Le problème de la patrouille multiagent. Étude de convergence de l'évaluation des stratégies cycliques

Patrolling an environment involves a team of agents whose goal consists in continuously visiting its most relevant areas as frequently as possible. The current research that tackles this complex multi-agent problem usually defines the environment as a graph upon which agents move and seek to optimize the worst idleness of a graph, that is the delay during which a node remains unvisited. This paper deals with the problem of evaluating cyclic multi-agent patrolling strategies, those where every agent ultimately visits the same set of nodes indefinitely. We provide theoretical results about the computation time required for the convergence in the evaluation of any such strategy. An evaluation algorithm is also presented. MOTS-CLÉS : patrouille multiagent, stratégies de patrouille cycliques, étude théorique.