Time History Gaussian Process Regression for Modeling the Dynamics of Ankle Exoskeleton User Comfort

Exoskeletons need to be comfortable. Further, users adapt over time, so the perception of comfort may change. To determine if short‐term time history directly influences comfort over a single experimental session, healthy subjects walked in a pair of ankle exoskeletons for twenty trials. Comfort can be modeled using Gaussian process regression (GPR) models with input features related to metabolic cost and joint angles for the current trial. This quasistatic model implicitly captures changes in the user's gait that influence comfort, but time and previous experiences are not explicitly captured. To determine if previous experience explicitly impacts comfort, 19 dynamic GPR models for each subject that included trial number and/or up to three previous trial's worth of input features and/or comfort were created. No previous trials’ input features were predictive of comfort. Including the previous trial's comfort score marginally improved model accuracy compared to the quasistatic model although this may be an artifact of how comfort was measured. Thus, on the timescale and protocol studied here, it appears that short‐term trial history provides limited added predictive value.