An Improved UAV Obstacle Avoidance Control Method Based on APF-DDPG: The DPEPM-KNN Approach

Aiming at the problem of poor stability and convergence of the obstacle avoidance algorithm of unmanned aerial vehicles, in order to improve the efficiency of flight missions in complex environments, this study proposes a UAV obstacle avoidance control method that integrates the deep deterministic policy gradient (DDPG) algorithm and the artificial potential field (APF) method. Based on the DDPG algorithm, an improved experience replay sampling method is introduced, and the K nearest neighbor (KNN) technology is used to manage and utilize the experience pool data, so that the experience pool can be used more efficiently. The experimental results show that compared with the traditional APF-DDPG algorithm, the average reward value of the proposed DPEPM-KNN algorithm is increased by 21.22%, the path efficiency is increased by 19.61%, and the path stability is increased by 95.6%. In addition, the network loss value fluctuates less and remains at a low level, indicating that the DPEPM-KNN algorithm is superior to the APF-DDPG algorithm in terms of path planning performance, training stability, and network convergence, proving that the proposed method can achieve higher path planning efficiency and lower training loss in complex UAV obstacle avoidance environments.