Simulated feedforward neural network coordination of hand grasp and wrist angle in a neuroprosthesis

This study presents a possible solution of the general problem of coordinating muscle stimulation in a neuroprosthesis when multiarticular muscles introduce mechanical coupling between joints. In a hand-grasp neuroprosthesis, extrinsic hand muscles cross the wrist joint and introduce large wrist flexion moments during grasp. In order to control hand grasp and wrist angle independently, a controller must take the mechanical coupling into account. In simulation, we investigated the use of artificial neural networks to coordinate hand and wrist muscle stimulation. The networks were trained with data that is easily obtained experimentally. Feedforward control showed excellent hand and wrist coordination when the properties of the system were fixed and there were known external loads. Predictable disturbances (e.g., gravity acting on the hand) can be compensated by sensing arm orientation. However, since wrist angle is sensitive to unpredictable disturbances (e.g., fatigue or object weight), voluntary intervention or feedback control may be required to reduce residual errors.