We study the online route planning problem for patrolling robots, to assign them to optimal routes to patrol in a large crime-prone area. To model the actively engaging, intelligent, and adversarial opponents, we use the Stackelberg Security Game between the patrolling robots and the attackers. We leverage a graph-based bandit algorithm [16] with adaptive adjustment of the reward for the robots in this game to perplex the best response attackers and gradually succeed over them. Our graph bandits can outperform other stochastic bandit algorithms [10] when a simulated annealing-based scheduling mechanism is incorporated to adjust the balance between exploration and exploitation. Hence our method can successfully assign a small group of patrolling robots to cover a large number of routes.
CITATION STYLE
Rahman, M., & Oh, J. C. (2018). Online learning for patrolling robots against active adversarial attackers. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10868 LNAI, pp. 477–488). Springer Verlag. https://doi.org/10.1007/978-3-319-92058-0_46
Mendeley helps you to discover research relevant for your work.