Real-time Classification of Diverse Reaching Motions Using RMS and Discrete Wavelet Transform Energy Values from EMG Signals for Human Assistive Robots.

With advancing technology, human assistive robots have been developed to enhance daily efficiency for users. Focusing on the reaching motions of the upper limb, this study aims to propose a motion classification method based on electromyographic (EMG) signals that can accurately and promptly differentiate among three distinct types of reaching motion-regular reaching, extended reaching, and weighted reaching-regardless of the motion direction. In the proposed method, the EMG signals of upper limb and torso muscles relevant to these reaching motions are used to identify pivotal features capable of clearly classifying these different reaching motions. A Gated Recurrent Unit (GRU) network is employed to train the model and infer user intentions based on the signal features. The results confirmed the efficiency in motion classification, which laid the foundation for the future application of human assist robots, enabling them to provide users with timely and precise responses.