CALIBRATION OF THE ROBOTIC ARM WITH CORRECTIONS USING LOCAL LINEAR NEURO-FUZZY MODELS

Abstract

The paper deals with the enhancement of the robotic arm calibration using corrections based on local linear neuro-fuzzy models. After the standard calibration of the geometric parameters in the robot's kinematic model, there are still residual errors between the measured positions and the positions predicted by the model. The source of these errors are various non-geometric parameters and nonlinear phenomena that traditional kinematic calibration models do not include. The neuro-fuzzy model based on a locally linear model tree can approximate the residual error as a function of the robot's joint angles. Adding this approximation to the output of the calibrated robot model significantly increases the accuracy of the endeffector position. The results of the described method were verified and compared with other approaches on a simulation model of a flexible planar two-link mechanism. Experimental verification was performed on an industrial robot Stäubli TX200 with data measured by Leica laser tracking device.