User-Informed Exoskeleton Design Metrics and Multi-Modal Exoskeleton Control Strategies

Abstract

Powered exoskeletons have long captivated both the public consciousness and scientific interest with their potential to transform the mobility of the public. In particular, augmentative lower-limb exoskeletons have had notable success in making walking more efficient; however, this has not yet resulted in these technologies achieving widespread adoption. One primary obstacle to this goal is that the ideal metrics by which augmentative exoskeletons should be designed and evaluated to enable dissemination are not yet clear. Without a better understanding of how to develop these devices in such a way that they are voluntarily adopted, exoskeletons will remain confined to the laboratory. Another major challenge is that, while existing exoskeleton controllers are largely designed for a controlled laboratory setting, widespread adoption of these technologies will require controllers that can adapt their assistance in response to the transitory, ever-changing terrains of the real world.

This dissertation aims to provide solutions to these two major challenges. The works contained in this dissertation can be categorized into 1) investigations into the metrics by which augmentative exoskeletons are developed and judged, and 2) the development of a controller that continuously learns the wearer's gait in real-time and applies appropriate assistance. These two categories are comprised of four primary projects.

In the first project, I calculated the Just Noticeable Difference (JND) of changes in metabolic rate, which is the ``gold standard'' physiological metric by which augmentative exoskeletons are designed and evaluated [117]. This study revealed that the average JND was nearly twice as high as the average energetic benefit from modern devices, which calls into question the primacy of a metric that yields imperceptibly beneficial devices. In the second project, I characterized the economic value of modern bilateral ankle exoskeletons. The goal of this work was to investigate economic value as a potential alternative metric that more effectively reflects user willingness to use an exoskeleton. Designing exoskeletons according to this metric may then result in devices that are likelier to translate outside the laboratory by virtue of providing economic value to potential users. Beyond this result, this work also revealed that the average net benefit, in dollars, of the use of the bilateral exoskeletons tested in this study is near-zero. In the third project, I investigated the fundamental problem of estimating the user's gait state---the parametrization of a person's locomotion---continuously throughout a variety of walking tasks using the Bayesian framework [118]. This project revealed that an Extended Kalman Filter (EKF) estimator was capable of tracking the gait state as it evolved throughout different speeds and inclines. Finally, in the fourth project, I built upon this work to develop a practical exoskeleton controller that learns the gait state in real-time and adapts biomimetic torque assistance accordingly [116]. This controller was validated both on controlled treadmill trials to establish a baseline, and on a series of ``stress-tests'' in which a participant walked in extreme outdoor terrain settings. Taken together, these projects provide a foundation for translating exoskeletons out of the laboratory and into the real world.