A Deep Reinforcement Learning Method for Lion and Man Problem

David Gale's lion and man problem, which is a fundamental pursuit-evasion game, attracts the interest of many scholars. We propose a deep reinforcement learning method based on Deep Deterministic Policy Gradient (DDPG) to solve this problem. We first improve the exploration strategy by adding guided exploration and dynamic spaces exploration strategies to the greedy algorithm. Then we introduce a learning reset mechanism to help the agents escape the traps in the learning process. With these improvements, our method achieves better performance than the classic DDPG algorithm in the lion and man problem. The simulation result shows that deep reinforcement learning method may be promising to solve this problem.