Adaptive control of prosthetic limbs using neural networks

It was suggested by E.A. Wan et al. (1990) that artificial neural networks be used to convert muscle control signals from the nerves of the human (or animal) arm into control signals for a prosthetic mechanical hand that has been attached in place of the original hand. The network would learn to translate the (wet) neural signal issued by the brain into electrical signals that would control the mechanical hand. A simplified architecture, the adaptive inverse prosthetic architecture, is presented for accomplishing the same purpose, along with an architecture for an artificial neural network that would learn to convert sensory input from the prosthetic hand into the appropriate sensory neural enervations in the nerves of the arm. The advantage of this approach is that the patient is an integral part of the training process, thereby reducing the need for specialized external computing and measurement equipment during training. This allows the patient to adapt the system at any time, providing a degree of adaptation that is satisfactory to the patient, compensating for nonidealities and changes in the hand, and allowing the patient to learn skills unanticipated by the hand´s designers