Vision-based end-effector position error compensation

This paper describes a computationally efficient algorithm that provides the ability to accurately place an arm end-effector on a target designated in an image using low speed feedback from a fixed stereo camera. The algorithm is robust to visual occlusion of the end-effector and does not require high fidelity calibration of either the arm or stereo camera. The algorithm works by maintaining an error vector between the locations of a fiducial on the arm's end-effector as predicted by a kinematic model of the arm and detected and triangulated by a stereo camera pair. It then uses this error vector to compensate for errors in the kinematic model and servo to the target designated in the stereo camera pair