What Cooperation Costs: Quality of Communication and Cooperation

Costs for Cooperative Vehicular Maneuvering in Large-scale Scenarios

Daniel Bischoff

1,3

, Florian Schiegg

, Tobias Meuser

, Dieter Schuller

, Nils Dycke

and Ralf Steinmetz

Active Safety Advanced Technology, Opel Automobile GmbH, R

usselsheim, Germany

Corporate Research, Robert Bosch GmbH, Hildesheim, Germany

Multimedia Communications Lab (KOM), TU Darmstadt, Germany

{tobias.meuser, ralf.steinmetz}@kom.tu-darmstadt.de

Keywords:

V2X, Cooperative Vehicular Maneuvering, MCM, Intersection, Quality of Communication, Cooperation

Cost.

Abstract:

With the rise of vehicles on the road, Cooperative Vehicular Maneuvering (CVM) is a crucial prospect to in-

crease the efﬁciency of future vehicular trafﬁc. Recent work proposes promising approaches for CVM using

Vehicle-to-Everything (V2X) communication to increase trafﬁc efﬁciency but evaluates its performance with

only a few vehicles involved and without considering realistic radio propagation channel models. CVM relies

on high quality of communication to coordinate cooperative maneuvers and increases trafﬁc efﬁciency primar-

ily under heavy vehicular trafﬁc load, which also challenges the V2X quality of communication in terms of

channel load and reliability. In this paper, we propose a novel computational efﬁcient CVM planning algo-

rithm specially designed for large-scale scenarios considering a realistic radio propagation channel model and

analyze the quality of communication and cooperation cost of CVM using ad hoc communication technology.

For our urban intersection scenario, we show that imperfect communication limits the earliest start of coop-

eration to 150 m and increases the average Age of Information (AoI) of CVM messages up to 400ms, which

motivates the need for more advanced V2X dissemination strategies.

1 INTRODUCTION

These days, vehicle efﬁciency and safety are consid-

ered increasingly important (Llatser et al., 2019). In

the past years, the ﬁeld of research primarily targeted

trafﬁc safety by assisting the human driver with Ad-

vanced Driver Assistance Systems (ADASs).

Recently, research activities also focus on trafﬁc

efﬁciency. A promising ﬁeld of research to address

trafﬁc safety and efﬁciency for human drivers as well

as for autonomous vehicles is V2X communication.

Sharing information among vehicles increases the en-

vironmental perception beyond the local sensor per-

ception of vehicles (Schiegg et al., 2019) and further

enables CVM (D

uring et al., 2014).

In order to illustrate the potential of CVM using

V2X communication, we consider an urban intersec-

tion. A Tagged Vehicle (TV), i. e., a vehicle equipped

with a CVM application, is about to perform a left-

turn. If there is no trafﬁc light control for this partic-

ular maneuver, the TV has to wait for an open gap in

the oncoming trafﬁc ﬂow. Waiting for an open gap

can lead to a trafﬁc jam because all vehicles behind

are forced to wait as well (Fu and Hellinga, 2000).

Already today, human drivers occasionally coop-

erate by giving priority to others to improve the over-

all trafﬁc efﬁciency. Such a vehicle that opens a gap

and emphasizes its willingness to cooperate with oth-

ers is referred to as Cooperation Vehicle (CV) for the

remainder of this paper. Unfortunately, CVM requires

coordination, which is limited for human drivers as

well as for autonomous vehicles (Llatser et al., 2019).

This motivates the need for V2X communica-

tion to reliably share information about the vehicles’

static and dynamic information over vast distances,

even under non-line-of-sight conditions (Boban et al.,

2018). Thus, V2X communication allows for novel

coordination capabilities, and the communication

range grants more time to the cooperation process

compared to other mechanisms (e. g., hand gestures).

V2X communication enables CVM by sharing the

vehicles’ planned maneuver via a so-called Maneu-

394

Bischoff, D., Schiegg, F., Meuser, T., Schuller, D., Dycke, N. and Steinmetz, R.

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale Scenarios.

DOI: 10.5220/0009592403940405

In Proceedings of the 6th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2020), pages 394-405

ISBN: 978-989-758-419-0

ver Coordination Message (MCM) (Lehmann et al.,

2018), e. g., left-turn for the TV. Hence, an oncom-

ing vehicle receives the planned maneuver, detects the

potential conﬂict with its planned path, and may co-

operate by adapting its planned path, e. g., by slowing

down. We assume that the TV already optimized its

planned path through the extended environmental per-

ception gained by V2X communication, i. e., the TV

cannot adapt its planned path to increase trafﬁc efﬁ-

ciency and requires cooperation from other vehicles.

Even if the environment is taken into considera-

tion to plan the cooperative maneuver, we cannot eas-

ily predict the trafﬁc ﬂow as it is highly dynamic (Shi

et al., 2008). Therefore, it is inevitable to contin-

uously adapt the currently planned maneuver to the

current trafﬁc state and share it with others.

V2X communication relies on radio propagation

to share information among vehicles. A radio prop-

agation channel describes the attenuation of wireless

information due to path loss, shadowing, and fading.

Once the attenuation of a signal is too high, the re-

ceiver cannot decode the packet. Path loss is the dom-

inant attenuation factor in radio propagation chan-

nels and foremost limits the communication distance.

In challenging environments, such as urban intersec-

tions, also channel congestion, slow-fading (shadow-

ing), and fast-fading (multi-path propagation) cause

the quality of communication to ﬂuctuate (Mecklen-

brauker et al., 2011) and reduces the number of re-

ceived messages for cooperation partners. Therefore,

it is inevitable to consider realistic radio propagation

effects for CVM simulations.

Considering our example above, let us assume the

cooperation immediately starts with the ﬁrst received

MCM. Due to fading (constructive interference), the

ﬁrst reception can be way ahead of the intersection.

Nevertheless, at vast distances, the Packet Error Rate

(PER) is very high, such that subsequent MCMs of

our TV are likely to get lost. Hence, the CV cannot

immediately react to any adaptations to the initially

planned maneuver and follows the last received ma-

neuver of the TV or is even forced to abort the coop-

eration. Intuitively, considering the latter, a coopera-

tion cost arises for our CV without any beneﬁt for the

trafﬁc efﬁciency. Thus, CVM needs to consider the

quality of communication.

However, assessing the quality of communication

for CVM is a computationally expensive task (D

uring

et al., 2014), as a large-scale simulation with hun-

dreds of vehicles equipped and acting with CVM is

required. Recent studies, as in (Xu et al., 2019), pri-

marily focused on small-scale scenarios with up to

three CVM enabled vehicles.

To evaluate the cost of CVM in large-scale sce-

narios, we consider the quality of communication and

contribute the following:

• We design an innovative and computational efﬁ-

cient planning algorithm for CVM application en-

abling large-scale simulations in Section 4.

• We show the performance of ad hoc direct com-

munication for CVM in Section 5 under realistic

radio propagation channel assumptions.

• We analyze the cooperation cost with quality of

communication during a cooperative maneuver.

More precisely, we think that the improvement

in efﬁciency by starting the cooperation as early

as possible cannot compensate for the uncertainty

caused by imperfect communication.

The remainder of this paper is structured as fol-

lows: Background concerning V2X communication

and CVM is given in Section 2. We formulate the

impact of the quality of communication on CVM efﬁ-

ciency, referring to a real-world intersection scenario

in Section 3. Existing approaches and initial studies

focusing on the requirements for quality of communi-

cation for CVM are analyzed in Section 6. We con-

clude the paper in Section 7.

2 BACKGROUND

We brieﬂy describe the V2X communication relevant

for this work and explain the basic principle of a de-

centralized CVM mechanism.

2.1 V2X Communication

The European Telecommunication Standard Institute

(ETSI) standardized ITS-G5 in (Intelligent Transport

Systems, 2019) and employed 802.11p on the access

layer. To ensure robustness in vehicular scenarios,

which is characterized by high mobility, the chan-

nel bandwidth was reduced to 10 MHz per channel

(Mecklenbrauker et al., 2011), where the center fre-

quency is allocated at 5.9 GHz. For safety-relevant

V2X applications, three different sub-channels have

been reserved, where the Control Channel (CCH)

is already in use for the Day-1 V2X applications

(Llatser et al., 2019).

On the networking layer, ITS-G5 employs

GeoNetworking. Besides the possibility to forward

messages over multi-hops to destinations out of the

vehicle’s immediate communication range, especially

safety-relevant applications use Single Hop Broadcast

(SHB) for message dissemination. SHB guarantees

low-latency communication within a limited commu-

nication range.

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale

Scenarios

395

As ITS-G5 relies on direct communication, there

is no central entity to control the channel access. De-

centralized Congestion Control (DCC) has been em-

ployed by ETSI in (Intelligent Transport Systems,

2018) to avoid channel congestion. DCC can reduce

the transmission rate or power, depending on the cur-

rently sensed channel state.

2.2 Decentralized CVM

CVM has gained increasing interest, as it is a promis-

ing mechanism to increase trafﬁc efﬁciency on the

road (Lehmann et al., 2018). In centralized ap-

proaches, an independent authority monitors the traf-

ﬁc within areas of interest. If cooperation is required,

the independent authority decides how to execute the

cooperation. Centralized approaches leverage their

increased environmental perception but have down-

sides on scalability. Further, they require an appropri-

ate infrastructure, such that the amount of cooperation

scenarios is limited.

In the paper at hand, we focus on decentralized

CVM using trajectories proposed by (Lehmann et al.,

2018). Each CVM enabled vehicle plans trafﬁc-rule

compliant and drivable trajectories, which we derive

from the vehicle’s strategic path. To obtain the strate-

gic path for human drivers can be a challenging task.

In the following, we assume that the strategic path is

given. Trajectories represent the vehicle’s future path

within a limited time horizon. The trajectory, which

is not interfering with static or dynamic objects (other

vehicles), can be taken as the planned trajectory. If

there is more than one collision-free trajectory, the ve-

hicle computes the cost for these trajectories and se-

lects the trajectory with the lowest cost. The vehicle

will follow this trajectory and continuously broadcast

it to other vehicles via V2X communication.

At junctions or motorway access roads, there

might be trajectories with less cost but overlapping

with others. As the right-of-way has priority, the ve-

hicle without right-of-way is forced to choose a more

cost expensive trajectory. With a second trajectory,

cooperation can be requested. A vehicle with a col-

lision of such a trajectory with its planned path can

adapt its trajectory, e. g., slowing down. If the colli-

sion is solved, the TV takes the desired as its planned

trajectory.

3 COOPERATION AT

INTERSECTIONS

First, we brieﬂy describe the real-world Kooperative

Perzeption (cooperative perception) (Ko-PER) inter-

Figure 1: Ko-PER Intersection scenario adopted from (Kra-

jzewicz et al., 2002).

section scenario, depicted in Figure 1, which we refer

to throughout this paper. Second, we detail why the

quality of communication signiﬁcantly impacts the

performance of CVM.

3.1 Ko-PER Scenario

As already mentioned, CVM aims at increasing trafﬁc

efﬁciency in challenging environments. We selected

an urban intersection scenario, as cities are challeng-

ing in terms of radio propagation as well as for ma-

neuver coordination. In (Strigel et al., 2014), an ur-

ban intersection scenario in Aschaffenburg, Germany,

was analyzed in detail. The authors tracked and eval-

uated the vehicular trafﬁc using video cameras. From

this study, valid scenarios were extracted in (Jesen-

ski et al., 2019). From these scenarios, we picked the

most interesting scenario for CVM, which is depicted

in Figure 1.

Vehicular trafﬁc from the north-west and south-

east side are stopping during the red light phase. We

consider these vehicles (not depicted in Figure 1), as

we are interested in the channel congestion within the

intersection caused by V2X communication. Vehi-

cles can pass the intersection during the green light

phase. In the following, we will focus on vehicles

from the north-east and south-west. We let vehi-

cles from the north-east only drive straight as this in-

creases the need for cooperation in our scenario. For

vehicles from the south-west, we consider vehicular

trafﬁc with the intention of different directions at the

intersection, i. e., a small portion of the vehicles want

to turn left, and the others want to drive straight. Our

TV is part of the trafﬁc ﬂow from the south-west to

turn left. These vehicles require cooperation to pass

the intersection. If no cooperation partner is found,

they have to wait in the intersection for a sufﬁcient

gap in the oncoming vehicular trafﬁc.

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

396

3.2 Impact of Imperfect

Communication

CVM requires a CV to change its planned path, e. g.,

change speed to open a sufﬁcient gap for the TV. This

change is usually connected to cooperation cost in

terms of speed loss for the CV. Starting the cooper-

ative maneuver at vast distances allows for early co-

operation and, hence, grants more time for the CV to

open a gap. This additional time prevents unneces-

sary deceleration and reduces cooperation costs. On

the other side, the quality of communication is high

at smaller distances, i. e., the path loss is highly cor-

related with the distance between the transmitter and

receiver. Additionally, the probability of shadowing

due to other objects (e. g., vehicles) increases with

distance.

For the intersection scenario introduced in Sec-

tion 2, we also consider a rich scattering environment

due to buildings and other vehicles. Compared to

motorway access roads, the intersection is character-

ized by relatively low speed and hence low Doppler

shifts (Mecklenbrauker et al., 2011). Shadowing due

to buildings might be reduced, considering V2X com-

munication with oncoming trafﬁc for cooperation -

at least for ad hoc communication. Even though the

speed in urban areas is relatively low, the vehicular

trafﬁc is often highly congested. These trafﬁc jams

lead to varying and unpredictable arrival times at the

intersection.

Considering the characterization of an intersec-

tion, as mentioned above, several challenges for CVM

arise: Constructive interference might lead to suc-

cessfully received messages at vast communication

distances. As of today, CVM applications are not de-

signed to predict message loss of subsequent MCMs

caused by a highly ﬂuctuating radio propagation

channel at vast communication distances. As stated

previously, the efﬁciency of cooperation leverages

vast cooperation distances. Hence, the CVM appli-

cation might immediately start the cooperation with

the ﬁrst received MCM. Due to the highly congested

vehicular trafﬁc in intersections, adaptations to the

planned maneuver are likely to be required. Due to

radio propagation effects, subsequent messages might

get lost, such that the CV and, more precisely, the

CVM application is unaware of any changes in the

planned maneuver of the TV, which increases with

increasing AoI.

In this paper, we aim at characterizing additional

costs, which arise from imperfect communication.

Therefore, we consider ad hoc communication tech-

nology. Further, considering the cooperation cost and

quality of communication, we design a novel CVM

planning algorithm, which can simulate maneuver co-

ordination in large-scale scenarios. In other words,

we make use of data provided by the trafﬁc simulator

to obtain the points reachable by cooperating vehicles

required for the trajectory generation.

4 SYSTEM DESIGN

CVM requires obtaining a set of trajectories to repre-

sent the future path of a vehicle. CVM algorithms are

considered to be computationally expensive (D

uring

et al., 2014). Thus, it could be challenging to analyze

the quality of communication in large-scale scenario

simulations with hundreds of vehicles. Our focus is to

reduce the computationally expensive trajectory gen-

eration for each simulated vehicle and simplify the

detection of trajectory collisions. Therefore, we inte-

grate a CVM application into a communication sim-

ulator, which is coupled with a trafﬁc simulator. For

the generation of trajectories, we make use of the data

provided by the trafﬁc simulator, such as route and

map information, to reduce the computational com-

plexity. That way, we can analyze the efﬁciency of

CVM under realistic V2X communication conditions

in large-scale scenarios.

4.1 System Architecture

The central entity of our architecture is the middle-

ware, which can be connected to multiple communi-

cation interfaces, e. g., cellular or ad hoc communica-

tion. Within this paper, we focus on ad hoc communi-

cation. As proposed by ETSI in (Intelligent Transport

Systems, 2010), the facility layer is composed of the i)

application support, ii) information support, and iii)

communication support facilities. The system archi-

tecture is depicted in Figure 2.

Figure 2: System Architecture of the communication stack.

The information support facility is composed of

the Map Data Provider (MDP), Vehicle Data Provider

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale

Scenarios

397

(VDP), and Local Dynamic Map (LDM). Among oth-

ers, the MDP stores the vehicle’s position and infor-

mation on the current road and lane as well as the

strategic path of the vehicle. Vehicle dynamic data

such as the current speed and acceleration are updated

in the VDP. The LDM stores received V2X messages

as objects, which are indicated by the application sup-

port facilities.

The application support facilities register and

connect themselves to the middleware. When pack-

ets are requested from the application support facil-

ities, the middleware selects the appropriate commu-

nication interface. We allow for a static and dynamic

interface selection scheme. For now, we focus on the

static interface selection scheme and leave dynamic

selection for future work. Within the application sup-

port facilities, messages are either build continuously

or on request.

We give details to the Maneuver Coordination

(MC) facility, describe our proposed MCM for-

mat, and explain the Maneuver Planning Application

(MPA).

4.2 Maneuver Coordination (MC)

Facility

The MC facility is responsible for the generation

of MCMs. According to the state of the triggering

conditions of the facility, the MC facility builds an

MCM based on the currently available data. The

VDP, MDP, and MPA provide data for the different

MCM containers. Upon data fetching and generation,

MCMs are timestamped to keep track of their AoI.

In this paper, the triggering conditions follow a con-

stant message generation rate. A constant triggering

condition allows us to better analyzing the impact of

quality of communication on CVM. More appropriate

data-driven generation rules can also be implemented.

4.3 Maneuver Coordination Message

(MCM)

For CVM, different message formats have been re-

cently proposed in the literature, e. g. (Correa et al.,

2019; Lehmann et al., 2018). Our adaptations to the

proposed formats aim at decreasing the message size

by removing unnecessary information. We adopt the

idea of a collision-free planned trajectory and an op-

tional desired trajectory from (Lehmann et al., 2018).

The proposed MCM is given in Table 1. Each tra-

jectory consists of a unique identiﬁer, the type of the

trajectory (desired or planned), the cost of the trajec-

tory, and a series of trajectory sections. In (Lehmann

et al., 2018), a trajectory section is composed of a

Table 1: Structure of the MCM.

Maneuver Coordination Message (MCM)

Container

Generation Delta Time

Absolute Reference Position

Trajectories

Identiﬁer

Trajectory

Type

Cost

Time Horizon

Sections

Position

Relative Position

Road Information

Polynomial

Coefﬁcients

Heading

Validity Range

longitudinal and lateral polynomial, where the for-

mer is the vehicle’s longitudinal position as a function

of time. The latter polynomial is the lateral position

as a function of the longitudinal position. In our ap-

proach, we neglect the lateral polynomial for the fol-

lowing reasons: In common trafﬁc simulators, the lat-

eral position within the lane is not reﬂected, because

its computation generates unnecessary overhead and

increases the message size.

For CVM, the lateral polynomial is essential to de-

tect lane changes, as the vehicle occupies two lanes

at the same time. We propose to extend the valid-

ity of the longitudinal polynomial on the current lane.

In parallel, our proposed trajectory also occupies the

other lane with a second longitudinal polynomial for

the time of the lane change, as depicted in Figure 3.

Figure 3: Proposed longitudinal Trajectory.

The trajectory type differentiates between the

planned and optional desired trajectory. The cost of

each trajectory is denoted in the proposed message,

where the cost of the optional desired trajectory is

always lower than the planned trajectory, i. e., the

desired trajectory improves the vehicle’s trafﬁc efﬁ-

ciency.

Each trajectory is composed of a set of trajectory

sections to cover the envisaged time horizon, where

the time horizon of a trajectory is also denoted in the

proposed message. Each trajectory section contains

the relative position to denote its starting point, refer-

ring to the absolute reference position. Further, each

trajectory section contains the respective road infor-

mation, including the unique road and lane identiﬁers

obtained from the MDP.

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

398

Each trajectory section is composed of a maxi-

mum of two polynomial functions, where each longi-

tudinal polynomial function has a maximum degree of

two. Both limitations guarantee to reduce the number

of possible trajectories and, thus, the computational

complexity of the trajectory generation. The head-

ing of each polynomial function describes the direc-

tion where the longitudinal path is pointing to. The

longitudinal validity of the polynomial is expressed

as validity range, i. e., start and end of the respective

polynomial function.

4.4 MPA

The Maneuver Planning Application (MPA) is com-

posed of three submodules: i) Trajectory Generation,

ii) Trajectory Processing, and iii) Cooperation Plan-

ning, which is adapted from the research within the

IMAGinE

project.

Trajectory Generation. A strategic path describes

the vehicle route from its current position to its ﬁ-

nal destination. We assume that any application from

the MDP can obtain the strategic path. This assump-

tion holds for fully automated vehicles. For human

drivers, the most probable path must be estimated.

The strategic path is composed of a list of roads

and junctions, where we assume that two roads are al-

ways connected to a junction. We store the informa-

tion of each road and junction such as right-of-way,

junction type (trafﬁc light, right-before-left,...), and

speed limits in a list of trails. Further, the maneuver

type (straight, left-turn,...) for each trail is obtained

from the strategic path. When passing a road or a

junction, the respective trail is removed from the list.

Each trail is composed of a list of segments. Seg-

ments describe the shape of roads and junctions. A

straight road can be described with one segment de-

noting the start and end of the road. For road curves,

the number of segments increases.

From each trail, a trajectory section is obtained.

Hence, the ﬁrst and last segment of a trail also de-

scribe the start and end of the respective trajectory

section. Using existing segment points provided by

the trafﬁc simulator decreases the computational com-

plexity to obtain reachable points for the trajectory

generation. While driving on a trail, the ﬁrst segment

is continuously adapted to the vehicle’s current posi-

tion.

As mentioned before, we limited the number of

polynomial functions. Hence, the ﬁrst longitudinal

polynomial function ranges from the start of the trail

https://imagine-online.de/en/home/

to one of the segments within the trail. The second

longitudinal polynomial function connects to the ﬁrst

polynomial and ranges until the end of the trail. We

can also obtain only one polynomial function, which

ranges from the start to the end of the whole trail. If

we have two longitudinal polynomial functions, the

ﬁrst is of the ﬁrst degree, and the second of second

degree, i. e., the ﬁrst longitudinal polynomial function

follows the trail starting with the speed of the last trail.

Each trajectory starts with the current speed of the ve-

hicle. The second polynomial function varies the tar-

get speed at the end of the trail, which also considers

the speed limit of the subsequent trail.

Further, we ensure to adjust the target speed de-

pending on the junction and maneuver type of the sub-

sequent trail, e. g., at junctions with right-before-left,

we reduce the target speed. From the target speed

mentioned above at the end of each trail, we reduce

the speed iteratively until we would stop at the end of

a trail to create trajectories with different speed pro-

ﬁles. That way, we ensure that a vehicle always ob-

tains at least one collision-free trajectory, i. e., a tra-

jectory that stops at the end of a trail in front of a

junction.

The absolute heading of each section is obtained

from the position of its start and end segment, to

get the direction where each longitudinal polynomial

function is pointing to.

To avoid trajectory collisions with preceding ve-

hicles, we also adapt the maximum speed of the tra-

jectory to the speed of preceding vehicles. As all ve-

hicles in the simulation are equipped with CVM, we

obtain the position and speed of other vehicles from

the LDM. If a vehicle is in close range and in front

of us, we limit our maximum trajectory speed to the

respective vehicle’s current speed. The speed can be

obtained from the ﬁrst polynomial of the planned tra-

jectory. That way, we avoid collisions with vehicles

in trafﬁc jams.

Trajectory Processing. The set of generated trajec-

tories are now scored and checked for collision with

other trajectories. In order to score the trajectories,

we denote the trajectory which represents the longest

path among all other generated trajectories within the

given time horizon as best, and we normalize the cost,

ranging from 0 to 1, where 1 denotes the worst trajec-

tory. The time horizon is equal for all generated tra-

jectories. After that, the trajectories are ﬁltered such

that we eliminate trajectories with similar costs, i. e.,

the ﬁltered trajectories differ from each other.

All generated trajectories are now checked for col-

lision with received trajectories, which are stored in

the LDM. We make use of the proposed collision de-

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale

Scenarios

399

tection from (Ericson, 2004). In this approach, both

the generated and received polynomial functions are

sampled in time steps. At each time step, the distance

between both sampled points is checked. If the dis-

tance is below the diameter of a circle, a collision is

detected. For the diameter of each circle, we con-

sider the width of the vehicle. In our approach, we

consider two circles, one representing the front and

the other the back of the vehicle to take the vehicle’s

length into account. This representation also consid-

ers a safety-margin for each vehicle.

The generated trajectories are grouped in i)

collision-free, ii) collision and right-of-way, and iii)

collision without right-of-way.

Cooperation Logic. From the ﬁrst two groups, the

trajectory with the lowest cost is denoted as reference.

In our simulation, we were able to show that this tra-

jectory is the most probable, i. e., our vehicle follows

this trajectory.

If the selected reference trajectory is from the sec-

ond group, we are requested to offer cooperation, i. e.,

we have detected a collision and have the right-of-

way. To enable cooperation, we select the cheapest

trajectory from the ﬁrst group - denoted as an alter-

native.

Further, we aim at improving our trafﬁc efﬁciency

by identifying potential cooperation. Therefore, we

select the cheapest trajectory from the third group and

denote this trajectory as desired, if its cost is lower

than the reference.

We deﬁne c

as the cooperation cost threshold for

trajectories, which is equivalent to all CVM vehicles.

First, the cooperation cost between the reference and

alternative trajectory is compared. Cooperation is

offered by selecting the alternative trajectory as our

planned trajectory, if c

alt

− c

ref

< c

holds, i. e., our

arising cooperation cost is acceptably low. Otherwise,

we implicitly decline cooperation. Second, coopera-

tion is requested by selecting the desired trajectory, if

ref

−c

des

> c

holds, i. e., we gain more from cooper-

ation than the respective CV is required to spend.

5 RESULTS

In this section, the proposed system is analyzed in de-

tail. Therefore, we ﬁrst describe our simulation setup

and explain the required assumptions for the commu-

nication protocol stack. Second, the metrics we use to

evaluate the system are detailed. In the end, we show

our simulation results for trafﬁc dynamics, quality of

communication, and cooperation costs.

Table 2: Parameters for the Maneuver Planning Algorithm.

Parameters Value

Simulation Time 300s

Message Frequency 5Hz

Target Speed Variation 2m/s

Maximum Acceleration ±4m/s

Speed Reduction Factor 2.6

Time Horizon 10s

Collision Detection Time Step 200ms

Cooperation Cost Threshold 0.1

Scale Cost Trajectories 0.2

Vehicle Density Route 1 550veh/h

Vehicle Density Route 2 450veh/h

Vehicle Density Route 3 150veh/h

5.1 Simulation Assumptions

The parameters for the evaluation are listed in Table 2.

The target speed variation denotes the iteration

from the maximum allowed speed at the next trail to

0, where more iterations can create more trajectories

when reducing this parameter. If a segment point is

very close to the subsequent trail, the acceleration can

be unbounded if the target speed is different from the

last considered speed. This parameter and the speed

reduction factor must be adjusted to the driver model

of the trafﬁc simulator. At junctions, the maximum

allowed speed does not change. Hence, to avoid colli-

sions with other vehicles, e. g., at junctions with right-

before-left, vehicles reduce the speed in front of junc-

tions. We analyzed the driver model of our trafﬁc sim-

ulator, where we found the speed reduction factor to

be roughly v · 2.6

−1

To detect collisions with other trajectories, we it-

erate over both trajectories within time steps. Increas-

ing the time step decreases the computational com-

plexity of the simulation, but also we might miss po-

tential collisions, as our sampling rate is too low. We

think that it is worth to adapt this parameter to the

speed of the vehicle and its context, i. e., we expect

more collisions at intersections and less on straight

roads.

Further, the post-ﬁltering of trajectories signiﬁ-

cantly impacts the computation time of the simula-

tion. Filtering too many trajectories decreases the

efﬁciency of CVM, as we might miss potential tra-

jectories for the cooperative maneuver. On the other

hand, this parameter impacts the computational com-

plexity of the simulation signiﬁcantly, i. e., all ﬁltered

trajectories are checked for collision with all received

trajectories. The vehicle density for each route is se-

lected such that cooperation is required, i. e., a vehicle

with the intention of a left-turn might require cooper-

ating with an oncoming vehicle. On the other side, the

chosen density does not induce too many trafﬁc jams.

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

400

We simulated our scenario for 300 s.

For the representation of vehicular trafﬁc ﬂow, we

use the Simulation of Urban Mobility (SUMO) 1.0

(Krajzewicz et al., 2002). We extracted the KoPER

intersection with Open Street Map (OSM) and con-

verted it to a SUMO readable ﬁle using Netconvert,

which is part of the SUMO package.

For the simulation of V2X communication, we

use the event-discrete network simulator OMNeT++

5.4.1 (Varga and Hornig, 2008). To connect SUMO

with OMNeT++, we use Veins 5.0 (Sommer et al.,

2011). Veins makes use of the bidirectional Trafﬁc

Command Interface (TraCI) to obtain trafﬁc and map-

related data. TraCI also allows controlling individ-

ual vehicles, e. g., change lane or speed. Furthermore,

Veins implements the ITS-G5 access layer 802.11p.

We extended this communication framework by a

Geo Networking Protocol (GNP) and Basic Transport

Protocol (BTP) (Intelligent Transport Systems, 2017)

layer, which we described in Section 2. The channel

model and communication parameters for ITS-G5 are

taken from (Bischoff et al., 2019).

5.2 Metrics

We use three metrics to evaluate the impact of imper-

fect communication on CVM: trafﬁc dynamic, AoI,

and required speed reduction for cooperation. Fur-

ther, we analyze the number of generated trajectories

for different vehicle routes within the considered sce-

nario.

The arrival time for vehicles, especially at the in-

tersection, is assumed to vary signiﬁcantly. To eval-

uate this variation, we deﬁne a measurement point

150m in front of the intersection. We measure the

speed of this vehicle at the measurement mentioned

above point and evaluate the mean and standard devi-

ation of the required time to the intersection.

The AoI captures how long received MCMs, saved

as objects in the LDM, are outdated. We measure and

obtain the average AoI as a function of the distance

to the intersection when a collision of trajectories is

detected. For collision detection, we consider the AoI

of the respective trajectory. To evaluate the required

speed reduction of a cooperative maneuver, we eval-

uate the difference in speed as mean and standard de-

viation of the alternative and reference trajectory as a

function of the distance to the intersection.

5.3 Evaluation

In the following, we analyze and discuss our results

considering the trafﬁc dynamics for the Ko-PER inter-

section, the number of generated trajectories for dif-

Figure 4: Variation of the time to intersection over the initial

speed at the measurement point.

ferent roads, the AoI, and the cooperation cost both

while coordinating the cooperative maneuver. For all

evaluations, we are only planning the maneuver and

not executing it.

5.3.1 Trafﬁc Dynamic

In Figure 4, the variation of the required time to

reach the intersection is depicted over the initial speed

recorded at the start of the measurement point. The

measurement point is 150 m in front of the intersec-

tion, which correlates with the maximum transmis-

sion range of two approaching vehicles. The mean

of the TTI ranges from 30 to 45s and is depicted as

a dark blue bold line. The mean TTI ﬁrst increases

from 6 to 6.5 m/s and slightly decreases to 30 s at

7m/s. The standard deviation is depicted as a light

blue area. At 6.5 m/s, the standard deviation is high-

est at approximately 25 s, which is more than half of

the mean TTI. This signiﬁcant standard deviation is

a result of occurring trafﬁc jams in front of the inter-

section, as vehicles with the intention of a left-turn

wait for an open gap in the oncoming trafﬁc ﬂow.

From Figure 4, we can conclude that, as the time to

arrive at the intersection varies signiﬁcantly, the vari-

ation of the arrival time at the intersection cannot be

considered in the disseminated trajectories. Hence,

we must continuously adapt the generated trajecto-

ries while approaching the intersection. Further, the

highly dynamic vehicular trafﬁc emphasizes the need

for reliable communication, as outdated information

might be unsuitable for planning a cooperative ma-

neuver.

5.3.2 Generated Trajectories

In Figure 5, the number of different generated trajec-

tories is denoted for all three considered routes. The

ﬁrst route represents vehicles of the oncoming traf-

ﬁc ﬂow. The second route represents vehicles driv-

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale

Scenarios

401

Figure 5: Number of generated trajectories for three differ-

ent routes.

ing from south-west to north-east, where the second

route is going straight, and the third represents a left-

turn. The lower and upper quartile for the ﬁrst route

range from 5 to 50 trajectories and the lower and up-

per whiskers from 0 to approximately 125. For the

second route, the lower quartile and whisker is simi-

lar to the ﬁrst route, where the upper whisker reaches

up to 200 trajectories. From this, we can conclude that

the complexity of the ﬁrst two routes is comparable as

both routes are going straight.

For the third route, the number of generated tra-

jectories is signiﬁcantly higher, where the median is

at 75, and the upper whisker ranges up to 350. The

complexity of this route is much higher compared to

the ﬁrst two routes, i. e., the shape of a left-turn re-

quires more segments than a straight road, where the

median of generated trajectories is below 25.

With an increasing amount of trajectories, the

computational complexity of the simulation in-

creases, which is contradictory when simulating

large-scale scenarios. In contrast to that, having more

trajectories increases the possible driving maneuvers

for the cooperative maneuver. For straight roads, we

are not expecting to require cooperation and cannot

improve our current maneuver through cooperation.

Thus, we only require different trajectories to offer

cooperation. In contrast to that, a vehicle with the

intention of a left-turn might require cooperation. In

order to ﬁnd the most efﬁcient trajectory, a large set

of different trajectories resulting in different driving

maneuvers is beneﬁcial.

5.3.3 Quality of Communication

Now, we evaluate the quality of communication while

coordinating a cooperative maneuver, i. e., at least one

of the generated trajectories collides with a received

trajectory of another vehicle. We evaluate the quality

of communication as AoI and the number of received

Figure 6: AoI and number of received messages over the

distance to the intersection while cooperating.

messages both as a function of the remaining distance

to the intersection. The left axis and the black dot-

ted line of Figure 6 depict the mean AoI. MCMs are

stored in the LDM, where MCMs with an AoI larger

than 5 s are removed from the LDM. The mean AoI

ranges from 0 to 1000 ms, where 0 s would represent a

direct reception and processing of the message (prop-

agation and access delay at the Medium Access Con-

trol (MAC) layer is considered but below 1 ms). Up

to 50 m in front of the intersection, the mean AoI is

approximately 500 ms. At a distance to the intersec-

tion between 50 to 100 m, the mean AoI ranges from

500 to 1000ms. From 100 to 175m, the mean AoI de-

creases again and ranges from 300 to 500 ms, which

is surprising at ﬁrst sight. These phenomena can be

explained with the number of considered MCMs. For

this scenario, the number of MCMs below 50 m with

a collision of at least one generated trajectory is above

3000 messages. The number of processed MCMs sig-

niﬁcantly decreases at a distance of 50 m in front of

the intersection and reaches 0 at approximately 175 m.

This is in accordance with the recent literature, e. g.

(Cunha et al., 2016; Bischoff et al., 2019), where the

authors indicated that a reliable V2X communication

using ad hoc communication is limited to a maximum

of 300m. Considering a similar distance to the inter-

section of the transmitter of the MCMs results in a

communication range of 300m.

5.3.4 Cooperation Cost

In the following, we evaluate the required speed re-

duction to enable a cooperative maneuver from the

perspective of the CV. As described in Section 4,

from all generated and ﬁltered trajectories, the CV

selects the cheapest trajectory as a reference, either

with or without a collision of other received trajecto-

ries, i. e., the CV has right-of-way for the upcoming

maneuver. If the planned trajectory has a collision, an

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

402

Figure 7: Required speed reduction for a cooperation offer

over the distance to the intersection.

alternative trajectory can be selected in order to offer

cooperation. In Figure 7, the difference in speed at

the end of both the reference and alternative trajec-

tory is depicted as a function of distance to the inter-

section. The mean required speed reduction decreases

from 6 m/s below 1 m/s at 150m. In contrast to the

trafﬁc dynamic in Figure 4, the standard deviation de-

creases with increasing distance to the intersection,

i. e., a considerable distance to the intersection grants

more time to reduce the speed and allows for more

alternative trajectories. At a distance to the intersec-

tion of 50 m, the standard deviation of the required

speed reduction reaches 0m/s. In this case, the al-

ternative trajectory is more expensive in terms of the

reachable distance within the time horizon. On the

other side, the alternative trajectory grants more time

for the TV to pass the intersection by reducing the

speed at the beginning of the trajectory and acceler-

ating again such that the speed at the end of both tra-

jectories is comparable. This also emphasizes that the

cost of each trajectory should not be obtained solely

from the speed at the end of each trajectory, as this

neglects speed variations within the trajectory.

6 RELATED WORK

There is a tremendous amount of research for

communication-based CVM. Promising approaches

targeting CVM are described in the following. Fur-

ther, we summarize existing work focusing on re-

quirements for different types of V2X applications.

6.1 CVM Approaches

CVM range from decentralized to centralized and

from scenario-speciﬁc to generic approaches.

First, we focus on decentralized CVM ap-

proaches. A reference architecture for cooperative

driving is proposed in (Franke et al., 2014). The au-

thors focus on message deﬁnition and the interaction

of system components to enable CVM. They divide

CVM in Sense, Model, Plan, and Act phases, which

provides a modular and ﬂexible architecture. On the

downside, this approach requires seven different types

of messages and explicit coordination, which limits

the applicability to speciﬁc scenarios.

uring et al., 2014) propose a more generic ap-

proach to increase road safety, comfort, and efﬁciency

using trajectories. The authors focus on path plan-

ning, including the generation of target points and tra-

jectory generation, also mentioning risk assessment

and maneuver execution by selecting the best trajec-

tory combination from their own and received trajec-

tories. The authors conclude that the approach applies

to different kinds of scenarios, but also, the algorithm

requires time-expensive computation.

The approach mentioned above is continued in

(Lehmann et al., 2018), introducing a planned and de-

sired trajectory and detailing the maneuver coordina-

tion process. That way, the authors realize implicit

coordination without conﬁrmation, which decreases

message overhead and coordination time.

(Xu et al., 2019) propose an extension of the con-

cept mentioned above and integrate it into a micro-

scopic simulation framework. The authors extend the

negotiation pattern to increase reliability due to con-

ﬂicting maneuvers or imperfect communication. Al-

though they show an increase in trafﬁc efﬁciency us-

ing CVM in a small-scale simulation scenario, the ap-

proach again introduces explicit coordination.

In (Correa et al., 2019), the authors build upon the

approach in (Lehmann et al., 2018) and propose to

extend the ETSI MCM format, also allowing support

of road infrastructure to increase trafﬁc safety further.

Also, the authors discuss and evaluate the static and

dynamic message generation rules for MCMs. In con-

trast to static generation rules, a dynamic rule can

adapt to the vehicular context and thus provides in-

formation that is more relevant but also reducing the

message overhead.

(Milanes et al., 2011) propose a centralized CVM

approach for on-ramp merging using fuzzy logic. A

central entity controls the arrival times for vehicles at

the merging zone, which aims at increasing the ve-

hicular trafﬁc ﬂow efﬁciency. Compared to the ap-

proaches mentioned above, the authors focus on the

longitudinal path of vehicles, which signiﬁcantly de-

creases the computational complexity and message

overhead.

In summary, the recent literature already pro-

vides promising approaches for CVM, focusing on

explicit and implicit coordination mechanisms. Un-

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale

Scenarios

403

fortunately, these approaches are not considering the

impact of the quality of communication on CVM.

Other approaches compensate for poor quality of

communication by introducing additional redundancy

for highly impactful messages but do not consider the

inﬂuence on CVM (Meuser et al., 2019). Further, ini-

tial evaluation studies focus on small-scale scenarios

with very few vehicles involved.

Our approach aims at improving the computa-

tional complexity of CVM. Thus, we can evaluate our

approach in a large-scale urban intersection scenario,

also considering the impact of the quality of commu-

nication using a realistic channel model.

6.2 Communication Requirements for

V2X Applications

Requirements targeting the quality of V2X commu-

nication are separated into the domain of road safety,

efﬁciency, and infotainment. General vehicular de-

mands targeting the network performance are given in

(Zhao et al., 2019). The authors focus on the commu-

nication delay and capable vehicle mobility and den-

sity to classify different communication technologies.

As these V2X applications are not further detailed,

speciﬁc communication requirements for CVM can-

not be derived.

(Zheng et al., 2015) survey communication re-

quirements for different safety and non-safety ap-

plications, primarily focusing on message frequency

and latency. The authors detail speciﬁc requirements

for cooperative lane change focusing on message fre-

quency and latency, thereby neglecting the reliability.

In (Mir and Filali, 2018), the authors classify co-

operative lane change as an active road safety applica-

tion and specify requirements for message frequency,

delay, and reliability. Both studies only focus on

cooperative lane change and leave requirements for

other CVM applications open.

For cooperative maneuvers in general, (Boban

et al., 2018) derive requirements for latency, reliabil-

ity, and communication range from (3GPP, 2019).

The recent literature already provides communi-

cation requirements for CVM, focusing on latency,

reliability, and message frequency separately. The

impact on the performance of CVM when violating

these requirements has not been thoroughly validated.

We think that, concerning the quality of commu-

nication, the performance of CVM relies on the AoI.

Therefore, our study provides a reasonable network

performance study, explicitly focusing on CVM for

ad hoc communication in an urban intersection sce-

nario.

7 CONCLUSION

In this paper, we introduce a maneuver planning algo-

rithm for a decentralized CVM application. In recent

literature, the evaluation of CVM was foremost eval-

uating for small-scale scenarios with less than three

vehicles. In order to allow for large-scale simula-

tions, we proposed an efﬁcient method to plan tra-

jectories for CVM and described how to identify col-

lisions with other received trajectories. Accordingly,

we proposed a novel message format for maneuver

coordination, which aims at reducing the message

size and complexity of the trajectory generation to en-

able large-scale simulations. The maneuver planning

application is composed of the generation of trajecto-

ries, the trajectory processing (ﬁltering and collision

detection), and the cooperation logic, where we ex-

plained the function of each module in detail. With

these tools at hand, we show that our planning al-

gorithm is capable of simulating CVM in large-scale

scenarios with more than hundreds of vehicles at the

same time, all equipped with a CVM application over

a simulation time of 300s.

Our simulation results show that the trafﬁc dy-

namic, i. e., the required time to reach the intersection,

varies signiﬁcantly, which motivates the need for high

quality of communication as the planned trajectory

requires high frequent updates. Further, our novel

planning algorithm is capable of generating hundreds

of different trajectories, where our design generates

more trajectories for complex scenarios such as a left-

turn compared to a straight road. When we analyze

the quality of communication and cooperation cost

while planning a cooperative maneuver with other ve-

hicles, we see that the standard deviation and mean

of the required speed reduction for the CV decreases

with increasing distance to the intersection. On the

other side, the quality of communication signiﬁcantly

decreases above 50 m to the intersection. Thus, imper-

fect communication causes CVM to plan on outdated

trajectories, where the AoI of trajectories can be up to

400ms during the cooperation process.

ACKNOWLEDGEMENTS

The German Federal Ministry for Economic Affairs

and Energy (BMWi) supports this research within

the project IMAGinE - Intelligente Man

over Au-

tomatisierung - kooperative Gefahrenvermeidung in

Echtzeit (Intelligent maneuver automation – cooper-

ative hazard avoidance in real time).

Furthermore, this work has been funded by the

DFG within the CRC 1053 - MAKI (B1).

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

404

REFERENCES

3GPP (2019). Enhancement of 3GPP support for V2X sce-

narios. Technical speciﬁcation, 3rd Generation Part-

nership Project.

Bischoff, D., Berninger, H., Knapp, S., Meuser, T.,

Richerzhagen, B., H

aring, L., and Czylwik, A. (2019).

Safety-relevant v2x beaconing in realistic and scalable

heterogeneous radio propagation fading channels. In

In Proceedings of the International Conference on

VEHITS, volume 5. SCITEPRESS.

Boban, M., Kousaridas, A., Manolakis, K., Eichinger, J.,

and Xu, W. (2018). Connected roads of the future:

Use cases, requirements, and design considerations

for Vehicle-to-Everything communications. IEEE Ve-

hicular Technology Magazine, 13(3):110–123.

Correa, A., Alms, R., Gozalvez, J., Sepulcre, M., Rondi-

none, M., Blokpoel, R., L

ucken, L., and Than-

davarayan, G. (2019). Infrastructure support for coop-

erative maneuvers in connected and automated driv-

ing. In IEEE Intelligent Vehicles Symposium, pages

20–25.

Cunha, F., Villas, L., Boukerche, A., Maia, G., Viana, A.,

Mini, R. A., and Loureiro, A. A. (2016). Data com-

munication in VANETs: Protocols, applications and

challenges. Ad Hoc Networks, 44:90–103.

uring, M., Franke, K., Balaghiaseﬁ, R., Gonter, M.,

Belkner, M., and Lemmer, K. (2014). Adaptive coop-

erative maneuver planning algorithm for conﬂict res-

olution in diverse trafﬁc situations. In ICCVE, pages

242–249.

Ericson, C. (2004). Real-Time Collision Detection. CRC

Press, Inc., USA.

Franke, K., Gonter, M., D

uring, M., Lemmer, K., Balaghi-

aseﬁ, R., and K

ukay, F. (2014). A reference archi-

tecture for CISS/CDAS within the ﬁeld of cooperative

driving. In ICCVE, pages 357–363.

Fu, L. and Hellinga, B. (2000). Delay variability at signal-

ized intersections. Transportation Research Record,

1710(1):215–221.

Intelligent Transport Systems (2010). Communications ar-

chitecture. European standard, ETSI.

Intelligent Transport Systems (2017). Basic Transport Pro-

tocol. Standard, ETSI.

Intelligent Transport Systems (2018). Decentralized Con-

gestion Control Mechanisms for ITS operating in the

5 GHz range. Technical speciﬁcation, ETSI.

Intelligent Transport Systems (2019). ITS-G5 access layer

speciﬁcation for its operating in the 5 GHz frequency

band. European standard, ETSI.

Jesenski, S., Stellet, J. E., Schiegg, F., and Z

ollner, J. M.

(2019). Generation of scenes in intersections for the

validation of highly automated driving functions. In

IEEE Intelligent Vehicles Symposium, pages 502–509.

Krajzewicz, D., Hertkorn, G., R

ossel, C., and Wagner, P.

(2002). SUMO (Simulation of Urban MObility); an

open-source trafﬁc simulation. In Proceedings of the

4th middle East Symposium on Simulation and Mod-

elling (MESM20002).

Lehmann, B., G

unther, H., and Wolf, L. (2018). A generic

approach towards maneuver coordination for auto-

mated vehicles. In 21st ITSC, pages 3333–3339.

Llatser, I., Michalke, T., Dolgov, M., Wildsch

utte, F., and

Fuchs, H. (2019). Cooperative Automated Driving

Use Cases for 5G V2X Communication. In IEEE 2nd

5G World Forum, pages 120–125.

Mecklenbrauker, C. F., Molisch, A. F., Karedal, J., Tufves-

son, F., Paier, A., Bernado, L., Zemen, T., Klemp,

O., and Czink, N. (2011). Vehicular channel char-

acterization and its implications for wireless system

design and performance. Proceedings of the IEEE,

99(7):1189–1212.

Meuser, T., Bischoff, D., Richerzhagen, B., and Steinmetz,

R. (2019). Cooperative ofﬂoading in context-aware

networks: A game-theoretic approach. In Proceedings

of the 13th ACM International Conference on Dis-

tributed and Event-Based Systems, DEBS ’19, page

55–66, New York, NY, USA. Association for Com-

puting Machinery.

Milanes, V., Godoy, J., Villagra, J., and Perez, J. (2011).

Automated on-ramp merging system for congested

trafﬁc situations. IEEE Transactions on Intelligent

Transportation Systems, 12(2):500–508.

Mir, Z. H. and Filali, F. (2018). Applications, requirements,

and design guidelines for multi-tiered vehicular net-

work architecture. In Wireless Days, pages 15–20.

Schiegg, F. A., Brahmi, N., and Llatser, I. (2019). Ana-

lytical performance evaluation of the collective per-

ception service in C-V2X Mode 4 networks. In IEEE

ITSC, pages 181–188.

Shi, W., Kong, Q., and Liu, Y. (2008). A GPS/GIS inte-

grated system for urban trafﬁc ﬂow analysis. In 11th

International IEEE Conference on Intelligent Trans-

portation Systems, pages 844–849.

Sommer, C., German, R., and Dressler, F. (2011). Bidi-

rectionally coupled network and road trafﬁc simula-

tion for improved IVC analysis. IEEE Transactions

on Mobile Computing, 10(1):3–15.

Strigel, E., Meissner, D., Seeliger, F., Wilking, B., and Diet-

mayer, K. (2014). The Ko-PER intersection laserscan-

ner and video dataset. In 17th ITSC, pages 1900–1901.

Varga, A. and Hornig, R. (2008). An Overview of the OM-

NeT++ Simulation Environment. In Proceedings of

the 1st International Conference on Simulation Tools

and Techniques for Communications, Networks and

Systems & Workshops, pages 60:1–60:10.

Xu, W., Willecke, A., Wegner, M., Wolf, L., and Kapitza,

R. (2019). Autonomous maneuver coordination via

vehicular communication. In 49th Annual IEEE/IFIP

International Conference on Dependable Systems and

Networks Workshops (DSN-W), pages 70–77.

Zhao, X., Li, X., Xu, Z., and Chen, T. (2019). An optimal

game approach for heterogeneous vehicular network

selection with varying network performance. IEEE In-

telligent Transportation Systems Magazine, 11(3):80–

92.

Zheng, K., Zheng, Q., Chatzimisios, P., Xiang, W., and

Zhou, Y. (2015). Heterogeneous vehicular network-

ing: A survey on architecture, challenges, and so-

lutions. IEEE Communications Surveys Tutorials,

17(4):2377–2396.

What Cooperation Costs: Quality of Communication and Cooperation Costs for Cooperative Vehicular Maneuvering in Large-scale

Scenarios

405