Using dynamic Bayesian optimization to induce desired effects in the presence of motor learning: a simulation study

Area of Science:

• Robotics and Human-Computer Interaction
• Computational Neuroscience and Motor Learning

Background:

• Human-in-the-loop optimization (HILO) is crucial for adaptive device control.
• Motor learning introduces dynamic changes in user output during optimization.
• Standard Bayesian optimization (BO) may struggle with time-varying user responses.

Purpose of the Study:

• To evaluate dynamic Bayesian optimization (DBO) for HILO of device control inputs.
• To assess DBO's suitability when user output changes due to motor learning.
• To compare DBO against standard BO in simulated adaptive control scenarios.

Main Methods:

• Simulations using time-dependent participant responses.
• Simulations incorporating state-space models of motor learning.
• Comparative analysis of DBO and standard BO convergence rates.

Main Results:

• DBO demonstrated superior convergence to optimal inputs and outputs compared to standard BO after a set number of iterations.
• DBO's performance advantage increased with sufficient iterations to differentiate learning from variability.
• Simulations confirmed DBO's efficacy in handling dynamic user responses.

Conclusions:

• Dynamic Bayesian optimization (DBO) is a suitable algorithm for HILO in adaptive device control.
• DBO offers improved performance over standard BO when dealing with motor learning and dynamic user states.
• Effective implementation of DBO requires sufficient iterations for accurate learning assessment.