A computer simulation test of feedback error learning-based FES controller for controlling random and cyclic movements

Abstract

Feedback control of movements by functional electrical stimulation (FES) can be useful for restoring motor function of paralyzed subjects. However, it has not been used practically. Some of possible reasons were considered to be in designing a feedback FES controller and its parameter determination, and nonlinear characteristics with large time delay in muscle response to electrical stimulation, which are different between subjects. This study focused on the hybrid controller that consists of artificial neural network (ANN) and fuzzy feedback controller. ANN was trained by feedback error learning (FEL) to realize a feedforward controller. Although FEL can realize feedforward FES controller, target movement patterns are limited to those similar to patterns used in the training. In this paper, FEL-FES controller was tested in learning both random and cyclic movements through computer simulation of knee joint angle control with 4 different training data sets: (1) sinusoidal patterns, (2) patterns generated by low pass filtered random values, (3) using both the sinusoidal and the LPF random patterns alternatively and (4) patterns that consisted of 3 random sinusoidal components. Trained ANNs were evaluated in feedforward control of sinusoidal and random angle patterns. Training with data set (1) caused delay in controlling random patterns, and training with data set (2) caused delay in controlling sinusoidal patterns. Training with data set (3) showed intermediate performance between those with data set (1) and (2). Training with data set (4) could control adequately both random and sinusoidal patterns. These results suggested that generating movement patterns using sinusoidal components would be effective for various movement control by FEL-FES controller.