Ambilateral Activity Recognition and Continuous Adaptation with a Powered Knee-Ankle Prosthesis.

Area of Science:

• Biomedical Engineering
• Robotics
• Rehabilitation Engineering

Background:

• Powered prosthetic legs require accurate activity classification for real-world use.
• Current systems often involve complex, high-dimensional feature spaces and transition limitations.

Purpose of the Study:

• To develop a straightforward activity classification system for powered prosthetic legs.
• To reduce the complexity of state transitions for enhanced prosthetic leg functionality.

Main Methods:

• Delegated sit/stand and variable-incline walking to a mid-level controller, reducing classification to four states.
• Implemented heuristic rules using kinematic features, ground contact, inclination, and ultrasonic sensor data.
• Tested with two transfemoral amputee subjects using a powered knee-ankle prosthesis.

Main Results:

• Achieved 99.2% ambilateral transition accuracy in self-paced and fatiguing conditions.
• Demonstrated a 100% recovery rate with backup logic and user-cued resets.
• Enabled continuous adaptation to inclines without explicit classification.

Conclusions:

• Simple heuristic transition logic enhances the reliability of powered prosthetic legs.
• The system supports a broad range of daily activities, improving prosthetic leg viability.
• Outdoor multi-terrain demonstrations confirm real-world applicability.