Enhancing Sleep Quality with Closed-Loop Autotuning of a Robotic Bed

Area of Science:

• Biomedical Engineering
• Control Systems
• Sleep Science

Background:

• Sleep is crucial for rehabilitation, enhancing motor learning, cognition, and well-being.
• Rocking beds offer non-invasive vestibular stimulation for improved sleep, but their effectiveness depends on precise control parameters.
• Manual tuning of rocking bed control parameters is time-consuming, subjective, and inefficient.

Purpose of the Study:

• To develop and evaluate an efficient method using Gaussian processes for automatic tuning of rocking bed PI (Proportional-Integral) control parameters.
• To compare the performance of Gaussian processes against random exploration for optimizing control parameters.
• To assess the effectiveness of the automated tuning method on a physical rocking bed system.

Main Methods:

• Simulated rocking bed kinematics to optimize control parameters based on desired acceleration profiles.
• Employed Gaussian processes for efficient parameter space exploration and optimization.
• Implemented and validated the Gaussian process method on a physical rocking bed, comparing results to manually tuned parameters.

Main Results:

• Gaussian processes achieved the control objective in a constant number of iterations, independent of search space size.
• Random exploration required a quadratically increasing number of iterations with search space size.
• Automated tuning discovered smoother motion parameters in under an hour, outperforming manual tuning, though motor noise increased.

Conclusions:

• Gaussian processes offer a significantly faster and more efficient method for optimizing rocking bed control parameters compared to manual tuning.
• The developed technique is transferable to real-world applications, improving rehabilitation through enhanced sleep.
• Future work should incorporate motor noise reduction into the optimization objective for sleep-related applications.