TOWARDS THE AGENTIFICATION OF A VIRTUAL SITUATED ENVIRONMENT FOR URBAN CROWD SIMULATION J. Demange*, S. Galland, A. Koukam *Multi-Agent Simulation Group Laboratoire Systèmes et Transports Université de Technologie de Belfort-Montbéliard, France jonathan.demange@utbm.fr http://www.multiagent.fr Keywords: Crowd Simulation, Multi-Agent System, Simulation, Environment Model. Abstract Research works in urban and crowd simulation tend to study systems, which are more and more complex, large and realistic. In such a case, the environment of the simulated system is a key concept which represents and manages the virtual world in which the agents are living. This paper starts from a modular object-oriented environment model, JaSIM, and proposes to agentify its components to enhance the modularity of the environmental model and its execution performances. 1. Introduction This paper is located in the domain of the simulation of urban systems with situated multi-agent systems. A multi-agent system (MAS) is a system composed of multiple interacting intelligent software agents. Multi-agent systems can be used to solve problems that are difficult or impossible for an individual agent or a monolithic system to solve. A multi- agent system is situated when the agents are immersed inside an environment. In the domain of urban system simulation, an agent is assumed to be a pedestrian, a vehicle, or any object that owns an autonomous decision-making process. The environment is then everything that is not an agent. Three different points of view may be adopted to study the notion of environment in situated MAS: (a) the part of the system which is outside the community of the agents; (b) the medium for coordination among these agents, or (c) the running infrastructure or platform [6]. According to Weyns and al. [7], confusion on the environment’s definition is mainly caused by mixing up concepts from points (a) and (b), and the infrastructure defined in point (c). Several authors refer to the environment as the logical entity of MAS in which agents and the other entities and resources are embedded. On another point of view, the notion of environment refers to the software infrastructure on which the multi-agent system is executed. Finally, it may even refer to the underlying hardware infrastructure. Odell and al. [6] distinguish between the physical environment and the communication environment. The physical environment refers to the laws, rules, constraints and policies that govern and support the physical existence of agents and the other entities. rest of this paper, only this aspect of the environment is taken. This paper presents a modular and domain-independent agent-oriented model for the simulation of situated urban environment. Here we differ from existing models such as JaSIM [4] in the sense that these models define the environment with an object-oriented point of view, and our model is agentifying the environment. This agentification of the environment model enables load-balancing, fault tolerance and dynamic adaptation to computational resources. Moreover, it allows introducing decision-making algorithms in the environment processes. Finally, it constitutes the first step to a holonic simulation model for situated environments. Our model is based on an object-oriented (OO) model JaSIM illustrated by Figure 2 [4]. JaSIM provides OO algorithms and data-structures required to model indoor and urban environments. The agent model corresponds to the intelligent agents simulated in the system (vehicle, pedestrian…). Each of them is associated to a physical representation in the environment: its body. This body permits to the agent to perceive its nearest environment and to act on it. Perceptions are computed by the perception generator and actions are gathered by the environment and any conflict among them is solved to obtain a valid reaction to the acts of the agents. The endogenous engine is maintaining the dynamic of the environment, which is not directly controlled by an agent. This paper is structured as follows. Firstly, agentification of the environment is one way to improve the simulation quality according to computational and simulation constraints, which is presented in the following section. Section 3 is describing several experiments with the agentified environment. Finally, we conclude this paper.