Model-Free Optimal Attitude Control of Spacecraft with External Disturbances and Input Saturation Based on DRL

Abstract

This work is devoted to optimal attitude control for rigid spacecraft in the presence of parametric uncertainties, control torque saturation, and external disturbances. Specifically, an optimal controller based on deep reinforcement learning (DRL) is proposed. The DRL controller is a model-free controller, which can continuously learn according to the feedback of the environment and achieve high-precision attitude control of the spacecraft, furthermore, there is no need to adjust the controller parameters repeatedly. Considering the continuity of the state space and action space for spacecraft, the Deep Deterministic Policy Gradient (DDPG) algorithm based on the ac-tor-critic architecture is adopted. Firstly, the general steps of designing the DRL controller are presented. Then, the controller based on DDPG algorithm is created and trained in the simulation environment. Finally, the effectiveness and robustness of the controller are verified by four groups of simulation examples. Experimental results show that after 462 episodes of training, the agent has learned the ideal control policy. The controller has strong robustness against uncertain inertial parameters and external disturbances and can achieve attitude stabilization control of spacecraft under continuous external disturbances.