Un modèle de mémoire dans un système multi-agent de géosimulation

Our research team is developing a multi-agent geosimulation environment (MAGS) to simulate crowd behaviour in a 3D virtual geographical environment. Currently, MAGS enables us to create simulations involving thousands of agents that can navigate in the 3D virtual world, perceive the objects and agents contained in it, memorise spatial knowledge and make decisions according to their goals. This paper presents the memory model that we developed for MAGS's agents. This model allows the perception and memorisation of spatial knowledge and is based on well-accepted models of human memory. We present a literature review of the main memory models developed in cognitive sciences and a selection of memory models implemented in multi-agent simulation systems. We show that the existing multi-agent systems have limited memory models. We present the characteristics of our memory model and show that it is more complete and efficient enough to enable real-time simulation involving thousands agents.