Applying Bayesian learning to multi-robot patrol

Abstract

Performing a patrolling mission with multiple mobile robots is a challenging task that requires effective coordination between agents. While predefined patrol circuits may lead to suitable routing performance, their deterministic nature eases the task of potential intruders. Therefore, the need to propose probabilistic strategies becomes evident. In this paper, a new multi-robot patrolling strategy is proposed, in which concurrent learning agents adapt their moves to the state of the system at the time, using Bayesian decision. When patrolling a given site, each agent evaluates the context and adopts a reward-based learning technique that influences future moves. Experiments show the potential of the approach, which outperforms several other state-of-the-art strategies.