An Improved Minimal Error Model Based on the Adjoint Transformation Using the POE

Abstract

Unreasonable weights of position and orientation data in the robot pose calibration model lead to low accuracy or even divergence of parameter identification. This article presents a kinematic parameter calibration method based on a minimal error model, separating orientation and position parameters in the calibration process. A robot kinematic model is built based on the Product of Exponentials (POE) formula and adjoint transformation. A non-singular model with complete parameters is obtained by removing the redundant parameters of the adjoint transformation. The linear mapping of the robot position error and the orientation error to the corresponding twist error is derived by differentiating the kinematic model, and the orientation and position error parameters are separately identified using the least-squares algorithm. The calibration experiments are designed and performed on the UR5 industrial robot, and the test results verify the validity of the calibration method.