spatial cognitive map and a human-like memory model: some solutions for the simulation of pedestrian navigation in virtual environments

Many articles dealing with agent navigation in an urban environment involve the use of various heuristics. Among them, one is prevalent: the search of the shortest path between two points. This strategy impairs the realism of the resulting behaviour. Indeed, psychological studies state that such a navigation behaviour is conditioned by the knowledge the subject has of its environment. Furthermore, the path a city dweller can follow may be influenced by many factors like his daily habits, or the path simplicity in term of minimum of direction changes. It appeared interesting to us to investigate how to mimic human navigation behavior with an autonomous agent. The solution we propose relies on an architecture based on a generic model of informed environment, a spatial cognitive map model merged with a human-like memory model, representing the agent's temporal knowledge of the environment, it gained along its experiences of navigation. The informed environment is described by a an informed hierarchical abstract graph in which are stored precomputed data such as potential visibility sets, oriented grids, or densities of people, which can be used individually by simulated entities. A hierarchical multicriteria path-planning has been defined that takes care of individual preferences as well as individual knowledge of the environment. Moreover, the path planning method we propose is reactive to some events, reflecting the perceived modification of the environment, which allows the entity to adapt its behaviour in consequence.