Improving the Inverse Dynamics Model of the KUKA LWR IV+ using Independent Joint Learning

Abstract

Abstract: In this paper, we discuss the improvement of the inverse dynamics models of the KUKA LWR IV+ by a recently proposed approach called Independent Joint Learning (IJL). In IJL, the error between the torques from the real robot and the torques from inaccurate dynamics model is estimated using only joint-local information. Due to the reduced model complexity IJL can be used for task-to-task transfer learning and to a task different from the trained tasks. In this paper, we implemented IJL to improve the accuracy of the already existing KUKA LWR IV+ inverse dynamics model and our results show a significant improvement. We also discuss IJL for different types of input datasets and compared them in terms of performance.