Distributed Decision-Theoretic Active Perception for Multi-robot Active Information Gathering

International audienceMultirobot systems have made tremendous progress in exploration and surveillance. However, information gathering tasks remain passive: the agents receive information from their sensors without acting in order to gather it. In this paper, we present a model and an algorithm for active information gathering using the infor-mation relevance concept. In this model, robots explore, assess the relevance, update their beliefs and communicate the appropriate in-formation to relevant robots. To do so, we propose a distributed de-cision process where a robot maintains a belief matrix representing its beliefs and beliefs about the beliefs of the other robots. This deci-sion process uses entropy and Hellinger distance in a reward function to access the relevance of their beliefs and their divergence with the other robots. This model allows to derive a policy for gathering in-formation to make the entropy low and a communication policy to reduce the divergence between robot's beliefs. An experimental sce-nario has been developed for an indoor information gathering mis-sion. Our model has been compared to two different systems : one without communication and one communicating each received ob-servation. The results show that our approach is more efficient than both systems