Two model-free schemes for solving kinematic tracking control of redundant manipulators using CMAC networks

Kinematic tracking control of robotic manipulators is a basic and vital issue in robotics. However, how to achieve the control of robotic manipulators with unknown kinematic models remains challenging. Existing studies show that cerebellum, a pivotal region of brain, plays a major role in human motion control, which motivates us to devise two schemes based on cerebellum model articulation controller (CMAC) for the tracking control of redundant manipulators with unknown kinematic models in this article. The first scheme is based on a CMAC network and a gradient neural dynamics (GND) algorithm. The CMAC network is utilized to generate approximate joint angle commands for the manipulator. By using the Jacobian matrix of the manipulator which is estimated by the GND algorithm, task space error is transformed into joint space error, which is taken as the teaching signal for training the CMAC. For the second scheme, another CMAC network replaces the GND algorithm to achieve the estimation of the Jacobian matrix of the manipulator. The effectiveness and robustness of the proposed control schemes are verified by both simulations and physical experiments.