Learning-based Modified Transpose Jacobian control of robotic manipulators

The modified transpose Jacobian (MTJ) algorithm has been proposed, based on an approximated feedback linearization approach, in which there is no need to a priori knowledge of the plant dynamics. In this paper, a Learning-based modified transpose Jacobian (LMTJ) control algorithm is presented for trajectory tracking of robotic manipulators in an iterative operation mode. In this new control method, the MTJ control scheme is integrated into a learning procedure; as such, the trajectory tracking converges very fast and the bettering performance is achieved after a few iterations. Unlike conventional learning techniques, control gains selection in LMTJ is easily performed. The LMTJ method possesses both MTJ and learning capabilities with a simple control structure. Obtained results show the merits of the new proposed LMTJ controller that its performance is comparable to that of model-based algorithms in terms of tracking error elimination, even in the presence of nonidentical initializations and un-repetitive disturbances.