MULTI-AGENT ARCHITECTURE FOR SIMULATION OF

TRAFFIC WITH COMMUNICATIONS

Pedro Fernandes and Urbano Nunes

Institute of Systems and Robotics, University of Coimbra, Polo II, Coimbra, Portugal

Keywords: Multi-agent systems, Traffic simulation, Network simulation, Intelligent Transportation Systems.

Abstract: Inter-vehicle communications, in the context of Intelligent Transportation Systems, will probably bring a

significant improvement in both traffic safety and efficiency. In order to evaluate in what measure this is

true, traffic simulations that take into account the communications between vehicles are needed.

In this paper, we propose an agent-based architecture, in which the simulation and management of the inter-

vehicle communications are integrated in the simulation of vehicles, in a hierarchical multi-agent environ-

ment. An overview of multi-agent methodologies, platforms, among other, is also presented.

1 INTRODUCTION

Human transport in urban spaces relies mostly on

individual vehicles, congesting the transportation

networks. Studies and simulations of traffic have

been made for decades, through macroscopic,

mesoscopic and microscopic traffic simulators.

Recently, in the context of Intelligent Transpor-

tation Systems (ITS), vehicle to-vehicle (V2V) and

vehicle to infrastructure communications (V2I) are

being developed, namely the DSRC (Dedicated

Short Range Communications), operating in 5.9

GHz band. The standardization process is almost

finished under IEEE 802.11p/IEEE 1609.x (also

designated by WAVE: Wireless Access in Vehicular

Environments) and IEEE 1556 standards. In EU, the

International Organization for Standardization

(ISO), under the Technical Committee TC204, is

working in similar standards – Communication Air

Interface Long and Medium Range (CALM) – to

ensure European-wide inter-vehicle communications

interoperability.

To study the impact that such systems may have

in the near future, efforts to integrate traffic and

network simulators have been pursued. However, a

useful solution has not been reached yet.

The integration of both traffic and network simu-

lations in a system may be considered a complex

task, due to a vast set of reasons, such as the intrinsic

complexity of traffic theory, the wireless network

transmissions involved, the real-time constraints and

the distributed nature of the system, among others.

At the present, traffic theory does not account to

driver behavior changes due to the existence of

communications. Therefore, equation-based model-

ing is not the most appropriate method to use in

simulation. Agent-based modeling allows the devel-

opment of a more adaptive system, and although

system validation may be more difficult, it can be

done at both system and individual levels.

2 RELATED WORK

The use of intelligent agents in traffic simulation is

an emergent area of research. Table 1 presents some

of the works in this area and simulators integration.

Table 1: Related work.

Vogel and

Nagel (2005)

Multi-agent simulation model with

application to Berlin traffic.

Hallé et al.,

(2004)

Agent-based architecture to develop

centralized and decentralized platoons.

Li et al.,

(2006)

Urban traffic control system using

multi-agent technology.

Dresner and

Stone (2005)

Agent-based simulation of a traffic

intersection.

Eichler et al.,

2005

Coupling traffic and network simula-

tors and a V2V messaging application.

Piorkowski

et al., 2006

Network- and application-centric

evaluation oriented architecture.

Avila et al.,

2005

Intersection warning system, coupling

traffic and network simulators.

215

Fernandes P. and Nunes U. (2008).

MULTI-AGENT ARCHITECTURE FOR SIMULATION OF TRAFFIC WITH COMMUNICATIONS.

In Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics - RA, pages 215-218

DOI: 10.5220/0001501202150218

 SciTePress

3 DEVELOPMENT ISSUES

According to Wooldridge (2002), “an agent is a

computer system that is situated in some environ-

ment, and that is capable of autonomous action in

this environment in order to meet its design objec-

tives”. Autonomy, situatedness, reactivity and proac-

tivity are some important characteristics of agents.

In a multi-agent architecture, issues like organiza-

tion, coordination and security are also relevant.

To develop a MAS system, a disciplined ap-

proach should be followed, and an appropriate plat-

form should be chosen, along with communication

standards between agents – preferably based on open

standards – and appropriate ontologies. The simula-

tion platform must also be selected or developed.

3.1 Methodologies

Several proposed methodologies to develop a MAS

may be considered. Prometheus, Gaia and Tropos

are some of the examples in the literature. However,

not all existing methodologies are appropriate for

every problem. Some of them aim at generality.

Others focus more on specific platforms and lan-

guages, gaining in detail and adaptability.

Prometheus methodology was proposed by

Padgham and Winikoff (2002). According to the

authors, the reason why they proposed a new meth-

odology was the methodology claimed detail, sup-

port of BDI (Beliefs, Desires and Intentions) agents,

scaling ability and tool support. To support design

and development of multi-agent systems using Pro-

metheus, Padgham and Winikoff developed the

Prometheus Design Tool, that implements the three

phases of Prometheus and process some consistency

checking.

Gaia methodology presents a general approach, to

allow its use for a broad type of agent-based sys-

tems. However, this characteristic, which is one of

its strengths, is also its most pointed weakness, since

the detailed design phase and implementation have

intentionally been left out.

Other methodologies appear in literature, namely

ROADMAP, Tropos, SODA, MESSAGE, MaSE,

MAS-CommonKADS, AOR, OPM/MAS, MAS-

SIVE, Ingenias, DESIRE, PASSI and AgilePASSI.

In Table 2, the phases of some methodologies are

presented.

Prometheus seems an appropriate methodology

for initial system development. All the relevant

phases are covered conveniently, and PDT tool al-

lows consistency and completeness checking

through the steps of each of the phases.

3.2 Platforms

Choosing the right platform for the problem domain

at hand is not a trivial task. The choice is closely

connected with the methodology adopted.

Follows a short description of some platforms:

Jade framework is probably the most used agent-

oriented middleware. Is an open source distributed

middleware system, compliant with FIPA specifica-

tions, that implements both white and yellow pages,

agent mobility, ontologies and content languages,

among other features. JADE does not provide, how-

ever, direct support to the development of BDI agent

architectures.

Jadex is a software framework for the development

of goal-oriented agents following the BDI model.

Since JADE platform does not allow direct imple-

mentation of this model, Jadex, using JADE, allows

the creation of rational agents. Jadex agents have

two main components: an agent definition file

(ADF), coded in XML, and Java code. Jadex BDI

metamodel is specified in XML Schema.

Jason is an interpreter of the an extended version of

AgentSpeak(L), allowing agents to be distributed

over the net using Simple Agent Communication

Infrastructure (SACI). Jason is available as open

source and uses jEdit (http://www.jedit.org) as IDE.

JACK

is a commercial agent platform, which uses

syntactic and semantic extensions of Java that allows

the implementation of BDI agents.

The use of an open source platform is preferable.

Moreover, the compliance with FIPA specifications

is important to allow interoperability of the systems.

JADE platform provides those and other features.

Table 3 presents some platform characteristics.

Table 2: Methodology phases.

Method-

ology

Phases

Prome-

theus

1-specifications; 2-architectural design;

3-detailed design; 4-implementation.

Gaia 1-requirements; 2-analysis; 3-design.

Road-

map

1-requirements; 2-analysis; 3-design.

OPM/

MAS

1-requirements; 2-analysis; 3-design;

4-deployment.

Table 3: Platform classification.

Platform

Open

source

BDI

Com-

pliance

White &

yellow pages

JADE Yes No FIPA Yes

Jadex Yes Yes FIPA* Yes*

Jason Yes Yes KQML No

JACK

No Yes FIPA Yes

* with JADE

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

216

3.3 Ontologies and Languages

Communication is a valuable tool for agents to in-

teract, exchange information and request services.

At the present, Ontology Web Language (OWL) is

the language of the Semantic Web that is being stan-

dardized by the World Wide Web Consortium.

An agent platform must allow the use of content

language (e.g. FIPA-SL Content Language Specifi-

cation), and communication languages (e.g. FIPA-

ACL Agent Communication Language).

3.4 Simulation

Multi-Agent Based Simulation is considered the

support of choice for the simulation of complex sys-

tems, replacing or integrating with other micro-

simulation techniques, most of them object-oriented.

4 THE MODEL

The model proposed consists of a novel multi-agent

system that manages the communications inside

each vehicle and simulates the communications be-

tween each of them and the infrastructure. Inter-

vehicle communications are managed by an agent-

based module that simulates real wireless communi-

cations between vehicles, using the appropriate stan-

dards. To allow interoperability, the platform sup-

porting the development of the proposed multi-agent

system complies with FIPA specifications.

The architecture will be tested in the context of

an intersection, where the management of communi-

cations and localization of the vehicles will have

both a distributed and a centralized component. This

option aims to provide simulation functionalities at

the communication level that, in the reality, would

be provided by the transmission media. Moreover,

localization of hazardous situations (vehicles with-

out communications, pedestrians) is better provided

by centralized facilities.

The architecture proposed to the multi-agent sys-

tem is depicted in Figure 1.

4.1 Multi-agent Architecture

A brief description of the main agents involved in

the proposed architecture follows:

Network Simulator: The main function of the Net-

work Simulator (NS) is to receive all communica-

tions between Communication Manager agents, and

simulate the network transmission between them,

considering the environment and the location of each

one. Appropriate communication standards must be

used by this agent, namely DSRC and CALM.

Figure 1: Multi-agent architecture.

Intersection Traffic Rules Arbiter (ITRA) must deal

with intersection control of traffic, recording all traf-

fic events and dealing with resolution of conflicts

between User Managers (UM). With low traffic

throughput, we may have a distributed control of

traffic, where UM may agree with the priority of

each other, always under ITRA supervision. As traf-

fic flow grows, ITRA will have to validate all UM

decisions, eventually overcoming some of them. In a

high traffic flow scenario, all traffic rules decisions

must be taken by ITRA, and vehicles become “data

probes” of the centralized traffic rule management.

Although this might seems contradictory with the

choice of an agent-based system, real-time con-

straints impose the option presented above.

Communication Manager (CM) manages commu-

nications between the vehicle and external systems,

such as the infrastructure and other vehicles. In both

cases, NS is used as an intermediary, to simulate

wireless network transmissions. Each vehicle com-

municates through its own CM.

Message Broker (MB) must manage all internal

messages, and has the incumbency of filtering and

its prioritization, ensuring that critical messages are

dealt first by the appropriate agents. In this scheme,

MB may delay low priority messages or, in some

cases, even discard such messages.

User Manager (UM) main function has to do with

decisions about the priority of the vehicle, with the

agreement of all vehicles in its direct neighborhood,

always under ITRA supervision. As stated before, as

traffic flow grows, the decisions are taken by ITRA,

MULTI-AGENT ARCHITECTURE FOR SIMULATION OF TRAFFIC WITH COMMUNICATIONS

217

in a centralized manner. To avoid deadlocks, all the

decisions must be taken with anticipation, allowing

the forecast of possible deadlocks and its resolution

before they actually occur.

Interface Manager (IM) agent deals with the selec-

tion of the most appropriate message interface to the

driver, taken in account the type of message.

Localization agent determines the localization of the

vehicle in the intersection map, using GPS data and

an intersection beam signal, and compares its posi-

tion with neighbor vehicles positions, periodically

transmitted through wireless communications. This

agent must decide whether the situation is critical,

based on position and vehicle data, and warn UM in

case of imminent danger.

Vehicle agent gathers vehicle data (e.g. speed, ac-

celeration, brakes, steering) and feeds Localization

agent with that information. UM receives also simi-

lar feedback. Moreover, this agent gets commands

issued by Driver agent.

Driver agent deals with the control of the whole ve-

hicle. It receives information, whether critical or not,

via IM agent and responds accordingly to that in-

formation and the type of driver modeled. For that

purpose, Driver agent maintains a driver type data-

base. This agent issues commands to Vehicle agent

directly and indirectly through IM.

Traffic Simulation Environment represents the en-

vironment where the agents evolve. One of its main

functions is to provide communications between

agents, in the platform level, allowing appropriate

management of agents’ percepts and actions.

Graphical presentation of simulation results will also

be directly connected with this component.

5 CONCLUSIONS AND FUTURE

WORK

In this paper we propose an architecture in which the

simulation and management of the inter-vehicle

communications are integrated in the simulation of

vehicles, in a hierarchical multi-agent environment.

We also present a short survey of existing method-

ologies, platforms, ontologies and languages, and

suggest some possible choices to allow appropriate

system implementation.

MAS development using the appropriate meth-

odology, the implementation of the solution in the

selected platform, the validation of the process and

final deployment will follow.

ACKNOWLEDGEMENTS

This work was supported by Institute of Systems and

Robotics and Fundação para a Ciência e Tecnologia

under contract NCT04:POSC/EEA/SRI/

58016/2004.

REFERENCES

Avila, A. et al., (2005). A complete Simulator Architec-

ture for Inter-vehicle Communication Based Intersec-

tion Warning Systems. In Proc. 8

Int. IEEE Conf. In-

telligent Transportation Systems, Vienna, Austria.

Braubach, L. Pokahr, A., & Lamersdorf W. (2006). Tools

and Standards, Chapter of Multi-agent Engineering -

Theory and Applications in Enterprises. Springer Se-

ries: Int. Handbooks on Information Systems, Editors:

S. Kirn, O. Herzog, P. Lockemann, O. Spaniol.

Dresner, K. and Stone, P., 2005. Multiagent Traffic Man-

agement: An Improved Intersection Control Mecha-

nism. In The Forth International Joint Conference on

Autonomous Agents and Multiagent Systems (AAMAS

05), pp. 471-477, Utrecht, The Netherlands.

Eichler, S et al. (2005). Simulation of Car-to-Car Messag-

ing: Analyzing the Impact on Road Traffic. In Proc. of

the 13th IEEE Int. Symposium on Modeling, Analysis,

and Simulation of Computer and Telecommunication

Systems, vol. 00, pp. 507 – 510.

Hallé, S., Laumonier, J., & Chaib-draa, B. (2004). A De-

centralized Approach to Collaborative Driving Coor-

dination. In IEEE Intelligent Transportation Systems

Conference, Washington, D.C., USA.

Li, Z., Wang, F., Miao, Q., & He, F. (2006). An Urban

Traffic Control System Based on Mobile Multi-

Agents. In Proc. of IEEE Int. Conference on Vehicular

Electronics and Safety, Shanghai, China.

Padgham, L., & Winikoff, M. (2002). Prometheus: A

Methodology for Developing Intelligent Agents. In

proceedings of the Third International Workshop on

Agent-Oriented Software Engineering, at AAMAS'02.

Piorkowski, M. et. al. (2006). Joint Traffic and Network

Simulator for VANETs. Poster Session, MICS 2006,

Zurich, Switzerland.

Sturm, A., & Shehory, Onn. (2003). A Framework for

Evaluating Agent-Oriented Methodologies. In M.

Winikoff P. Giorgini, B. Henderson-Sellers, editor,

Proc. of the Int. Bi-Conference Workshop on Agent-

Oriented Information Systems, AOIS.

Vogel, A., & Nagel, K. (2005). Multi-agent based simula-

tion on individual traffic in Berlin, In Proc. of the Int.

Conference on Computers in Urban Planning and Ur-

ban Management (CUPUM’05), London.

Wooldridge, M. (2002). An Introduction to multi-agent

Systems. John Wiley & Sons, Chichster, U.K.

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

218