Hyper Q-Learning Algorithm for Pursuit-Evasion Games with Misperceptions

We consider solving the optimal strategies for players in a pursuit-evasion game with misperceptions, where the pursuers aim to capture the evaders but possess misperceptions about the evaders, while the evaders seek to escape the pursuer with the knowledge of the rivals' misperceptions. We first characterize the misperceptions via hypergame frameworks in a discrete-action pursuit-evasion game and propose a core concept of optimal strategies for exploiting these misperceptions. Then we propose a novel Q-learning-based algorithm for the proposed optimal strategy. Experimental results illustrate the convergence and effectiveness of the algorithm.