A Reinforcement Learning based End-to-End Algorithm for Confrontation Problem

In this paper, a confrontation problem between two agents is investigated in which one agent is required to find then eliminate the other agent through projecting bullets. A reinforcement learning based algorithm is designed to realize an end-to-end strategy for one agent to confront the other. Self-play method is used to enhance the algorithm. Simulation results show the effectiveness of the proposed algorithm in confrontation with FSM based strategy and human player.