Autoencoder-based myoelectric controller for prosthetic hands.

Area of Science:

• Biomedical Engineering
• Robotics
• Neuroscience

Background:

• Linear dimensionality reduction techniques like Principal Component Analysis (PCA) have limitations in simplifying complex myoelectric control of high-dimensional prosthetic hands.
• Nonlinear methods, specifically Autoencoders, demonstrate superior performance in compressing and reconstructing intricate hand kinematics data.

Purpose of the Study:

• To introduce and evaluate a novel Autoencoder-based controller for intuitive myoelectric control of a high-dimensional virtual hand.
• To assess the controller's efficacy in enabling users to manipulate a 17-dimensional virtual hand through a reduced 2-dimensional space.

Main Methods:

• Development of a novel Autoencoder-based controller for mapping high-dimensional hand data to a low-dimensional control space.
• Validation experiment involving four unimpaired participants controlling a virtual hand using the developed myoelectric interface.

Main Results:

• Participants significantly reduced the time required to match target gestures with the virtual hand.
• Improved path efficiency was observed in three out of four participants, indicating enhanced control accuracy.
• The Autoencoder controller demonstrated potential for higher accuracy compared to traditional PCA methods in prosthetic hand manipulation.

Conclusions:

• The Autoencoder-based controller shows significant promise for enhancing myoelectric control of high-dimensional prosthetic hands.
• Further research is needed to optimize learning strategies for Autoencoder controllers in real-world prosthetic applications.