Multi-robot adversarial patrolling: facing coordinated attacks

The use of robot teams is common for performing patrol tasks, in which the robots are required to repeatedly visit a target area (perimeter, in our case) controlled by an adversary, in order to detect penetrations. Previous work has focused on determining the optimal patrol algorithm when facing a general adversary that tries to penetrate once through the patrol path. There, the robots' goal is to detect penetrations, i.e., the robots do not change their behavior once a penetration is detected. Requiring the robots to physically inspect penetration attempts can have far reaching consequences on the performance of the patrol algorithm. Specifically, it creates vulnerability points along the patrol path that a knowledgeable adversary can take advantage of. In this work we investigate the problem of coordinated attacks, in which the adversary initiates two attacks in order to maximize its chances of successful penetration, assuming a robot from the team will be sent to examine a penetration attempt. We suggest an algorithm that computes the optimal robot strategy for handling such coordinated attacks, and show that despite its exponential time complexity, practical run time of the algorithm can be significantly reduced without harming the optimality of the strategy.