混合余事象分布を内包した確率ニューラルネットに基づく5指駆動型ロボットハンドのEMG制御

抄録

Many man-machine interfaces controlled by electromyogram (EMG) signals such as the myoelectric prosthetic hand have been proposed. General classifiers do not cover unintended motions in the training phase and misclassify those inevitably. Since the misclassification can cause dangerous incidents, an interface with high security is required. To solve this problem, this paper proposes a novel control method of man-machine interfaces that can treat unlearned motions. In the experiments, the proposed method was applied to forearm and finger motion classification to evaluate the validity. The outcomes showed that the approach produces higher and more stable classification performance than comparative methods.