應用ANFIS於上肢外骨骼機器人之即時操控

Abstract

為了因應高齡化社會與勞動力不足所帶來的衝擊，穿戴式輔具應用於醫療復健或生活幫扶的相關研究應運而生。如何使輔具操控的方式更加直覺性，且使系統與人體達到更為協調的運動，一直是此領域的研究重點。因此，本論文提出基於EMG訊號作為控制命令輸入源，來操控雙軸的上肢外骨骼機器人，EMG訊號是肌肉收縮時發出的生理訊號，包含了動作的資訊與意圖。根據EMG訊號的高度非線性和模糊性，提出應用Adaptive Neural Fuzzy Inference System(ANFIS)之類神經網路學習方法建立EMG訊號的意圖偵測模型，並搭配力矩輔助控制器減輕使用者的負擔，達到動作幫扶的效果。為了驗證系統的可行性，我們設計相關的測試任務，由實驗可知，由於ANFIS具有良好的適應性，所建立好的力矩輔助系統確實能適用於所有受測者，達到較為省力的輔助效果，且應用到醫療復健或者生活輔助上都能有所助益。

In facing the coming of an aging society and decreasing number of labor force, the exoskeleton types of robots are thus introduced for assistance of rehabilitation and daily activities. Researches have been devoted to simplify the way of manipulation and let the system comply with human body. The thesis proposes a strategy based on the Electromyography (EMG) to govern a two-DOF upper-limb exoskeleton robot. EMG is a physiological signal generated during muscle contraction, which is related to muscle movement and the tension. As the EMG signal is highly non-linear and varying, Adaptive Neural Fuzzy Inference System (ANFIS) is adopted to construct the intention detection model, with a torque assistance controller developed to alleviate the load of the user. To verify system feasibility, we design several tasks for testing. From the experimental results, through the proposed system, especially the adaptability of ANFIS, every user can govern the exoskeleton robot by their intention with less effort. It demonstrates its potential applications for rehabilitation and daily activities.