Robust neural decoding for dexterous control of robotic hand kinematics

Area of Science:

• Biomedical Engineering
• Neuroscience
• Robotics

Background:

• Loss of hand dexterity due to neuromuscular injuries limits daily tasks.
• Current assistive robotic hands lack dexterous, real-time control of multiple degrees of freedom.
• Advanced prosthetic control remains a significant challenge.

Purpose of the Study:

• To develop an efficient and robust neural decoding approach for real-time prosthetic hand control.
• To enable continuous decoding of intended finger dynamic movements.
• To improve dexterity and functionality of assistive robotic hands.

Main Methods:

• Acquired high-density electromyogram (HD-EMG) signals during finger movements.
• Implemented a deep learning neural network to map HD-EMG features to neural-drive signals.
• Used predicted neural-drive signals for real-time control of prosthetic finger kinematics.

Main Results:

• The neural-drive decoder achieved significantly lower prediction errors for joint angles compared to other methods.
• Decoder performance remained stable and robust to EMG signal variations.
• Demonstrated superior finger separation with minimal error in unintended finger predictions.

Conclusions:

• The developed neural decoding technique provides a novel and efficient neural-machine interface.
• High-accuracy prediction of robotic finger kinematics enables dexterous control.
• This approach significantly advances the capabilities of assistive robotic hands.