Balancing Unpredictability and Coverage in Adversarial Patrolling Settings

We present a novel strategy for a patroller defending a set of heterogeneous assets from the attacks carried by an attacker that through repeated observations attempts to learn the strategy followed by the patroller. Implemented through a Markov chain whose stationary distribution is a function of the values of the assets being defended and the topology of the environment, the strategy is biased towards providing more protection to valuable assets, yet is provably hard to learn for an opponent. After having studied its properties, we show that our proposed method outperforms strategies commonly used for this type of problems.