Real-Time Gait Phase Estimation Based on Neural Network and Assistance Strategy Based on Simulated Muscle Dynamics for an Ankle Exosuit
Zhuo Wang, Chunjie Chen, Fangliang Yang, Yao Liu, Gang Li, Xinyu Wu
Abstract
Soft exosuits have been demonstrated to possess the potential to reduce metabolism and muscle fatigue in human locomotion. How to maximize the user’s metabolic benefit and reduction of muscle fatigue are the primary objectives of assistance patterns optimization. For the exosuit controller, accurate gait phase estimation and appropriate assistance force profile are crucial factors. To gain an accurate real-time gait phase estimation, a neural network algorithm based on multi-modal information fusion was developed. The eigenvector of which was composed of Euler angles and gyroscope data of four Inertial Measurement units fixed on thigh and calf, and gait period estimated by force sensing resistor installed on foot insoles. The mean squared error of the model is below 3 ms and the regression coefficient R exceeds 0.999 when walking at a speed of 1.5 m/s. For the assistance force profile, which was designed based on the characteristics of soleus muscle dynamics. Wherein the biomechanics of walking were derived from simulation with OpenSim software. Compared with wearing an exosuit without assistance and with assistance, the metabolic demand was reduced by an average of 7.5%, and the average peak value and RMS of calf muscles involved in palntarflexion were decreased to different degrees. Our research illustrates the feasibility of neural network-based real-time gait phase estimation and assistance strategy design obtained by simulating muscle dynamics in the OpenSim software.