Gait Cycle-Inspired Learning Strategy for Continuous Prediction of Knee Joint Trajectory from sEMG

• DOI: 10.48550/arxiv.2307.13209
• Publication Date: 2023-01-01
• Authors: Xueming Fu; Hao Zheng; Luyan Liu; Wenjuan Zhong; Haowen Liu; Wenxuan Xiong; Yuyang Zhang; Yi-Feng Chen; Wei Dong; Mingjie Dong; Yefeng Zheng; Mingming Zhang
• Keywords: Motion Capture; Gait cycle

Predicting lower limb motion intent is vital for controlling exoskeleton robots and prosthetic limbs. Surface electromyography (sEMG) attracts increasing attention in recent years as it enables ahead-of-time prediction of intentions before actual movement. However, the estimation performance human joint trajectory remains a challenging problem due to inter- intra-subject variations. The former related physiological differences (such height weight) preferred walking patterns individuals, while latter mainly caused by irregular gait-irrelevant muscle activity. This paper proposes model integrating two gait cycle-inspired learning strategies mitigate challenge predicting knee trajectory. first strategy decouple angles into amplitudes exhibit low variability show high among individuals. By through separate network entities, manages capture both common personalized features. In second, principal activation masks are extracted from cycles prolonged walk. These used filter out components unrelated raw sEMG provide auxiliary guidance more gait-related Experimental results indicate that our could predict with average root mean square error (RMSE) 3.03(0.49) degrees 50ms ahead time. To knowledge this best relevant literatures has been reported, reduced RMSE at least 9.5%.