Moving Target Shooting Control Policy Based on Deep Reinforcement Learning

Robots are playing a more and more important role in people’s production and life, recently. However, robot control in dynamic environment is still a difficulty. With the great breakthrough of deep reinforcement learning in the field of video games, this method is also extended to the field of robots. Due to the gap between the simulation environment and the real environment, the deep reinforcement learning algorithm trained in the simulation environment is difficult to be applied to the real environment. Aiming at the gimbal control with two degrees of freedom (DOF), a pipline combining system identification and deep reinforcement learning is proposed. On the one hand, the shooting accuracy of the gimbal to moving objects is improved through deep reinforcement learning algorithm. On the other hand, the gap between simulation and reality is reduced through system identification. The method is verified in the RoboMaster University AI Challenge (RMUA) shooting system. The results show that the shooting accuracy is better than the classical control method.