Human-in-the-loop Bayesian optimization of wearable device parameters.

Area of Science:

• Robotics
• Biomechanics
• Control Systems

Background:

• Exoskeletons and powered prosthetics require advanced control strategies.
• Human-in-the-loop optimization uses real-time physiological data for tuning.
• Metabolic cost is a key metric but challenging to measure accurately.

Purpose of the Study:

• Evaluate Bayesian optimization for efficient control parameter tuning.
• Minimize metabolic cost in human walking by optimizing step frequency.
• Compare Bayesian optimization against gradient descent methods.

Main Methods:

• Utilized Bayesian optimization, a sample-efficient and noise-tolerant global optimization technique.
• Focused on optimizing walking step frequencies in human subjects to reduce metabolic cost.
• Compared performance against a gradient descent approach for parameter tuning.

Main Results:

• Bayesian optimization achieved near-optimal step frequency identification faster than gradient descent (12 minutes).
• Demonstrated reduced inter-subject variability in convergence time (± 2 minutes).
• Observed significantly lower overall energy expenditure with Bayesian optimization.

Conclusions:

• Bayesian optimization is a highly effective method for rapidly tuning exoskeleton control parameters.
• This approach overcomes the limitations of traditional methods in metabolic cost optimization.
• Enables more efficient and individualized control strategies for walking assistance devices.