Abstract
The cybernetic interface through which users can communicate with computers “as we may think” is the dream of human-computer interactions. Aiming at interfaces where machines adapt themselves to users' intention instead of users' adaptation to machines, we have been applying a neural network to realize electromyographic (EMG)-controlled slave hand. This paper proposes that EMG patterns can be analyzed and classified by a neural network. Through experiments and simulations, it is shown that recognition of finger movement and joint angles in dynamic finger movement can be successfully accomplished.
A 3-layred back-propagation network is used for finger action recognition from 1 or 2ch surface EMG. In the case of static fingers' motions recognition, 5 categories were classified by the neural network and the recognition rate was 86%. In the case of joint angles estimation in continuous finger motion, the root mean square error was under 25 degrees for 5 fingers 10 joints angles' estimations.
Cyber Finger with 5 fingers 10 joint angles was realized to be controlled by 2ch surface EMG. The slave hand was controlled smoothly and voluntarily by a neural network.
