Abstract: Aiming at the problem of autonomous obstacle avoidance and target tracking of unmanned aerial vehicle (UAV), based on the Deep Q Network (DQN) algorithm, a Multiple Pools Deep Q Network (MP DQN) algorithm is proposed to optimize the success rate of UAV obstacle avoidance and target tracking and the convergence of the algorithm. Furthermore, the environmental perception ability of UAV is given, and the directional reward function is designed in the reward mechanism, which improves the generalization ability of the UAV to the environment and the overall performance of the algorithm. The simulation results show that, compared with DQN and Double Deep Q Network (DDQN) algorithms, MP DQN algorithm has faster convergence speed, shorter tracking path and stronger environmental adaptability.

Key words: Unmanned aerial vehicle (UAV), Deep reinforcement learning, Autonomous obstacle avoidance, Target tracking, Environmental perception