本研究的目標在建立一套可控制的肢體同動機械手臂系統，可藉由使用者的肌電訊號強弱控制機械手臂完成預期的動作，如：舉起、放下至任意角度、支撐重物等方面，未來可作為肢體障礙患者或復健的使用以及軍事和醫學用途上的機械輔具。 研究利用倒傳遞類神經演算法，將肌電訊號做為輸入資料，進行手臂輸出扭力的估測，並將類神經加以改良，同時結合判斷器方法，對肌電訊號進行扭力大小的分類，再輸入到不同的類神經系統進行訓練，提高估測的準確性以及實用性，並與基本類神經及最小平方法做比較，驗證改善的效果是否如預期。 The purpose of this research is to develop a controllable mechanical arm which can move with human’s arm simultaneously. The user can control the mechanical arm to finish prospective motion, such as raise it or lay it down at any angle, or sustain a heavy thing on it by user’s electromyographic signal. In the future, this controllable mechanical arm can help those physical disabled patients to move and recovery. This can be used in military or medical perspective, too. The research is used the backpropagation neural network method, and use the electromyographic signal as its input parameter in order to estimate torque which is made by the arm. We also improve the backpropagation neural network method and combine it with a judging method. It can classify the electromyographic signal by torque. After that, we put the signal into the different system to estimate. In that way, we can improve the accuracy and practicability of our estimate. Besides, we will compare our method with basic backpropagation neural network and least-square method to see if the research is better.