A relation-learning method between motions of both arms aiming at application to upper limb prosthesis

We proposed a relation-learning method by using Volterra neuron networks in order to represent cooperative relationships between motions of each part in a human body. Though it is considered that these relationships are associated with skillful motion, they are often nonlinear and cannot be formulated. A Volterra neuron network, which one of the authors has presented in previous work, can deal with nonlinear signal processing better than TDNN since each neuron has a Volterra filter used for nonlinear filtering. Then we assumed that the relation-learning method was applied to upper limb prosthesis for an above-elbow amputee. A difference from conventional myoelectric prosthesis is a motion generation method which the cooperative relationships is applied to. Conceptually the prosthesis does not take any surgery or training. Preliminary experiments were performed in order to check if Volterra neuron networks can learn several kinds of measured motions. Rotation angles of both arm's joints were Used as signals of the neural networks. In the best result, average of the error margin rate was approximately 5.5%. From the results, it was shown that the relation-learning method by using a Volterra neuron network can be useful for representation of cooperative relationships between motions.