Toward improving the training of pattern recognition based myoelectric control
| dc.contributor.advisor | Scheme, Erik | |
| dc.contributor.advisor | Englehart, Kevin | |
| dc.contributor.author | Wright, Kadie | |
| dc.date.accessioned | 2023-03-01T16:26:40Z | |
| dc.date.available | 2023-03-01T16:26:40Z | |
| dc.date.issued | 2018 | |
| dc.date.updated | 2023-03-01T15:02:17Z | |
| dc.description.abstract | Decades of advancements in the development of myoelectric signal processing techniques have made prosthetic devices an effective means of functional replacement for upper limb amputees. One of the control approaches that has been widely researched in this field is pattern recognition (PR) based control using electromyography (EMG) signals, which has only recently become commercially available. There are many opportunities to improve the user experience using PR control. One important issue is optimizing the training of the PR controller, which requires the collection of appropriate data. Although the inclusion of confounding factors (such as varying limb position) in the training data has been shown to significantly improve the performance of the pattern recognition approach, little work has focused on how to actually elicit the training contractions themselves. This work examined two existing training techniques that are currently being used in the field (ramp contractions, and Velocity Guided Training), and introduces two new alternative training methods; Position Guided Training and an alternate position guided training (Position-Reset Training) approach to mimic the prompts for Prosthesis Guided Training (PGT). The comparison of approaches was motivated by a desire to incorporate more dynamic motion into the training process, which may better reflect the actual use case compared to existing methods. It was hypothesized that the new methods would provide more relevant training data which would result in improvements in real-time performance and usability in a virtual target acquisition task. Thirteen able-bodied subjects (9 male and 4 female, mean age 24 +/- 2.1 years) completed a Fitts' Law based usability test using controllers trained with each of the training methods. For each method, EMG data representative of five different motions (hand open, hand close, wrist pronation, wrist supination, and no motion) were recorded and used to train the controller, before completing 24 repetitions of the target acquisition task. Comparison of real-time performance metrics showed no significant difference between the ramp, Position Based Training and Position-Reset Training approaches. Velocity Guided Training, however, the currently employed method of Prosthesis Guided Training, obtained significantly better movement efficiency (p<0.05). No significant differences were found in throughput, a Fitts' law summary metric, which combines speed and accuracy into a single measure. These results suggest that, although other training approaches may offer more intuitive training prompts, Velocity Guided Training more effectively informs the training of pattern recognition based myoelectric control. Future work could include consideration of cognitive load and motivation on the part of the user, to help form a more complete picture of training and usability. | |
| dc.description.copyright | © Kadie Wright, 2019 | |
| dc.format | text/xml | |
| dc.format.extent | xii, 87 pages | |
| dc.format.medium | electronic | |
| dc.identifier.uri | https://unbscholar.lib.unb.ca/handle/1882/13853 | |
| 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 improving the training of pattern recognition based myoelectric control | |
| dc.type | master thesis | |
| thesis.degree.discipline | Electrical and Computer Engineering | |
| thesis.degree.fullname | Master of Science in Engineering | |
| thesis.degree.grantor | University of New Brunswick | |
| thesis.degree.level | masters | |
| thesis.degree.name | M.Sc.E. |
Files
Original bundle
1 - 1 of 1