Adaptive neural network control of manipulators with uncertain kinematics and dynamics

A manipulator system may have both kinematics and dynamics uncertainties, which pose difficulties in controller design. To solve the above problem, this study proposes an adaptive double neural networks sliding mode control (ADNSMC). First, the gradient method is utilized to derive the update rate of the Jacobian matrix and is combined with a recurrent neural network (RNN) to realize compensation for unknown kinematics. Then, a fully tuned radial basis function neural network (RBFNN) is designed to approximate and compensate for unknown dynamics. Moreover, both the neural networks are incorporated into a nonsingular fast terminal sliding mode control (NFTSMC) framework to improve the accuracy and convergence speed of the control system. The Lyapunov theory is employed to verify the system stability. Finally, simulation experiments based on a KUKA LBR iiwa 14 R820 manipulator are performed, which demonstrate the effectiveness and superiority of ADNSMC, verifying its engineering applications.