A Semiparametric Model-Based Friction Compensation Method for Multijoint Industrial Robot

Abstract

Industrial robots have received enormous attention due to their widespread uses in modern manufacturing. However, due to the frictional discontinuous and other unknown dynamics in a robotic system, existing researches are limited to simulation and single- or double-joint robot. In this paper, we introduce a semiparametric controller combined with a radial basis function neural network (RBFNN) and a complete physical model considering joint friction. First, to extend the neural network (NN) controller to real-world problems, the continuously differentiable friction (CDF) model is adopted to bring physical information into the learning process. Then, RBFNN is employed to approximate the model error and other unmolded dynamics, and the parameters of the CDF model are updated online according to its learning ability. The stability of the robot system can be guaranteed by the Lyapunov theory. The primary parameters of the CDF model are determined by the identification experiment and subsequently iteratively updated by the NN. Real-time tracking tasks are performed on a six-degree-of-freedom (DoF) manipulator to follow the desired trajectory. Experimental results demonstrate the effectiveness and superiority of the proposed controller, especially at low speed.