A Kinematic Calibration Method Based on Residual Network Combining Joint Angles and Robot Pose

Abstract

Robot performance metrics like their absolute positioning accuracy have a significant impact on their industrial applications. This research introduces a kinematic calibration approach for industrial robots based on the residual network that combines joint angles and robot pose. To compensate for the geometric error, a geometric parameter identification algorithm founded on the MDH model and error model is suggested. The residual network uses joint angles and robot pose as network inputs to compensate for non-geometric error together with the results of geometric parameter identification. Experimental validation is conducted on the Rokae XB7S robot, with a dataset constructed from joint angle changes and robot pose variations. The experimental results show that the robot’s position error decreases from $0.6549\mathrm{~mm}$ to $0.1011\mathrm{~mm}$ after compensation, verifying the suggested approach’s efficiency.