Multi-Rotor UAV Autonomous Tracking and Obstacle Avoidance Based on Improved DDPG

Abstract

To solve the problem of multi-rotor UAV autonomous tracking dynamic ground targets in obstacles environment, we used Markov decision process (MDP) to establish an autonomous maneuvering model of multi-rotor. Considering the obstacle avoidance requirements of UAV during the tracking process, we integrated the Long Short-Term Memory (LSTM) neural network with memory unit and time series data processing characteristics into the Deep Deterministic Policy Gradient (DDPG) algorithm framework, so that the Actor network can fully refer to the prior state information when making decisions. Finally, the performance test was implemented on the UAV 3D simulation platform based on Robot Operating System (ROS). The results show that the method proposed in this paper can enable the UAV to complete the whole process of autonomous tracking of the ground dynamic target. Compared with the traditional DDPG algorithm, the DDPG algorithm combined with LSTM has stronger accuracy and real-time performance, and can better meet the tracking and obstacle avoidance mission requirements of the multi-rotor UAV.