A robot geometric error identification method based on a hybrid LR-QNLM algorithm

In order to improve the poor accuracy of industrial robots, the calibration technique for robot geometric parameters has emerged as a mainstream approach. However, existing calibration algorithms do not perform as expected due to initial large residual errors. To solve the problem, this work proposes a hybrid LR-QNLM algorithm based on the modified quasi-Newton (MQN) and Levenberg-Marquardt-Fletcher (LMF) methods. Firstly, we build the robot kinematic error identification model considering extended axes as the objective equation. Then, hybrid strategies are used to incorporate two methods above for a more accurate approximation of the Hessian matrix and well-local convergence property. Moreover, a non-monotonic linear search algorithm is used to achieve sufficient descent ability with a control criterion used to achieve more precise iteration. Finally, calibration experiments show that the robot positioning accuracy and robustness are well improved, which verifies the feasibility of the proposed method.