A Comparative Study of Reinforcement Learning in Spacecraft Orbit Control

This paper addresses the close-range rendezvous in spacecraft on-orbit services, focusing on orbit control under perturbations like the non-spherical earth perturbation and controller noise. By using Deep Reinforcement Learning algorithms (Proximal Policy Optimization, Two-delay Deep Deterministic Policy Gradient, and Soft Actor-Critic), the simulations demonstrate superior that the method used in this paper performs well in terms of control accuracy, stability, and disturbance rejection capability. A detailed analysis of the merits and limitations of three algorithms is presented, ultimately concluding with reinforcement learning applications in autonomous space missions.