Multi-Agent Based Framework for Cooperative Trafﬁc Management in

C-ITS System

Ameni Aloui

1 a

, Hela Hachicha

1,2

and Ezzeddine Zagrouba

1 b

University of Tunis El Manar, Higher Institute of Computer Science, Laboratory of Informatics,

Modeling and Information and Knowledge Processing (LIMTIC), Ariana 2080, Tunisia

University of Jeddah, College of Computer Science and Engineering, Al Faisaliah, Jeddah, Saudi Arabia

Keywords:

C-ITS, Multi-Agent Systems, Trafﬁc Management, Communication, Congestion Detection, Decision-Making.

Abstract:

The continual growth in road trafﬁc poses signiﬁcant challenges to effective trafﬁc management, necessitat-

ing innovative solutions such as Cooperative Intelligent Transport Systems (C-ITS). This paper introduces a

novel multi-agent based model designed to address road trafﬁc management in C-ITS systems. Our approach

aims to reduce congestion and enhance driver decision-making by leveraging dynamic communication and

information exchange between vehicles and infrastructure. Our multi-agent system is intricately designed to

play speciﬁc roles in managing trafﬁc ﬂow. Through real-time execution using a C-ITS road safety case study

focused on warning accidents, we evaluate the performance of our architecture through key metrics including

mean travel time and mean speed in the C-ITS system. The innovative aspects of our approach lie in the

integration of multi-agent systems in such a system, providing a signiﬁcant advancement in the ﬁeld of C-ITS

road trafﬁc management. By detailing the instantiation of our system and emphasizing concrete services, we

contribute to the broader goal of improving road safety and trafﬁc efﬁciency in urban environments.

1 INTRODUCTION

For decades, the transportation system has tirelessly

pursued advancements in efﬁciency, environmental

protection, and safety in trafﬁc management. How-

ever, the escalating number of vehicles over the years,

as highlighted by recent statistics (Davis and Boundy,

2021), claims that “since 1900, the number of ve-

hicles per 1,000 persons in the United States has

increased dramatically. After reaching a peak of

844.5 in 2007, the number fell but then started to

rise in 2012. In the United States in 2018, there

were 836.3 vehicles per 1,000 persons”. The surge

in vehicle numbers necessitates a paradigm shift to-

wards improved road trafﬁc management, a critical

aspect given its direct impact on daily life and en-

vironmental quality. Better road trafﬁc management

must be implemented due to the rise in vehicle num-

bers over time. To ensure that trafﬁc is maintained,

trafﬁc management is largely concerned with service

management. Services are the main focus of C-ITS

systems. C-ITS, a cutting-edge extension of Intel-

https://orcid.org/0009-0009-7146-8106

https://orcid.org/0000-0002-2574-9080

ligent Transport Systems (ITS), emerges as a piv-

otal technology for augmenting trafﬁc efﬁciency, road

safety, and environmental sustainability and provid-

ing comfort needs for end-users (driver) (Li et al.,

2022). C-ITS leverages Information and Commu-

nication Technologies (ICT) to facilitate real-time

information exchange among vehicles (Vehicle-to-

Vehicle, V2V) and between vehicles and infrastruc-

ture (Vehicle-to-Infrastructure, V2I). This communi-

cation is facilitated by the On-Board Units (OBU) in

connected vehicles and the Road Side Units (RSU)

within the infrastructure, representing the C-ITS sub-

systems. Therefore, C-ITS represents a transforma-

tive approach to better transportation through the in-

tegration of smart C-ITS sub-systems and advanced

communication technologies that specify how they in-

teract. In this interaction, C-ITS sub-systems com-

municate and cooperate based on international stan-

dards such as CEN, ISO, and ETSI. Therefore, C-

ITS makes it possible to go a step further in provid-

ing real-time information and tailored control strate-

gies to speciﬁc drivers. C-ITS excels in collecting

and disseminating real-time messages about the road

environment directly to drivers, thereby enhancing

decision-making. This connectivity can ameliorate

420

Aloui, A., Hachicha, H. and Zagrouba, E.

Multi-Agent Based Framework for Cooperative Trafﬁc Management in C-ITS System.

DOI: 10.5220/0012468300003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 1, pages 420-427

ISBN: 978-989-758-680-4; ISSN: 2184-433X

road safety and efﬁciency, exempliﬁed by services of-

fering advice, warnings, or actions. Despite the wide-

ranging beneﬁts of C-ITS development and deploy-

ment, challenges in trafﬁc monitoring, C-ITS com-

munication, and security persist as trafﬁc manage-

ment issues. This paper focuses speciﬁcally on the

C-ITS trafﬁc management challenge, recognizing its

potential to impede trafﬁc efﬁciency and compromise

service delivery to drivers if not adeptly addressed.

Our key contribution lies in leveraging a multi-agent

system to model and simulate the complex dynamics

of road trafﬁc. Unlike traditional approaches, which

often rely on centralized control mechanisms, our

methodology empowers individual agents to make au-

tonomous decisions based on real-time information.

This decentralized approach enables a more adaptive

and responsive system, particularly crucial during un-

foreseen events such as accidents or road works. Our

Contributions are summarised as follows:

• Propose a C-ITS architecture with a cluster de-

composition to facilitate trafﬁc management and

increase road safety and trafﬁc efﬁciency by con-

trolling and monitoring vehicles.

• Propose a new Multi-agent framework for co-

operative trafﬁc management in C-ITS systems

to alleviate trafﬁc congestion and improve driver

decision-making

This paper is structured in four sections: the ﬁrst sec-

tion introduces our paper. The second section enu-

merates and details the different related work as well

as a brief description of our contribution. The third

section describes our architecture. The experimenta-

tion is presented in section fourth. We conclude our

research work in the last section with some future per-

spectives.

2 RELATED WORK

Literature explores Multi-Agent Systems (MAS) to

support C-ITS, addressing communication issues, es-

pecially in Vehicle Ad-Hoc Networks (VANET) as

the core part of such a system and trafﬁc control for

better trafﬁc management. For instance, in (Devan-

gavi and Gupta, 2017) agents collaborate on path dis-

covery using Bezier curves for multipath routing and

mobile agents for reducing communication costs. The

model outperforms in transmission time, number of

multipath computed, communication overhead, and

packet delivery ratio, but the short simulation time of

600 seconds warrants further testing in dynamic traf-

ﬁc scenarios and complex situations like closed lanes

or segments.

Some research focuses on using MAS to address C-

ITS trafﬁc management in all the systems includ-

ing its components. For instance, in (Zarari et al.,

2018) a generic architecture for the deployment of C-

ITS within a MAS dedicated to VANET is proposed.

The model utilises commonly stationary and mobile

agents and is based on a formal representation but

lacks evaluation of performance metrics and C-ITS

service management which is the main concern of the

C-ITS system. In the work of (Zouari et al., 2021)

a Cooperative MAS for Road Trafﬁc Decision Mak-

ing based on Hierarchical Interval Type-2 Fuzzy Sys-

tem (HIT2FS) is proposed, showing positive results

in path ﬂow and mean travel time, yet the short 360-

second simulation time calls for longer testing.

Other research works focus on speciﬁc components

of C-ITS using MAS. In (Gu

eriau et al., 2016), the

main focus was on the RSU. An agent-based model

focusing on the decision-making of the C-ITS sys-

tem to RSU components as discretized agents to de-

liver messages to vehicles. They propose a reinforce-

ment learning process model to send these messages

based on a k-means classiﬁcation. Further details are

needed in this work such as reinforcement learning

algorithms used. In the research work of (Hamdani

et al., 2022), the main focus was on Smart Road Signs

(SRS) which is responsible for Trafﬁc monitoring and

sending warnings to drivers. They propose route guid-

ance to reduce travel time based on mobile and sta-

tionary agents. However, this work was not evalu-

ated. The work of (Belbachir et al., 2019) focuses on

trafﬁc lights and proposes a self-adaptive mechanism

to regulate trafﬁc lights via I2I communication. This

work lacks scalability, it is speciﬁc at intersections to

deal with speciﬁc situations(congestion). In the work

of (Naderi et al., 2023), authors propose a hierarchi-

cal trafﬁc light-aware routing scheme using reinforce-

ment learning at the two-level RSU and SDN to ad-

just their policies depending on the variation of time.

In (Teixeira et al., 2020) the focus is on autonomous

vehicles using Belief-Desire-Intention (BDI) agents

to make decisions within their cognitive capacities,

with scalability concerns and limited scenario testing.

Added to that, BDI agents are expensive in time exe-

cution. If the number of vehicles increases, the exe-

cution time will be much higher.

Other research work is focusing on congestion man-

agement due to warnings. We cite as an exam-

ple (Hamidi and Kamankesh, 2018) (Perez-Murueta

et al., 2019). For example, MAS was applied in

(Hamidi and Kamankesh, 2018) to deal with emer-

gency warnings. Their proposal was based on the in-

crease in the quality of the entire path network. Sim-

ilarly in (Perez-Murueta et al., 2019), a model was

Multi-Agent Based Framework for Cooperative Trafﬁc Management in C-ITS System

421

designed to deal with congestion issues based on a

MAS. Authors utilize real-time probe vehicle data and

deep learning for trafﬁc state prediction. The rerout-

ing process is done with entropy entropy-balanced k

shortest path for vehicles. However, the lack of eval-

uation against existing research limits the assessment

of their approach’s effectiveness.

The main difference between our proposal and the

approaches analyzed is that we will delve into the

rich value that multi-agent systems bring to the mod-

eling, exploration, and optimization of the complex

problems inherent in the different C-ITS subsystems.

Our proposal focuses on the limits of the analyzed

approaches for enhancing C-ITS architectures using

MAS, emphasizing adaptability for ﬂexible trafﬁc

management and services. This approach goes be-

yond cooperation, aiming for a holistic solution to the

challenges of modern transportation systems.

3 PROPOSED ARCHITECTURE

3.1 Hierarchical C-ITS Architecture

The proposed hierarchical C-ITS architecture, in-

spired by successful deployment projects, and with

respect to the standardization deﬁned in the C-ITS

reference architecture (Dajsuren et al., 2017), implies

a set of components (C-ITS subsystems), which are

responsible for the operational aspect of the C-ITS

system based on their features and communication,

is structured into three levels (Fig. 1): Center System,

Infrastructure, and Urban Road. Each level plays a

crucial role in ensuring efﬁcient trafﬁc management

and enhancing road safety.

Figure 1: An overview of the proposed C-ITS system archi-

tecture.

1. Center System: is the highest level of the archi-

tecture aiming to control trafﬁc. It comprises sev-

eral Base Controllers (BC), each one representing

a city. BCs facilitate communication with each

other through C2C communication, with RSUs in

the same city (I2C/C2I). Geographic considera-

tions ensure that each cluster represents a coherent

and manageable section of the road network.

2. Infrastructure. The Infrastructure level includes

RSUs as essential ﬁxed components for guiding

and monitoring trafﬁc. Their signiﬁcance lies in

supporting local trafﬁc management, addressing

intersection management, speed control, warning

systems, parking management, and overall trafﬁc

monitoring. Details about RSU subsystems and

functionalities are elaborated in (Dajsuren et al.,

2017). Each RSU covers a speciﬁc zone, serving

as a data collection point for real-time monitoring

and analysis of road conditions.

3. Urban Road: is the low level of the architecture

equipped with a range of sensors such as loop de-

tectors, cameras, etc, and comprises a decomposi-

tion of the road to clusters. More details about

cluster decomposition are presented in our pre-

vious research work (Aloui et al., 2021). Each

cluster refers to one BC and comprises a set of

RSUs and vehicles equipped with the C-ITS sys-

tem. This implementation of C-ITS in vehicles

enables interaction between infrastructure via the

RSUs and vehicles via their OBUs (V2I/I2V) as

well as the interaction between vehicles (V2V).

RSUs actively provide data related to trafﬁc ﬂow,

congestion, and incidents, which is then processed

collaboratively with BCs to set and adjust rules ef-

fectively.

The development and deployment strategy of the C-

ITS systems hinges on the implementation of spe-

ciﬁc C-ITS applications within each sub-system. A

C-ITS application is a speciﬁc use case that falls un-

der a particular C-ITS service. To optimize these ap-

plications, our proposal advocates for the utilization

of agents as the fundamental building blocks of C-

ITS sub-systems. Each C-ITS application is concep-

tualized, designed, and implemented as a multi-agent

system, harnessing the power of agent-based technol-

ogy to enhance the system’s efﬁciency, communica-

tion, and decision-making. This innovative approach

aligns with the dynamic nature of trafﬁc management,

leveraging agent-based systems to create a respon-

sive, collaborative, and adaptable framework for C-

ITS. The subsequent sections delve deeper into the

speciﬁcs of the multi-agent system implementation.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

422

3.2 Urban Road Architecture Based on

the Multi-Agent System in C-ITS

The adoption of a MAS in the proposed C-ITS ar-

chitecture is strategically justiﬁed for its capacity to

enhance decentralized decision-making, adaptability

to local contexts, and efﬁcient information sharing

among intelligent agents. The MAS enables de-

centralized decision-making, allowing agents to au-

tonomously respond to local conditions.

Figure 2: The distributed C-ITS architecture based on

MAS.

Table 1: Roles of each agents.

Agents Role

AMBC Have a global view in the entire trafﬁc.

It aims to propose the relevant rules to

prevent congestion due to warnings.

RAP Check trafﬁc conditions. It is responsi-

ble for detecting congestion situations.

It communicates with both AMV and

AMBC and informs them every prede-

ﬁned time t about the trafﬁc condition

AMV Checks the subscription and unsub-

scription of drivers in their covered

zone. It communicates with RAP,

AMBC and disseminates services to

subscribed DA

DA Interact with AMV for receiving warn-

ings, advice, and rules and transmitting

them to other DA as well as proposing

recommendations.

Our proposed MAS employs a set of stationary

agents, which handle the core functionalities within

each C-ITS subsystem, and we propose the integra-

tion of mobile agents in case of connection interrup-

tion in different C-ITS subsystems.

Figure (Fig. 2) illustrates the architecture of the

C-ITS system based on MAS Our multi-agent-based

architecture comprises four agents that cooperate to

manage and deploy C-ITS services.

Within the vehicle subsystem, the DA (Driver Agent)

is tasked with receiving services and facilitating real-

time information exchange with other DAs. In the

RSU subsystem, the AMV (Agent Manager Vehicle)

takes on the responsibility of receiving services and

sharing trafﬁc-related data with the RAP (Road Agent

Parent). Similarly, within the BC subsystem, the

AMBC (Agent Manager Base Controller) analyzes

trafﬁc data, receives real-time information and collab-

orates with BCs ’ agents and RSUs’ agents to enhance

overall system efﬁciency. This agent-based architec-

ture ensures dynamic and responsive road safety, fos-

tering effective communication and information shar-

ing across the entire system. The role of each agent is

described in (tab. 1). We deﬁne the operation of the

C-ITS system as consisting of the following different

steps represented in (Fig. 3).

Figure 3: Flowchart of the proposed method.

A. Subscription of Vehicles

In the dynamic landscape of C-ITS, the ”Subscrip-

tion of Vehicles” process plays a pivotal role, forming

a crucial link in the efﬁcient management of trafﬁc

and the delivery of essential C-ITS services. Vehicles,

upon entering the RSU-covered zone, have the option

to either subscribe to or unsubscribe from C-ITS ser-

vices. Subscription entails an agreement to share and

receive real-time information, fostering a cooperative

and informed trafﬁc environment. Thus, when a ve-

hicle enters a new cluster, the nearest AMV in RSU

is responsible for detecting every vehicle. Each sub-

scribed vehicle exchanges information with the DA as

well as its parameters (ID, speed, destination. . . .) to

monitor and manage trafﬁc information in its covered

zone. Unsubscribed vehicles will be detected by road

sensors. Every vehicle v

follows a set of routes rep-

resented by the following equation:

R = {r

, r

......r

} (1)

Each subscribed DA sends a message to the corre-

Multi-Agent Based Framework for Cooperative Trafﬁc Management in C-ITS System

423

sponding AMV as follows:

msgi = {id

, speed

, position

, destination

, route

}

(2)

Every subscribed vehicle will deploy a C-ITS

application according to the corresponding rules.

The AMV agent is responsible for detecting and

monitoring the set of vehicles as well as deploying

C-ITS applications based on rules to the subscribed

ones within their DAs.

B. Trafﬁc Monitoring

As vehicles traverse the road network, subscribed

DAs continuously exchange real-time information

with the corresponding AMV in the RSU. The

AMV, responsible for monitoring and managing

trafﬁc information in its covered zone, detects events

through the analysis of received data. Events can

include sudden changes in speed, unexpected stops,

or deviations from the planned route. For every

predeﬁned interval time t, the AMV computes the

density k as follows:

k =

(3)

where N is the number of vehicles and L is the maxi-

mum number of vehicles (in units of vehicles per km)

computed as follows:

L =

Lengtho f road

avgvehiclelength + min − gap

(4)

where min−gap is the safe inter-distance between ve-

hicles and the mean speed V

for edges ed

where Ed

is presented as follow:

ED = {ed

, ed

......ed

} (5)

The mean speed is presented as follows:

∑

n=1

(6)

In fact V

is the mean of speeds of vehicles passing

an edge ed

C. Congestion Detection

Based on the computed trafﬁc density and mean

speed received by the AMV, the RAP deploys

dynamic rules to manage congestion effectively. The

RAP aims to detect congestion levels in the RSU

coverage area using an Interval Type-1 Fuzzy Logic

model (IT1FL). Figure (Fig. 4) illustrates the model

with inputs including maximum speed in ed

[0,

120] and the density of ed

[0,100], each possessing

’low,’ ’medium,’ and ’high’ membership functions.

Complementing this, Figure (Fig. 5) showcases

fuzziﬁed input membership functions, providing a

visual representation of input membership degrees.

We generate the graphical input memberships within

the juzzy library (Wagner, 2013). The output of

the fuzzy model is the level of congestion in ed

[0,1] which has three membership functions ‘light’,

‘moderate’, and ‘heavy’.

Figure 4: IT1FL model for rule generation of one evalua-

tion.

Figure 5: Fuzziﬁed Input membership functions.

Table 2: Rules of the IT1F proposed model.

Rules Fuzzy Rules

1 IF Density is medium AND Mean

Speed is high THEN Level Congestion

is heavy.

2 IF Density is medium AND Mean

Speed is medium THEN Level Conges-

tion is moderate

3 IF Density is medium AND Mean

Speed is low THEN Level Congestion

is light

Then, the RAP will deﬁne the corresponding rules

to the detected congested edges mentioned in (tab. 2).

In a practical scenario, these rules guide the

RAP in assessing congestion. Once congested edges

are identiﬁed, messages are dispatched to both the

AMBC and AMVs. The AMV, in turn, notiﬁes

subscribed DAs about the congestion, fostering

informed decision-making among drivers.

D. Rule Generation

In the rule generation phase, our approach combines

insights from congestion detection and Mean Travel

Time (MTT) analysis to formulate a dynamic set

of rules for efﬁcient trafﬁc management. Using

congestion data from the RAP, the AMBC identiﬁes

congested edges, categorizing them by severity.

Simultaneously, MTT serves as a metric to assess

overall travel time performance, establishing thresh-

olds (δ) for acceptable travel times in different

scenarios. The AMBC receives from the AMV

responsible for the warning event edge position (

position) about the density (k). Once responding,

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

424

The AMBC computes MTT every interval time t

which is synchronized with the AMV interval time

t. We use Greenshield’s model (Banks, 2002) for the

MTT estimation since it proves its efﬁciency with

well-use by transportation researchers. The idea was

that there is a relation between the density and the

speed on an edge. Thus, it is computed as follows:

(7)

where

= V

(1 −

jam

) (8)

is the length or the same edge ed

, V

is the esti-

mated road speed. K

is the trafﬁc density (vehicles

per meter) on ed

and K

jam

is the trafﬁc jam density.

The integrated approach dynamically adjusts MTT

thresholds based on congestion severity, ensuring

adaptive decision-making. For instance, heavy

congestion prompts dynamic lane management or

rerouting, while exceeding MTT thresholds trig-

gers actions like suggesting alternative routes. By

combining congestion and MTT-driven rules, our

system achieves comprehensive trafﬁc management,

addressing immediate congestion concerns and

anticipating potential issues based on travel time

trends.

E. Adaptive Decision-Making

The DA receiving the warning service enables drivers

to make informed decisions and adapt their travel

plans accordingly. Thus, it will compute the distance

(distance) between its actual position and the warning

event position as follows:

distance

i j

(pd

− pd

)

+ (pw

− pw

)

(9)

where pd

and pd

are the current 2D position of

the driver (pd) and pw

and pw

are the current 2D

position of the warning (pw) where distance

> 0,

and (pd

, pd

) ̸= (pw

, pw

). If distance

i j

> β where

is a limit value deﬁned by the system, the DA will get

the alternative routes, if the distance

i j

< β of limit

deﬁned by the system, it will apply imposed rules,

the decision-making of the vehicle is executed by the

following algorithm.

Driver Agent Algorithm

Begin

1. Each DA has an Origin and Destination.

2. While origin != Destination

3. DA i Send parameters to AMV when

entering its covered zone

4. AMV subscribe DA i via its parameters

(id, position, speed, destination) and

deploy the C-ITS application

5. If distance between DA i AND event at

time t > beta

6. Asking the AMV about real-time roads

conditions

7- AMV transmits the message to the AMBC

8- the AMBC send the roads

condition about MTT

9- DAi look for the possible alternative

roads and choose the k-shortest path

10. else Apply imposed rules

End.

4 CASE STUDY: C-ITS ROAD

HAZARD WARNING

To illustrate our proposed model, we choose a C-ITS

road hazard warning service to illustrate an accident

warning as a C-ITS application. It is important to note

here, that the proposed architecture is not speciﬁed

for accident warning only, but can also be extensible

with any other C-ITS application. We aim to demon-

strate how to manage trafﬁc information to deliver the

drivers the relevant services.

The signiﬁcance of addressing accidents becomes ap-

parent when considering statistics such as those pro-

vided by the National Highway

Trafﬁc Safety Ad-

ministration, which anticipates a 7.342% increase in

road accident fatalities to 50 per 1 million inhabitants

in the United States in 2022. Similarly, Eurostat

re-

ported 42.1 road accident-related fatalities per million

people in 2020. These statistics underscore the im-

pact of accidents on road safety and trafﬁc efﬁciency,

emphasizing the need for effective solutions. An ac-

cident scenario serves as a challenge to the efﬁciency

of our approach.

4.1 Simulation

For our simulation scenario, we have chosen to repli-

cate the case study previously described in (Bedogni

et al., 2015), focusing on the city of Bologna (Fig. 6).

As per our architecture, Bologna City is conceptual-

ized as a cluster.

To simulate incidents, such as accidents or road

disruptions, we use the capabilities of SUMO

. In-

cidents, in the real world, can encompass collisions,

road works, adverse weather causing slow speeds, or

simply high trafﬁc ﬂows. There are a few options in

SUMO to simulate an accident. In our simulation, we

replicate an accident by instructing a vehicle to stop

https://www.nhtsa.gov/

https://ec.europa.eu/eurostat

https://eclipse.dev/sumo/

Multi-Agent Based Framework for Cooperative Trafﬁc Management in C-ITS System

425

Figure 6: Bologna Map city.

at a deﬁned point along its route and specifying the

duration of the halt. TRACI

is an API that facili-

tates real-time interaction between the SUMO inter-

face and the trafﬁc simulation. The simulation runs

for a timestep of 60 minutes. Parameters for this

example are outlined in (tab. 3). Our MAS is im-

Table 3: Maps parameters.

Parameters Value

Network area 25 km

Total number of road segments 2856

Total number of Trafﬁc lights 99

Total number of junctions 1539

Total number of Vehicles 22000

plemented in the JADE platform (Bellifemine et al.,

2005). We selected the TRASMAPI (Tim

oteo et al.,

2010) API due to its capability to enable real-time in-

teraction between MAS and the SUMO interface. It

establishes a higher level of abstraction between the

SUMO API (TRACI) and JADE, facilitating commu-

nication between the two tools via TCP sockets. The

MAS conﬁguration for handling accident warnings

features three designated containers:

• Main Container. Represents the center system,

with one AMBC agent acting as the agent respon-

sible for the Bologna city cluster.

• Infrastructure Container. Encompasses RSUs,

where two associated agents: the AMV and the

RAP function for the operational aspect of an

RSU. The Bologne map includes 5 RSUs

• Urban-Road Container. Contains DAs created

randomly for the simulation scenarios.

4.2 Results and Evaluation

In the experimentation, we are focusing on the strat-

egy proposed for our distributed C-ITS multi-agent

based architecture. When the BC detects an accident,

the ﬁrst step is to decide for which RSU it will send

http://sumo.dlr.de/wiki/TraCI

the rules. Drivers receiving the warning event will

execute the algorithm of the driver agent described in

section 3.2. For the K shortest path, the Dijkstra Algo-

rithm (1959) is integrated into the SUMO simulator.

The simulation output presents several statistics

for each car, each edge, and each lane. We chose the

MTT and the mean speed as metrics in our simula-

tion experiment to evaluate our approach compared

to the Original Trafﬁc Trace (OTT) with no coopera-

tion in the sense that the system is centralized. This

measurement reveals that our approach leads to bet-

ter trafﬁc management and ﬂuid trafﬁc. Figure 7 and

Figure 8 show the efﬁciency of our proposed archi-

tecture compared to an architecture without the coop-

eration between the agents. The MTT is smaller and

the mean speed is higher in the proposed model than

in the other cases which proves how our architecture

is beneﬁcial to enhance trafﬁc conditions. The results

show how our architecture and placement strategy can

impact the MTT and mean speed.

Figure 7: Vehicles mean travel time (from origin to destina-

tion).

Figure 8: Mean Speed.

5 CONCLUSIONS

In this paper, we propose a generic distributed multi-

agent architecture to support C-ITS systems. Our

main focus is to manage trafﬁc cooperatively in an ur-

ban road environment. MAS in our approach provides

distributed trafﬁc monitoring and control as well as a

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

426

better management of services. We also integrated

the IT1FL model as a decision-making knowledge

representation to detect congestion situations due to

warning events. It enables handling and modeling un-

certainty and makes our system more robust in real-

world scenarios. To prove the validation of our ap-

proach, we elaborate on a scenario of an accident

warning. Our approach outperforms mean travel time

and mean speed. There are also other possible re-

search directions to improve the proposed architec-

ture. Additional scenarios are needed to improve the

evaluation such as considering more data sets and per-

formance metrics (eg. CO2 emission). We can also

include the prediction of unplanned events by using

learning methods. In this case, communication at the

C-ITS sub-system to identify such a service can be

avoided by detecting an event before it even happens.

Besides, extending our proposal by considering the

security aspect is another challenging and crucial re-

search direction.

REFERENCES

Aloui, A., Hachicha, H., and Zagrouba, E. (2021). A mobile

agent-based framework to monitor and manage inter-

operability of c-its services in vehicles. In Agents and

Multi-Agent Systems: Technologies and Applications

2021: Proceedings of 15th KES International Confer-

ence, KES-AMSTA 2021, June 2021, pages 89–101.

Springer.

Banks, J. H. (2002). Introduction to transportation engi-

neering, volume 21. McGraw-Hill New York.

Bedogni, L., Gramaglia, M., Vesco, A., Fiore, M., H

arri, J.,

and Ferrero, F. (2015). The bologna ringway dataset:

Improving road network conversion in sumo and val-

idating urban mobility via navigation services. IEEE

Transactions on Vehicular Technology, 64(12):5464–

5476.

Belbachir, A., El Fallah-Seghrouchni, A., Casals, A., and

Pasin, M. (2019). Smart mobility using multi-agent

system. Procedia Computer Science, 151:447–454.

Bellifemine, F., Bergenti, F., Caire, G., and Poggi, A.

(2005). Jade—a java agent development framework.

Multi-agent programming: Languages, platforms and

applications, pages 125–147.

Dajsuren, Y., Karkhanis, P., Kadiogullary, D., and Fuen-

frocken, M. (2017). C-mobile d3. 1 reference archi-

tecture. Technical report, Technical report.

Davis, S. C. and Boundy, R. G. (2021). Transportation en-

ergy data book: Edition 39. Technical report, Oak

Ridge National Lab.(ORNL), Oak Ridge, TN (United

States).

Devangavi, A. D. and Gupta, R. (2017). Adaptive conges-

tion controlled multipath routing in vanet: A multia-

gent based approach. International Journal of Agent

Technologies and Systems (IJATS), 9(1):43–68.

eriau, M., Armetta, F., Hassas, S., Billot, R., and

El Faouzi, N.-E. (2016). A constructivist approach for

a self-adaptive decision-making system: application

to road trafﬁc control. In 2016 IEEE 28th Interna-

tional Conference on Tools with Artiﬁcial Intelligence

(ICTAI), pages 670–677. IEEE.

Hamdani, M., Sahli, N., Jabeur, N., and Khezami, N.

(2022). Agent-based approach for connected vehicles

and smart road signs collaboration. Computing and

Informatics, 41(1):376–396.

Hamidi, H. and Kamankesh, A. (2018). An approach to

intelligent trafﬁc management system using a multi-

agent system. International Journal of Intelligent

Transportation Systems Research, 16:112–124.

Li, S., Edwards, S., Isik, M. O., Zhang, Y., and Blythe,

P. T. (2022). Qualitative examination of cooperative-

intelligent transportation systems in cities to facilitate

large-scale future deployment. Sensors, 22(21):8423.

Naderi, M., Mahdaee, K., and Rahmani, P. (2023). Hi-

erarchical trafﬁc light-aware routing via fuzzy rein-

forcement learning in software-deﬁned vehicular net-

works. Peer-to-Peer Networking and Applications,

16(2):1174–1198.

Perez-Murueta, P., G

omez-Espinosa, A., Cardenas, C., and

Gonzalez-Mendoza Jr, M. (2019). Deep learning sys-

tem for vehicular re-routing and congestion avoid-

ance. Applied Sciences, 9(13):2717.

Teixeira, M., d’Orey, P. M., and Kokkinogenis, Z.

(2020). Simulating collective decision-making for

autonomous vehicles coordination enabled by vehic-

ular networks: A computational social choice per-

spective. Simulation Modelling Practice and Theory,

98:101983.

Tim

oteo, I. J., Ara

ujo, M. R., Rossetti, R. J., and Oliveira,

E. C. (2010). Trasmapi: An api oriented towards

multi-agent systems real-time interaction with multi-

ple trafﬁc simulators. In 13th International IEEE Con-

ference on Intelligent Transportation Systems, pages

1183–1188. IEEE.

Wagner, C. (2013). Juzzy-a java based toolkit for type-2

fuzzy logic. In 2013 IEEE Symposium on Advances in

Type-2 Fuzzy Logic Systems (T2FUZZ), pages 45–52.

IEEE.

Zarari, C., Balbo, F., and Ghedira, K. (2018). Agent execu-

tion platform dedicated to c-its. Procedia Computer

Science, 126:821–830.

Zouari, M., Baklouti, N., Kammoun, M. H., Ayed, M. B.,

Alimi, A. M., and Sanchez-Medina, J. (2021). A

multi-agent system for road trafﬁc decision making

based on hierarchical interval type-2 fuzzy knowledge

representation system. In 2021 IEEE International

Conference on Fuzzy Systems (FUZZ-IEEE), pages 1–

6. IEEE.

Multi-Agent Based Framework for Cooperative Trafﬁc Management in C-ITS System

427