A Systematic Study on Electromyography-Based Hand Gesture Recognition for Assistive Robots Using Deep Learning and...

Area of Science:

• Robotics
• Biomedical Engineering
• Machine Learning

Background:

• Upper limb amputation significantly impacts daily living and autonomy.
• Robotic hand prostheses offer improved functionality but face challenges in user control and adaptation.
• Surface electromyography (sEMG) is a key sensing technology for myoelectric prostheses, yet signal noise and computational demands complicate its use.

Purpose of the Study:

• To develop and evaluate motion intention classifiers for transradial amputees using machine learning and deep learning models.
• To systematically analyze the impact of time-domain features and pre-processing parameters on classifier performance.
• To investigate the influence of amputation level and conditions on classifier generalization.

Main Methods:

• Implementation of various machine learning and deep learning models for EMG-based motion intention classification.
• Benchmarking classifier performance based on generalization across different classes.
• Systematic study on the effect of sliding window variations on feature-based and non-feature-based classification models.

Main Results:

• Ensemble learning and deep learning algorithms demonstrated superior performance compared to classical machine learning methods.
• The study identified correlations between sliding window trends and amputation levels.
• Classifier performance varied based on individual amputation history and conditions, highlighting the need for personalized approaches.

Conclusions:

• Machine learning and deep learning approaches show significant promise for enhancing the control of robotic hand prostheses.
• Understanding the impact of features, pre-processing, and amputation characteristics is crucial for developing effective assistive robotic systems.
• These findings are vital for advancing the development of intelligent, user-friendly prosthetic devices for amputees.