Iterative Learning Control of Functional Electrical Stimulation Electrode Arrays

Stroke often causes weakness, paralysis, or loss of coordination in the hand and wrist, making it difficult to perform everyday tasks. Current rehabilitation approaches do not adequately assist patients in regaining their lost function; however, it is possible to produce accurate hand and wrist gestures by artificially stimulating muscles using functional electrical stimulation (FES) applied to multielement electrode arrays. This has been possible using iterative learning control (ILC); however, it required lengthy model identification tests, and accuracy degraded due to fatigue, spasticity, and changes in array position. This article develops a new FES electrode array control framework that maintains high accuracy despite uncertain and potentially time-varying dynamics. First, a model of stimulated hand and wrist dynamics embedding FES array misalignment is developed, and robust stability properties are derived using the gap metric. A compensating controller is then proposed to ameliorate array misalignment, and this is integrated within a powerful framework termed estimation-based multiple-model ILC (EMMILC), which automatically updates the underlying model to maintain performance in the presence of uncertain and changing dynamics. It is shown that EMMILC can remove the need for model identification, whilst maintaining high performance. This significantly improves the usability of FES arrays and opens up the possibility of bringing effective therapy to millions of patients in their own homes. Experimental results reveal that the proposed controller reduces the average converged error norm to 31.3% of that obtained using the existing model-based ILC.