Evaluation of a neural network-based control strategy for a cost-effective externally-powered prosthesis

Area of Science:

• Robotics
• Biomedical Engineering
• Artificial Intelligence

Background:

• Prosthetic hands often face challenges with inexpensive sensors and hardware, leading to nonlinearities that affect control.
• Effective sensory feedback and force control are crucial for prosthetic hand dexterity and object manipulation.
• Existing nonlinear models can be complex to implement and adjust for long-term prosthetic use.

Purpose of the Study:

• To develop and evaluate a cost-effective control strategy for prosthetic hands that compensates for sensor nonlinearities.
• To implement a neural network-based approach for robust force control and slippage detection.
• To enhance the adaptability and reliability of prosthetic hands in activities of daily living.

Main Methods:

• A neural network-based force control strategy was developed, incorporating sensory feedback for slippage detection.
• The strategy was implemented on a multi-digit, underactuated prosthetic hand.
• Experiments involved adjusting finger forces and evaluating object displacement under varying mass conditions and grasping configurations.

Main Results:

• The control strategy successfully detected dynamic changes in object mass.
• The prosthetic hand adjusted grasping force in real-time to prevent object slippage and drops.
• Performance was validated across different grasping configurations and object masses, simulating daily activities.

Conclusions:

• The proposed neural network control strategy effectively compensates for nonlinearities in low-cost prosthetic hardware.
• This approach offers an adaptable and easily adjustable solution for prosthetic hand control.
• The system demonstrates improved dexterity and reliability in object manipulation tasks relevant to daily living.