Identification of muscle-activation-dependent human-exoskeleton coupling parameters.

This paper proposed a muscle-activation-dependent human-exoskeleton model for predicting human-exoskeleton coupling parameters to improve the studies of coupling dynamics. With a newly designed platform and the help of 20 volunteers (10 males and 10 females, age: 24.45 ± 2.31 years old, height: 167.70 ± 8.35 cm, weight: 66.50 ± 18.74 kg), coupling parameters were identified with surface electromyographic (EMG) signals monitored to represent muscle activation. Then convolutional neural network (CNN) was used to predict coupling parameters with six EMG features as inputs:mean absolute value (MAV), mean absolute value slope (MAVSLP), waveform length (WL), Willison Amplitude (WAMP), variance (VAR), and auto regressive (AR) coefficients. Finally, sensitivity analysis of the CNN's performance identified AR, MAV, and VAR as the key determinants of the coupling parameters. Further analysis unveiled strong correlation between coupling stiffness and both MAV and VAR. The novelty and contribution are the design of coupling experimental platform and the establishment of muscle-activation-dependent human-exoskeleton coupling model which provides a possibility to obtain coupling parameter identification form complex human-exoskeleton interaction scenarios.