Optimal phase-based gait guidance control on a lower-limb exoskeleton

Abstract

Lower-limb exoskeletons, often used for assistance and rehabilitation, operate using a variety of control strategies, such as phase-based gait guidance controllers. Phase is a configuration-dependent scalar representing the progression of gait. In purely robotic systems, the definition of phase is largely arbitrary so long as it satisfies monotonic and diffeomorphic conditions. However, in applications involving an active user, the volitional inputs of the operator can alter system behavior. This can negatively influence the progression of phase throughout locomotion, impacting reference gait generation and user comfort. This paper investigates defining phase in a way that considers the influence of the user. Specifically, a series of walking experiments is conducted where recorded data from each experiment is used to optimally define phase for the subsequent experiment. In each iteration, phase is optimally defined to minimize its sensitivity to natural variability in gait. To evaluate whether the proposed method is able to improve phase progression and system behavior, three walking experiments were performed using a bilateral mixing gait guidance controller on an Indego Explorer exoskeleton with an able-bodied subject. The controller combines two single-support gait tracking controllers plus gravity compensation where the gravity compensation is proven to be preserved during double-support. In the final experiment, the root-mean-square error between the experimental and theoretical progression of phase was 0.0136 where phase nominally progresses monotonically from 0 to 1. For comparison, a walking experiment using the horizontal hip coordinate as phase, normalized from 0 to 1, resulted in a root-mean-square error of 0.0995. The progression of phase improved by 86% from the non-optimized to iteratively optimized definition. As a result, the commanded references tended to better resemble the desired nominal gait pattern. In conjunction with the gait guidance controller with optimized phase definition, an adaption law adjusted the reference gait based on walking speed estimated from foot displacement and timing. Experimental results show how the adaptive controller is able to predict walking speed and adjust the commanded reference accordingly.