Prosthetic Control by Learning: A Multi-Agent Cooperative Game Framework.

Area of Science:

• Robotics and Human-Computer Interaction
• Biomedical Engineering and Rehabilitation Technology

Background:

• Traditional lower-limb powered prostheses use rigid, phase-based controllers, restricting adaptability to varied human movements.
• Effective prosthetic control requires seamless coordination and reciprocal adaptation between the human user and the prosthetic device.

Purpose of the Study:

• To reframe powered prosthetic control as a cooperative multi-agent game.
• To develop a model-free reinforcement learning framework for adaptive prosthetic control.

Main Methods:

• A cooperative multi-agent reinforcement learning framework was developed.
• The model-free approach enables the prosthesis to learn adaptive policies through interaction.

Main Results:

• Simulated results show the framework generates human-like motions across walking and complex tasks.
• The cooperative policy learning approach demonstrated adaptability to diverse human movement patterns.

Conclusions:

• Viewing prosthetic control as a multi-agent cooperative game enhances adaptability and intuitiveness.
• This approach paves the way for advanced prosthetic systems that naturally synchronize with user intentions.