On adaptive vision feedback control of robotic manipulators

An adaptive vision feedback control system for a manipulator for grasping a moving object is described. The pose (position and orientation) of the object is determined from camera images and its time history is described by an AR (autoregressive) model. Because of the inherent delay caused by the image processing unit, the object pose is predicted for the controller using the AR model. An adaptive controller is designed using the discrete-time Lyapunov theory. The error which is defined in the joint space as the difference between the pose of the end-effector of the manipulator and the predicted pose of the object is shown to approach zero asymptotically by the second method of Lyapunov. The construction of the controller also gives an updating scheme for estimating unknown constant parameters of the manipulator model