Stereo visual servo of robot manipulators by estimated image features without 3D reconstruction

Abstract

A new visual servo system using estimated image features on the stereo image plane without 3D reconstruction is proposed. In visual servo systems, the sampling period of the image processing is longer than that of manipulator control. In order to compensate for the image processing delay, image features having no delays are estimated on the stereo image plane from the image Jacobian matrix and deviation of the joint angles of the robot manipulators. The visual servo system carries out all processes only on the stereo image plane in real time without an inner loop controller and 3D reconstruction, i.e., a torque reference is generated from position, velocity, and acceleration commands on the stereo image plane. Therefore, it is possible to control manipulators quickly and precisely and to decrease the complex process of 3D reconstruction. Results of position control using a three joint manipulator in the stereo image plane are presented.