Multi-Robot Real-time Game Strategy Learning based on Deep Reinforcement Learning
Ki Deng, Yanjie Li, Songshuo Lu, Yongjin Mu, Xizheng Pang, Qi Liu · 2022 IEEE International Conference on Robotics and Biomimetics (ROBIO) · 2022
With the booming development of artificial in-telligence, the autonomous decision-making of robots on the battlefield has become a future trend. In the field of en-tertainment games, some institutions have already attained outstanding results in multi-agent decision-making tasks by using reinforcement learning algorithms. But in the field of robot confrontation game, the robot confrontation strategies are still based on the simple logic of artificial design in most cases, which hinders the further improvement of robot strategies. In this paper, a reinforcement-learning-based multi-robot control method is proposed to improve the strategy performance. We enhanced the reinforcement learning algorithm PPO with the self-play method by easy-to-implement tricks for multi-agent training and improve training efficiency. In simulated experiments based on the rules of the RoboMaster University AI Challenge, we show our method has a much higher winning percentage than traditional approaches like behavior tree.