Toward Biosignals-Free Autonomous Prosthetic Hand Control via Imitation Learning

Area of Science:

• Robotics
• Biomedical Engineering
• Artificial Intelligence

Background:

• Limb loss significantly impacts millions worldwide, necessitating advanced prosthetic solutions.
• Traditional prosthetic control methods (surface electromyography, semi-autonomous) impose high physical and mental demands on users.
• Existing systems often require conscious effort for each control action, limiting natural interaction.

Purpose of the Study:

• To develop a fully autonomous prosthetic hand control system.
• To enable automatic grasping and releasing of diverse objects using only a camera.
• To reduce the cognitive load and physical strain associated with prosthetic use.

Main Methods:

• Implemented a vision-based autonomous control system for a prosthetic hand.
• Utilized a teleoperation system to collect human demonstration data.
• Employed imitation learning to train the prosthetic hand control model on human actions.

Main Results:

• The autonomous system demonstrated high success rates in grasping and releasing various objects.
• The imitation learning model generalized effectively to new users and unseen objects.
• The system automatically adjusts grip force based on object properties and environmental context.

Conclusions:

• A fully autonomous prosthetic hand control system is feasible and effective.
• Imitation learning provides a robust method for training prosthetic control models.
• This technology offers a more intuitive and less demanding prosthetic experience for users with limb loss.