Vision-based control of the Manus using SIFT

Abstract

The rehabilitation robot Manus is an assistive device for severely motor handicapped users. The executing of all day living tasks with the Manus, can be very complex and a vision-based controller can simplify this. The lack of existing vision-based controlled systems, is the poor reliability of the computer vision in unstructured environments. In this paper, a computer vision solution is presented, which can estimate realtime the pose of an object and co-operate with a vision-based controller. The computer vision is robust to illumination changes, a varying scale and rotation and is robust to occlusion. The computer vision is mainly based on the SIFT-algorithm and the usage of a 3D-model of an object. Important steps to create this 3D-model are discussed. The detection and recognition of the required SIFT-keypoints, has become realtime, by exchanging redundancy against calculation time. With a position-based visual servoing controller, the Manus can be positioned with respect to an object.