Simulation-Based Evaluation of Cooperative Maneuver Coordination and Its Impact on Traffic Quality

Abstract

Original available at SAE Mobilus _____ Cooperative Maneuver Coordination (CMC) is one of the cornerstone technologies on the way to automated and connected driving of the future. The goal of this technology is to increase traffic safety and efficiency, by solving potential conflict situations on the road, based on cooperative maneuver planning, negotiation and decision-making, with the aid of inter-vehicular communication. This innovative topic represents a wide area for research projects, e.g., such as German funded project IMAGinE. A variety of different approaches for CMC has already been implemented and evaluated in the related work of the recent past. However, due to the high system complexity in general and vast testing effort in particular of these solutions, their actual impact on the traffic quality has not yet been extensively addressed, and therefore must be further investigated. In order to fulfill this task, one needs a methodology with appropriate evaluation metrics alongside with a suitable testing environment, in order to obtain comprehensive results. The scientific contribution of the work at hand involves a simulation-based evaluation methodology for CMC. For this, we will propose a novel CMC algorithm, which is built upon a direct trajectory exchange via inter-vehicular communication and a decentralized decision-making process, suitable for both manually operated as well as autonomous vehicles. In order to examine this algorithm, we will introduce a co-simulation environment with automatic scenario generation and multi-instances capability, which consists of a coupled simulation of traffic flow and vehicle dynamics. Eventually, we will present a set of metrics, which we determined in order to evaluate effectiveness and efficiency of our CMC algorithm, considering its impact on various aspects of the traffic quality.

Full text

2021-01-0171 Published 06 Apr 2021
3159
Simulation-Based Evaluation of Cooperative
Maneuver Coordination and Its Impact on
Trac Quality
Viktor Lizenberg, Daniel Bischo, Youssef Haridy, and Ulrich Eberle Opel Automobile GmbH
Steen Knapp Saarland University of Applied Sciences
Frank Koester German Aerospace Center
Citation: Lizenberg, V., Bischoff, D., Haridy, Y., Eberle, U. et al., “Simulation-Based Evaluation of Cooperative Maneuver Coordination
and Its Impact on Traffic Quality,” SAE Int. J. Advances & Curr. Prac. in Mobility 3(6):3159-3169, 2021, doi:10.4271/2021-01-0171.
This article was presented at the WCX World Congress Experience Digital Summit, April 13-15, 2021.
Abstract
Cooperative Maneuver Coordination (CMC) is one of the
cornerstone technologies on the way to automated and
connected driving of the future. e goal of this tech-
nology is to increas e trac safety a nd eciency, by solving poten-
tial con ict situations on the road, bas ed on cooperative maneuver
planni ng, negotiation and decision-making , with the aid of inter-
vehicu lar communic ation. is in novative topic represents a w ide
area for research projects, e.g., such as German funded project
IMAGinE. A v ariety of dierent approaches for CMC has already
been implemented and evaluated i n the related work of the recent
past. However, due to the high system complexity in general and
vast testing eort in particular of these solutions, their actual
impact on the t rac quality ha s not yet been exten sively addressed,
and therefore must befurther investigated. In order to fulll this
task, one needs a methodology with appropriate evaluation
metrics a longside with a suitable testing env ironment, in order to
obtain comprehensive results. e scientic contribution of the
work at hand i nvolves a simulation-ba sed evaluat ion methodology
for CMC. For this, wewil l propose a novel CMC algorithm, wh ich
is built upon a direct trajectory exchange via inter-vehicular
communication and a decentralized decision-making process,
suitable for both manually operated as well as autonomous
vehicles. In order to examine this algorithm, wewill introduce a
co-simulation environment with automatic scenario generation
and multi-insta nces capability, which consists of a coupled simula-
tion of trac ow and vehicle dynamics. Eventually, wewill
present a set of metrics, w hich wedetermined in order to evaluate
eectiveness and eciency of our CMC algorithm, considering
its impact on various aspects of the trac quality.
 Introduction
Vehicle-to-Everything (V2X) communication is paving
its way towards Intelligent Transportation Systems
(ITS) of the future by being gradually researched,
developed and standardized [1]. Supported by the V2X tech-
nology, the Cooperative Maneuver Coordination (CMC),
particularly where the vehicles exchange their driving trajec-
tories and, thus, perform cooperative maneuver planning,
negotiation as well as decision-making, has a high potential
to increase safety and eciency in various trac situations.
An exemplary trac situation on a highway entrance,
where the trajectory-based CMC between vehicles leads to a
successful cooperation, is illustrated in Figure 1. ere, at the
initial state t1, all vehicles follow their original trajectories
with no conicts. At t2, as soon as vehicle A intends to perform
a lane change, it broadcasts a new trajectory that causes a
conict with a trajectory of vehicle B. By waiving its right-of-
way, vehicle B may cooperate by decelerating, in order to
increase the gap for merging. At t3, aer vehicle B coopera-
tively adapted and transmitted its new trajectory, the conict
is resolved. At t4, vehicle A can follow its trajectory to merge in.
Nowadays, the development of cooperative driving func-
tions for automated and connected vehicles, which are carried
out through V2X communication and CMC algorithms,
represents an attractive eld of research for numerous projects.
One of them is notably the German funded project IMAGinE
[2], which features among others a consortium of prominent
car manufacturers and suppliers. Major objective of the
project IMAGinE is to elaborate a common (i.e., cross-
company) specication, continued by design, implementation
Keywords
Intelligent Transportation Systems (ITS), Vehicle-to-
Everything (V2X) Communication, Automated and
Connected Driving, Cooperative Maneuver Coordination
(CMC), X-in-the-Loop (XiL), Robot Operating System (ROS),
Traffic Quality, Co-Simulation, Automatic Scenario
Generation (ASG), Static and Dynamic Coupling, Long-Term
and Short-Term Scenarios, IMAGinE
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3160
and evaluation of state-of-the-art cooperative driving func-
tions based on CMC. Herewith, testing activities of these
functions are collaboratively performed in simulation as well
as in eld tests.
1.1. Motivation
In the development process of CMC, a big challenge is posed by
the in-depth testing and performance evaluation procedure, due
to the high complexit y of the involved systems and their, including
cooperative, interactions. In order to extensively test CMC in the
majority of relevant trac situations, an exceptional variety of
scenarios must beconsidered, which is reasoned by the large
parameter s pace involved. Furthermore, since CMC has i nuence
on multiple dierent aspec ts of trac at the sa me time, it is dicu lt,
but important to eva luate this inuence from mult iple perspect ives
in a suitable approach. Since the cooperation algorithms are also
computationally intensive, it is necessary to distinguish relevant
aspects f rom less relevant, resulting i n a diculty of  nding appro-
priate Level of Detail (LoD). erefore, even virtual testing with
modern simulation technologies requires new methodical
approaches, in order to overcome the chal lenge of prolonged simu-
lation execution and, thus, inecient generation of results. With
this contribution, wewant to address these issues and to propose
a simulation-based me thodology for cooperation algorit hms, with
focus on the evaluation of their impact on the trac quality.
1.2. Outline
is article is structured as follows. In Section 2, wewill
describe our simulation-based evaluation methodology for
cooperation algorithms. Hence, in Section 3, wewill intro-
duce our novel cooperation algorithm, including its main
characteristics and modules, which we used in order to
demonstrate the practicability of our evaluation methodology.
ereaer in Section 4, wewill present our simulation envi-
ronment, consisting of a co-simulation of multiple compo-
nents, which wearranged specically for this purpose. In
Section 5, wewill dene the metrics and the scenarios, which
wedetermined as a part of our methodology. ereaer in
Section 6, we will present the evaluation results, which
weobtained from our simulations. Finally, wewill complete
this article with conclusion and outlook in Section 7.
 Methodology
In order to evaluate the qua lity of the cooperation algorithms
regarding their inuence on distinct trac aspects, wefurther
enhanced the methodology which weoriginally described in
[3]. e goal of this methodology is to establish a generic
simulation-based workow, which creates a possibility to rate
and to compare dierent CMC algorithms with each other in
a computationally reasonable manner. e corresponding
methodology is schematically shown in Figure 2, whereas the
working steps of this methodology will beexplained in detail
further on.
2.1. Algorithm-Under-Test
Algorithm-under-test, in this case a CMC algorithm, which
needs to beexamined, must beintegrated into the test system
and connected with the simulation environment in accor-
dance to consistent interfaces. us, the algorithm can exert
action on the simulation, as well as the simulation can exert
reaction on the algorithm (and vice versa). Consequently, an
ongoing interaction bet ween the algorithm and the simulation
is achieved, creating a closed feedback loop. e algorithm
itself can exist in form of a model, soware or hardware.
FIGURE 1 Illustration of an exemplary trac situation on a
highway entrance (reduced to one lane) with cooperative
maneuver, which is facilitated by trajectory-based CMC.
Trajectory colors will beexplained in Section 3.1.
© SAE International.
FIGURE 2 Evaluation methodology for
cooperation algorithms.
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3161
erefore, the methodology is generically applicable to various
steps of X-in-the-Loop (XiL) testing procedure. An exemplary
cooperation algorithm, which wedeveloped and used as an
algorithm-under-test for our methodology, will bepresented
in Section 3 of this contribution.
2.2. Co-Simulation
Since on the one hand, the computationally intensive CMC
algorithms are typically applied to a limited amount of coop-
eration participants on a small-scale, and on the other hand,
one is interested in eva luating their eect (e.g., on trac ow)
in various trac situations on a large-scale, this creates a
conict between LoD and performance of the simulation. As
a solution to this, weelaborated a combination of two simula-
tors, which are statically and dynamically coupled (see [4, 5]),
running synchronously and, therefore, result in one common
co-simulation environment. In this particular case, weutilize
a trac ow simulator, which is responsible for execution of
a single long-term scenario (with lower LoD), and a vehicle
dynamics simulator, which is responsible for execution of
multiple short-term scenarios (with higher LoD).
e long-term scenario (0in Figure 2) is created by the
logic of static parameter variation, mainly involving the road
topology (e.g., shape, number of lanes) and the demanded trac
quantities (e.g., trac ow). e short-term scenarios (1, 2, 3in
Figure 2) are situational ly derived from the long-term scenario
by the logic of dynamic parameter variation, whereas the
number and the constellation (incl. positions, velocities, etc.)
of the participating cooperative vehicles dene initial states
(i.e., scenes) of the resulting short-term scenarios. e overall
co-simulation environment, including the concept of automatic
scenario generation, will befurther described in Section 4.
2.3. Test Results
Once the co-simulation, including the integrated
algorithm-under-test, is running, both the long-term and the
short-term scenarios can beanalyzed with the corresponding
macroscopic, microscopic, nanosomic and individual metrics
(see [4, 5]). In case of a long-term scenario, the CMC algorithm
can be evaluated regarding its eciency (e.g., travel time,
energy consumption, etc.) and impact on trac quality, i.e.,
continuousness of trac ow. Additionally, in case of a short-
term scenario, the evaluation of the CMC algorithm is suited
towards its eectiveness (e.g., robustness, safety, comfort, etc.).
Aer the complete co-simulation is nished, if necessary, the
test results can be used in order to adapt the parameters
through the static parameter variation and, thus, to prepare
a new long-term scenario for the next simulation. Trac
quality metrics, which are relevant for this contribution, will
beexplained in Section 5.
 CooperationAlgorithm
As a part of the research project IMAGinE, wedesigned and
implemented the so-called “Opel Core” cooperation algo-
rithm, which acts in this work as an algorithm-under-test for
our methodology. is algorithm belongs to the group of
decentralized intention-based CMC (see [3]), comparable to
approaches [6, 7], meaning that the participating vehicles can
directly exchange their intentions in form of trajectories via
ad-hoc or cellular V2X communication, in order to coopera-
tively coordinate their maneuvers.
3.1. Characteristics
A vehicle equipped with “Opel Core” algorithm is required
to broadcast only one trajectory at a time, in order to preserve
the amount of data exchanged via V2X and, thus, to reduce
the channel load. is characteristic also makes the algorithm
suitable for use in mixed trac, i.e., between autonomous and
manually driven vehicles, since the cooperation here is based
only on one driving trajectory per participant. In case of
manual driving, the trajectory is derived from the observation
of vehicle’s movement and driver’s behavior. Cooperative
maneuvers with “Opel Core” include deceleration, accelera-
tion and lane changes, as well as support cascaded (i.e.,
multiple successive) negotiation processes. Similar to approach
[7], a trajectory in “Opel Core” can belong to one of three
dierent types, depending on the intention it symbolizes.
ese trajectory types are:
•Reference– default, when no cooperation is active
•Request– asking for cooperation (without right-of-way)
•Oer– accepting cooperation (with right-of-way)
If a non-communicating vehicle, which cannot transmit
any trajectory, is present in a cooperative situation, then its
most probable trajectory must be estimated (e.g., based on
vehicle dynamics) and used as a replacement for its reference
trajectory. e principle of CMC with “Opel Core” algorithm
is shown in the illustration of the exemplary merging scenario
(Figure 1). ere, reference trajectories are displayed in red,
request in green and oer in blue colors.
“Opel Core” algorithm is independently executed in every
vehicle (further called ego-vehicle) which is actively partici-
pating on the cooperation. is results in a decentralized
negotiation and decision-making process, meaning that each
ego-vehicle can autonomously, yet cooperatively or non-coop-
eratively, determine its course of action based on its own
maneuver planner and the information from V2X.
3.2. Modules
e CMC algorithm “Opel Core” consists of several modules,
whose execution is serially arranged and cyclically repeated.
e overview of these modules (incl. their inputs and outputs)
is shown in Figure 3. e task of the rst module, trajectory
planning, is to generate a list with a wide diversity of possible
driving trajectories for the ego-vehicle. Herewith, the planning
horizon (i.e., temporal range of trajectories) is a variable
parameter. e output of this module is an unsorted list of
trajectories, which is then processed by the next module. In
the cost function analysis, the trajectories in the list are
assessed and ranked according to their cost values. In case of
“Opel Core”, these values result from the cost functions in
respect to driving behavior (e.g., deviation from desired
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3162
velocity, distance to other vehicles, lane changes) and safety
(e.g., violation of road boundaries and vehicle’s limits). Further
possible cost functions for CMC can be found in [8, 9].
Consequently, the list of trajectories becomes sorted and is
then passed to the next module. In the collision check, trajec-
tories of the ego-vehicle are inspected for collisions with
trajectories which are received via V2X communication from
other vehicles. Colliding trajectories are marked in the list,
and aerwards, the list is handed over to the next module.
In the conict resolution, which represents the most
signicant module of this CMC approach, the algorithm
makes a decision, depending on the right-of-way rule and the
acceptability of the maneuver. e logic of this module is
depicted as owchart in Figure 4. First, the algorithm deter-
mines, whether the ego-vehicle possesses the right-of-way or
not. If yes, the algorithm decides, whether the maneuver is
acceptable for the ego-vehicle or not, based on a certain cost
threshold, which is applied on its sorted list with marked
trajectories. As a result, the ego-vehicle can choose either best
collision-free (i.e., cooperative) or best collision-aicted (i.e.,
non-cooperative) trajectory. In contrast, if the ego-vehicle
possesses no right-of-way, it has to wait for a certain amount
of cycles, duri ng which the conict may beresolved. ereaer,
in any case, the algorithm chooses best collision-free (i.e.,
cooperative) trajectory. e nal trajectory, which is selected
by the conict resolution this way, is then transmitted to other
vehicles via V2X. Aer that, the algorithm repeats a new cycle,
by beginning from the trajectory planning.
 Co-Simulation
Environment
As a part of the evaluation methodology for CMC algorithms,
we established a unique co-simulation environment. In
contrast to comparable frameworks [10, 11, 12], our
co- simulation environment supports simultaneous simulation
of a large-scale long-term scenario with lower LoD and a
small-scale short-term scenarios with higher LoD, thus,
oering a comprehensive opportunity to evaluate the impact
of cooperation on trac quality.
4.1. Structure
As shown in Fig ure 5, the co-simulation environment consists
of a CMC algorithm, vehicle dynamics simulator and trac
ow simulator. e CMC algorithm, in this case “Opel Core”,
which weimplemented as a proof-of-concept soware in
Robot Operating System (ROS) [13], is responsible for control
FIGURE 4 Flowchart of the conflict resolution module.
© SAE International.
FIGURE 5 Structure of the co-simulation environment with
CMC algorithm, vehicle dynamics and trac flow simulation.
© SAE International.
FIGURE 3 Module overview of the cooperation algorithm
“Opel Core”, which is decentrally executed as independent
instances in each ego-vehicle entity.
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3163
of cooperative vehicles in the simulation. Herewith, the simu-
lation of vehicle dynamics is realized by CarM a k e r 7.1.2
(provided by IPG Automotive) [14] with ROS interface exten-
sion, whereas the simulation of trac ow is realized by
SU M O 1.7.0 (provided by German Aerospace Center) [15],
both exchanging data via so-called Traffic Control
Interface (TraCI).
CarMaker and SUMO are statically and dynamically
coupled, similar to solutions [4, 16], meaning that they share
the same road topology and exchange data about vehicles in
a synchronous simulation process. Herewith wedierentiate
between cooperative ego-vehicles (white), which are controlled
by CMC through CarMaker, and non-cooperative trac
vehicles (gray), which are controlled entirely by SUMO. On
the one hand, SUMO is running as a single instance, which
performs the simulation of trac vehicles with lower LoD in
large-scale. On the other hand, CarMaker is running in
multiple instances (one per each ego-vehicle) and simulates
realistic movement of the corresponding ego-vehicles, which
are par ticipating in cooperative maneuvers in small-sca le. e
maneuver control of the ego-vehicles is carried out by the
CMC algorithm, which is also present in multiple instances
(one per each ego-vehicle) a nd passes trajectories to CarMaker.
e exchange of trajectories between cooperating vehicles is
achieved by a simplified V2X communication model
within ROS.
The graphical visualization capability of each
co- simulation component: ROS, CarMaker and SUMO; is also
demonstrated in Figure 5. In case of ROS, the visualization
is accomplished by soware [17], which weextended with an
option to visualize V2X trajectories (colors correspond to
trajectory types).
4.2. Automatic Scenario
Generation
In the soware of our co-simulation environment, the logic
of Automatic Scenario Generation (ASG) acts as an interme-
diate step in the SUMO-CarMaker coupling and is illustrated
in Figure 6. As long as no cooperative maneuver is running,
only SUMO is simulating the long-term scenario with full
control of all the vehicles in it. Once ASG detects a potential
cooperative situation, it triggers generation of a so-called
snapshot, which is utilized as a scene (i.e., initial state) for the
short-term scenario. Herewith, we define snapshot as a
captured constellation of vehicles (incl. their positions and
velocities) at a given point in time. When creating a snapshot,
certain number of vehicles, which are expected by ASG to
participate in a cooperative maneuver, are promoted to ego-
vehicles (white), whereas others remain trac vehicles (gray).
e snapshot is then loaded in CarMaker, which starts a simu-
lation of the resulting short-term scenario. During execution
of the short-term scenario on top of the long-term scenario,
multiple instances of CarMaker with CMC algorithms are
controlling the ego-vehicles, whereas SUMO continues to
control the trac vehicles, which act as non-cooperative
moving obstacles in the short-term scenario. Once the simula-
tion of the short-term scenario is nished, SUMO retracts the
control of all vehicles and continues to simulate the long-term
scenario alone, until the next cooperative situation is detected,
triggering generation of a new short-term scenario by ASG.
In case of a highway entrance (Figure 1), trigger for a snapshot
may be, for example, an appearance of a vehicle on the accel-
eration lane, which may potentially lead to a cooperative
merging maneuver.
According to our methodology, ASG undertakes here the
task of the dynamic parameter variation by continuously
generating new short-term scenarios that are dierent from
each other. Herewith, the long-term scenario allows for a
simpler, but longer evaluation due to lower LoD, whereas the
short-term scenarios allow for more complex, but shorter
evaluations due to higher LoD. Since the goals of both simula-
tors SUMO and CarMaker are dierent, their results can
beevaluated with dierent metrics. In this work, wewill focus
on the evaluation of a long-term scenario with metrics of
trac qua lity. e conguration of the co-simulation environ-
ment, which is further used for this contribution, is listed in
Table 1.
 MetricsandScenarios
In this section, wewill dene the evaluation metrics and the
corresponding long-term scenarios, in order to quantitatively
assess the impact of CMC on dierent aspects of the trac
quality. An overview of these metrics is illustrated in Figu re7
and will befurther explained. e metrics are only applied to
FIGURE 6 Logic of automatic scenario generation
with snapshots.
© SAE International.
TABLE 1 Configuration of the co-simulation with essential
parameters and arguments, used for this contribution (on
standard commercial workstation).
Parameter Argument
Planning horizon of CMC algorithm 10 s
Cycle frequency of CMC algorithm 5Hz
Cycle frequency of CarMaker simulation 1000Hz
Cycle frequency of SUMO simulation 20Hz
Synchronization between SUMO & CarMaker 20Hz
Average simulation speed of SUMO 1.0 real-time
Average simulation speed of SUMO & CarMaker 0.1 real-time
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3164
vehicles which are temporarily located in a stationary Region
of Interest (RoI). is RoI includes a segment of the road where
frequent cooperative maneuvers are expected to occur (e.g.,
at highway entrance). Nomenclature, needed for the formulas
in this section, can befound in the end of the article.
5.1. Metrics Definitions
5.1.1. Trac Density, Velocity and Flow e rst
evaluation metrics indicates the relation between traffic
density k, space mean trac velocity v and trac ow q [18],
which is dened as follows:
qkv
(1)
Herewith, the traffic flow q describes the vehicle
throughput of the road and is oen used as an indicator of
trac quality. In general, higher q should beachieved via
lower k as well as higher v (and not vice versa). Lower k and
higher v correspond to free ow, whereas higher k and lower
v result in congested ow, even if q is the same. is metrics
is collectively applied to all vehicles in the RoI.
5.1.2. Coecient of Variation e second evaluation
metrics is Coecient of Variation (CV)– CVi [19], which can
becalculated as a quotient of the velocity standard deviation
σvi and the arithmetic mean velocity
vi
of each vehicle i, thus,
yielding following equation:
CV v
i
i
i


v (2)
In general, higher CVi values indicate poorer traffic
quality due to higher σvi, which means t hat vehicles frequently
have to adapt their velocities (by accelerating and deceler-
ating), in order to stay synchronized with the overall trac
ow. Furthermore, due to direct proportionality, CVi punishes
lower and favors higher
vi
, since higher vehicle velocities are
more benecial for trac quality. is metrics is indiv idually
applied to each vehicle in the RoI.
5.1.3. Time-Exposed Time-to-Collision We derive
the next evaluation metrics from Time-to-Collision (TTC)–
TTCij [20], which is dened as a quotient of the relative longi-
tudinal position (i.e., distance) ∆pij and the relative
longitudinal velocity (i.e., speed dierence) ∆vij between two
vehicles i and j:
T
TC
p
v
ij
ij
ij



(3)
In this contribution, since TTCij does not deliver an
integral output value, weuse an extended measure called
Time-Exposed Time-to-Collision (TETTC)- TETTCi [20].
Taking into account only positive TTCij of every vehicle i (i .e.,
TTCi), as well as by specifying a minimal acceptable threshold
TTC∗, the TETTCi of every vehicle i can be calculated
as follows:
T
ETTC t
i
t
T
i





0

(4)

i
i
i
t
TTC tTTC
TTC tTTC















1
0 (5)
Herewith, the TETTCi is summed up for every time point
t, with time step τ, from 0 to T of the total evaluation time,
when the corresponding TTCi lies below the threshold TTC∗.
Normally, TTC and TETTC are used as safety metrics, never-
theless they also suit for assessment of trac quality, since
both metrics indicate the continuousness of traffic f low.
Ideally, when all vehicles move with the same velocity without
changing distances between them, this results in TTCij→∞
and TETTCi→0, consequently. Both metrics are individually
applied to each vehicle in the RoI.
5.1.4. Spatiotemporal Patterns Last metrics, the
so-called spatiotemporal patterns, which weuse in this contri-
bution for assessment of trac quality, does not contain any
specic formula and serves mainly for graphical evaluation
(see [21, 22]). e spatiotemporal patterns can becreated by
observing the movement of all vehicles in the RoI and
recording their trajectories (positions pi and velocities vi over
time t). Hence, the recorded pi and vi can be displayed as
spatiotemporal patterns in form of 2- or 3-dimensional
diagrams. This way, an overview of traffic quality can
beachieved, allowing evaluation of trac ow according to
three-phase trac theory [21]. In addition, trac shockwaves,
i.e., abrupt braking of several vehicles due to disturbances
(e.g., merging vehicles), can bedetected when examining the
spatiotemporal patterns.
5.2. Simulation Scenarios
In order to demonstrate the plausibility of the chosen metrics,
wewill focus on three concrete scenarios which we derive
from the logical scenario of highway entrance, as shown in
Figure 1. ese long-term scenarios are synthetically gener-
ated with SUMO and labeled as follows, depending on the
trac state they aim to recreate:
•‘Congested ’
•‘with CMC’
•‘Free’
FIGURE 7 Overview of trac quality metrics, applied on a
highway entrance, with ego-vehicles (white) and trac
vehicles (gray).
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3165
In case of ‘Congested ’ and ‘Free’ long-term scenarios, the
CMC algorithm is deactivated, in order to produce edge cases
for later evaluations. In such an edge case, only SUMO simula-
tion is executed, meaning that no short-term scenarios for
CarMaker are being generated. On the one hand, in the
‘Congested’ long-term scenario, merging vehicles always drive
until the end of the acceleration lane before they begin to
un-cooperatively merge in, thus, causing signicant shock-
waves each time (worst-case scenario). On the other hand, in
the ‘Free’ long-term scenario, there are no merging vehicles
at all, meaning that the trac on the highway can ow undis-
turbedly (best-case scenario). For the long-term scenario ‘with
CMC’, the cooperation algorithm and the whole toolchain of
ASG are activated, resulting in multiple cooperative merging
maneuvers as short-term scenarios. is way, the outcome of
‘with CMC’ scenario is expected to lie in-between the outcome
of ‘Congested’ and ‘Free’ scenarios.
The common configuration of all three long-term
scenarios is listed in Table 2. Hereby, in order to achieve more
apparent and distinctly interpretable results with our metrics,
the trac behavior in SUMO simulation, concerning the car-
following model and the velocity keeping, is set to ideal.
Furthermore, weassume perfect V2X communication, i.e.,
unlimited range with no latencies and losses.
 Results
Aer having presented the methodology, the cooperation
algorithm, the co-simulation environment as well as the
metrics and the scenarios, wewill demonstrate and interpret
our simulation results in this section. Herewith, we will
successively focus on each evaluation metrics, according to
Section 5. While doing so, we will graphically support
our argumentations.
6.1. Trac Density,
Velocity and Flow
e evaluation results considering trac quality in time
domain are displayed as graphs in Figure 8. ere, trac
density k, trac velocity v (incl. one standard deviation) and
trac ow q are displayed over simulation time t for all three
long-term scenarios ‘Congested’, ‘with CMC’ and ‘Free’.
Herewith, k and v are collectively determined form the
number and the mean velocity of vehicles in RoI, whereas q
is then calculated with Equation 1. As can beseen, in case of
‘Congested’ scenario, three un-cooperative merging maneu-
vers cause signicant reduction of trac velocity, followed by
an increase of trac density, which therefore results in oscil-
lation of trac ow. In contrast, in case of ‘Free’ scenario,
trac density, velocity and ow remain always constant. At
this point, in case of ‘with CMC’ scenario, the positive impact
of CMC on trac quality becomes apparent, since the coop-
eration allows for much smoother merging, which then results
in almost constant (i.e., optimal) trac density, velocity
and ow.
Identic simulation results can bedisplayed in a so-called
fundamental diagram [18], as shown in Figure 9. Herewith,
the points in the diagram represent the relations between
trac density k, velocity v and ow q. Basically, the closer the
points are located to the edge-case ‘Free’ (yellow zone), the
FIGURE 8 Results– trac density, velocity and flow in
time domain.
© SAE International.
TABLE 2 Configuration of the scenarios with essential
parameters and arguments, used for this contribution.
Parameter Argument
Duration of a long-term scenario 90 s
Duration of a short-term scenario 30 s (3 times)
Number of ego-vehicles per
short-term scenario
3 veh
Number of lanes on the highway 1
Speed limit on the highway 100 km/h
Demanded trac flow 7200 veh/h
Length of the acceleration lane 250m
Total RoI length 500m
© SAE International.
FIGURE 9 Results– trac density, velocity and flow as
fundamental diagram.
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3166
better is the trac quality. It can beseen, that points of ‘with
CMC’ scenario all lie in the proximity of ‘Free’ scenario,
whereas points of ‘Congested’ scenario are more scattered,
due to higher uctuations in trac density, velocity and ow.
6.2. Coecient of Variation
As a next step, wewill present the results of our evaluation
considering the CV of vehicle velocities in the simulation.
Herewith, the mea n velocity
vi
and the correspond ing standa rd
deviation σvi is determined for each vehicle, which is driving
in the RoI during the runtime of a long-term scenario.
Consequently, CVi is calculated with Equation 2. Aerwards,
every CVi value is portrayed over
vi
value in a scatter plot, as
shown in Figure 10, where each point represents one vehicle.
In case of ‘Free’ scenario, one can notice that all corre-
sponding points are located in one place, meaning that all
vehicles in the simulation were ideally driving with 100km/h
without deviations. In case of ‘Congested’ scenario, the points
are scattered, indicating lower mean velocities with higher
variations for many vehicles. In comparison to this, ‘with
CMC’ scenario shows in-between values, meaning that the
cooperation algorithm visibly improves the CVi, due to
smoother merging maneuvers on the highway entrance. In
general, for better trac quality, higher velocities and lower
variations are preferable, which is represented by the yellow
zone in the diagram. is results in better travel time and
better energy eciency of trac.
6.3. Time-Exposed
Time-to-Collision
As a next metrics for trac quality, wewill evaluate TETTC.
The corresponding results are shown as a bar chart in
Figu re11. In order to produce this diagram, werstly calcu-
late TTCi for each vehicle at each time step of the simulation,
which is then summed up to TETTCi according to Equations4
and 5. Herewith, weset the TTC∗ threshold to a relatively high
value of 25s, which is reasoned by the ideal driving behavior
(car-following model) of trac in the simulation. ereaer,
weenumerate the numbers of vehicles with equal TETTCi
(rounded values), which are then displayed as bars. At the
same time, weskip vehicles with TETTCi=0. In the resulting
diagram, one can see that in case of ‘Congested’ scenario,
many vehicles demonstrate high TETTCi, denoting that they
had to reduce the distances to preceding vehicles for longer
periods of time (over several seconds), being obliged to do so
due to un-cooperative merging maneuvers. In contrast, in
case of ‘Free’ scenario, all vehicles have TETTCi=0. Eventua lly,
in case of ‘with CMC’ scenario, the evaluation delivers
in-between values, thus, indicating that CMC allows for more
consistent longitudinal distances between the vehicles
during merging.
In general, less vehicles with less TETTCi characterize
better trac quality, which is represented by the yellow zone
in the diagram. In case of more realistic trac simulation (i.e.,
non-ideal driving behavior), which is also achievable with our
co-simulation environment, the threshold TTC∗ can be set
lower, even into the range of safety relevant values. is way,
the TETTC evaluation would beable to deliver information
regarding trac safety in addition to trac quality.
6.4. Spatiotemporal Patterns
As a nal step of our evaluation, wewill present the spatio-
temporal patterns, which are displayed as separate diagrams
for each scenario ‘Congested’, ‘with CMC’ and ‘Free’ in
Figu re12 . Herewith, the individual positions of vehicles pi in
the RoI are displayed as their recorded trajectories over simu-
lation time t, as well as the velocities vi are denoted with
dierent colors. In all diagrams, the beginning of the accelera-
tion lane corresponds to 0km, whereas the end corresponds
to 0.25km. In case of ‘Free’ scenario, the spatiotemporal
patterns demonstrate ideal trac quality, where all vehicles
can always freely follow their ways. In case of ‘Congested’
scenario, it is easy to recognize the strong decreases in velocity,
which are caused by the three un-cooperative merging
FIGURE 10 Results– CV over vehicle mean velocity.
© SAE International.
FIGURE 11 Results– number of vehicles with
corresponding TETTC.
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3167
maneuvers in the simulation. ese velocity breakdowns result
in intense shockwaves, starting at the merging site and propa-
gating upstream, implying that multiple vehicles were forced
to brake. Aer the actual merging is fullled and the shock-
wave dissolves, the trac recovers to the free ow condition.
In case of ‘with CMC’ scenario, the deviations in velocity are
almost unnoticeable, since the merging vehicles are quickly
synchronized with the overall t rac ow, whereas the merging
itself occurs much earlier, causing almost no perturbances for
the trac on the highway. Hence, wecan clearly distinguish
the positive impact of CMC on the trac quality.
 ConclusionandOutlook
In this work, we delivered an in-depth presentation of our
simulation-based evaluation methodology for CMC, consid-
ering its impact on the traffic quality. In the process,
wedescribed the cooperation algorithm, the co-simulation
environment, as well as the metrics and the scenarios, which
weused in order to reveal the pract icabilit y of our methodology.
As a part of our contribution, weintroduced a novel
decentralized trajectory-based CMC algorithm “Opel Core”
with its main characteristics and modules, which acted as
algorithm-under-test for our methodology. Considering the
co-simulation environment, wepresented the static and the
dynamic coupling of trac ow simulator SUMO with vehicle
dynamics simulator CarMaker. Concurrently, wedescribed
the logic of ASG, alongside with denition of long-term and
short-term scenarios. ereaer, wefocused on evaluation
metrics that wedetermined as eligible for the assessment of
trac quality. Eventually, wedemonstrated and comprehen-
sively discussed the corresponding evaluation results.
During our work, weprepared several synthetic simula-
tion scenarios, in order to get more distinct evaluation results
with our metrics. Hence, for the sake of generality, the meth-
odology has yet to beapplied on more complex and realistic
scenarios (incl. modeling of vehicle and driver behavior, as
well as V2X communication) in the future. As a consequence,
one of the biggest challenges herewith will lie in a correct
detection of the cooperative trac situations in a long-term
scenario and converting them to short-term scenarios. For
this, one possible solution is to utilize articial intelligence,
which would classify the trac situations as relevant and non-
relevant for CMC. Furthermore, the co-simulation should
becompleted with an ability to execute multiple overlapping
short-term scenarios at the same time. Another challenge,
which has to be faced in the future work, is posed by the
restrained performance of the co-simulation environment
with CMC algorithm. For instance, in worst-case, weran our
simulations with 0.1 real-time on a regular commercial work-
station (2.8−3.8GHz 4/8 cores/threads CPU, 32GB RAM).
erefore, further soware optimization is needed, both for
the algorithm-under-test and for the co-simulation. For this,
one generic approach would beto decrease the serialism and
to increase the parallelism in the computational processes.
Alternatively, it would bepossible to combine the co- simulation
components into one soware, thus, reducing the amount and
the delay of data being exchanged between them.
In conclusion, with our contribution wefound a meth-
odological way to evaluate the impact of innovative CMC
algorithms on different aspects of the traffic quality.
Universally, being suitable for XiL testing, our methodology
can beutilized to assess and to compare the performance of
diverse CMC algorithms against each other, moreover during
dierent stages in the development process.
References
1. Festag, A., “Standards for Vehicular Communication- From
IEEE 802.11p to 5G,” e&i Elektrotechnik und
Informationstechnik 132(7):409–416, 2015, doi:10.1007/
S00502-015-0343-0.
FIGURE 12 Results– spatiotemporal patterns.
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021 3168
2. IMAGinE, “IMAGinE– Solutions for Cooperative Driving,”
European Center for Information and Communication
Technologies (EICT GmbH), www.imagine-online.de.
3. Lizenberg, V., Knapp, S., Mannale, R., Wendel, V., and
Köster, F., “Simulationsbasierte Bewertungs- und
Vergleichsmethodik für Abstimmungsverfahren in
kooperativen Fahrfunktionen,” in AAET: Automatisiertes
und vernetztes Fahren, 49–65, Braunschweig, Germany,
2019, doi:10.13140/RG.2.2.28250.90569.
4. Hallerbach, S., Xia, Y., Eberle, U., and Köster, F.,
“Simulation-Based Identication of Critical Scenarios for
Cooperative and Automated Vehicles,” SAE International
Journal of Connected and Automated Vehicles 1(2):93–106,
2018, doi:10 .4271/2018-01-1066.
5. Hallerbach, S., “Simulation-Based Testing of Cooperative and
Automated Vehicles,” Dissertation, University of Oldenburg,
Germany, 2020.
6. Lehmann, B., Günther, H.-J., and Wolf, L., “A Generic
Approach towards Maneuver Coordination for Automated
Vehicles,” in IEEE ITSC: 21st International Conference on
Intelligent Transportation Systems, 3333–3339, Maui, USA,
2018, doi:10.1109/ITSC.2018.8569 442.
7. Llatser, I., Michalke, T., Dolgov, M., Wildschütte, F., and
Fuchs, H., “Cooperative Automated Driving Use Cases for
5G V2X Communication,” in IEEE 5GWF: 2nd 5G World
Forum, 120–125, Dresden, Germany, 2019, doi:10.1109/5G
WF. 2019.8911628.
8. Düring, M., “Fahrzeugübergreifendes kooperatives
Fahrerassistenzund Sicherheitssystem für automatische
Fahrzeuge,” Dissertation, Technical University of
Braunschweig, Germany, 2017, doi:10.24355/DBBS.084-
201710021309.
9. Knies, C., Fank, J., and Diermeyer, F., “How to Measure
Cooperation? Cost Functions for Cooperative Maneuver
Planning on Highways,” in IEEE ITSC: Intelligent
Transportation Systems Conference, 2483–2489, Auckland,
NewZealand, 2019, doi:10.1109/ITSC.2019.8917075.
10. Franke, K., Balaghiase, R., Düring, M., and Günther, H.-J.,
“A Cooperative Driver Assistance System: Decentralization
Process and Test Framework,” in 7. Tagung
Fahrerassistenzsysteme, München, Germany, 2015.
11. Massow, K. and Radusch, I. (2018). “A Rapid Prototyping
Environment for Cooperative Advanced Driver Assistance
Systems”. Journal of Advanced Transportation,
2018(2586520), Hindawi, doi:10.1155/2018/2586520.
12. Schiegg, F., Krost, J., Jesenski, S., and Frye, J., “A Novel
Simulation Framework for the Design and Testing of
Advanced Driver Assistance Systems,” in IEEE VTC-Fall:
90th Vehicular Technology Conference, Honolulu, USA, 2019,
doi:10.1109/ VTCFall.2019.8891221.
13. Quigley, M., Gerkey, B., Conley, K., Faust, J., Foote, T.,
Leibs,J., Berger, E., Wheeler, R., and Ng, A., “ROS: An Open-
Source Robot Operating System,” in IEEE ICRA: Workshop
on Open Source Soware in Robotics, Kobe, Japan, 2009.
14. An, N. and Specka, F., “Entwicklung und Test kooperativer
Fahrfunktionen in virtueller Umgebung,” in AAET:
Automatisiertes und vernetztes Fahren, 292–305.
Braunschweig, Germany, 2019.
15. Krajzewicz, D., Erdmann, J., Behrisch, M., and Bieker, L.,
“Recent Development and Applications of SUMO-
Simulation of Urban Mobility,” International Journal on
Advances in Systems and Measurements 5(3&4), 2012.
16. Kaths, J. and Krause, S. “Integrated Simulation of
Microscopic Trac Flow and Vehicle Dynamics,” in IPG
Apply & Innovate, Karlsruhe, Germany, 2016.
17. Lizenberg, V., Büchs, B., Knapp, S., Mannale, R., and
Köster,F. “Graphical Data Visualization for Vehicular
Communication Systems in Real and Virtual Test
Environments,” in AmE: Automotive meets Electronics (11th
GMM-Symp os iu m), 65–70, Dortmund, Germany, 2020,
doi:10.13140/RG.2. 2.34961.792 05.
18. Friedrich, B., “Verkehrliche Wirkung autonomer Fahrzeuge,”
. In: Maurer, M., Gerdes, C., Lenz, B., and Winner, H.,
editors. Autonomes Fahren: Technische, rechtliche und
gesellschaliche Aspekte, (Springer, 2015), 331–350,
doi:10.1007/978-3-662-45854 -9_16.
19. Zhu, W., “Microscopic Trac Flow Quality of Service from
the Drivers’ Point of View,” in IEEE International Conference
on Intelligent Transportation, 47–52, Seattle, USA, 2007,
doi:10.1109/ ITSC.2007.4357 790.
20. Minderhoud, M. and Bov y, P., “Extended Time-to-Collision
Measures for Road Trac Safety Assessment,” Accident
Analysis & Prevention 33 (1):89 –97, 2001, doi:10.1016/S0 001-
4575(00)00019-1.
21. Kerner, B., Introduction to Modern Trac Flow eory and
Control: e Long Road to ree-Phase Trac eory
(Springer, 2009), doi:10.1007/978-3-642-02605-8.
22. Treiber, M. and Kesting, A., Trac Flow Dynamics: Data,
Models and Simulation (Spr inger, 2013), doi:10.10 07/978-3-
642-32460-4.
Acknowledgments
The research activities leading to this contribution were
supported as a part of project IMAGinE “Intelligent Maneuver
Automation– cooperative hazard avoidance in real-time”,
thankfully funded by the German Federal Ministry for
Economic Aairs and Energy, within consortium of project
partners BMW, Bosch, Continental, Hessen Mobil, IPG,
MAN, Mercedes-Benz, Nordsys, Opel, TU Munich,
Volkswagen and WIVW. Proof-of-concept implementation
and evaluation of the CMC algorithm “Opel Core” were elabo-
rated in cooperat ion with KOM– Multi media Com munications
Lab, headed by Ralf Steinmetz (Technical University
of Dar mstadt).
Contact Information
Viktor Lizenberg
Opel Automobile GmbH
Bahnhofsplatz
D-65423 Rüsselsheim amMain
Germany
Viktor.Lizenberg@opel-vauxhall.com
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021

© 2021 SAE I nternational. Al l rights reserve d. No part of this p ublication may be re produced, store d in a retrieval system , or transmitted , in any form or by any mean s,
electronic, me chanical, photocopying, re cording, or otherwise, without the prior written p ermission of SAE International.
Positions and opinions advanced in this work are those of the author(s) and not necessarily those of SAE International. Responsibility for the content of the work lies
solely with the author(s).
e-ISSN 2641-9645
3169
Lizenberg et al. / SAE Int. J. Advances & Curr. Prac. in Mobility, Volume 3, Issue 6, 2021
Nomenclature
Symbol/
Acronym Description Unit
V2X Vehicle-to-Everything
ITS Intelligent Transportation Systems
CMC Cooperative Maneuver Coordination
LoD Level of Detail
XiL X-in-the-Loop
ROS Robot Operating System
TraCI Trac Control Interface
ASG Automatic Scenario Generation
RoI Region of Interest
CV Coecient of Variation
TTC Time-to-Collision
TETTC Time-Exposed Time-to-Collision
tSimulation time s
τSimulation time step s
TTotal simulation time s
qTrac flow veh/h
kTrac density veh/km
vTrac velocity km/h
CViCV metrics -
σviStandard deviation of vehicle
velocity
m/s
i
v
Mean vehicle velocity m/s
TTCij TTC metrics s
∆pij Relative vehicle position (distance) m
∆vij Relative vehicle velocity (speed
dierence)
m/s
TETTCiTETTC metrics s
piVehicle position m
viVehicle velocity m/s
iof vehicle with index i-
ij between vehicles with indices i and j-
© SAE International.
Downloaded from SAE International by Ulrich Eberle, Friday, December 10, 2021