A Comprehensive On-Load Calibration Method for Industrial Robots Based on a Unified Kinetostatic Error Model and Gaussian Process Regression

Abstract

Industrial robots are widely used in various manufacturing processes due to their flexibility and versatility. However, the robot’s absolute accuracy is significantly impacted by inaccurate kinematic parameters, joint compliance and other nonlinear factors. To improve the robot’s absolute accuracy, a comprehensive calibration method is proposed in this paper. Based on the local product-of-exponential formula and force Jacobian mapping, the forward kinematics of the loaded serial robot is established. Thereby, a unified kinetostatic error model is obtained using matrix differentiation and adjoint mapping. This model achieves the simultaneous calibration of geometric and deformation errors. To further improve the robot’s absolute accuracy, a Gaussian process regression model based on Bayesian optimization is proposed to compensate for residual errors. Experiments were conducted on the ABB-IRB 4400 industrial robot. The results under various loads demonstrate that, compared with state-of-the-art methods, the proposed method can enhance the calibration accuracy by approximately 8.1% to 54.8%, thus verifying its effectiveness.