RISE Controller Tuning and System Identification Through Machine Learning for Human Lower Limb Rehabilitation via Neuromuscular Electrical Stimulation

TL;DR

This paper presents a novel machine learning-based control methodology for NMES in lower limb rehabilitation, improving stability and personalization while reducing muscle fatigue for SCI patients.

Contribution

It introduces a robust control and system identification framework using machine learning and past data, specifically applying RISE control with neural networks for SCI rehabilitation.

Findings

Better control performance than empirical tuning

Reduced muscle fatigue in SCI patients

Effective use of past data for system modeling

Abstract

Neuromuscular electrical stimulation (NMES) has been effectively applied in many rehabilitation treatments of individuals with spinal cord injury (SCI). In this context, we introduce a novel, robust, and intelligent control-based methodology to closed-loop NMES systems. Our approach utilizes a robust control law to guarantee system stability and machine learning tools to optimize both the controller parameters and system identification. Regarding the latter, we introduce the use of past rehabilitation data to build more realistic data-driven identified models. Furthermore, we apply the proposed methodology for the rehabilitation of lower limbs using a control technique named the robust integral of the sign of the error (RISE), an offline improved genetic algorithm optimizer, and neural network models. Although in the literature, the RISE controller presented good results on healthy subjects, without any fine-tuning method, a trial and error approach would quickly lead to muscle fatigue for individuals with SCI. In this paper, for the first time, the RISE controller is evaluated with two paraplegic subjects in one stimulation session and with seven healthy individuals in at least two and at most five sessions. The results showed that the proposed approach provided a better control performance than empirical tuning, which can avoid premature fatigue on NMES-based clinical procedures.