Iterative Learning Control and System Identification of the Antagonistic Knee Muscle Complex During Gait Using Functional Electrical Stimulation

Abstract

Functional Electrical Stimulation (FES) can be used to support the gait of stroke patients. By measuring joint angles and adjusting the stimulation intensities automatically to the current need of the patient, setup times can be reduced and time-variant effects like muscle fatigue can be compensated. This was achieved in recent publications by using Iterative Learning Control (ILC) on the ankle complex. In this paper we consider FES of the antagonistic knee muscle complex (quadriceps and hamstring muscles) that controls knee flexion/extension. We used a coactivation strategy in order to map the two stimulation channels to a single control input. A large class of dynamic models was obtained by system identification based on data from two experiments: one with standing subjects and one with subjects walking on a treadmill while being stimulated during different time segments of the gait cycle. Time delays, system poles, and in particular the system gains were found to vary largely. Furthermore, large differences were observed between muscle dynamics in standing pose and during walking. We designed an iterative learning controller that is stable for almost all models. In experiments with eight healthy subjects walking on a treadmill, the ILC was found to reduce deviations from a reference trajectory to about five degrees within two strides.