Utilization of Classification Learning Algorithms for Upper-Body Non-Cyclic Motion Prediction.

Area of Science:

• Biomedical Engineering
• Robotics
• Machine Learning

Background:

• Exoskeletons often suffer from low fluency due to control system delays.
• This impacts energetic efficiency and user comfort.
• Predictive control is needed to overcome these limitations.

Purpose of the Study:

• To explore classification algorithms for predicting non-cyclic upper-body motions.
• To improve the fluency and reduce the energetic inefficiency of exoskeletons.
• To investigate the use of surface electromyography (sEMG) signals for motion prediction.

Main Methods:

• Utilized k-nearest neighbor (KNN) and deep learning models.
• Processed sEMG signals from elbow-related muscles to detect activation changes.
• Classified motion characteristics based on sEMG signal slopes and continuous categorization.

Main Results:

• Both KNN and deep learning models predicted voluntary non-cyclic motions beyond electromechanical delay.
• The deep learning model achieved >90% certainty in predicting motion characteristics before muscle activation.
• Classification algorithms demonstrated potential for predicting upper-body motions.

Conclusions:

• Machine learning classification algorithms can predict upper-body non-cyclic motions.
• These predictions can potentially enhance machine interfacing fluency in exoskeletons.
• Further research into regression models and wearable applications is warranted.