Individualizing and Monitoring Ankle Assistance for People Poststroke with Soft Exoskeletons and Exosuits Using Limited Data

Abstract

Exoskeletons and exosuits are wearable robots that can augment the performance of gait in unimpaired users and restore mobility in in people with gait impairments. Understanding how people who use exoskeletons and exosuits utilize these devices is critical to improving their design and control. Advances in our understanding of how humans interact with wearable robots on a physiological and biomechanical level have led to the design of rigid and soft exoskeletons and exosuits that can specifically target a single joint or a single activity. This thesis focuses on the control of exoskeletons and exosuits for assisting people poststroke. Nearly 80% of people poststroke experience gait impairments, which can lead to low community involvement, an increased risk of trips and falls, and a reduced quality of life. These impairments are particularly apparent at the ankle joint, where people poststroke present with reduced ankle power and torque generation, as well dorsiflexor weakness. Therefore, increasing the quality of gait is a common rehabilitation goal of people poststroke. Exoskeletons and exosuits assisting people poststroke, particularly those assisting the ankle, have been shown to improve the quality of gait in research settings. However, their translation into clinical practice has been slow. A key obstacle to clinical adoption is the limitation of quality data available to inform the control of these devices. Gait in people poststroke is highly heterogeneous and unstable; while there is a large body of literature on sensing gait in unimpaired populations, these approaches may not perform well in people poststroke. We first present a technique for monitoring gait in people poststroke using inertial measurement units (IMUs) placed on either feet. We use instantaneous information from both IMUs to correct for bias and drift in acceleration data. Using this corrected data, we calculate foot position in three dimensional space, providing clinically relevant gait metrics such as walking speed, stride length, foot clearance, and circumduction. We validate this technique in the lab against optical motion capture data. Further, we demonstrate the this technique can be used to monitor exosuit-induced changes in gait as a person walks in the community. A second obstacle to the clinical adoption of exoskeletons and exosuits is the need for individualized control. It is typical for exoskeleton and exosuit users poststroke to fall into categories of "responders" and "nonresponders"; modes of assistance that work well for one user may not work at all for another. Here, again, there are limitations in the data available to inform individualization. Lengthy experiment times can often preclude the use of individualization techniques that work well in unimpaired populations. To enable individualization, we first present a method of optimizing exosuit assistance profiles offline using previously-recorded data. We validate that this technique can be used to selectively apply either positive or negative augmentation power to the ankle. We then demonstrate the people poststroke utilize positive and negative augmentation power differently. Third, we present a method that streamlines exoskeleton and exosuit design by lowering the number of IMUs required to segment the gait cycle from two to one. We use a gradient-boosting strategy to train a machine learning model that can predict when the paretic foot is on the ground using information solely from an IMU on the paretic foot. More, we develop an approach to update this machine learning model with recent data as a user walks in new environments, individualizing gait segmentation to each user's gait patterns. We validate this approach in one person poststroke walking both on the treadmill and overground. Together, this thesis presents a series of developments that can be used to (i) inform clinicians and researchers if and how exoskeleton-based interventions change gait in the community, (ii) optimize exosuit assistance to examine the biomechanical and physiological impact of exosuit assistance, and (iii) support community and clinical adoption by simplifying exoskeleton design and adapting gait detection to new environments and interventions.