Coordinated learning control for multiple manipulators holding an object rigidly

Abstract

A learning controller for coordination of multiple manipulators holding an object rigidly is proposed based on the principle of joint-space orthogonalization. It is devised by referring to (1) the fact that dynamics of multiple manipulators resembles that of a single robot under geometric endpoint constraint except the term of the sum of forces exerted from manipulators to the object, (2) the passivity of the dynamics of multiple manipulators, (3) the use of feedback signals of velocity error and force error which are projected to their respective manifolds to be orthogonal to each other in joint-space, and (4) the passivity of the total error system. The convergence of tracking position and force errors is proved theoretically provided that multiple manipulators hold the object rigidly and are initialized at the same posture at the beginning of each learning trial. The effectiveness of this controller is ascertained by computer simulation by using two manipulators with three joints.