Abstract:The space adhesive climbing robot can be attached to the outer surface of the spacecraft and complete the external inspection and operation tasks independently, which is an important way to realize the long-term unmanned in-orbit service of the spacecraft. In order to solve the problem of insufficient generalization ability of the control strategy of the adhesive climbing robot after unexpected changes in spacecraft surface characteristics, the mechanism of adhesion force is constructed under the framework of reinforcement learning, and the intensive reward function is constructed by combining the “follow-update” mechanism of the foot contact force, and the proximal policy optimization-clip (PPO-clip) algorithm is used to train and generate the adhesion crawling strategy of the robot in microgravity environment. The results show that the strategy convergence rate increases by about 14.81% under the “follow-update” mechanism of foot contact force. The climbing strategy obtained can maintain the adhesion stability of the robot on a flat surface, and has the ability to reach the target position with an arrival error of less than 0.1m. On surfaces with an unpredictable height change of ±40mm and an unpredictable slope change of ±18°, the climbing strategy obtained on the flat surface can achieve stable adhesion climbing of the robot.