This paper explores a methodology that integrates the proportional integral derivative (PID) method with deep reinforcement learning (DRL). Constructing a compensatory controller using DRL to enhance the control performance of a PID controller. The proposed method is utilized in developing a fixed-wing unmanned aerial vehicle (UAV) flight controller to achieve longitudinal flight control. The compensatory controller is constructed using the Deep Deterministic Policy Gradient (DDPG) algorithm, tailored with a state-action space selection specifically designed for UAV dynamics and tracking targets. Meanwhile, the penalty term for tracking error and the sparse reward for completing the goal are introduced, and the construction scheme of the reward function is given. Simulation results demonstrate that the DRL-based compensatory controller can improve control performance when PID controller parameters are not optimally tuned to some extent, effectively eliminating pitch angle tracking overshoot and reducing setting time.
Longitudinal Control and Optimization of Fixed-Wing UAV Based on Deep Reinforcement Learning
Lect. Notes Electrical Eng.
International Conference on Guidance, Navigation and Control ; 2024 ; Changsha, China August 09, 2024 - August 11, 2024
2025-03-04
11 pages
Article/Chapter (Book)
Electronic Resource
English