Path Planning of Redundant Robot Manipulator for Obstacle Avoidance Using Reinforcement Learning - Reduction of Searched Configuration Space Using SOM and Two-stage Learning

摘要

It is necessary for a robot manipulator to plan a path, which can adapt to unknown and changeable environment. A method of path planning using the reinforcement learning is proposed. By restructuring a configuration space using self-organizing maps (SOM), convergence time is reduced. Two-stage learning for obstacle avoidance of redundant robot is proposed, the first stage of which searches a path without obstacle, the second one searches a path with obstacle. The second search is carried out in the configuration space near the first path, which reduces searched space and convergence time. The proposed method is confirmed by a computer simulation.