Deep Reinforcement Learning and L1 Adaptive Control Algorithm-Based Attitude Control of Fixed-Wing UAVs

Abstract

The fixed-wing UAV is a nonlinear strongly coupled system. Although the traditional linear controller can achieve attitude control of the UAV, the controller parameters tuning process is tedious and the robustness of the system is weak, which greatly limits the performance of the UAV. To address the above problems, this paper proposes a comprehensive control scheme that uses the deep reinforcement learning algorithm DDPG to train the controller parameters to achieve automatic tuning of these parameters on the structure of the classical controller, while ensuring system stability by limiting the range of the control parameters, and then utilizes the L1 adaptive controller to improve the robustness of the system. First, we construct a fixed-wing UAV model and a PID controller framework as the training environment, and generate the value function of the reward feedback to the agent; then, the DDPG algorithm is applied to train the controller parameters to achieve stable control of the UAV; finally, the L1 adaptive algorithm is applied to enhance the robustness of the UAV attitude controller, and simulation results are presented. The results show that the DDPG-L1 adaptive control scheme designed in this paper can effectively solve the problem of tedious tuning of controller parameters and enhance the robustness of the system while ensuring the stability of the system.