A Classical Machine Learning Approach For Emg-Based Lower Limb Intention Detection For Human-Robot Interaction Systems

Abstract

Surface Electromyography (sEMG)-based intention-detection systems of lower limb can intelligently augment human- robot interaction (HRI) systems to detect subject’s walking direction prior-to or during walking. Ten Subject-Exclusive (Subj-Ex) and Generalized (Gen) Classical Machine Learning (C-ML)-based models are employed to detect direction intentions and evaluate inter-subject robustness in one knee/foot- gesture and three walking-related scenarios. In each, sEMG signals are collected from eight muscles of nine subjects during at least nine distinct gestures/activities. Linear Discriminant Analysis (LDA) and Random Forest (RF) classifiers, applied to the Time-Domain (TD) feature set (of the four input sets), provided the best accuracy. Subj-Ex approach achieves the highest prediction accuracy, facing occasional competition from the Gen approach. In knee/foot gesture scenario, LDA reaches an accuracy of 91.67%, signifying its applicability to robotic-assisted walking, prosthetics, and orthotics. The overall prediction accuracy among walking- related scenarios, though not as remarkably high as in the knee/foot gesture recognition scenario, can reach up to 75%.