ANTI-INTERFERENCE TRAJECTORY TRACKING CONTROL OF QUADROTOR UAV BASED ON REINFORCEMENT LEARNING

This study presents a quadrotor UAV tracking control algorithm that addresses the issues of parameter uncertainty and external disturbances during trajectory tracking. The algorithm combines the Q-Learning reinforcement learning algorithm with a nonsingular terminal sliding mode controller. Firstly, a four-rotor UAV model based on tracking error is defined, and the coupling and external interference between channels are converted into lumped interference. Extended state observers are designed for estimation and compensation of lumped interference by the outer loop position subsystem and the inner loop attitude subsystem. At the same time, a fast non-singular terminal sliding mode UAV controller is constructed, which includes an outer loop position controller and an inner loop attitude controller. Then, a Q-learning algorithm based on fuzzy strategy is proposed to realize the adaptive adjustment of the key parameters of the controller and the observer. By designing the reward function, the Q values of the UAV under different flight states are iteratively optimized and the Q table is constantly updated. Finally, the trained Q table is used for drone control. The algorithm can not only save the complicated process of manual parameter adjustment, but also realize the adaptive adjustment of key parameters in the face of different flight environments and flight states. The simulation and comparison experiments show that the proposed algorithm has a higher degree of fit with the reference trajectory in the trajectory tracking control process and has good robustness.