Spatial error prediction method for industrial robot based on Support Vector Regression

The high-end intelligent manufacturing field has put forward higher requirements for the absolute pose accuracy of industrial robots in high-accuracy application scenarios. This paper investigated the improvement of robot accuracy performance based on Support Vector Regression（SVR）. Kinematic modeling and error analysis were performed on the Staubli TX60 series industrial robot. A robot measurement experiment platform was established based on the Leica AT960 laser tracker，and a large number of spatial position points were measured. The SVR model was trained and optimized based on real data sets. The actual pose error of the robot is predicted by Support Vector Regression Model，which avoids the complicated error modeling in the model-based robot accuracy improvement method. The average position error and average attitude error of the robot are reduced from（0.706 1 mm，0.174 2°）to（0.055 6 mm，0.024 6°）before compensation，respectively，and the position error is reduced by 92. 12% and the attitude error is reduced by 85. 88%. Finally, the comparison with BP neural network, Elman neural network and traditional LM geometric parameter calibration method verified the effectiveness and balance of spatial error prediction based on SVR model in reducing robot position and attitude errors. © 2024 Chinese Academy of Sciences. All rights reserved.