Validation of Dynamic Bayesian Optimization for Human-in-the-Loop Optimization of Exoskeleton Control at User-Driven...

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Human-in-the-loop optimization (HILO) is crucial for personalized performance augmentation.
• Existing HILO methods may not adequately address time-varying human responses, limiting their effectiveness in rehabilitation.

Purpose of the Study:

• To evaluate dynamic Bayesian optimization (DBO), a modified Bayesian optimization (BO) algorithm, for identifying subject-specific optimal walking assistance parameters.
• To assess DBO's ability to account for non-stationarity in human responses during gait assistance.

Main Methods:

• Sixteen healthy participants received hip exoskeleton assistance.
• Torque parameters were optimized using HILO with either DBO or BO.
• Validation iterations were used to compare optimizer performance over time.

Main Results:

• Both DBO and BO significantly increased walking speed compared to baseline.
• DBO demonstrated superior efficacy, modeling accuracy, and personalization compared to BO.
• DBO-driven assistance led to greater improvements in walking speed and personalization than BO.

Conclusions:

• DBO is an improved HILO approach over standard BO for gait rehabilitation applications.
• DBO's strength lies in its ability to adapt to non-stationary human responses.
• This adaptive capability enhances personalized assistive solutions for gait improvement.