利用前饋神經網路與數位雙生學習增強機器手臂的運動學校正演算法

Abstract

本論文採用了一種創新的方法來解決機械手臂精度控制中的挑戰。通過結合雅可比矩陣(Jacobian matrix)校正和前饋神經網路(Feedforward neural network, FNN)，我們提出了一個新型的平行架構模型，旨在提高機械手臂在現實世界中的運動學準確性。傳統的雅可比矩陣校正方法在捕捉機械手臂系統複雜性方面存在局限性，因此我們引入了數位雙生(Digital twin)技術和轉移學習(Transfer learning)的理念，以優化模型。透過這種平行架構，我們旨在識別出最貼近實際機械手臂行為的模型，從而提高精度控制，以滿足工業和醫療應用對精準度不斷增長的需求。 這一研究的貢獻在於，它不僅修正了傳統運動學模型的缺陷，還引入了數位雙生和轉移學習的思想，使得模擬環境中的演算法能夠直接應用到實際機械手臂的問題中。這不僅提高了機械手臂在現實應用中的準確性和效率，也拓展了機械手臂技術在工業和醫療領域的應用範疇。透過本研究提出的方法，可以更好地應對機械手臂精度控制面臨的挑戰，為相關領域的發展提供新的思路和解決方案。

This paper adopts an innovative approach to address challenges in precision control of robotic arms. By combining Jacobian matrix correction and feed-forward neural networks, we propose a novel parallel architecture model aimed at enhancing the kinematic accuracy of robotic arms in real-world scenarios. Traditional Jacobian matrix correction methods have limitations in capturing the complexity of robotic arm systems; therefore, we introduce the concept of digital twin technology and transfer learning to optimize the model. Through this parallel architecture, we aim to identify models that closely mimic actual robotic arm behavior, thereby improving precision control to meet the increasing demands in industrial and medical applications. The contribution of this research lies in its dual approach: correcting the shortcomings of traditional kinematic models and integrating the concepts of digital twins and transfer learning. This allows algorithms developed in simulation environments to be directly applied to real-world robotic arm challenges. Consequently, this approach not only enhances the accuracy and efficiency of robotic arms in practical applications but also broadens the scope of robotic arm technology in industrial and medical fields. The method proposed in this study provides a robust solution to the challenges in precision control of robotic arms and offers new perspectives and solutions for the advancement of related fields.