Functional error assessment in gait event discrimination for FES-assisted locomotion

Discrimination of gait events is required for control of functional electrical stimulation (FES) used to assist with locomotion after lesions in the central nervous system. The authors have previously demonstrated that adaptive logic networks (ALNs) can be used in gait event discrimination (GED) and automatic design of control algorithms for FES-assisted locomotion. Machine learning techniques for supervised learning generalize well in cloning the skill of the subject controlling the stimulation manually. However, safe operation of a GED-based control system applied to a highly variable human walking may require more redundancy than ALNs alone can provide. A practical method was designed to provide additional safety for the subject. Functional errors of the FES-control system were defined to evaluate its performance, and restriction rules (RRs), based on a priori knowledge about the subject´s walking, were introduced to verify decisions made by ALNs and to eliminate infrequent functional errors. The authors´ results demonstrated that RRs can completely eliminate critical functional errors if a certain level of ALN training is achieved. The authors´ future efforts are directed toward expanding the training error assessment method to define the training level required for error-free performance