Construction of numerical potential fields with reactive agents

Abstract

In the context of unknown environments, this paper deals with autonomous robots navigation for exploration and foraging tasks. This remains a hard problem when the environment contains complex obstacles (dead ends, mazes,..). Artificial Potential Fields (APF) is a well known approach to deal with navigation of autonomous agents. They are computed to define attractive forces towards goals and repulsives ones from obstacles. However, the APF approach has a critical drawback: the possible existence of local minima [1] (e.g. positions where the different forces equalize). The only solution to avoid this drawback is to know/perceive the whole environment [2]. As a consequence, the solution for autonomous agents equiped with limited perception consists in marking the environment in order to use it as a shared memory. This approach, inspired by pheromones mechanisms in ant colonnies. provides good solutions but requires a lot of agents and time for their emergence.