Robot Positioning Error Compensation Method Based on Machine Learning and Reduction of Manual Operation Using Estimation of Compensation Error

Abstract

Recently digital twins are widely utilized in various manufacturing stage, and various methods are proposed to compensate the error between digital model and real world. This paper discusses the types of errors that occur in the digital model of a robot and their correction methods. We focus on machine learning methods that can be applied to the target robot with a minimal number of measurement points or without the need for a mechanism model. We state that the RBF interpolation method is generally applicable among the interpolation methods that can effectively utilize the high repeatability accuracy of the robot. We also describe the error estimation process using similarity measure such as k-NN and demonstrate through simulation that the number of teaching points can be efficiently reduced by performing teaching correction in order of the teaching points with the maximum estimated error. Additionally, we proposed a use case based on the automotive spot-welding line and demonstrated that applying this method could potentially reduce the manual teaching points by one-third.