Management of Intelligent Vehicles: Comparison and Analysis

Guilhem Marcillaud

, Valerie Camps

, St

ephanie Combettes

, Marie-Pierre Gleizes

and Elsy Kaddoum

Institut de Recherche en Informatique de Toulouse, Universit

e Paul Sabatier, Toulouse,France

Institut de Recherche en Informatique de Toulouse, Universit

e Jean Jaures, Toulouse, France

Keywords:

Connected and Autonomous Vehicles, Cooperative Interaction, Intelligent Transportation System.

Abstract:

The main purpose of Connected and Autonomous Vehicles (CAVs) is to ensure optimal safety while improving

user comfort. Many studies address the problem of CAVs to improve speciﬁc driving situations (intersection

management, trafﬁc ﬂow management, etc.). In this paper, we propose both a comparison of these systems

according to a set of criteria and an analysis to assist the development of CAVs ﬂeets. This analysis shows that,

among other, the ad-hoc algorithms use similar data (position, speed, etc.) and that the decisions for vehicles

are based on cooperative processing for speciﬁc situations. The objective of this paper is to provide a guide

for the design of CAVs ﬂeet capable of managing all trafﬁc situations.

1 INTRODUCTION

Over the past years, the population of large urban ar-

eas has increased signiﬁcantly. The resulting trafﬁc

jams cause stress, considerable time loss and harm-

ful pollution (Pulter et al., 2011). In addition, the

majority of road accidents are due to human error

such as inattention or drunk driving (French govern-

ment, 2019). Level 3 to 5 Autonomous Driving Sys-

tem (ADS) equipped vehicles (SAE On-Road Auto-

mated Vehicle Standards Committee, 2018) are useful

to prevent accidents related to those errors.

Vehicles have now the ability to send speciﬁc

messages to other vehicles and road infrastructures

(Sharif et al., 2018); V2V (Vehicle To Vehicle) and

V2I (Vehicle To Infrastructure) communications are

called ”connected”. Such a vehicle controlled by

a computer system is called a Connected and Au-

tonomous Vehicle (CAV).

Although many CAVs ﬂeet have been developed

in recent years, it is difﬁcult to obtain an artiﬁcial

system that can efﬁciently address all situations en-

countered during a journey. This problem is increased

by the necessary use of simulators to test these CAV

ﬂeet. Road trafﬁc encompasses many different situa-

tions. It is then very complicated to ﬁnd a simulator

able to model all encountered situations including a

large number of vehicles, with characteristics close to

reality (Sobieraj et al., 2017). In this paper, different

CAVs ﬂeets are ﬁrst compared. Then based on this

analysis, a guideline for a CAVs ﬂeet capable of man-

aging all road situations is proposed.

The paper is organised as follow: section 2 deﬁnes

the requirements and criteria required. Then, sections

3 to 5 describe existing CAV ﬂeet dedicated to each

trafﬁc situation and analyse them based on the previ-

ously deﬁned criteria. section 6 presents the synthesis

of all the studied system. Before concluding, section

7 proposes a guideline for designing a robust CAV

ﬂeet.

2 REQUIREMENTS OF A CAVs

FLEET

As a driver, a human or an ADS, has the duty to ensure

safety of all passengers. We assume that this point is

covered in all the studied systems. The comparison

is based on the situations addressed in these systems.

Multiple criteria such as the purpose of the system,

the trafﬁc mix, and the volume of communications are

used to study them.

2.1 Focus on the Addressed Situations

During a journey, a vehicle can encounter many dif-

ferent situations such as: (i) trafﬁc congestion, (ii)

intersections and (iii) lane merging. These three

situations are focused in the current research works.

258

Marcillaud, G., Camps, V., Combettes, S., Gleizes, M. and Kaddoum, E.

Management of Intelligent Vehicles: Comparison and Analysis.

DOI: 10.5220/0009117802580265

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 1, pages 258-265

ISBN: 978-989-758-395-7; ISSN: 2184-433X

Whatever the encountered situation, a vehicle has the

objective of taking the best decision in coordination

with other vehicles to manage the situation in an opti-

mal and safe manner. During the experimentation of

a system, scenarios are designed; their consideration

is detailed during the analysis.

2.2 Criteria of Comparison

For highlighting the strengths and weaknesses of the

studied systems, we propose eight relevant criteria

identiﬁed in the literature.

1 - System Objectives: among the compared

systems, three main purposes stand out: minimising

travel time (Agarwal and Paruchuri, 2016), reducing

energy consumption and pollution (Pulter et al., 2011)

and smooth driving (Di Vaio et al., 2019). It should

be noted, that these issues are not independent.

2 - Locality of the Decision: (Barthelemy and

Carletti, 2017) being in the context of a CAVs ﬂeet,

the implementation of the control process is studied:

distributed or centralised.

3 - Communication Volume: (Sharif et al., 2018)

connected vehicles have the ability to receive and pro-

vide data through increasingly efﬁcient communica-

tion. The amount of data ﬂow can become very large

and be a problem.

4 - System Robustness: (Ioannou and Zhang,

2018) robust system can react to unexpected events.

The ability of CAVs to act in the event of disruptions

is assessed.

5 - Knowledge of the Scenario: the experimenta-

tion of a system takes place within the framework of

a speciﬁc scenario that CAVs might know in advance.

This knowledge inﬂuences the reaction of a CAV after

it has detected unexpected events in the scenarios.

6 - Mixed Trafﬁc: (Rios-Torres and Malikopou-

los, 2016) the possibility of the presence of non ADS-

equipped vehicles and other types of users (pedestri-

ans, cyclists, ...) must be taken into account.

7 - Scaling: the number of vehicles inﬂuences the

reliability of the system and may reveal speciﬁc prob-

lems that need to be considered.

8 - Simulation Model: (Sobieraj et al., 2017) to

know its effectiveness, a system is evaluated under

certain conditions. Microscopic models are used to

observe a situation with a few numbers of vehicles

with a close representation of the vehicles’ dynamics.

Macroscopic models aim at observing a large number

of vehicles (the trafﬁc in a big city).

The following section presents and evaluates 16

CAVs systems identiﬁed in the literature that address

a solution to a road problematic. The purpose is to

extract a guideline and the requirements for a system

capable of managing all road situations. Each sys-

tem is studied according to the eight criteria deﬁned

in section 2.2.

3 PANORAMA OF CAV FLEET

FOR TRAFFIC FLOW

The objective of these CAVs ﬂeet is to maintain a

smooth trafﬁc in the 8 following situations.

3.1 Learning the Best Path based on

Congestion

Large urban areas experience congestion problems on

major roads can be mitigated by the use of CAVs

and learning mechanisms (Barthelemy and Carletti,

2017). The authors present a Multi-Agent System

(MAS) with CAV agents. Some CAV agents are

called strategic and choose the best path to improve

travel time using a neural network. Each agent uses its

knowledge on nearby roads to take its decision. The

experiment focuses on the Chicago city network, by

varying the proportion of strategic CAV agents in the

system. The authors observe that an increase of the

strategic CAV agents also increases the trafﬁc ﬂow.

Using local data only provides a low volume of com-

munications and agents do not need to know the sce-

nario to learn and choose the best route. In this exper-

iment, Matlab is used to simulate at the macroscopic

level a large number of vehicles and therefore to suc-

cessfully scale up.

3.2 Choosing the Best Path using

Pheromone Deposit

(Mouhcine et al., 2018) propose a MAS in which

agents follow a pheromone based strategy to in-

dicate congestion on an axis. In their MAS

called ”Distributed Vehicle Trafﬁc Routing System”

(DVTRS), pheromones are dropped by stopped ve-

hicles. Through communicating infrastructures, non

grounded vehicles are aware of the densest roads and

can therefore choose a more ﬂuid road. The advan-

tage of this system is that vehicles require few knowl-

edge and communication to take their decisions. On

the other hand, this MAS is not deﬁned in a hybrid

context: the CAVs will then lack information about

non connected vehicles. In this DVTRS, each vehi-

cle is an agent with a total control over its own de-

cisions/actions and can be abstracted from the par-

ticularities of each scenario through the use of lo-

cal knowledge only. The DVTRS experiment at the

Management of Intelligent Vehicles: Comparison and Analysis

259

macroscopic level is located in the road network of a

large city.

3.3 Reduce Shock Waves using

Communications

A shock wave is often the result of intense actions

such as emergency braking and it results in slow-

downs and an increase of trafﬁc density. (Di Vaio

et al., 2019) propose a control strategy with MAS in

which CAVs coexist with non autonomous vehicles,

both being connected. CAV agents use the exchanged

data to adapt their speed and to be informed as quickly

as possible about the actions of others. A better antici-

pation before the shock wave, prevents vehicles to act

urgently: vehicles thus have a more ﬂuid and more

ecological driving behaviour. CAVs make local de-

cisions using certain data (velocity, direction, target)

obtained every 20 ms. The experiment simulates vehi-

cles dynamics in a microscopic simulation and shows

that this strategy goes to scale with a high number

of vehicles.

3.4 Reduction of Consumption by

Anticipation

In the case of CAVs, communications can provide

and process many more information than a human

driver. The use of a predictive control model provides

to the vehicle anticipation capabilities. This type of

model proposed by (Kamal et al., 2015) combines

lane change, acceleration and braking data. In a con-

text of perfect communications, CAVs accelerate op-

timally and predict the best time to change lane: they

save fuel while reducing travel time. But as commu-

nications must be perfect and data up to date, it re-

quires a large volume of communications and in case

of incorrect or missing communications, the system

cannot work. A CAV does not have any knowledge of

the scenario used for the experiment. This one simu-

lates six vehicles, on two lanes using a microscopic

simulator, the scaling has not been proved.

3.5 Selection of the Least Congested

Intersections

The distribution of vehicles at intersections can be un-

even and result in trafﬁc congestion. Through the

use of intelligent communications and infrastructure

controlling intersections, CAV agents can decide to

change the route as shown by (Lin and Ho, 2019).

This MAS approach allows a better distribution of

CAV agents between intersections, leading to a re-

duction of intersection congestion. Each agent de-

cides locally even if a server may ask it to change its

route. Although communication is essential for the

proper functioning of the system, the volume of in-

formation exchanged is not high. The CAV agent

has knowledge about the scenario through the graph

representing all the intersections given as an input at

the entrance of the system. The presented experimen-

tation concerns the macro level and is scalable.

3.6 Cooperation of CAVs in Squads

Platooning is a squad based vehicle formation in

which the front vehicle is the leader. The advantage of

such a strategy is the fast and secure information shar-

ing between the vehicles of the squad (Llatser et al.,

2015). Thus, the vehicles follow the leader and their

speed is aligned its. Vehicles have the possibility to be

closer than usual and do not risk collisions as they all

act in coordinated way, the driving is smooth. This

type of training was experimented with real vehicles

as part of the ”Grand Cooperative Driving Challenge

2016” (Englund et al., 2016) in which two CAVs

squads aimed to merge into a single squad. The merg-

ing action is initiated by leaders of each squad, and

from that moment each vehicle wishing to join the

squad binds by communication to one of the present

vehicles. Vehicles then insert themselves one behind

the other. The platooning method centralises deci-

sions (speed and travel) at the leader’s level but each

vehicle has autonomy for some actions, such as de-

ciding to leave the squad when desired.

3.7 Emergency Vehicle in Heavy Trafﬁc

Driving emergency vehicles (EV) in heavy or con-

gested trafﬁc is often problematic, as they are slowed

down even if vehicles trapped in trafﬁc try to clear the

way for the EV. (Agarwal and Paruchuri, 2016) pro-

pose a strategy for the EV that consists in choosing the

fastest lane at one time and sticking to it, while asking

other Connected Vehicles (CV) to clear that lane. This

strategy, constraining other CVs, is not efﬁcient when

the trafﬁc density is too high. Another strategy con-

sists in choosing the best lane according to the number

of vehicles visible by the EV and the possible speed

on the lane. It thus allows EVs to reduce their travel

time. As only one message is exchanged between the

EV and each vehicle, the volume of communication is

low. The experiment of both strategies is carried out

on a two kilometres road with the SUMO simulator

(Behrisch et al., 2011) allowing microscopic simula-

tion of vehicles.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

260

3.8 Anticipation of Lane Change in the

Event of an Incident

When a vehicle has an incident that brings to a stand-

still on a lane, the reduction of available lan might

lead to a trafﬁc congestion. As a consequence, in-

tense braking and lane changes at lower speeds may

arise. That is why (Ioannou and Zhang, 2018) pro-

pose to inform the CAVs of the lane impracticability

in advance. The CAV agent, the incident initiator or

one CV receiving the message, can in turn dissem-

inate the information. Although lane reduction still

has an impact on trafﬁc, congestion is reduced. Each

CAV agent takes its own decision based on the infor-

mation received and can communicate with other ve-

hicles, ADS-equipped or not. As only the presence of

the incident is exchanged, the communication volume

is low. A microscopic digital experiment is presented

and this system succeeds in scaling up.

This study highlights the contribution of commu-

nications, and the use of a local control for a CAV in

order to take a better decision.

4 PANORAMA OF CAVs FLEET

FOR INTERSECTION

MANAGEMENT

Until now, the road code and trafﬁc signs were suf-

ﬁcient to guarantee the safety of users to manage in-

tersection when all the rules are respected. An inter-

section may be seen as a dangerous area. In this sec-

tion, the studied CAVs systems addressing this issue

are using two different approaches: the centralised

approach with an agent in charge of the intersection,

Infrastructure Management Agent (IMA), which con-

centrates decisions, and the decentralised approach in

which CAV agents cooperate with each other to ﬁnd

a consensus, with an IMA possibly helping this coop-

eration.

4.1 Coordination of Vehicles by

Reservation in an Intersection

In the centralised approach, an IMA communicates

with vehicles wishing to cross the intersection by in-

dicating how to proceed, and uses the notion of reser-

vation (Jin et al., 2012; Dresner and Stone, 2008; Pul-

ter et al., 2011). (Jin et al., 2012) propose a Multi-

Agent Intersection Management System (MAIMS)

using reservations. A CAV agent approaching the

intersection has to request a reservation by sending

its objective. The IMA reserves a slot that allows

the CAV to cross the intersection without being in

conﬂict with other vehicles. The particularity of the

MAIMS is that it uses a vehicle scheduling approach

instead of a FIFO one. All the near CAV agents re-

port their speed, position and destination to the IMA,

then it organises their crossing order. Authors have

noticed that with a large number of CAVs, the volume

of communications becomes problematic. The ex-

periments carried out on SUMO with MAIMS show

a clear improvement of the vehicle ﬂow compared

to the FIFO-based approach. However, this system

shows no improvement in fuel consumption: the au-

thors suggest that it comes from vehicle trajectories.

The presented experimentation uses a large number of

vehicles: the scaling for this scenario is validated.

4.2 Optimisation of Trajectories in an

Intersection

(Kamal et al., 2014) propose a Vehicle-Intersection

Coordination Scheme (VICS) to obtain a more ﬂuid

trafﬁc at an intersection and a reduced fuel consump-

tion. To reach these objectives, they optimise the

trajectory of each vehicle agent using an IMA lo-

cated in the intersection, which schedules the CAVs.

The authors experiment VICS on Matlab, which is

distant from real vehicle conditions and show that

VICS eliminates almost all the stops at the intersec-

tion: this signiﬁcantly improves trafﬁc ﬂow and

fuel consumption for each vehicle. An experiment

with a large number of CAVs shows the transition to

system-wide. However, VICS operates in the context

of a trafﬁc composed solely of CAVs that are coop-

erative with the IMA and the volume of communica-

tions may become too large.

A centralised approach leads to signiﬁcant data

exchange and to possible congestion of communica-

tion ﬂow. CAVs, equipped with V2X perception and

communication devices, can also negotiate to facili-

tate the resolution of non-cooperative situations.

4.3 Negotiations between CAVs in an

Intersection

(Gaciarz et al., 2015) propose a negotiation mecha-

nism between CAV agents wishing to cross an inter-

section. In this mechanism, the intersection is divided

into cells and CAV agents communicate in order to

obtain a coherent conﬁguration. The IMA of this sys-

tem has a greater knowledge than CAV agents and

has the ability to propose conﬁgurations that improve

the overall ﬂuidity without impacting local agent sat-

isfaction. Each agent is free to accept or refuse the

Management of Intelligent Vehicles: Comparison and Analysis

261

IMA proposition according to its preferences and ob-

jectives. Negotiation includes all the agents present

in the intersection at the current time and it does not

generate a high volume of communications; however

the presence of non-cooperative vehicles is not con-

sidered. Experimentation with this mechanism shows

that the length of queues in the intersection is reduced.

The simulation using threads does not allow to have

a simulation close to reality but the number of vehi-

cles is sufﬁcient to validate the scalability.

These studies underline that the CAVs can negoti-

ate together the appropriate time to cross an intersec-

tion with a low volume of communications. Cooper-

ation between CAVs and infrastructure improves the

smoothness of the crossing.

5 PANORAMA OF CAVs FLEET

FOR THE MANAGEMENT OF

LANE MERGING

The third situation that a CAVs must be able to handle

is the lanes merging. It is difﬁcult because a vehicle

that wants to join a lane has to adapt its speed to the

other vehicles in that lane and to ﬁnd the best place to

insert into it with a minimum of discomfort for other

vehicles.

5.1 Coordination of CAVs at Merging

Point

The merging zone of the two lanes may be man-

aged by an agent present in a connected infrastruc-

ture. (Rios-Torres and Malikopoulos, 2016) propose a

controller that organises vehicles in FIFO for this area

by giving each of them an identiﬁer to organise them.

This system seeks to optimise fuel consumption by

calculating the most appropriate time and speed to in-

sert smoothly and safely at the merging point. A con-

troller ordering the insertion allows the management

of CVs but not CAVs. Experiments with this system,

carried out under Matlab, have shown a reduction in

fuel consumption and a more ﬂuid merging regard-

less of the number of vehicles. However, the authors

highlight a limit when the vehicle speed is too high.

5.2 Cooperation Upstream of the

Merging Point

(Wang et al., 2018b) propose a protocol for cooper-

ation between vehicles before reaching the merging

point of the lanes. Each CAV agent sends its data

(acceleration, speed and position) to an infrastructure

agent which sequences the vehicles and indicates to

each of them which vehicle is in front of it. A CAV

agent receives data from the vehicle in front of it in or-

der to adapt its speed to reach the merging point at the

optimal time, even if both are on two different lanes.

A large volume of data is required to build this sys-

tem. An agent-based simulation made of six vehicles

in Unity (Wang et al., 2018a) compares the use of this

system to human conductors; it results in a reduction

of travel time and energy savings for CAV agents.

Although the previous approach presents very in-

teresting results, it requires the presence of a con-

troller and therefore an infrastructure to host it. Some

decentralised approaches avoiding infrastructures are

presented.

5.3 Decentralised Cooperation before

the Merging Point

(Mosebach et al., 2016) propose a decentralised lane

merging control algorithm in which CAVs use com-

munications to determine if a vehicle is present in the

area preceding the lane merging. If a CAV has de-

tected vehicles, it slows down and calculates a tra-

jectory to cross the merging point without collision.

When the CAV considers that this trajectory is cor-

rect, it follows it by re-accelerating. The required

communications concern the other vehicles as well

as their speed. The CAVs know the road typology

and cross the merging zone in FIFO. The simulation

of two lanes merging with 40 vehicles shows that the

CAVs are inserted smoothly while maintaining a safe

distance.

5.4 Decentralised Merging in Mixed

Trafﬁc

(Sobieraj, 2018) proposes a cooperation protocol be-

tween CAVs. In an area, known by the CAV agents

and close to the merging point, the CAVs of two dif-

ferent lanes communicate to form pairs. Then, the

CAV agents of each couple coordinate their accelera-

tion to position themselves at a safe distance one be-

hind the other. The resulting communication vol-

ume is low and acceleration calculations are based

on models that allow ﬂuid insertion. The experi-

mentation consists in merging two lanes crossed by

a dense vehicles ﬂow. The presence of CAV agents

improves trafﬁc ﬂow and waiting times have almost

disappeared. This approach can be used in a mixed

trafﬁc despite lower performance.

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

262

5.5 Change of Cooperative Path with

Conﬁdence Index

(Monteil et al., 2013) have designed a mechanism for

cooperation between CAVs agents by considering un-

reliable communications. This method is based on

a conﬁdence index that each CAV agent calculates

for each vehicle communicating with it, thus it may

evaluate the quality of the communications. In the

case of missing or unreliable communications, the

CAV can only use its own perceptions to make deci-

sions: the presence of unconnected vehicles is con-

sidered. The local control and the ability to use

only its perceptions validate the robustness of this

method. Adding to this mechanism a connected in-

frastructure (Gu

eriau et al., 2016), provides access to

additional information to the CAV agents with an in-

crease of the volume of communications. The sim-

ulation of a ﬂow of a large number of vehicles shows

that the presence of CAVs and infrastructure smooths

the merging zone.

6 SYNTHESIS AND ANALYSIS

Table 1 summarises and synthesises the results (with

ratings ranging from - - to ++; - - means not at all and

++ totally, while NA means not addressed).

We note that the majority of systems only work in

a fully connected and cooperative environment. The

presence of unconnected, so eventually uncoopera-

tive vehicles, makes difﬁcult for CAVs to determine

their behaviour. Nevertheless, CAVs and non ADS-

equipped Vehicles will very probably have to coexist

before the trafﬁc becomes fully autonomous.

First of all, only the system of (Wang et al., 2018b)

has an explicit inﬂuence on the travel time, the pol-

lution emitted and the ﬂuidity of the trafﬁc ﬂow.

However, it only concerns the insertion in a lane

merging. Systems that aim to ﬁnd the best route

mainly inﬂuence travel time even if it is not clearly

stated.

Managing an intersection requires a substantial

volume of communications when a large number of

vehicles are present. The studied systems need that ei-

ther IMA or CAV agents have some scenario knowl-

edge. Thus, each intersection must be speciﬁed and

the modiﬁcation structure of an intersection requires

a new speciﬁcation.

Only systems whose objective is to ﬁnd the best

path are robust because it does not require to make a

quick decision. This is due to the complexity of the

real vehicles and all the factors that need to be con-

sidered when decisions are made within a very short

time frame.

In all three situations, the fact that a CAV agent

shares its objectives, speed and position allows other

vehicles to have a better anticipation and understand-

ing of this vehicle. Most of the conﬂict situations that

a vehicle may encounter in these systems involve the

sharing of space.

The presence of unconnected entities is often not

considered in the studied systems (see Table 1). If we

consider that a not connected entity in a trafﬁc is a

mobile or an unpredictable obstacle, then a CAV will

only have to avoid that entity.

A CAV that only needs local information and a

small amount of knowledge can adapt to unexpected

situations. As shown in Table 1, only systems with a

positive assessment for these two criteria also have a

positive assessment for robustness.

Among the 16 systems presented, 12 are MAS in

which CAV agents use a cooperative process when

confronted with one of the three situations.

7 A GUIDELINE TO SATISFY

THE REQUIREMENTS OF A

CAVs FLEET

From the requirements (Section 2) and the analysis of

the different systems, we propose a guideline to de-

sign CAVs ﬂeet, which consists in high level design

principles concerning decentralisation, communica-

tion, robustness, knowledge of the scenario, mixed

trafﬁc and scalability.

All the CAVs in the studied systems have the

same design and consequently the same behaviour

in a given situation. However, in real life they have

several designs leading to different behaviours in the

same situation. Thus, the results given in these studies

have to be taken with hindsight.

Because centralised systems use a lot of informa-

tion considered as safe, regardless of uncooperative

behaviour or incidents, a local control of the CAV

using partial perceptions and able to adapt to unex-

pected situations is mandatory.

In order to avoid bottlenecks of communications,

the use of a centralised system for communications

should be avoided. However, each CAV should not

share all its data to all other CAVs because the volume

of communications will explode. It is required for a

CAV to learn which information is useful to share

for a situation in order to limit their amount and to

facilitate the decision process.

Even faced to unknown situation, a robust CAV

must be able to take a correct decision. Moreover,

Management of Intelligent Vehicles: Comparison and Analysis

263

Table 1: Synthesis of Connected and Autonomous vehicles Systems.

System

Locality

Travel Time

Pollution

Smooth Driving

Data volume

Robustness

Scenario Knowledge

Mixed Trafﬁc

Scaling

Simulation Model

Trafﬁc ﬂow : Choose the best path

or Smooth driving

Barthelemy et al.

++ + NA NA + + + + - + Macro

Mouhcine et al.

+ + NA NA + - + - + Macro

Di Vaio et al.

++ NA + + - + + + + - Micro

Taguchi et al.

++ NA + + - - + - - Micro

Lin et al.

+ + NA NA + - - - + Micro

Englund et al.

+ - NA NA + + - + - + - - Real

Agarwal et al.

- - + NA + + + - - - - Micro

Ioannou et al.

++ + + NA + + + - + Micro

Intersection

Jin et al. - - NA - + + - - + - + Micro

Kamal et al. - - NA + + + - - + - + Micro

Gaciarz et al. + + NA + + - + + - - + Macro

Lane merging

Rios-Torres et al. + - NA + + + - - + - - Micro

Wang et al. + - + + + - - - + - - Micro

Mosebach et al. ++ NA NA + ++ + - - + - Micro

Sobieraj et al. ++ NA NA + + - + ++ + Micro

Gueriau et al. ++ + NA + + - + - ++ + Micro

it should be able to learn which behaviour is better

for the next time a similar situation will be encoun-

tered. Because the amount of information available is

continuously evolving and it seems unlikely that the

same data would be expressed in the same way by

two different vehicles, a CAV should learn continu-

ously the usefulness of information without seman-

tic. At ﬁrst, the CAV would drive safely, like a new

driver, and it will gradually acquire experience while

driving, improving its behaviour. A CAV that aims at

managing every road situation needs to be able to ex-

change its data and to negotiate with other vehicles.

Moreover, it needs to be able to cooperate with other

connected entities to reﬁne its knowledge of the con-

text and thus, it will lead to have a better behaviour

(Di Vaio et al., 2019; Barthelemy and Carletti, 2017).

It seems that situation awareness can be a barrier

because it is impossible to model all the possible sit-

uations in beforehand. It is desirable for a CAV to

be able to understand the current situation so that it

can adapt its behaviour to it. Coordination between

vehicles will make it possible to solve many trafﬁc

situations. So, cooperation and negotiation must

be implemented in each CAV (Llatser et al., 2015;

Gaciarz et al., 2015; Wang et al., 2018b).

The presence of non CAV could be addressed by

considering them as non cooperative entities with un-

predictable behaviour. The CAV must be able to iden-

tify them and learn how to interact with them, with

avoidance strategy (Degas et al., 2019) and the ex-

change of information without communication.

In the validation step, the evaluation of the CAVs

ﬂeet must be done simultaneously at the micro and

macro levels. The ﬁrst one is necessary to observe

how a CAVs ﬂeet manages a precise situation while

the second one would be used to observe the result at

a larger scale.

If the CAVs ﬂeet design follows the guideline, an

improvement of the road trafﬁc (pollution, travel time

and smooth driving) will be observed.

8 CONCLUSION

This paper proposes a state of the art of CAVs sys-

tems. An analysis of the papers is made according

to a set of criteria that are presented beforehand. It

is shown that communication exchange and coordina-

tion between vehicles allows them to better manage

the situations they encounter. After comparing and

analysing different systems, a guideline is proposed,

mainly: local control and interactions, cooperation

and negotiation between CAVs and lifelong learning.

ACKNOWLEDGEMENTS

This work is part of the neOCampus opera-

tion of the Universit

e Toulouse III Paul Sabatier.

www.neocampus.org

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

264

REFERENCES

Agarwal, A. and Paruchuri, P. (2016). V2V communica-

tion for analysis of lane level dynamics for better ev

traversal. In 2016 IEEE Intelligent Vehicles Sympo-

sium (IV), pages 368–375. IEEE.

Barthelemy, J. and Carletti, T. (2017). A dynamic

behavioural trafﬁc assignment model with strategic

agents. Transportation Research Part C: Emerging

Technologies, 85:23–46.

Behrisch, M., Bieker, L., Erdmann, J., and Krajzewicz,

D. (2011). Sumo–simulation of urban mobility: an

overview. In Proceedings of SIMUL 2011, The Third

International Conference on Advances in System Sim-

ulation. ThinkMind.

Degas, A., Rantrua, A., Kaddoum, E., Gleizes, M.-P., and

Adreit, F. (2019). Dynamic collision avoidance using

local cooperative airplanes decisions. CEAS Aeronau-

tical Journal, pages 1–12.

Di Vaio, M., Fiengo, G., Petrillo, A., Salvi, A., Santini, S.,

and Tufo, M. (2019). Cooperative shock waves miti-

gation in mixed trafﬁc ﬂow environment. IEEE Trans-

actions on Intelligent Transportation Systems.

Dresner, K. and Stone, P. (2008). A multiagent approach

to autonomous intersection management. Journal of

artiﬁcial intelligence research, 31:591–656.

Englund, C., Chen, L., Ploeg, J., Semsar-Kazerooni, E.,

Voronov, A., Bengtsson, H. H., and Didoff, J. (2016).

The grand cooperative driving challenge 2016: boost-

ing the introduction of cooperative automated vehi-

cles. IEEE Wireless Communications, 23(4):146–152.

French government (2019). Key ﬁgures for french

road trafﬁc. https://www.statistiques.developpement-

durable.gouv.fr/. Accessed: 2019-02-18.

Gaciarz, M., Aknine, S., and Bhouri, N. (2015). Automated

negotiation for trafﬁc regulation. Advances in Social

Computing and Multiagent Systems, pages 1–18.

eriau, M., Billot, R., El Faouzi, N.-E., Monteil, J.,

Armetta, F., and Hassas, S. (2016). How to assess the

beneﬁts of connected vehicles? a simulation frame-

work for the design of cooperative trafﬁc management

strategies. Transportation research part C: emerging

technologies, 67:266–279.

Ioannou, P. and Zhang, Y. (2018). Reducing truck emis-

sions and improving truck fuel economy via intelli-

gent transportation system technologies. Retrieved

from https://escholarship.org/uc/item/2rr3b8mz.

Jin, Q., Wu, G., Boriboonsomsin, K., and Barth, M. (2012).

Multi-agent intersection management for connected

vehicles using an optimal scheduling approach. In

2012 International Conference on Connected Vehicles

and Expo (ICCVE), pages 185–190. IEEE.

Kamal, M. A. S., Imura, J.-i., Hayakawa, T., Ohata, A., and

Aihara, K. (2014). A vehicle-intersection coordina-

tion scheme for smooth ﬂows of trafﬁc without using

trafﬁc lights. IEEE Transactions on Intelligent Trans-

portation Systems, 16(3):1136–1147.

Kamal, M. A. S., Taguchi, S., and Yoshimura, T. (2015).

Efﬁcient vehicle driving on multi-lane roads using

model predictive control under a connected vehicle

environment. In 2015 IEEE Intelligent Vehicles Sym-

posium (IV), pages 736–741. IEEE.

Lin, S.-H. and Ho, T.-Y. (2019). Autonomous vehicle rout-

ing in multiple intersections. In Proceedings of the

24th Asia and South Paciﬁc Design Automation Con-

ference, pages 585–590. ACM.

Llatser, I., K

uhlmorgen, S., Festag, A., and Fettweis, G.

(2015). Greedy algorithms for information dissem-

ination within groups of autonomous vehicles. In

2015 IEEE Intelligent Vehicles Symposium (IV), pages

1322–1327. IEEE.

Monteil, J., Billot, R., Sau, J., Armetta, F., Hassas, S.,

and Faouzi, N.-E. E. (2013). Cooperative highway

trafﬁc: Multiagent modeling and robustness assess-

ment of local perturbations. Transportation research

record, 2391(1):1–10.

Mosebach, A., R

ochner, S., and Lunze, J. (2016). Merging

control of cooperative vehicles. IFAC-PapersOnLine,

49(11):168–174.

Mouhcine, E., Mansouri, K., and Mohamed, Y. (2018).

Solving trafﬁc routing system using VANet strategy

combined with a distributed swarm intelligence opti-

mization. Journal of Computer Science, page 14.

Pulter, N., Schepperle, H., and B

ohm, K. (2011). How

agents can help curbing fuel combustion: a perfor-

mance study of intersection control for fuel-operated

vehicles. In The 10th International Conference on

Autonomous Agents and Multiagent Systems-Volume

2, pages 795–802. International Foundation for Au-

tonomous Agents and Multiagent Systems.

Rios-Torres, J. and Malikopoulos, A. A. (2016). Auto-

mated and cooperative vehicle merging at highway

on-ramps. IEEE Transactions on Intelligent Trans-

portation Systems, 18(4):780–789.

SAE On-Road Automated Vehicle Standards Committee

(2018). Taxonomy and deﬁnitions for terms related

to on-road motor vehicle automated driving systems.

SAE Standard J, 3016:1–16.

Sharif, A., Li, J. P., and Saleem, M. A. (2018). Internet

of things enabled vehicular and ad hoc networks for

smart city trafﬁc monitoring and controlling: a re-

view. International Journal of Advanced Networking

and Applications, 10(3):3833–3842.

Sobieraj, J. (2018). Methodes et outils pour la concep-

tion de Syst

emes de Transport Intelligents Cooperat-

ifs. PhD thesis, Universite Paris-Saclay.

Sobieraj, J., Hutzler, G., and Klaudel, H. (2017).

Mod

elisation et simulation de la coop

eration dans les

Syst

emes de Transport Intelligents : un comparatif. In

Journ

ees Francophones sur les Syst

emes Multi-Agents

(JFSMA 2017), Caen, France.

Wang, Z., Kim, B., Kobayashi, H., Wu, G., and Barth,

M. J. (2018a). Agent-based modeling and simulation

of connected and automated vehicles using game en-

gine: A cooperative on-ramp merging study. arXiv

preprint arXiv:1810.09952.

Wang, Z., Wu, G., and Barth, M. (2018b). Distributed

consensus-based cooperative highway on-ramp merg-

ing using v2x communications. Technical report, SAE

Technical Paper.

Management of Intelligent Vehicles: Comparison and Analysis

265