Date of Award

2022

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Graduate Group

Mechanical Engineering & Applied Mechanics

First Advisor

Michelle J. Johnson

Abstract

With the shortage of rehabilitation clinicians in rural areas and elsewhere, remote rehabilitation (telerehab) fills an important gap in access to rehabilitation. We have developed a first of its kind social robot augmented telepresence (SRAT) system --- Flo --- which consists of a humanoid robot mounted onto a mobile telepresence base, with the goal of improving the quality of telerehab. The humanoid has arms, a torso, and a face to play games with and guide patients under the supervision of a remote clinician.

To understand the usability of this system, we conducted a survey of hundreds of rehab clinicians. We found that therapists in the United States believe Flo would improve communication, patient motivation, and patient compliance, compared to traditional telepresence for rehab. Therapists highlighted the importance of high-quality video to enable telerehab with their patients and were positive about the usefulness of features which make up the Flo system for enabling telerehab.

To compare telepresence interactions with vs without the social robot, we conducted controlled studies, the first to rigorously compare SRAT to classical telepresence (CT). We found that for many SRAT is more enjoyable than and preferred over CT. The results varied by age, motor function, and cognitive function, a novel result.

To understand how therapists and patients respond to and use SRAT in the wild over long-term use, we deployed Flo at an elder care facility. Therapists used Flo with their own patients however they deemed best. They developed new ways to use the system and highlighted challenges they faced.

To ease the load of performing assessments via telepresence, I constructed a pipeline to predict the motor function of patients using RGBD video of them doing activities via telepresence. The pipeline extracts poses from the video, calculates kinematic features and reachable workspace, and predicts level of impairment using a random forest of decision trees.

Finally, I have aggregated our findings over all these studies and provide a path forward to continue the evolution of SRAT.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS