Cooperative multi-robot patrol with Bayesian learning

Patrolling indoor infrastructures with a team of cooperative mobile robots is a challenging task, which requires effective multi-agent coordination. Deterministic patrol circuits for multiple mobile robots have become popular due to their exceeding performance. However their predefined nature does not allow the system to react to changes in the system's conditions or adapt to unexpected situations such as robot failures, thus requiring recovery behaviors in such cases. In this article, a probabilistic multi-robot patrolling strategy is proposed. A team of concurrent learning agents adapt their moves to the state of the system at the time, using Bayesian decision rules and distributed intelligence. When patrolling a given site, each agent evaluates the context and adopts a reward-based learning technique that influences future moves. Extensive results obtained in simulation and real world experiments in a large indoor environment show the potential of the approach, presenting superior results to several state of the art strategies.