Gait phase prediction for lower limb exoskeleton robots

Abstract

Robust gait phase prediction is crucial to exoskeleton robots, as it detects the intention of users and improves the lag of motion signals. Therefore, this paper predicts gait phases from two perspectives, including one perspective of spatial features and the other of spatio-temporal features. We employ two machine learning models from the two perspectives to predict. One is support vector machine (SVM) optimized by particle swarm optimization (PSO) algorithm, and it only focuses on joint information. Another is nonlinear autoregressive models with external inputs (NARX), and it utilizes previous data to predict present status. As for input, four goniometers built into the exoskeleton robot are used to collect hip and knee joint angles during the walking process. To obtain gait phases, a multi-pressure sensor network composed of three force sensitive resistors (FSRs) is set up and four gait phases, including heel-contact, foot-flat, heel-off and toe-high, are determined according to plantar pressure distribution. Experimental results show that both SVM and NARX are capable of predicting gait phases. Specifically, NARX outperforms SVM in terms of accuracy, since it uses FSRs data to correct the wrong predictions. Consequently, it is better to predict gait phases based on space and time dimensions simultaneously.