Calibration of Serial Robots through Integration of Local POE Formula and Artificial Neural Networks

Abstract

This paper presents a calibration method for serial robots to improve pose accuracy. In this method, the complicated calibration of the actual robot containing various error sources is converted to a simple one of the equivalent robot only containing configuration-dependent joint motion errors. For the lower-mobility robot, which has n (less than 6) degree of freedom, other 6-n virtual joints need to be introduced into the equivalent robot to meet the completeness requirement. A simplified local POE formula is used to build the relationship between the pose error and joint motion errors, and an artificial neural network is used to approximate the relationship between joint motion errors and nominal joint variables. By integrating the two models, the error model of the equivalent robot can be deduced and then used for calibration. Simulation results on a typical serial robot show that the proposed method can reduce pose errors significantly.