Elbow Joint Angle Estimation from Surface Electromyography Using Hammerstein-Wiener Models

Abstract

Estimation of a model that relates the surface electromyography (sEMG) to the joint angle is of major importance to design human robot interfaces, to control biomimetic mechanisms and to model muscular systems. However, it is challenging to obtain information from electromyographic signals due to nonlinearities, unmodeled muscle dynamics, noise and interference. In this paper, we applied system identification methods to estimate the parameters of a Hammerstein Wiener model that relates sEMG from three muscles of the arm (biceps brachii, triceps brachii and brachioradialis) to the elbow joint angle. In order to determine an estimation model and a calibration procedure, a set of experiments were carried out with two subjects. The experiment consisted of a series of elbow flexion and extension movements with different weights (Okg, 1.5kg and 3kg) and two different types of movement: continuous and intermittent, with controlled velocity. The sEMG and angular data were recorded during the experiment. The chosen model for system identification was the Hammerstein-Wiener with Wavelet Network as input and output nonlinearities. A second experiment was conducted in a different day in order to validate the estimated models. The results show a procedure to estimate a model for the EMG-to-Joint Angle, showing high correlations and coefficient of determination as well as low Root-Mean-Square Errors (RMSE) with respect to the measured data: correlations of 94.90±.92%, coefficients of determination of 0.8814 ±0.0978 and RMSE values of 10.82 ±3.73°. These models are interesting candidates to help in the control of actuated prosthetic and orthotic devices.