An Elbow Joint Angle Estimation Method Based on sEMG Feature and the Extra IMF Feature

Abstract

With the tendency of aging society, the assistant robot for the aged has become a research hotspot, such as exoskeleton robot. Surface electromyography (sEMG) is a kind of biological signals generated by muscle, and the joint angle estimated by sEMG can be used to construct efficient control strategy for the assistant robot. Estimating joint angle requires large amount of sEMG training data; however, due to the muscle fatigue, it is inapplicable to collect sEMG signals from each subject for a long time. To improve the angle estimation accuracy based on less data, this paper proposes a novel joint angle estimation method with empirical mode decomposition (EMD), and selects the elbow joint as the research joint. This estimation method utilizes extra intrinsic mode functions (IMF) feature extracted from IMF and conventional sEMG features including auto-regressive model (AR model) and root mean square (RMS). This research adopts support vector regression (SVR) to estimate elbow joint angle by IMF features and conventional sEMG features. The experimental results show that with the help of extra IMF features the accuracy of estimation result is substantially improved. To be specific, the average regression mean square error (MSE) is reduced from 0.2459 to 0.1712 and the average R2 is improved from 0.7546 to 0.8242. According to the experimental results, it can be concluded that the estimation method with extra IMF feature can estimate the elbow joint angle efficiently, which paves a solid foundation for the future control strategy of the assistant robots.