Collision Avoidance for a Visuo-motor System Using Multiple Self-organizing Maps

Collision avoidance for a visuo-motor system in unstructured and cluttered environment is described. The achievement of collision avoidance is based on a simplified path planning system and motion control performed by self-organizing maps. The self-organizing maps are learned to determine joint angles of a redundant manipulator. Since the learning algorithm promises to make the manipulator reach targets precisely with obstacle-free poses, the path planning system only needs to plan a collision-free path for the end effector of the manipulator in the image spaces. By means of the cooperation of two self-organizing maps, the system solves occlusion problems successfully. Simulation results are presented to demonstrate the effectiveness of the proposed approach.