Pedestrian Agent Based Model Suited to Heterogeneous Interactions Overseen by Perception

Most of the work about pedestrian simulations concern situations of homogeneous interactions between them as for planning and for designing in transportation studies. These studies assume that pedestrians broadly share common behavior and goal (reaching out of a building, crossing a street, following target ...). They do not take into account the many inter-pedestrian interactions induced by a dynamical environment requiring a contextual adaptation, for instance a situation with a danger perceived. In order to model different interactions between pedestrians in complex and dynamic context/situation such as urban area, we propose an exploratory work which concerns different types of interaction between pedestrians and other actors. Thus the model will be able to take simultaneously different kind of interaction. Technically, to deviate from a dangerous pedestrian, one will accept to get closer to other normal pedestrians whereas in panic situation, pedestrians may forget there initial destination. Thus, one alters for a while his interactions or his destination based on his perception of the environment and of the context. Our framework is the Multi-Agent Systems, within which the pedestrian travels can be seen as an agent coordination problem, coordination which is often competitive, occasionally cooperative. Thus, we propose an Agent-based model with a hierarchical architecture driven by perception, articulated with some controllers suited for various situations. The model is generic since it use for different task as road crossing, collision avoidance, danger avoidance and group interaction. First scenarios were tested to check our model