Performance of a Novel Muscle Synergy Approach for Continuous Motion Estimation on Untrained Motion

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Continuous motion estimation (CME) models based on surface electromyography (sEMG) face reduced prediction accuracy on untrained motions.
• Existing methods struggle when user movements deviate from the model's training data.

Purpose of the Study:

• To develop a novel CME method that improves prediction accuracy for untrained motion tasks.
• To introduce muscle synergy as a robust feature for enhanced motion estimation.

Main Methods:

• Introduced deep non-smooth NMF (Deep-nsNMF) for efficient synergy extraction.
• Developed a redundancy classification (RC) algorithm to optimize synergy identification, improving upon the redundancy segmentation (RS) algorithm.
• Utilized a NARX neural network for regression modeling.

Main Results:

• The proposed CME method using muscle synergy outperformed traditional time-domain features on untrained motions.
• Achieved a highest accuracy of 99.7% using Deep-nsNMF with RS.
• Deep-nsNMF with RC demonstrated comparable accuracy and superior stability across subjects and motions.

Conclusions:

• Muscle synergy-based CME shows significant potential for improving prediction accuracy in complex, real-world human-robot interaction scenarios.
• The developed Deep-nsNMF and RC algorithms offer a robust approach to feature extraction for motion estimation.
• Further research is needed to address the correlation between prediction error and actual movement values.