Fault-tolerant control of space manipulators moving object capture task based on deep reinforcement learning
Jian Zhao, Yanjiang Chen, Hao Zhang, Guangwei Yu, Tongtong Li · Journal of Physics Conference Series · 2025
Abstract Space manipulators and ground cooperative manipulators exhibit dynamic differences, including the effects of zero gravity, the lower joint stiffness of space manipulators, their slightly weaker dynamic response characteristics, and the fact that they are often considered to have floating bases. These differences result in higher control difficulty for space manipulators compared to ground manipulators. At the same time, due to the limitations of in-orbit hardware computing power, the recognition cycle of visual positioning algorithms is long, making mobile target grasping one of the most challenging tasks for space manipulators. Joints are the core components of space manipulators. Solving the problem of joint jamming and preventing the normal use of space manipulators can effectively improve the service life and quality of space manipulators in orbit. This paper proposes a deep reinforcement learning-based method for executing mobile target capture tasks when a single joint of a space manipulator becomes stuck and loses one degree of freedom. In a simulation environment, a model of a faulty space manipulator is built. Dynamic characteristics and signal delay conditions that are close to the real state are input to train the mobile target capture task. The obtained model meets the requirements of task implementation in the simulation environment.