Control for neural prostheses: neural networks for determining biological synergies

Abstract

The neural prostheses (NPs) for grasping were developed to assist some daily living activities of hemiplegic subjects after a stroke. A NP that also controls the elbow joint movements could benefit even more to some hemiplegic subjects. NP users require an effective automatic control and practical command interface. The control that we developed is based on the following hypotheses: once the task and preferred strategy for movement are selected, then by using the voluntary (natural) control that drives the proximal segment (shoulder joint), the synergistic (artificial) control drives the distal segment (elbow joint). We confirmed in experiments that reproducible synergies between the shoulder and elbow joint movement exist. Here, we describe a method for determining synergies between joint movements while reaching by applying an inductive learning (IL) technique. This method relies on the hierarchical mutual information classifier algorithm. The synergy is a map obtained by IL between the flex ion/extension (F/E) angular velocities at the shoulder and elbow joints. As the two other shoulder joint rotations are independent from the F/E synergy; thus the results of this study are applicable to a general 3D movement.