Human-in-the-loop Bayesian optimization of a tethered soft exosuit for assisting hip extension

Abstract

Advances in wearable devices have led to an increased need to develop sophisticated and individualized control strategies. To address this problem, several researchers have begun exploring human-in-the-loop optimization methods that automatically adjust control parameters in a wearable device using real-time physiological measurements. A common physiological measurement, metabolic cost, poses significant experimental challenges due to its long measurement times and low signal-to-noise ratio. This study addresses the challenges by using Bayesian optimization—an algorithm well-suited to optimizing noisy performance signals with very limited data—to perform control adaptation online. When applied to a soft exosuit designed to provide hip assistance, optimized control parameters were found in 24 min with a significant reduction in metabolic cost. These results suggest that this method could have a practical impact on improving the performance of wearable robotic devices.