Answers are grounded in this paper and cite their sources.
Robotic exoskeletons require precise assistance tailored to specific locomotion modes for optimal performance. Thus, accurate identification of locomotion modes and transitions is essential to assure appropriate power deployment. We evaluated the accuracy of machine learning models based on different feature sets (EMG, IMU, EMG+IMU) and different classifier structures (single-stage and cascade structures) in predicting locomotion modes, by using gait data (n = 14) from inertial measurement units (IMUs) and surface electromyography (EMG). Four gait modalities (level-ground walking, stair ascending, stair descending, and standing still) and 10 gait transition types were classified. Machine learning models were evaluated using feature sets of IMU data, EMG data or IMU+EMG. Moreover, we compared the performance of single-stage classifier to two cascade classifiers (CCT1 and CCT2). The results show that classifying gait modalities using IMUs presents greater performance (84.58%) when compared to EMG (48.36%) and EMG+IMU (83.53%). Moreover, increasing the level of classifier structures improved the accuracy of transition classification from 58.31% to 84.40% (26%) while decreasing the inference time by as much as 60.5%. In conclusion, cascade structure contribute to the classification of transition gaits and IMU is much more useful than EMG in locomotion modes classification.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2025) studied this question.