Continuous Exoskeleton Adaptation for Activities of Daily Life

Studies have shown that Human-in-the-loop (HIL) optimization with Bayesian optimization(BO) of exoskeleton parameters has succeeded in finding the optimal control strategy, resulting in improved performance of the device and reducing the physical effort of the user. However, a subject’s gait has been shown to change due to various factors such as training or acclimatization, change in environment, time, etc. Such changes in gait would result in a change in the optimal control strategy. Repeated HIL trials using the current state-of-the-art are impractical due to the long trial time and continuous optimization has been known to fail due to the inability of the Gaussian Process to handle large variations within the observed dataset. I employed a new variation of the BO, Rank-Weighted Gaussian Process Ensemble (RGPE)- based BO that can fasten the optimization time and is more noise-resistant thereby allowing continuous optimization. Performance was first evaluated using numerical simulations using test functions that showed that the RGPE-based BO located a better optimal point (p=0.007).