Research on Autonomous Obstacle Avoidance and Target Tracking of UAV Based on Improved Dueling DQN Algorithm

In order to overcome the shortcomings of slow convergence, low success rate and sub-optimal tracking path of deep reinforcement learning algorithms in solving the problems of unmanned aerial vehicle (UAV) autonomous obstacle avoidance and target tracking. This paper adopts Dueling Deep Q-network (Dueling DQN) algorithm with improved network structure based on the Deep Q-network (DQN) algorithm, combined with ε-inspire exploration strategy and experience replay buffer idea, a Multi Pool Dueling Deep Q-network (MP-Dueling DQN) algorithm is proposed to optimize the success rate and path of UAV target tracking. Furthermore, UAV is endowed with the ability of environment perception, and the environmental generalization ability of UAV is improved by designing a continuous reward function. The simulation results show that MP-Dueling DQN algorithm has shorter motion path, stronger environmental adaptability and better performance compared with DQN, Double Deep Q-network (DDQN) and Dueling DQN.