Probabilistic multiagent patrolling

Abstract

Patrolling refers to the act of walking around an area, with some regularity, in order to protect or supervise it. A group of agents is usually required to perform this task efficiently. Previous works in this field, using a metric that minimizes the period between visits to the same position, proposed static solutions that repeats a cycle over and over. But an efficient patrolling scheme requires unpredictability, so that the intruder cannot infer when the next visitation to a position will happen. This work presents various strategies to partition the sites among the agents, and to compute the visiting sequence. We evaluate these strategies using three metrics which approximates the probability of averting three types of intrusion - a random intruder, an intruder that waits until the guard leaves the site to initiate the attack, and an intruder that uses statistics to forecast how long the next visit to the site will be. We present the best strategies for each of these metrics, based on 500 simulations. © 2008 Springer Berlin Heidelberg.