Coordination d'agents à l'aide d'un algorithme en-ligne pour les POMDPs

This paper presents an online method for POMDPs based on a look-ahead search to find the best action to execute at each cycle in an environment. The basic idea of our ap- proach, called RTBSS (Real-Time Belief Space Search) , is to avoid computing a complete policy. Our approach is especially motivated by real-time environments where the state space is too large to consider traditional algorithms . We first describe the formalism of our online method, fol- lowed by some results on standard POMDPs. Then, we present an adaptation of our method for multiagent envi- ronments on an example presenting a way to manage the agent's interactions with the dynamic parts of the world and a coordination method based on the reward function.