Global asymptotic stabilization of visually-servoed manipulators

Abstract

In this paper, the problem of controlling the relative pose between a robot camera and a rigid object of interest is solved using nonlinear system theory. The camera-object visual interaction model is derived in the state space form in terms of image points. Then it is shown that the image-based visual system is completely controllable. Moreover, as a consequence of nonlinear controllability, global uniform asymptotic stability of the image reference set-point is formally proved using Lyapunov's direct method. Active contours are used to track the 2-D projection of the visible object's surface in the image plane, and 3-D estimation procedure based on prediction errors is used to cope with the unknown depth of the object. Experimental results with a 6-DOF robot manipulator endowed with a camera on its wrist validate the framework.