The Uniformed Patroller Game
This work addresses adversarial patrolling scenarios for security applications, presenting an incremental modification to existing game theory models.
The paper tackles the problem of patrolling games by introducing a modification where the Attacker can observe the Patroller's presence and delay attacks, showing that this extra information can reduce thwarted attacks by up to a factor of four in specific models. The main result is that optimal strategies involve the Attacker starting in the second period the Patroller is away and the Patroller avoiding consecutive visits to the same location.
Patrolling Games were introduced by Alpern, Morton and Papadaki (2011) to model the adversarial problem where a mobile Patroller can thwart an attack at some location only by visiting it during the attack period, which has a prescribed integer duration m. Here, we modify the problem by allowing the Attacker to go to his planned attack location early and observe the presence or the absence there of the Patroller (who wears a uniform). To avoid being too predictable, the Patroller may sometimes remain at her base when she could have been visiting a possible attack location. The Attacker can then choose to delay attacking for some number of periods d after the Patroller leaves his planned attack location. As shown here, this extra information for the Attacker can reduce thwarted attacks by as much as a factor of four in specific models. Our main finding, is that the attack should begin in the second period the Patroller is away (d = 2) and that the Patroller should never attack the same location in consecutive periods.