LSTM-DDPG-Based Dynamic Obstacle Avoidance for UAVs in Power Distribution Networks Using Velocity Obstacle Modeling

Abstract

This paper addresses the problems of obstacle avoidance and route planning in the autonomous inspection of distribution network drones and proposes an intelligent control algorithm based on deep reinforcement learning. By integrating the velocity obstacle method and the LSTM - DDPG framework, dynamic obstacle avoidance decisions in complex environments are achieved. Simulation experiments on the Gazebo platform show that, compared with the traditional DWA and VOM algorithms, this solution reduces the average obstacle avoidance time to 0.13s and the path length by 8.2% and 2.0%, respectively, while achieving an obstacle avoidance success rate of 98.2%. Field tests verify the practicality and robustness of the algorithm in the complex environment of distribution networks.