Title:
Online Adaptation of User State Estimation in a Powered Hip Exoskeleton using Machine Learning

dc.contributor.advisor Young, Aaron
dc.contributor.author Kunapuli, Pratik A.
dc.contributor.committeeMember Inan, Omer
dc.contributor.committeeMember Bloch, Matthieu
dc.contributor.department Electrical and Computer Engineering
dc.date.accessioned 2021-09-15T15:50:37Z
dc.date.available 2021-09-15T15:50:37Z
dc.date.created 2020-08
dc.date.issued 2020-07-28
dc.date.submitted August 2020
dc.date.updated 2021-09-15T15:50:38Z
dc.description.abstract Powered exoskeletons are a wearable technology that seeks to use robotic power to augment the capabilities of the human using them. Lower limb exoskeletons in particular are designed to aid in mobility tasks, rehabilitation, and other forms of human ability augmentation. Hip exoskeletons apply assistance at the hip joint, one of the largest contributors to positive mechanical work during walking tasks and aim to reduce the energetic expenditure of the human by supplementing robotic power to this joint. Recent exoskeleton technology has grown immensely, and as hip exoskeletons are being developed, so are their control systems. These control systems rely on understanding information about the user and correspondingly applying robotic power in the form of an assistance profile to aid the user in walking. This user state estimation is a crucial part of exoskeleton control, as it directly affects the ability of users to walk and the efficacy of the exoskeleton. In this thesis, a study of novel user state estimation techniques in various settings is presented. The first hypothesis investigated is that a machine learning-based gait phase estimation technique will reduce estimation error compared to time-based estimation techniques in treadmill walking with able-bodied subjects wearing the powered hip exoskeleton. The second hypothesis presented is that using biological sensor information with machine learning-based estimators for walking speed and inclination angle will reduce estimation error compared to machine learning-based estimators using only mechanical sensor data, specifically with elderly subjects with the powered hip exoskeleton. The final hypothesis presented is that online adaptation of deep learning-based models for gait phase and walking speed estimation will reduce estimation error compared to user-independent deep learning-based models for overground walking of able-bodied subjects wearing the powered hip exoskeleton. Human subject testing is presented to test all of these hypotheses. The results from the first experiment show that machine learning-based gait phase estimation outperforms time-based estimation in dynamic walking tasks such as acceleration or deceleration, and performs the same on steady-state walking. The results from the second experiment show that using biological sensor information reduces estimation error for walking speed and inclination angle compared to using only mechanical sensors for both able-bodied and elderly subjects. Finally, the preliminary results from the third experiment show that online adaptation of deep learning-based models reduces the estimation error for gait phase and walking speed estimation compared to user-independent models.
dc.description.degree M.S.
dc.format.mimetype application/pdf
dc.identifier.uri http://hdl.handle.net/1853/65123
dc.language.iso en_US
dc.publisher Georgia Institute of Technology
dc.subject Machine learning
dc.subject Supervised learning
dc.subject Online learning
dc.subject Hip exoskeleton
dc.subject State estimation
dc.subject Human augmentation
dc.title Online Adaptation of User State Estimation in a Powered Hip Exoskeleton using Machine Learning
dc.type Text
dc.type.genre Thesis
dspace.entity.type Publication
local.contributor.advisor Young, Aaron
local.contributor.corporatename School of Electrical and Computer Engineering
local.contributor.corporatename College of Engineering
relation.isAdvisorOfPublication 7f9a67d3-b78f-45e2-a5e9-d9a1650849db
relation.isOrgUnitOfPublication 5b7adef2-447c-4270-b9fc-846bd76f80f2
relation.isOrgUnitOfPublication 7c022d60-21d5-497c-b552-95e489a06569
thesis.degree.level Masters
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
KUNAPULI-THESIS-2020.pdf
Size:
4.08 MB
Format:
Adobe Portable Document Format
Description:
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
LICENSE.txt
Size:
3.87 KB
Format:
Plain Text
Description: