Reactive Humanoid Motion Planning for Reaching Tasks

This paper addresses a reactive motion planning framework that allows a humanoid robot, supposedly teleoperated, to perform reaching tasks in complex environments with uncertainty like a damaged plant using measured information like voxel map or point clouds. Since sufficient reactiveness is required for smooth human operation, we have developed an efficient computation method to update the environment information and to plan or replan a feasible whole-body reaching path within a second when necessary. This highly responsive planning scheme benefits from rapid computation of whole-body stable posture using approximated center of gravity and analytical inverse kinematics, combined with effective representation of 3D environment using sphere tree that can be rapidly updated when environmental changes occur. We validate the proposed method in a cluttered plant environment with moving object.