Motion planning of eye, hand and body of humanoid robot

Abstract

This paper presents an algorithm for planning the motions of a sensor and a manipulator that are kinematically connected. Humanoid robots or other service robots have a vision system and one or two arms, and they are usually mounted on a single body. Given a task to be performed, the robot needs to plan a short, collision-free motion for the arm while placing the camera for optimal viewing conditions so the task progress can be monitored. Since they are kinematically connected, motion of the camera affects the motion of the arm, or the arm may block the view of the camera. Our algorithm plans the body posture, camera motion and the manipulator motion that allows occlusion-free viewing of a target point and placement of the hand at a target hand location. It uses robot's kinematic redundancy to avoid collisions and occlusion of the view point. Since the ensemble of the camera actuator and the manipulator has many joints, the search is performed in the world space rather than in the high-dimensional joint space. The global nature of our search algorithm enables finding a solution in most practical cases.