Recovery strategy identification throughout swing phase using kinematic data from the tripped leg

Area of Science:

• Biomedical Engineering
• Robotics
• Biomechanics

Background:

• Falls pose a significant risk for individuals with lower limb amputations.
• Current powered prosthetics lack active balance recovery mechanisms to prevent falls after perturbations.

Purpose of the Study:

• To investigate the feasibility of a real-time pattern recognition system for identifying trip recovery strategies.
• To classify different recovery strategies like walking, elevating, and lowering the leg during a trip.

Main Methods:

• Able-bodied subjects were repeatedly tripped during the swing phase of gait.
• Linear discriminant analysis was used to classify kinematic data from the tripped leg.
• A two-stage classifier architecture (trip detection then strategy identification) was compared to a single-stage approach.

Main Results:

• The two-stage classifier achieved a median accuracy of 92% (range 88%-96%), outperforming the single-stage classifier (88% median accuracy).
• Classification accuracy plateaued within 100 ms after the trip, with most errors occurring immediately post-trip.
• Optimal window length varied by classification stage, while window increment did not impact accuracy.

Conclusions:

• Real-time pattern recognition algorithms can accurately identify trip recovery strategies.
• Sensor data from robotic assistive devices can trigger active balance restoration strategies following trips.
• This technology holds promise for enhancing the safety and stability of prosthetic devices.