Reinforcement learning strategy for spacecraft attitude hyperagile tracking control with uncertainties

Abstract

Spacecraft attitude tracking control is a challenge for different space moving target observation tasks, especially when uncertainties exist. Traditional controllers have poor adaptability, and iterative learning controllers are time consuming. This paper focuses on the spacecraft attitude tracking control problem with uncertainties, including an unknown inertia matrix, desired attitude and desired angular velocity. To improve the attitude tracking accuracies under uncertainties, a reinforcement learning (RL)-based sliding mode observer is designed in which the uncertainties are estimated by the learning process using RL. To guarantee the convergence of the prescribed sliding mode surface, a controller based on the combination of traditional feedback control and RL is developed. The traditional feedback controller is used to speed up the learning process and to reject the aperiodic disturbance, while RL is applied to solve the uncertainty and to reject the periodic disturbance. The proposed controller can maintain high attitude tracking accuracies for different tracking missions in which the unknown inertia matrix, the desired attitude and attitude angular velocity are different without adjusting parameters. Finally, comparison results of the numerical simulation illustrate the effectiveness of the proposed method.