On line-processing of position and force measures for contour identification and robot control

The problem of controlling a robot whose task is to follow the profile of an unknown workpiece, with specified tangential velocity and normal force, is considered. At the same time, geometrical and physical characterizations of the workpiece are obtained as a byproduct. Local identification of the object shape must be achieved online since the control scheme requires the direction of normal and tangent spaces to the surface at each contact point. Information from position and force sensors is available but corrupted by noise. Moreover, elasticity at the contact and at the robot joints affects the accuracy of position measurements, and contact friction disturbs the geometrical informations obtained from force measurements. The control and the identification goals are obtained by processing both position and force measurements using a recursive-least-squares procedure. Experiments that validate the proposed identification scheme have been performed on a three-jointed robot arm