Day-To-Day User Adaptation to a Robotic Ankle-Foot Prosthesis.

Area of Science:

• Biomechanics
• Rehabilitation Engineering
• Human-Computer Interaction

Background:

• Assistive devices like robotic ankle-foot prostheses (AFP) require user adaptation for effective locomotion.
• Personalization frameworks like Bayesian Optimization (BO) and Human-in-the-Loop (HIL) are used but may not capture dynamic adaptation.
• Understanding day-to-day changes in amputee gait is crucial for optimizing prosthetic performance.

Purpose of the Study:

• To investigate short-term, day-to-day adaptation in lower limb amputees using robotic ankle-foot prostheses.
• To model socket pressure distributions and identify shifts in optimal prosthetic stiffness.
• To evaluate the impact of adaptation on personalization strategies for assistive devices.

Main Methods:

• Secondary analysis of data from three unilateral transtibial amputees over a four-day protocol.
• Utilized Gaussian processes to model socket pressure distributions under varying AFP stiffness.
• Employed Kullback-Leibler (KL) divergence and mean absolute error (MAE) to quantify adaptation.

Main Results:

• Subject-specific shifts in optimal AFP stiffness were observed between discrete parameter sweeps and HIL optimization days.
• Significant changes in the socket pressure landscape indicated adaptation in gait and limb dynamics.
• Average KL divergence of 1.15 and MAE loss of 0.59 quantified the extent of adaptation.

Conclusions:

• Human adaptation to robotic ankle-foot prostheses occurs even over short time scales (days).
• Personalization methods should incorporate dynamic adaptation, not just static performance snapshots.
• Future research should focus on adaptive control strategies that account for user's evolving gait patterns.