A continuous control model of satellite attitude on DSAC-T

The development of the aerospace technology enables countries around the world to launch a large number of satellites to carry out extraterrestrial missions. This also poses greater challenges to the control of satellites. The extremely long communication time between satellites and the ground has led to the necessity of autonomous onboard control capability for satellites. The rise of reinforcement learning (RL) technology has driven research on applying RL technology to satellite control problems. This paper attempts to apply the Distributed Soft Actor Critic with Three Improvements (DSAC-T) algorithm to the adaptive continuous control problem of satellite attitude. The problem is described as a reinforcement learning model that controls the satellite to perform a large-angle slew maneuvers and maintain a desired orientation. By setting up rotation experiments, the results show that the DSAC-T algorithm can control satellite to perform rotation tasks more quickly and smoothly than existing continuous control algorithms.