Comparison of Multivariate Linear and Nonlinear Models for FES Cycling

A common rehabilitation approach for individuals with SCI is the use of functional electrical stimulation (FES)-assisted cycling. However, there are current efforts to overcome it limitations, especially in relation to fatigue and the nonlinearity of the muscle system. Model based FES controllers have been proposed to improve accuracy of cyclic tasks, such cycling or stepping. But, these controllers needs accurate models to obtain good predictions and calculate bounded and constrained control actions. To improve model accuracy, algorithms tests with linear (ARX) and non-linear (NARX) can be used to represent the highest degree of reproducibility for human physiological movement. Then, the objective of this work is to determine the minimum model linear and nonlinear structure needed to represent the relationship between FES and angle in FES cycling activity. For that, two subjects (one without disability and the other with spinal cord injury) were sat in an armchair coupled to a cycle ergometer reproducing the pedaling movement at a speed of 30 RPM with its movement being generated exclusively by FES (pulse width 300 μs and frequency 35 Hz). Meanwhile, an IMU sensor collected position and movement data from the right lower limb. The collected data showed that the NARX model, despite having more parameters, better represented the relationship between FES and IMU in both cases. With this, we assume that these parameters can be used in the MPC FES to improve pedaling ability and make it safer for clinical application in subjects with SCI.