Application of Interacting Models to Estimate the Gait Speed of an Exoskeleton User

Abstract

This paper outlines steps toward a framework for model-based user intent detection to enable fluent human-robot interaction in assistive exoskeletons. An interacting multi-model (IMM) estimation scheme is presented to address state estimation for lower-extremity exoskeletons and to handle their hybrid dynamics. The proposed IMM scheme includes new approaches that enable it to estimate states of hybrid systems with dynamics that are unique to each phase. Traditional IMMs only consider the probabilistic likelihood of being in each phase, while the implementation in this work has been modified to consider physical likelihood as well. The IMM compares exoskeleton sensor readings to multiple candidate gaits from a template model of walking. Candidate gaits are generated using a numerical optimization procedure applied to a Bipedal Spring-Loaded Inverted Pendulum (B-SLIP) model. The framework was tested with sensor data acquired from walking trials in an Ekso GT exoskeleton, and was used to estimate gait phase and center of mass velocity. It is shown that the standard IMM filtering approach results in incorrect estimates of gait phase, while the proposed addition to the IMM estimator using physical likelihood improves the estimates. Results with human subject data further show the ability to estimate gait phase and speed in experimental settings.