Multi-Context, User-Independent, Real-Time Intent Recognition for Powered Lower-Limb Prostheses

Bhakta, Krishan; Maldonado-Contreras, Jairo; Camargo, Jonathan; Zhou, Sixu; Compton, William; Herrin, Kinsey R.; Young, Aaron

Title:

Multi-Context, User-Independent, Real-Time Intent Recognition for Powered Lower-Limb Prostheses

dc.contributor.author	Bhakta, Krishan
dc.contributor.author	Maldonado-Contreras, Jairo
dc.contributor.author	Camargo, Jonathan
dc.contributor.author	Zhou, Sixu
dc.contributor.author	Compton, William
dc.contributor.author	Herrin, Kinsey R.
dc.contributor.author	Young, Aaron
dc.contributor.corporatename	Georgia Institute of Technology. George W. Woodruff School of Mechanical Engineering	en_US
dc.contributor.corporatename	Georgia Institute of Technology. Institute for Robotics and Intelligent Machines	en_US
dc.contributor.corporatename	Georgia Institute of Technology. Exoskeleton and Prosthetic Intelligen Controls Lab	en_US
dc.date.accessioned	2023-03-13T16:52:29Z
dc.date.available	2023-03-13T16:52:29Z
dc.date.issued	2023
dc.description	Supplementary material for manuscript to be published in Science Robotics	en_US
dc.description.abstract	Community ambulation is a critical component in maintaining a healthy lifestyle but has numerous task demands that can be challenging for individuals with limb loss. In wearable robotics, specifically powered prostheses, a need exists to provide intuitive and seamless assistance to the user. We developed a user-independent and multi-context, intent recognition system that was deployed in real-time to an open-source knee and ankle powered prosthesis (OSL). The intent recognition system predicted user intent and environment attributes using embedded sensing and control. Eleven individuals with transfemoral amputation were recruited for this study, in which 7 individuals were used for real-time validation. Here, we proposed a hierarchical control framework in which the intelligent prosthesis would first predict locomotion mode and subsequently estimate an environmental variable (i.e., walking speed or slope). Two main conclusions were found: 1) the user-independent (IND) performance across mode, speed, and slope was not statistically different from user-dependent (DEP) models in real-time, even though the offline performance of the IND system was worse 2) IND walking speed estimates showed ~0.09 m/s average error and slope estimates showed ~0.95 deg average error, which provided acceptable performance for modulating ankle and knee assistance across multi-context scenarios. Our study suggests that intelligent controllers can generalize to individuals and can perform well in real-time. In addition, we made our training dataset and the developed machine learning models publicly available to an open-source repository. This approach provides novel prosthesis users with autonomous and task-dependent functionality across real-world walking tasks.	en_US
dc.description.sponsorship	Prosthetics Outcomes Research Award No. W81XWH-17-1-0031	en_US
dc.identifier.uri	http://hdl.handle.net/1853/70300
dc.publisher	Georgia Institute of Technology	en_US
dc.subject	Prostheses	en_US
dc.subject	Machine learning	en_US
dc.subject	Intent recognition	en_US
dc.subject	Wearable robotics	en_US
dc.subject	Robotic control	en_US
dc.title	Multi-Context, User-Independent, Real-Time Intent Recognition for Powered Lower-Limb Prostheses	en_US
dc.type	Dataset	en_US
dspace.entity.type	Publication
local.contributor.author	Young, Aaron
local.contributor.corporatename	George W. Woodruff School of Mechanical Engineering
local.contributor.corporatename	College of Engineering
relation.isAuthorOfPublication	7f9a67d3-b78f-45e2-a5e9-d9a1650849db
relation.isOrgUnitOfPublication	c01ff908-c25f-439b-bf10-a074ed886bb7
relation.isOrgUnitOfPublication	7c022d60-21d5-497c-b552-95e489a06569