Vision-based posture estimation and null space control for redundant manipulator

Abstract

This paper presents an approach for a posture control method based on the vision-based posture estimation for a redundant manipulator. The redundant manipulator can provide dexterous task capability depending on the environmental information. In this paper, a hybrid motion controller including both a posture controller based on visual feedback and an end-effector motion controller based on encoder signal is proposed. In the proposed posture controller, the manipulator posture is estimated by using a particle filter, only based on the visual information. This means that posture control is completely achieved only by a vision sensor signal in the null space. The control performance of vision-based null space control is experimentally evaluated for several motions using the redundancy. A suitable control model for obstacle avoidance is also proposed by using an optical flow field, which is detected based on the visual information. The validity of the proposed method is evaluated by several experimental results.