Integrating Shared Information into the Sensorial Mapping of

Connected and Autonomous Vehicles

Filipo Studzinski Perotto, Stephanie Combettes, Valerie Camps, Elsy Kaddoum,

Guilhem Marcillaud, Pierre Glize and Marie-Pierre Gleizes

IRIT, University of Toulouse, France

ﬁlipo.perotto, stephanie.combettes, valerie.camps, elsy.kaddoum,

Keywords:

Multi-Agent Coordination, Connected and Autonomous Vehicles, Cooperative Perception.

Abstract:

A connected and autonomous vehicle (CAV) needs to dynamically maintain a map of its environment. Even

if the self-positioning and relative localization of static objects (roads, signs, poles, guard-rails, buildings,

etc.) can be done with great precision thanks to the help of hd-maps, the detection of the dynamic objects

on the scene (other vehicles, bicycles, pedestrians, animals, casual objects, etc.) must be made by the CAV

itself based on the interpretation of its low-level sensors (radars, lidars, cameras, etc.). In addition to the need

of representing those moving objects around it, the CAV (seen as an agent immersed in that trafﬁc environ-

ment) must identify them and understand their behavior in order to anticipate their expected trajectories. The

accuracy and completeness of this real-time map, necessary for safely planning its own maneuvers, can be

improved by incorporating the information transmitted by other vehicles or entities within the surrounding

neighborhood through V2X communications. The implementation of this cooperative perception can be seen

as the last phase of perception fusion, after the in-vehicle signals (coming from its diverse sensors) have al-

ready been combined. In this position paper, we approach the problem of creating a coherent map of objects

by selecting relevant information sent by the neighbor agents. This task requires correctly identifying the posi-

tion of other communicant agents, based both on the own sensory perception and on the received information,

and then correcting and completing the map of perceived objects with the communicated ones. For doing so,

the precision and conﬁdence on each information must be taken into account, as well as the trust and latency

associated with each source. The broad objective is to model and simulate a ﬂeet of vehicles with different

levels of autonomy and cooperation, based on a multi-agent architecture, in order to study and improve road

safety, trafﬁc efﬁciency, and passenger comfort. In the paper, the problem is stated, a brief survey of the

state-of-the-art on related topics is given, and the sketch of a solution is proposed.

1 INTRODUCTION

Road trafﬁc accidents cause approximately 1.3 mil-

lion deaths worldwide per year, and between 20 and

50 million non-fatal injuries (Chen et al., 2019).

The promise is that the arrival of connected and au-

tonomous vehicles (CAVs) will help to reduce those

war-like numbers considerably (Litman, 2020). There

are important and growing investments both in indus-

trial and academic research aiming to develop such

technologies, and almost all the giants of automo-

bile construction and information technology are in-

volved. The estimated value of this billionaire CAV

global market is projected to increase exponentially

* This work is supported by the C2C project, ﬁnanced

by the French Regional Council of Occitanie.

in the next coming years (Jadhav, 2018). But if

this boom is relatively recent, the building blocks

for modern autonomous vehicles started to appear

several decades ago, with anti-lock brakes, traction

control, power assisted steering, adaptive cruise con-

trol, etc., when mechanical components have been re-

placed by electrical, then by electronic successors.

Now, as these components are being successfully

tied together, along with lidar, radar, cameras, high-

deﬁnition mapping, pattern recognition, and AI-based

control and navigation mechanisms, the car is becom-

ing a kind of thinking machine.

In the literature, it is usual to classify CAVs ac-

cording to their level of autonomy, ranging from fully

manual (level 0 on the SAE taxonomy) to fully auto-

mated vehicles (level 5) (Terken and Pﬂeging, 2020).

Due to legal and especially technological limitations,

454

Perotto, F., Combettes, S., Camps, V., Kaddoum, E., Marcillaud, G., Glize, P. and Gleizes, M.

Integrating Shared Information into the Sensorial Mapping of Connected and Autonomous Vehicles.

DOI: 10.5220/0010387604540461

In Proceedings of the 13th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2021) - Volume 1, pages 454-461

ISBN: 978-989-758-484-8

no commercial vehicle is currently capable of au-

tonomously performing all the driving tasks without a

minimum of supervision, not even the most advanced

prototypes. In the perspective of automobile manu-

facturers, the last level of autonomy should be reached

soon, but in fact several locks are still to be lifted be-

fore it happens, which include: preparing people and

cities for the arrival of new and smart transportation

modes, deﬁning a legal and regulatory framework at

national and international levels, adapting telecom-

munication networks to support the amount of data

exchanged by CAVs, improving the technology on

sensors while reducing costs, and above all, increas-

ing the robustness and reliability of the AI algorithms

that control these vehicles.

For safety and fault-tolerant reasons, CAVs must

ensure their essential driving capabilities without any

communication. However, their behavior can be

greatly improved by the possibility of exchanging in-

formation with other connected vehicles and road in-

frastructures in the neighborhood (Sharif et al., 2018).

Vehicular ad-hoc networks (VANETs) are sponta-

neous wireless networks of vehicles, generally orga-

nized according to vehicle-to-everything (V2X) com-

munication architectures, with the objective of help-

ing navigation and providing diverse roadside ser-

vices (Sommer and Dressler, 2014). In addition to

that dedicated short-range communication (DSRC)

network, the tendency is that V2X will heavily relay

on the 5G cellular network (which allows low latency

and large data bandwidths), in a vehicle-to-network-

to-everything (V2N2X) ﬂavor (Hakeem et al., 2020).

To be able to drive autonomously, a CAV needs

to dynamically construct and update a map of its en-

vironment: based on its low-level sensors (radars,

lidars, cameras, etc.), the agent represents the ob-

jects around (identity, position), their current trajec-

tories (speed, direction, acceleration), and supposed

itineraries (road driving steps) (Arnold et al., 2019).

In addition to identifying the roads and lanes, and

estimating its self-position with centimeter precision,

the map of detected moving objects is necessary for

planning the CAV trajectory and for choosing adapted

maneuvers. An accurate perception of the surround-

ing environment is necessary to ensure a reliable be-

havior, which implies transforming sensory data into

semantic information. 3D object detection is a funda-

mental function of this perceptual system.

Supposing that standard and common communi-

cation protocols should be established among differ-

ent car constructors and road trafﬁc authorities, CAVs

will be able to constantly share their own percep-

tions with other surrounding agents, communicating

their high-level updated map of detected objects and

Figure 1: The image represents a CAV perceiving its sur-

rounding environment. Based on its low level sensors (cam-

eras, radars, lidars, etc.) and hd-maps, the vehicle identiﬁes

roads, lanes, and its own position on the street, and projects

a map with the objects perceived in the scene and their re-

spective trajectories.

Figure 2: The vehicle and the neighbor entities that it can

detect and identify, given the range of its sensors and its

ﬁeld of view.

corresponding estimated dynamics. Another alterna-

tive is the transmission of low-level signals, with the

drawback of requiring both very large (then expen-

sive) data transfer bandwidth, and intense processing

effort of the ego-vehicle to interpret its own sensors

and, in addition, the other agents communicated sig-

nals. Since self-driving is a real-time safety-critical

decision problem, and a CAV needs to be able to an-

swer rapidly to unexpected events, the possibility of

adopting the ﬁrst alternative (sharing high-level infor-

mation) should be preferred.

In this paper, we approach the problem of creat-

ing a coherent map of objects by selecting relevant

and reliable information given by the neighbor agents.

The task implies to correctly identify the position of

other communicant agents based both on the own sen-

sory perception and on communicated geolocation in-

formation, then correcting and completing the pro-

jected scene based on matching the objects (perceived

and communicated), and eventually ﬁnding the un-

perceived ones. This procedure can offer two advan-

Integrating Shared Information into the Sensorial Mapping of Connected and Autonomous Vehicles

455

Figure 3: The image represents the map of objects projected

by a CAV, where the blue arrows indicate the positions and

trajectories of objects estimated by the agent, and the or-

ange arrows, the ones communicated by another agent in

the neighborhood. This information can be used, if trusted,

to improve the precision of the map, and to anticipate the

presence of unperceived objects.

tages: increasing the precision of the reconstructed

map of objects by correcting the estimated positions,

and also helping to incorporate non-detected objects.

This functioning is exempliﬁed in Figures 1 to 3.

Since the CAV network constitutes an open con-

nected environment, the possible presence of non-

cooperative or even malicious agents, which send in-

correct data, cannot be ignored. In this case, the trust

on the source, and the conﬁdence on the information

must be considered. When several connected agents

receive and interpret the same signals, a possible solu-

tion to that issue is estimating reliability by measuring

local inconsistencies. In the rest of the paper, Section

2 presents the validation strategy, Section 3 exposes

an overview on the state-of-the-art, Section 4 intro-

duces the proposed solution, and Section 5 concludes

the paper.

2 VALIDATION

Once reliable CAVs will be available, it will be essen-

tial to study how a ﬂeet will be able to interact in order

to maximize collective safety, passenger comfort, and

produce intelligent mobility. Indeed, it is important

to identify the underlying barriers to human accept-

ability of CAVs (Fraedrich and Lenz, 2016), in order

to ensure the proper deployment of this new mode of

transportation. Between humans, in a context of road

interaction, body and facial cues are sent back and

mutually interpreted. A CAV will not provide humans

with such clues, making it more difﬁcult for them to

understand the decisions made by the AI and to adopt

appropriate behavior.

In this work, we would like to include this er-

gonomic dimension into the evaluation of the pro-

Figure 4: CompactSim, using SCANeR Studio (AVS),

adapted to simulate the situation where a human is trans-

ported by a CAV.

posed model, tackling two complementary objectives:

(i) to study the problem of coordinating a collective

of VACs in mixed trafﬁc; (ii) to study their accept-

ability and appropriation by the various human road

users with whom those VACs interact. The use of

an immersive simulation tool is an acceptable alter-

native (Sovani, 2017) to perform preliminary tests

with humans. For this work, a physical simulator in-

tegrating the SCANeR Studio software (Fig. 4) will

be used to implement our validation tests. SCANeR

Studio is one of the high-end simulators currently

available, implementing microscopic trafﬁc simula-

tion, such as SUMO (Lopez et al., 2018), GAMA

(Taillandier et al., 2019), or MATISSE (Torabi et al.,

2018). The platform simulates object detection tasks

performed by the CAV, artiﬁcially introducing impre-

cision, false-detection, and conﬁdence estimates, like

shown in Fig. 5.

Figure 5: Self-positioning, object and road detection, en-

dowed by the SCANeR Studio simulation platform.

In this way, in addition to classic metrics (Friedrich,

2016) into which the performance of the proposed

model is compared to the performance of road trafﬁc

without it, considering average time-travel, and aver-

age number of dangerous or damaging events (acci-

dents, emergence breaks, etc.), the proposed model

will also be tested through the use of simulation to

perform ergonomic experiments with humans trans-

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

456

ported by VACs in order to measure and under-

stand the situations and behaviors that can cause dis-

comfort or fear. From the results of these experi-

ences, machine-human communication strategies can

be proposed, tested, then validated using the simula-

tor, with the aim of explaining the actions taken by

the AI and thus promoting acceptability. Moreover,

this human evaluation of the system as a whole can

enable a subjective metric to compare the efﬁciency

of different cooperation strategies.

3 STATE-OF-THE-ART

The exchange of local sensing information with other

vehicles or infrastructures via wireless communica-

tions by which the perception range can be extended

or augmented up to the boundary of connected vehi-

cles is called cooperative perception. It can provide

oncoming trafﬁc information beyond the line-of-sight

and ﬁeld-of-view of the ego-vehicle, helping the CAV

on its decision making and planning tasks by improv-

ing vehicle awareness on the road, which can promote

safer and smoother autonomous driving (Kim et al.,

2015). The recent literature is rich in examples of the

use of cooperative perception applied to self-driving

problems, and an exhaustive survey is not in the scope

of this paper. Instead of it, we would like to present

in this section some illustrative samples.

Several overviews on the state-of-the-art concern-

ing CAVs have been published recently, such as (Yurt-

sever et al., 2020; Marcillaud et al., 2020; Cavazza

et al., 2019), concerning a domain that is evolving

fast. Roughly speaking, current CAV research on co-

operative models focus on two different sets of prob-

lems: (a) managing the ﬂow of vehicles to maintain

a smooth trafﬁc, and (b) road and lane trajectory co-

ordination. The ﬁrst topic concerns meso and macro-

trafﬁc organization, and approaches problems such as

the reduction of the average travel time, reduction

of congestion, reduction of consumption, itinerary

re-planing to balance trafﬁc density, etc. The sec-

ond topic concerns maneuvers such as overtaking,

lane merging, roundabout merging, road intersection

crossing, etc.

Trafﬁc jams and slowdowns are problems that

concern several cities in the world, and several stud-

ies focus on the use of CAVs and learning mecha-

nisms to reduce them (Barthelemy and Carletti, 2017;

Mouhcine et al., 2018), both based on local data

(low volume of communication shared among near

agents), and on global (city-scale) data (diffused by

central trafﬁc authorities) (Marcillaud et al., 2020).

The presence of accidents, obstacles, or blocked vehi-

cles reduces the number of available lanes, which can

also lead to trafﬁc jams. If the vehicles can commu-

nicate and take into account such events during their

decision processes, disseminating the information and

coordinating actions, such jams can be reduced (Kor-

donis et al., 2020).

Another particular approached problem concerns

reducing shock waves using communication. A shock

wave is often the result of exceptional actions, such

as emergency braking (Marcillaud et al., 2020), which

can cause slowdowns and increase trafﬁc density. The

occurrence of shock waves can be reduced with the

help of multiagent coordination (Vaio et al., 2019).

CAV agents use the information exchanged with

neighbor vehicles to adapt their speed and maneu-

vers. More generally, anticipating events also enables

to avoid accidents, save fuel, and reduce car wear (Ka-

mal et al., 2015b).

Many slowdowns and jams are caused by the pres-

ence of intersections with dense trafﬁc, which put ve-

hicles in conﬂict, increasing the chances of having an

accident (Marcillaud et al., 2020). To avoid collisions,

intersections are regulated by trafﬁc rules and infras-

tructures that determine priority. However, in case

of heavy trafﬁc, the distribution of vehicles at inter-

sections can also be uneven, causing slowdown and

stress. Through the use of intelligent coordination,

CAV agents can decide to change route (Lin and Ho,

2019), allowing a better distribution of vehicles over

the different possible itineraries, or even inﬂuence the

behavior of trafﬁc lights.

Vehicle-intersection coordination (Kamal et al.,

2015a; Gaciarz et al., 2015) allows to obtain a more

ﬂuid trafﬁc and to reduce fuel consumption. Each

CAV optimizes its trajectory after accepting its pas-

sage order. When well-coordinated, the number of

situations into which the vehicles must stop at inter-

section can be greatly reduced. As a trafﬁc light in an

intersection controls the ﬂow of vehicles in the axes,

anticipating the state of an intersection managed by a

trafﬁc light helps to avoid unnecessary stops and in-

tensive braking. Communicating trafﬁc lights can in-

form when they will change their states, allowing the

CAVs to anticipate those events and then adapt their

actions (Almannaa et al., 2019). At limit, given cer-

tain conditions, trafﬁc light-free intersection control

can be envisioned, only based in communication and

intelligent coordination (Zhang et al., 2020).

Another situation that a CAV must be able to

handle is lane merging, which requires ﬁne speed

adaption to the other vehicles. The merging zone

of two lanes may be managed through infrastruc-

ture (Rios-Torres and Malikopoulos, 2017) or by self-

organization (Wang et al., 2018; Mosebach et al.,

Integrating Shared Information into the Sensorial Mapping of Connected and Autonomous Vehicles

457

2016). If a CAV detects other vehicles, it slows down

and calculates a trajectory to cross the merging point

without collision. The involved vehicles can com-

municate their intentions in order to better coordinate

their actions.

The work presented in (Vasic et al., 2016) intro-

duces an overtaking decision algorithm for networked

intelligent vehicles based on cooperative tracking and

sensor fusion. The ego-vehicle is equipped with lane

keeping and lane changing capabilities, and with a

forward-looking lidar sensor which feeds a tracking

module that detects other vehicles, such as the vehi-

cle that is to be overtaken (leading) and the oncoming

trafﬁc. Based on the estimated distances to the lead-

ing and the oncoming vehicles and their speeds, a risk

is calculated and a corresponding overtaking decision

is made. The performance of that overtaking algo-

rithm, in which it fuses object estimates received from

the leading car, which also has a forward-looking li-

dar, overcomes the case when the ego-vehicle only

relies on its own individual perception.

High-deﬁnition (HD) maps are essential for the

safe operation of CAVs. They allow a CAV to de-

termine its exact position in real time, ensuring self-

localization, helping to detect ﬁxed objects and terrain

details by matching collected data and stored data. It

liberates the CAV to concentrate its efforts on detect-

ing the dynamic objects on the scene. GPS, radar,

ultrasonic technology, and normal cameras combine

with lidar sensors to create a centimeter accurate 3D

image, required for safe navigation. An additional se-

mantic layer is superimposed onto the 3D image. Se-

mantic maps include information such as lane bound-

aries, speed limits, turn restrictions, and stopping

points (Badue et al., 2021).

Like the human perception, which relies on dif-

ferent senses (vision, hearing, touching, etc.), a CAV

needs to fuse data provided by different sources for

obtaining coherent information. The sensors that

equip a CAV work in a collaborative way: the analy-

sis of a speciﬁc sensor could not be sufﬁcient to make

a decision and should be coupled with the analysis of

other sources. A lidar sensor, for example, is good at

calculating distances, but cannot see colors. A video

camera would bring this missing information to lidar.

In a general scheme, the interaction between several

sensors can occur at different stages of the process

(Elmenreich and Leidenfrost, 2008): point, feature,

or detected object level, like shown in Fig. 6. Com-

prehensive reviews of the state-of-the-art of CAV per-

ception technology concerning sensing, localization

and mapping methods can be read in (Van Brumme-

len et al., 2018; Rosique et al., 2019; Hu et al., 2020).

Depending on the conﬁguration, data from differ-

Figure 6: The fusion of data coming from different sensors

can present interactions from the low to the high-level anal-

ysis (Elmenreich and Leidenfrost, 2008).

ent sources can be fused in complementary, compet-

itive, or cooperative combinations (Elmenreich and

Leidenfrost, 2008). A fusion is called complementary

if the information are independent, but can be com-

bined in order to create a more complete model of the

environment, providing a spatially or temporally ex-

tended view. Generally, fusing complementary data

can be made by addition, or juxtaposition (e.g. the

employment of multiple cameras to build up a 360 de-

grees picture of the environment). Competitive fusion

is used for fault-tolerant systems, where each sensor

delivers independent measurements of the same tar-

get, providing redundant information. In this case,

a more robust information can be achieved by cor-

recting erroneous sources (e.g. reduction of noise by

combining two overlaying camera images). Finally,

cooperative fusion provides an emerging view of the

environment by combining non redundant informa-

tion, but the result is sensitive to inaccuracies coming

from any of the sources (i.e. the combined model can

present decreased accuracy and reliability).

4 PROPOSED APPROACH

Exchanging information is necessary to implement

cooperation between vehicles. In this work, we are

not searching to integrate data from the raw sensors of

other cars, but rather high-level information: e.g. the

presence of a pedestrian or a vehicle in an invisible

segment, emergency braking impossible to detect lo-

cally, etc. Communication can signiﬁcantly improve

vehicle perceptive capabilities, working as an addi-

tional sensing, combining information collected from

multiple cars in a cooperative perspective.

However, extracting good information from the

data communicated by other agents can be challeng-

ing as the agent may receive uncertain information

from unknown agents (Cholvy et al., 2017). One so-

lution is, based on a priori trusted information, to rank

the agents and representing reliability through a total

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

458

Figure 7: Diagram of modules which compound a con-

nected and autonomous vehicle.

preorder. The overall communication set is ﬁrst eval-

uated with the help of inconsistency measures. Next,

the measures are used for assessing the contribution

of each agent to the overall inconsistency of the com-

munication set.

Techniques for merging raw information have

been studied extensively. Two different approaches

exist: the ﬁrst one considers sources (i.e. agents) in an

equal way and merging techniques such as majority

merging, negotiation, arbitration merging or distance-

based merging for solving conﬂicts raised by contra-

dictory information. The second one takes sources re-

liability into account, providing weights for discount-

ing or ignoring pieces of information whose source

is not sufﬁciently reliable. This factor can be used

to weaken the importance of information provided by

unreliable sources (Cholvy et al., 2017).

In the proposed approach, a CAV must: (1) inte-

grate the information received via V2X into its own

high-level perception of the scene; (2) assess the rele-

vance of the information to be communicated to other

vehicles; and (3) send to other vehicles its own inten-

tions, and receive the other vehicles intentions, to co-

ordinate actions. Such elements are shown in Fig. 7.

In this paper, we offer a reﬂection on the ﬁrst two

problems. The ﬁrst one concerns how can a VAC

combine its own perceptual data (interpreted from

camera, radar, lidar) with communicated information

from multiple nearby cars using self-localization to

build an improved perception. Many data can be un-

certain and inaccurate. In other words, the problem

is how to combine information from multiple agents,

given the potential uncertainty and inaccuracy of the

information, and based on the trust and reputation ac-

corded to the sources. The second question concerns

how critical or important each isolated piece of infor-

mation from each neighbor vehicle can be in assisting

the vehicle, and how the information can be hierarchi-

cally set up to provide the critical information to the

vehicles, minimizing data transfer.

Even if the industry works with the perspective

of sharing raw sensors data from different neighbor

Figure 8: The information coming from different sources

must be integrated into the vehicle’s model of the scene,

creating an additional fusion stage.

vehicles, it can be very complicated for each vehicle

to fuse all this data, assessing conﬁdence and trust.

The approach proposed in this paper concerns a high-

level integration: each CAV realizes its own interpre-

tation concerning self-localization, environment de-

tection, and mobile objects detection, and shares this

interpreted information. The bandwidth necessary to

do so is then greatly reduced, but the agent must still

make choices concerning how the communicated in-

formation can be integrated into its own interpretation

of the scene, dealing with inconsistency, imprecision,

and conﬁdence factors.

One of the difﬁculties in fusing data from multiple

vehicles involves their relative localization. The cars

need to be able to know precisely where they are in

relation to each other as well to objects in the vicinity.

For example, if a single pedestrian appears to differ-

ent cooperating cars in different positions, there is a

risk that, together, they see two pedestrians instead

of one. Using directly other cars raw signals is very

hard due to the amount of data that must be processed

in real time, while the vehicles is in motion and tak-

ing decisions. We can list some different sources of

positioning information which must be treated in dif-

ferent levels of conﬁdence: (a) my position given by

an hd-map, (b) my position given by the gps or de-

duced from proprioception, (c) my position accord-

ing to the view of other agents and communicated to

me, (d) the position of other vehicles or pedestrians

according to my own sensory interpretation, (e) their

position according to their own self-positioning meth-

ods and communicated to me, (f) the position of other

vehicles or pedestrians informed by another VACs or

infrastructures.

When detecting a given object j, a set of prop-

erties θ

is also perceived (type, position, direction,

speed, etc.), and the agent can assess the conﬁdence

on its own detection w

0, j

. From the point of view

of the ego-vehicle (i = 0), and based on the history

of recent interactions, a trust measure t

can be as-

sociated to each of the neighbor vehicles i. Then,

an aggregation function selects the answer follow-

ing a simple weighted majority rule, calculated by

Integrating Shared Information into the Sensorial Mapping of Connected and Autonomous Vehicles

459

∑

i=0

i, j

: ∀i, j, where s

is the score of can-

didate answer for object j, and m is the number of

vehicles communicating in the considered time.

Those simple scheme can be improved by includ-

ing the known reputation of the sources on the weight-

ing function, which must be informed by recognized

road authorities. Another possibility is to consider the

different ﬁelds of perception of the sources (e.g. an

agent giving information about detected objects in an

area outside the ﬁeld-of-view of the ego-vehicle). Fi-

nally, the sources and pieces of information can be

ﬁltered, labeled as inlier or outlier, depending on how

distant they are from the majority, or from a priori re-

liable data, ensuring that bad sources of information

will not disturb the projected map of detected objects.

Concerning the relevance of a given information

to a given vehicle, we consider that each detected ob-

ject is a piece of information that can be communi-

cated. Two evidences can be used to rank the set of

possible messages in terms of relevance in three pref-

erence categories: (1) if the vehicle directly asks for

the information related to a given portion of the space

where the object was detected, (2) if the neighbor ve-

hicle judges that the information is important to the

ego-vehicle (because it anticipates a possible colli-

sion), and (3) the other objects. When an agent ex-

presses a need of information, it is suggesting a ﬁrst

criterion to assess usefulness. The degree of useful-

ness of a piece of information is, however, a multi-

faceted notion which takes into account the fact that

it represents potential interest and trustability (Saurel

et al., 2019). Another straightforward metric to de-

termine relevance of a detected object is given by

its probability of intercepting the ego-vehicle’s trajec-

tory, and the distance from the current position to the

intercept position. For example, a vehicle that is near,

but which is moving away, is less important to pay

attention than a vehicle that is farther, but which is

approaching.

5 CONCLUSION

In this article, we investigated how cooperative per-

ception can impact decision making and planning of

autonomous vehicles. Particularly, we proposed the

sketch of a model for fusing the list of detected objects

detected by neighbor vehicles into the list of detected

objects of the ego-vehicle. We also suggested how a

validation strategy can include humans in the loop in

order to contemplate an ergonomic metric for evaluat-

ing the comfort produced by a solution, in addition to

other metrics focused in reducing overall travel time

and accidents.

Beyond safety and comfort, cooperation can be

used to optimize trajectories, save energy, and im-

prove trafﬁc ﬂows. Another important opportunity

made possible by vehicle communication is the shar-

ing of intentions, which allows the agents to ﬁnd ar-

rangements and negotiated solutions to their eventual

conﬂicts, leading to an increased collective perfor-

mance. The next steps of this research include pre-

cising and reﬁning the aggregation method in order

to consider source reputation and local trust, informa-

tion reliability given by the source, and the conﬁdence

on it estimated by the ego-vehicle depending on the

context.

REFERENCES

Almannaa, M., Chen, H., Rakha, H., Loulizi, A., and El-

Shawarby, I. (2019). Field implementation and testing

of an automated eco-cooperative adaptive cruise con-

trol system in the vicinity of signalized intersections.

Transp. Res. D: Transp. Environ., 67:244–262.

Arnold, E., Al-Jarrah, O., Dianati, M., Fallah, S., Oxtoby,

D., and Mouzakitis, A. (2019). A survey on 3d object

detection methods for autonomous driving applica-

tions. IEEE Trans. Intell. Transp. Syst., 20(10):3782–

3795.

Badue, C., Guidolini, R., Carneiro, R. V., Azevedo, P., Car-

doso, V., Forechi, A., Jesus, L., Berriel, R., Paix

ao,

T., Mutz, F., Veronese, L., Santos, T., and Souza, A.

(2021). Self-driving cars: a survey. IEEE Expert Syst.

Appl., 165:113816.

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

behavioural trafﬁc assignment model with strategic

agents. Transp. Res. C: Emerg. Technol., 85:23–46.

Cavazza, B., Gandia, B., Antonialli, F., Zambalde, A.,

Nicola

ı, I., Sugano, J., and Neto, A. (2019). Manage-

ment and business of autonomous vehicles: a system-

atic integrative bibliographic review. Int. J. Automot.

Technol. Manag., 19(1/2):31–54.

Chen, S., Kuhn, M., Prettner, K., and Bloom, D. (2019).

The global macroeconomic burden of road injuries:

estimates and projections for 166 countries. Lancet

Planet. Health, 3(9):e390 – e398.

Cholvy, L., Perrussel, L., and Th

evenin, J. (2017). Using

inconsistency measures for estimating reliability. Int.

J. Approx. Reason., 89:41–57.

Elmenreich, W. and Leidenfrost, R. (2008). Fusion of het-

erogeneous sensors data. In Int. Workshop on Intell.

Solut. in Embed. Syst. (WISES), pages 1–10. IEEE.

Fraedrich, E. and Lenz, B. (2016). Societal and individual

acceptance of autonomous driving. In Autonomous

Driving: Technical, Legal and Social Aspects, pages

621–640. Springer, Berlin, Heidelberg.

Friedrich, B. (2016). The effect of autonomous vehicles

on trafﬁc. In Autonomous Driving, pages 317–334.

Springer.

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

460

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

negotiation for trafﬁc regulation. Advances in Social

Computing and Multiagent Systems, pages 1–18.

Hakeem, S., Hady, A., and Kim, H. (2020). 5g-v2x: stan-

dardization, architecture, use cases, network-slicing,

and edge-computing. Wireless Netw., 26:6015–6041.

Hu, J., Zheng, B., and Wang, C. (2020). A survey on multi-

sensor fusion based obstacle detection for intelligent

ground vehicles in off-road environments. Front. In-

form. Technol. Electron. Eng., 21:675–692.

Jadhav, A. (2018). Autonomous Vehicle Market by Level of

Automation (...) Global Opportunity Analysis and In-

dustry Forecast, 2019-2026. Allied Market Research.

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

hara, K. (2015a). A vehicle-intersection coordina-

tion scheme for smooth ﬂows of trafﬁc without us-

ing trafﬁc lights. IEEE Trans. on Intell. Transp. Syst.,

16(3):1136–1147.

Kamal, M., Taguchi, S., and Yoshimura, T. (2015b). Efﬁ-

cient vehicle driving on multi-lane roads using model

predictive control under a connected vehicle environ-

ment. In Intelligent Vehicles Symposium (IV), pages

736–741. IEEE.

Kim, S., Liu, W., Ang, M. H., Frazzoli, E., and Rus, D.

(2015). The impact of cooperative perception on de-

cision making and planning of autonomous vehicles.

IEEE Intell. Transp. Syst. Mag., 7(3):39–50.

Kordonis, I., Dessouky, M. M., and Ioannou, P. A. (2020).

Mechanisms for cooperative freight routing: Incen-

tivizing individual participation. IEEE Trans. Intell.

Transp. Syst., 21(5):2155–2166.

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.

Litman, T. A. (2020). Autonomous Vehicle Implementa-

tion Predictions: Implications for Transport Plan-

ning. Victoria Transport Policy Institute.

Lopez, P., Behrisch, M., Bieker-Walz, L., Erdmann, J.,

otter

od, Y.-P., Hilbrich, R., L

ucken, L., Rummel, J.,

Wagner, P., and Wießner, E. (2018). Microscopic traf-

ﬁc simulation using sumo. In 21

Int. Conf. Intell.

Transp. Syst. (ITSC), pages 2575–2582. IEEE.

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

Kaddoum, E. (2020). Management of intelligent ve-

hicles: Comparison and analysis. In Int. Conf. on

Agents and Artif. Intell. (ICAART), pages 258–265.

SCITEPRESS.

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. J. Comp. Sci., 14:1499–1511.

Rios-Torres, J. and Malikopoulos, A. A. (2017). Automated

and Cooperative Vehicle Merging at Highway On-

Ramps. IEEE Trans. on Intell. Transp. Syst., 18(4).

Rosique, F., Navarro, P. J., Fern

andez, C., and Padilla, A.

(2019). A systematic review of perception system and

simulators for autonomous vehicles research. Sensors,

19(3):648.

Saurel, C., Poitou, O., and Cholvy, L. (2019). Assessing the

usefulness of information in the context of coalition

operations. In Information Quality in Information Fu-

sion and Decision Making, pages 135–154. Springer.

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. Int. J. of Advanced Networking and Applica-

tions, 10(3):3833–3842.

Sommer, C. and Dressler, F. (2014). Vehicular Networking.

Cambridge University Press.

Sovani, S. (2017). Simulation accelerates development of

autonomous driving. ATZ worldwide, 119(9):24–29.

Taillandier, P., Gaudou, B., Grignard, A., Huynh, Q.-N.,

Marilleau, N., Caillou, P., Philippon, D., and Dro-

goul, A. (2019). Building, composing and experi-

menting complex spatial models with the gama plat-

form. Geoinformatica, 23(10):299–322.

Terken, J. and Pﬂeging, B. (2020). Toward shared control

between automated vehicles and users. Automot. In-

nov., pages 53––61.

Torabi, B., Al-Zinati, M., and Wenkstern, R. (2018). Ma-

tisse 3.0: A large-scale multi-agent simulation system

for intelligent transportation systems. In Advances

in Practical Applications of Agents, Multi-Agent Sys-

tems, and Complexity (PAAMS), pages 357–360.

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

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

gation in mixed trafﬁc ﬂow environment. IEEE Trans.

on Intell. Transp. Syst.

Van Brummelen, J., O’Brien, M., Gruyer, D., and Naj-

jaran, H. (2018). Autonomous vehicle perception:

The technology of today and tomorrow. Transp. Res.

C: Emerg. Technol., 89:384 – 406.

Vasic, M., Lederrey, G., Navarro, I., and Martinoli, A.

(2016). An overtaking decision algorithm for net-

worked intelligent vehicles based on cooperative per-

ception. In Intelligent Vehicles Symposium, pages

1054–1059. IEEE.

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

consensus-based cooperative highway on-ramp merg-

ing using v2x communications. In World Congress

Experience (WCX). SAE.

Yurtsever, E., Lambert, J., Carballo, A., and Takeda, K.

(2020). A survey of autonomous driving: Common

practices and emerging technologies. IEEE Access,

8:58443–58469.

Zhang, Y., Liu, L., Lu, Z., Wang, L., and Wen, X.

(2020). Robust autonomous intersection control ap-

proach for connected autonomous vehicles. IEEE Ac-

cess, 8:124486–124502.

Integrating Shared Information into the Sensorial Mapping of Connected and Autonomous Vehicles

461