Assisting and Evaluating Upper Extremity Movements with Wearables Systems

Abstract

Upper extremity impairments from injuries and illnesses can result in physical disabilities, medical expenses, and productivity loss in labor intensive occupations. In recent years, wearable sensing and assistive robots have risen as promising tools for general kinematics tracking, injury prevention, and rehabilitation. These wearable options offer a new realm of solutions avoiding issues such as limited clinician availability and expensive infrastructural changes in the workplace. For wearable robotics specifically, soft robots also have the potential for extra comfort, minimal restrictions and harm from joint misalignments as compared to their rigid counterparts. This thesis seeks to bridge the gaps in current wearable assistive and rehabilitative technologies through introducing control and evaluation strategies for a soft inflatable shoulder robot to assist with industrial overhead work and presenting a motor function estimation algorithm with wearable inertial sensors. The thesis begins with the development of a kinematics-based state machine controller for a soft inflatable shoulder wearable robot for industrial work. The controller is aimed to provide assistance to the shoulder quickly and accurately when needed during overhead industrial work. The state machine was designed to classify user intent using shoulder and trunk kinematics estimated with body-worn inertial measurement units. Through human subject experiments, we evaluated the controller’s intent classification accuracy and response times, by using the users’ reactions to cues as their ground truth intentions. On average, we found that the kinematics controller had 99% classification accuracy, and responded 0.8 seconds after the users reacted to the cue to begin work and 0.5 seconds after the users reacted to a cue to stop the task. In addition, we implemented an EMG-based controller for comparison, with state transitions determined by EMG-based thresholds instead of kinematics. Compared to the EMG controller, the kinematics controller required similar time to detect the users’ intentions to stop overhead work but an additional 0.17 seconds on average for detecting users’ intentions to begin. We also implemented an online adaptive tuning algorithm for the kinematics controller to speed up response time while ensuring accuracy during offset transitions. Post assessment of the controller, we evaluated the robot’s performance with this controller with a portable system. The updated portable robot is worn like a shirt with integrated textile pneumatic actuators, inertial measurement units (IMUs), and a portable actuation unit. It can provide up to 6.6 newton-meters of torque to support the shoulder and cycle assistance on and off at six times per minute. From human participant evaluations during simulated industrial tasks, the robot reduced agonist muscle activities (anterior, middle, and posterior deltoids and biceps brachii) by up to 40% with slight changes in joint angles of less than 7% range of motion while not increasing antagonistic muscle activity (latissimus dorsi) in the current sample size. Comparison of controller parameters further highlighted that higher assistance magnitude and earlier assistance timing resulted in statistically significant muscle activity reductions. During a task circuit with dynamic transitions among the tasks, the kinematics-based controller of the robot showed more robustness to misinflations (96% true negative rate and 91% true positive rate), indicating minimal disturbances to the user when assistance was not required. A preliminary evaluation of a pressure modulation profile also highlighted a trade-off between user perception and hardware demands. Finally, five automotive factory workers used the robot in a pilot manufacturing area and provided feedback. Building upon the robot controller and biomechanics evaluation, we aimed to improve the adaptability of the robot through developing a dynamic controller with the purpose of providing assistance during quasi-static motion and allowing for minimal restrictions during fast movements. Furthermore, we designed an on-body measurement tool to enable direct measurements of elevation torque during unconstrained tasks for the first time. With the development of these tools, we evaluated the effect of various support types on user biomechanics and perception. We found that the improved robot can provide 7.3 newton-meters of elevation torque with controller gains being effective in modulating the absolute torque delivered. The robot reduced muscle activity in the agonist muscles without affecting the antagonist ones during controlled and functional activities. Comparing the amount of support delivered to the biomechanical outcomes, we observed a pseudo-linear relationship between the percentage of support to the user and the percentage of muscle activation reduction. Finally, the users were able to perceive the differences among the assistance types, in terms of perceived robot support, transparency and precision, with a main preference for sufficient support during loaded conditions. In addition to control and evaluation methods for assistive robots, this thesis aims to contribute to wearable sensing algorithms. In a final demonstration of an application of wearable sensing, we present an estimation algorithm for estimating upper extremity Fugl-Meyer Assessment (FMA-UE) scores, a well-accepted post-stroke recovery metric for motor function assessments. To address the need for simplified automated FMA-UE assessments, we present an estimator which can make score predictions for a subset of the full assessment using data from inertial measurement units placed on the hands, arms and the trunk from a minimal number of volitional reaching motions representative of functional daily activities. To develop the estimator, we collected a dataset of eleven stroke participants performing a key subset of FMA-UE motions, and three reaching motions. The FMA-UE of each participant was assessed by an occupational therapist providing the labeled score for the training data. The estimator was trained on windowed data during FMA-UE motions and was able to make score estimates from reaching motions. Through leave-one-subject-out cross validation, the estimator achieved a normalized RMSE of 7%, which is comparable to or below the established minimal clinically important difference and minimal detectable change of FMA-UE for individuals with chronic stroke. Comparison experiments of various model designs also revealed the importance of trunk-based features inspired by compensation strategies common post stroke and features extracted from the hand sensor. Together, this thesis contributes to control and evaluation methods for wearable systems for both healthy and clinical populations targeting the use cases of injury reduction and rehabilitation progress tracking respectively.