Toward an enhanced understanding of rejection in pattern recognition-based myoelectric control
| dc.contributor.advisor | Scheme, Erik | |
| dc.contributor.advisor | Englehart, Kevin | |
| dc.contributor.author | Robertson, Jason William | |
| dc.date.accessioned | 2023-03-01T16:17:37Z | |
| dc.date.available | 2023-03-01T16:17:37Z | |
| dc.date.issued | 2019 | |
| dc.date.updated | 2023-03-01T15:01:23Z | |
| dc.description.abstract | The goal when replacing an amputated limb with an artificial one is for the prosthesis to respond as smoothly and naturally as the original limb, and for its use to be as simple and intuitive to learn as possible. After decades of development, myoelectric control has begun to fulfill that promise, with pattern recognition (PR) allowing a greater range of potential motions, accessible in a more straightforward interface, than ever before. However, the technology remains prone to difficulties from a number of sources, leading many users to abandon their devices due to poor control. The recent development of rejection – a simple paradigm in which uncertain movement decisions are ignored to prevent potentially costly errors – has shown promise in a controlled setting, however many questions remain about its utility and practicality. The exploration of this emerging aspect of control resulted in several research objectives for this thesis. The first was to demonstrate the usability of rejection for different classification schemes and to systematically establish a threshold for optimal controllability. The second was to implement and evaluate rejection as part of a practical error-reduction task. The third was to examine what effect rejection may have on a user’s ability to learn and adapt to a PR controller, as well as how rejection affected the way the user interacted with a PR controller. The results establish that rejection can be used not just with the gold-standard classifier, linear discriminant analysis (LDA), but with classifiers based on support vector machines (SVM) and support vector regression (SVR), as well. Controllability ii was found to be equally viable for a range of rejection thresholds, and even outside that range, using rejection remained superior to not using it at all. Finally, it was found that users did not change their behaviour, nor did they internalize or adapt to PR controllers any differently, whether they used a rejection scheme or not. These findings collectively present a strong case that, when properly tuned, rejection has broad utility as a post-processing technique that improves the controllability and user experience of PR without the host of drawbacks commonly incurred with other techniques. | |
| dc.description.copyright | ©Jason William Robertson, 2019 | |
| dc.description.note | Electronic Only. | |
| dc.format | text/xml | |
| dc.format.extent | xviii, 134 pages | |
| dc.format.medium | electronic | |
| dc.identifier.uri | https://unbscholar.lib.unb.ca/handle/1882/13352 | |
| dc.language.iso | en_CA | |
| dc.publisher | University of New Brunswick | |
| dc.rights | http://purl.org/coar/access_right/c_abf2 | |
| dc.subject.discipline | Electrical and Computer Engineering | |
| dc.title | Toward an enhanced understanding of rejection in pattern recognition-based myoelectric control | |
| dc.type | doctoral thesis | |
| thesis.degree.discipline | Electrical and Computer Engineering | |
| thesis.degree.fullname | Doctor of Philosophy in Electrical and Computer Engineering | |
| thesis.degree.grantor | University of New Brunswick | |
| thesis.degree.level | doctoral | |
| thesis.degree.name | Ph.D. |
Files
Original bundle
1 - 1 of 1