A positional error compensation method for industrial robots combining error similarity and radial basis function neural network

Abstract

To solve the problem of low absolute position accuracy for industrial robots in application, a positional error compensation method combing error similarity and RBF neural networks is proposed. The positional errors experience error similarity when describing the degree of error similarity developed with the error model based on a robot kinematic model. The experimental semivariogram is fitted by using a set of robot joint angles and corresponding positional errors. The bandwidth of the RBF neural network is modified by using the parameter of semivariogram. Then, an RBF neural network is constructed to estimate the positional errors of the target positions. The estimated positional errors are used to modify the target position. The modified position is given to the robot controller. To verify the proposed method, a simulation study and experiments are respectively carried out with a simulated robot and a KUKA KR210 industrial robot. The experimental results show that, after compensation, the average residual positional error is reduced by 91.99% from 1.361 mm to 0.109 mm and the maximum residual positional error is reduced by 85.41% from 1.741 mm to 0.254 mm. In addition, the proposed method can enhance the absolute position accuracy of industrial robots.