A Path Planning Algorithm for Space Manipulator Based on Q-Learning

Abstract

An improved Q-Learning autonomous learning algorithm is proposed to solve the problem of the adaptive path planning of the space manipulator in the unknown environment. After simplification of the manipulator and obstacle model, the grid model of the environment is established, and the position of the manipulator and obstacles are randomly deployed in the grid map. Based on the analysis of the basic principle of reinforcement learning and the state generalization method, the improved Q-Learning algorithm is used to carry out the path planning. In this algorithm, the reward and punishment strategies in the path planning of the manipulator are designed, and the approximate greedy and continuous micro Botlzmann distribution behavior selection strategy is adopted. According to the autonomous learning of Q-table, the manipulator can guide its follow-up action selection and path planning, reduce the number of manipulator movement, and reduce the blindness of the learning process. The results show that the algorithm has the advantages of simple calculation, strong self-learning ability, and can successfully complete the adaptive path planning in unknown environment.