Optimal posture control for object on active vision robot

Abstract

The paper describes way of determining the optimal posture of active vision robot (AVRobot) for recognizing unidentified objects. In the object recognition with visual information by the AVRobot, it is required to measure the accurate positions of feature points of a target object. For the precise measurement, it is preferable, for example, that several feature points exist in a captured image for calculation of relative positions and distances among them and that the position of feature point on the image moves boldly in small motion of the camera attached on the robot. In this paper, the motion control law based on visual servoing and posture design technique with null-space vector is proposed so as to satisfy such requirements. The control law simultaneously achieves two control requirement in our approach; the first one is to make the position of primary feature point of the target locate center of the image, and the second to induce the camera position where let the image transition sensitive to the infinitesimal change of the camera motion referring the positions of the secondary features, so as to suppress the measurement errors of the feature positions.