Studies, University of California, Berkley
Liu, H., Kan, X., Shladover, S.E., Lu, X.-Y., Ferlis, R.E., 2018b. Impact of cooperative adaptive cruise
control on multilane freeway merge capacity. Journal of Intelligent Transportation Systems 22
(3), 263-275.
Liu, H., Lu, X.-Y., Shladover, S.E., 2019. Traffic signal control by leveraging cooperative adaptive cruise
control (cacc) vehicle platooning capabilities. Transportation research part C: emerging
technologies 104, 390-407.
Liu, H., Shladover, S.E., Lu, X.-Y., Kan, X., 2020. Freeway vehicle fuel efficiency improvement via
cooperative adaptive cruise control. Journal of Intelligent Transportation Systems, 1-13.
Loeb, B., Kockelman, K.M., Liu, J., 2018. Shared autonomous electric vehicle (saev) operations across
the austin, texas network with charging infrastructure decisions. Transportation Research Part C:
Emerging Technologies 89, 222-233.
Mahdavian, A., Shojaei, A., Oloufa, A., Year. Assessing the long-and mid-term effects of connected and
automated vehicles on highways’ traffic flow and capacity. In: Proceedings of the International
Conference on Sustainable Infrastructure 2019: Leading Resilient Communities through the 21st
Century, pp. 263-273.
Arvin, Khattak, Kamrani, Rios-Torres
23
Mahdinia, I., Arvin, R., Khattak, A.J., Ghiasi, A., 2020. Safety, energy, and emissions impacts of adaptive
cruise control and cooperative adaptive cruise control. Transportation Research Record,
0361198120918572.
Milakis, D., Van Arem, B., Van Wee, B., 2017. Policy and society related implications of automated
driving: A review of literature and directions for future research. Journal of Intelligent
Transportation Systems, 1-25.
Milanés, V., Shladover, S.E., 2014. Modeling cooperative and autonomous adaptive cruise control
dynamic responses using experimental data. Transportation Research Part C: Emerging
Technologies 48, 285-300.
Milanés, V., Shladover, S.E., 2016. Handling cut-in vehicles in strings of cooperative adaptive cruise
control vehicles. Journal of Intelligent Transportation Systems 20 (2), 178-191.
Milanés, V., Shladover, S.E., Spring, J., Nowakowski, C., Kawazoe, H., Nakamura, M., 2013.
Cooperative adaptive cruise control in real traffic situations. IEEE Transactions on intelligent
transportation systems 15 (1), 296-305.
Mintsis, E., 2018. Modelling, simulation and assessment of vehicle automations and automated vehicles’
driver behaviour in mixed traffic. TransAID.
Motamedi, S., Wang, P., Zhang, T., Chan, C.-Y., 2020. Acceptance of full driving automation: Personally
owned and shared-use concepts. Human factors 62 (2), 288-309.
Nodjomian, A.T., Kockelman, K., 2019. How does the built environment affect interest in the ownership
and use of self-driving vehicles? Journal of Transport Geography 78, 115-134.
Nowakowski, C., Shladover, S.E., Cody, D., Bu, F., O'connell, J., Spring, J., Dickey, S., Nelson, D.,
2011. Cooperative adaptive cruise control: Testing drivers' choices of following distances.
Olia, A., Razavi, S., Abdulhai, B., Abdelgawad, H., 2018. Traffic capacity implications of automated
vehicles mixed with regular vehicles. Journal of Intelligent Transportation Systems 22 (3), 244-
262.
Papadoulis, A., Quddus, M., Imprialou, M., 2019. Evaluating the safety impact of connected and
autonomous vehicles on motorways. Accident Analysis & Prevention 124, 12-22.
Parsa, A.B., Shabanpour, R., Mohammadian, A., Auld, J., Stephens, T., 2020. A data-driven approach to
characterize the impact of connected and autonomous vehicles on traffic flow. Transportation
Letters-The International Journal of Transportation Research.
Rahman, M.S., Abdel-Aty, M., 2018. Longitudinal safety evaluation of connected vehicles’ platooning on
expressways. Accident Analysis & Prevention 117, 381-391.
Rahman, M.S., Abdel-Aty, M., Lee, J., Rahman, M.H., 2019. Safety benefits of arterials’ crash risk under
connected and automated vehicles. Transportation Research Part C: Emerging Technologies 100,
354-371.
Rios-Torres, J., Malikopoulos, A.A., 2017a. Automated and cooperative vehicle merging at highway on-
ramps. IEEE Transactions on Intelligent Transportation Systems 18 (4), 780-789.
Rios-Torres, J., Malikopoulos, A.A., 2017b. A survey on the coordination of connected and automated
vehicles at intersections and merging at highway on-ramps. IEEE Transactions on Intelligent
Transportation Systems 18 (5), 1066-1077.
Saccomanno, F., Cunto, F., Guido, G., Vitale, A., 2008. Comparing safety at signalized intersections and
roundabouts using simulated rear-end conflicts. Transportation Research Record: Journal of the
Transportation Research Board (2078), 90-95.
Seiler, P., Pant, A., Hedrick, K., 2004. Disturbance propagation in vehicle strings. IEEE Transactions on
automatic control 49 (10), 1835-1842.
Sheng, S., Pakdamanian, E., Han, K., Kim, B., Tiwari, P., Kim, I., Feng, L., Year. A case study of trust on
autonomous driving. In: Proceedings of the 2019 IEEE Intelligent Transportation Systems
Conference (ITSC), pp. 4368-4373.
Shladover, S., Su, D., Lu, X.-Y., 2012. Impacts of cooperative adaptive cruise control on freeway traffic
flow. Transportation Research Record: Journal of the Transportation Research Board (2324), 63-
70.
Shladover, S.E., Nowakowski, C., Lu, X.-Y., Ferlis, R., 2015. Cooperative adaptive cruise control:
Arvin, Khattak, Kamrani, Rios-Torres
24
Definitions and operating concepts. Transportation Research Record 2489 (1), 145-152.
Singh, S., 2015. Critical reasons for crashes investigated in the national motor vehicle crash causation
survey. Traffic Safety Facts Crash Stats National Highway Traffic Safety Administration,
Washington, DC.
Stanek, D., Milam, R.T., Huang, E., Wang, Y.A., 2018. Measuring autonomous vehicle impacts on
congested networks using simulation. Transportation Research Board.
Sun, W., Zheng, J., Liu, H.X., 2017. A capacity maximization scheme for intersection management with
automated vehicles. Transportation research procedia 23, 121-136.
Talebpour, A., Mahmassani, H.S., 2016. Influence of connected and autonomous vehicles on traffic flow
stability and throughput. Transportation Research Part C: Emerging Technologies 71, 143-163.
Van Arem, B., Van Driel, C.J., Visser, R., 2006. The impact of cooperative adaptive cruise control on
traffic-flow characteristics. IEEE Transactions on Intelligent Transportation Systems 7 (4), 429-
436.
Virdi, N., Grzybowska, H., Waller, S.T., Dixit, V., 2019. A safety assessment of mixed fleets with
connected and autonomous vehicles using the surrogate safety assessment module. Accident
Analysis & Prevention 131, 95-111.
Wali, B., Khattak, A.J., 2020. Harnessing ambient sensing & naturalistic driving systems to understand
links between driving volatility and crash propensity in school zones–a generalized hierarchical
mixed logit framework. Transportation Research Part C: Emerging Technologies 114, 405-424.
Wali, B., Khattak, A.J., Bozdogan, H., Kamrani, M., 2018. How is driving volatility related to
intersection safety? A bayesian heterogeneity-based analysis of instrumented vehicles data.
Transportation Research Part C: Emerging Technologies 92, 504-524.
Wali, B., Khattak, A.J., Karnowski, T., 2019. Exploring microscopic driving volatility in naturalistic
driving environment prior to involvement in safety critical events—concept of event-based
driving volatility. Accident Analysis & Prevention 132, 105277.
Werner, C., 2010. Integration von fahrzeugfolge-und fahrstreifenwechselmodellen in nachtfahrtsimulation
luciddrive. Master's thesis, Universität der Informationsgesellschaft, Paderborn.
Wiedemann, R., 1974. Simulation des strassenverkehrsflusses. Schriftenreihe des Instituts für
Verkehrswesen der Universität Karlsruhe, Band 8, Karlsruhe, Germany.
Wu, Y., Abdel-Aty, M., Wang, L., Rahman, M.S., 2019. Combined connected vehicles and variable speed
limit strategies to reduce rear-end crash risk under fog conditions. Journal of Intelligent
Transportation Systems, 1-20.
Xiao, L., Wang, M., Van Arem, B., 2017. Realistic car-following models for microscopic simulation of
adaptive and cooperative adaptive cruise control vehicles. Transportation Research Record 2623
(1), 1-9.