筋電義手のための両腕協調動作生成システムによる日常生活で行う動作の生成

抄録

The joints of a myoelectric upper limb prosthesis are driven by electromyographic signals from the wearer's residual muscles. An above-elbow amputee, who has lost an elbow or shoulder, has few residual muscles in the arm and requires many active joints to drive prosthetic movement. It is very difficult for above-elbow amputees to operate myoelectric upper limb prostheses. In order to support the daily life activities of above-elbow amputees and supplement the function of their myoelectric prostheses, we have focused on the coordinated relationship between movements of both arms. We have previously developed a motion generator in which a neural network learns bimanual coordinated movements and calculates the desired prosthetic movements based on the movements of the healthy arm. Although wearers strive to move their healthy arm precisely as they do a movement, there is always some variation. The neural network of the motion generator requires a generalization capability to generate desired movements of the prosthesis in accordance with speed variations in the motion of the healthy arm. The goal of our study was to train the neural network with 12 types of movements typical of those performed during daily living and to generate desired prosthetic movements based on the variable-speed movements of the healthy arm. Our results showed that the motion generator can calculate desired prosthetic movements with sufficient accuracy based on the movements of the healthy arm, with speeds ranging from 0.5 to 2.0 times that used for training of the neural network.