CME-EPC: A coarse-mechanism embedded error prediction and compensation framework for robot multi-condition tasks

Abstract

While industrial robots are widely used in various fields owing to their large workspace and high flexibility, significant errors constrain their application in high-precision scenarios. Though there have been notable achievements in mechanism modeling for different working conditions, they are complex, work-dependent, and difficult to apply conveniently to multiple operating conditions. Therefore, a coarse-mechanism embedded error prediction and compensation (CME-EPC) framework for robot multi-condition tasks is proposed, combining knowledge-rich coarse mechanism models and intelligent algorithms. These modules are proposed in CME-EPC, including coarse mechanism embedded simulation domain construction, active learning-based labeling of few-shots, and clustering-guided balanced domain adaptation transfer learning. These modules perform jointly to achieve accurate prediction and reliable compensation of errors. The proposed framework is experimentally validated in four tasks under three conditions, achieving superior performance compared to the other six methods with a conventional coarse-mechanism model and 10 % of the labeled samples. The compensated error is significantly reduced compared with other methods, with a maximum reduction of 94.31 %. Further analysis revealed that the CME-EPC is efficient, stable, and robust against the uncertainty of the mechanism model, thus highlighting the future potential of the robot's high-precision applications.