Sliding Surface-Based Integral Reinforcement Learning for Optimal Tracking Control of Quadcopters Considering Uncertainties

A sliding surface-based integral reinforcement learning (IRL) control scheme is proposed for quadcopter trajectory tracking. The proposed controller combines IRL with the sliding surface approach, improving trajectory tracking performance. The proposed controller addresses the limitations of model-based controllers by leveraging the advantages of RL, particularly in dealing with uncertainties. It enables accurate trajectory tracking even in the presence of system model uncertainties, providing robust performance. The proposed controller needs significantly less data for training than existing RL approaches, including IRL. The proposed controller also incorporates Euler angle constraints in a simplified manner, distinguishing it from other constrained control methods. The online learning nature of the controller enables real-time adaptation and robustness against uncertainties. The performance of the proposed controller is compared with that of standard model-based controllers via numerical simulation, demonstrating its effectiveness and robustness.