Repeatability of grasp recognition for robotic hand prosthesis control based on sEMG data.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Surface electromyography (sEMG) is a key technology for controlling prosthetic hands.
• Current sEMG control systems often lack robustness, limiting functional use for amputees.
• Limited research exists on the repeatability of sEMG classification for hand grasps.

Purpose of the Study:

• To investigate the repeatability of sEMG data for hand grasp classification.
• To create and release a publicly available database of sEMG repeatability experiments.
• To assess the impact of different data acquisitions and subjects on classification accuracy.

Main Methods:

• Recorded sEMG data from 10 subjects performing 7 grasps, repeated multiple times over 5 days.
• Utilized Mean Absolute Value and Waveform Length features with a Random Forest classifier.
• Compared classification accuracy between training/testing on the same vs. different data acquisitions and across subjects.

Main Results:

• Training and testing on different acquisitions resulted in a 27.03% average decrease in accuracy compared to using the same acquisition.
• Classification accuracy achieved when training on previous acquisitions suggests their utility for algorithm training.
• Significant inter-subject variability was observed, impacting repeatability and classification accuracy.

Conclusions:

• sEMG data exhibits variability across different acquisition times and subjects, impacting prosthetic control robustness.
• The findings suggest that previously recorded sEMG data can be leveraged to train more robust control algorithms.
• The released repeatability database will facilitate further research into improving sEMG-based prosthetic hand control systems.