TRAFFIC SIMULATION AT JUNCTION: NON-NORMATIVE PRACTICES VS. DEADLOCK Arnaud Doniec 1 , Stéphane Espié 2 , René Mandiau 3 , Sylvain Piechowiak 4 ABSTRACT Simulating traffic at junction is quite complex. The common approach simplifies drastically the process: the traffic “solver” lets vehicles enter into the junctions only when their trajectories are not in conflict. This solution is sometimes sufficient, but is not acceptable when the question is to mimic actual behaviour. In particular these common approaches don’t take the non-normative practices usually performed by real driver into account. Our aim is to improve traffic simulation models for driver's behaviour at junctions by using a multi-agent approach. The framework is the INRETS-MSIS ARCHISIM behavioural traffic simulation model, which is based on psychological fundings. In this model, actors of the traffic system interact between themselves in order to produce traffic phenomena. Each simulated driver has its own knowledge, goal and strategies. In crossroad situation for example, they decide to stop or go according to their own assessment of their relative priorities. This fully distributed algorithm gives good results, but leads in some cases to deadlocks (particularly for situations with many left-turn manoeuvres). A second attempt was to improve this algorithm by taking account of anticipation mechanism: simulated drivers recognize the situation and enter into the junction if and only if they anticipate not to create a deadlock situation. The use of this algorithm removes nearly all the deadlocks while “mimicking” actual behaviour. We will first present the problematic of traffic in junctions. We will then present the framework of our work. We will discuss the game theory based algorithm used for the simulation of the go / no go process, and the anticipation mechanism we have developed to ensure a mitigation of deadlocks. We will conclude by explaining our prospects and the possible use of such a simulation. KEY WORDS Multi-agent system, coordination, anticipation, traffic simulation. 1 Phd Student, MSIS Department, INRETS, 2 avenue du Général Malleret-Joinville, 94114 Arcueil, France, Phone +33 1 47 40 70 03, FAX +33 1 45 47 56 06, doniec@inrets.fr 2 Director, MSIS Department, INRETS, 2 avenue du Général Malleret-Joinville, 94114 Arcueil, France, Phone +33 1 47 40 70 23, FAX +33 1 45 47 56 06, espie@inrets.fr 3 Professor, LAMIH, University of Valenciennes, Le Mont Houy, 59313 Valenciennes, France, Phone +33 3 27 51 14 38, FAX +33 3 27 51 13 16, mandiau@univ-valenciennes.fr 4 Professor, LAMIH, University of Valenciennes, Le Mont Houy, 59313 Valenciennes, France, Phone +33 3 27 51 14 38, FAX +33 3 27 51 13 16, piechowiak@univ-valenciennes.fr June 14-16, 2006 - Montréal, Canada Joint International Conference on Computing and Decision Making in Civil and Building Engineering Page 2564