This decommissioned ERA site remains active temporarily to support our final migration steps to https://ualberta.scholaris.ca, ERA's new home. All new collections and items, including Spring 2025 theses, are at that site. For assistance, please contact erahelp@ualberta.ca.
- 91 views
- 142 downloads
Position-Aware Control of Myoelectric Prostheses
-
- Author / Creator
- Williams, Heather E.
-
As reported in 2020, millions of individuals worldwide have some form of upper limb amputation or congenital loss of limb that impedes execution of everyday actions like reaching for, grasping, transporting, and releasing objects. Although lost function can be artificially replaced via prostheses, there remains a problem—state-of-the-art robotic (myoelectric) devices continue to cause frustrating control challenges for users.
The work herein focuses on mitigating the sources of myoelectric prosthesis control problems for those with below elbow (transradial) amputation. Wrist and hand movements of such devices are controlled by captured electromyographic (EMG) signals that are generated by muscles in a user’s residual limb. Advanced pattern recognition-based controllers now make device operation intuitive for users, with software models trained to recognize EMG signal patterns. Still, devices can become unreliable when limb position changes are introduced during device use. This specific problem is known as the limb position effect. Here, when a user attempts to use their prosthesis in a position not included in model training, the captured EMG signals are unrecognizable and misinterpreted by the controller. This scenario can cause control malfunctions that result in unexpected device movements. Understandably, unreliable control leads to user frustration.
The broad goal of this thesis, therefore, is to reimagine upper limb myoelectric prosthesis control—offering persons with transradial amputation reliable, robust, and personalized device operation. Its specific objective is to develop an improved prosthesis control solution that mitigates the limb position effect.
The work in this thesis addresses the limb position effect problem by implementing position-aware myoelectric prosthesis control. For such control, EMG data must be augmented with details about a user’s limb position during device use. This was accomplished by capturing positional data via an inertial measurement unit (IMU) device worn on a user’s forearm, with blended EMG and IMU data handled using a deep neural network. Several complex recurrent convolutional neural network (RCNN) models for pattern recognition-based control were explored in this thesis, as they offer two major advantages: (1) they can be trained using large multimodal datasets (including EMG and IMU data), and (2) their architectures can learn features directly from input data to yield accurate movement predictions. After assessing the accuracy of both classification- and regression-based RCNN models, the three most promising were compared to a commonly implemented baseline classification model. Comparative testing sessions required participants without upper limb impairment to perform functional tasks in multiple limb positions. They wore a simulated prosthesis (controlled by each model), as a proxy for device use by individuals with limb loss. An RCNN regression-based model yielded the best task performance versus the baseline alternative. Given this, a participant with amputation was recruited to assess the novel control solution, confirming that it could indeed provide control across multiple limb positions that is accurate, simultaneous, and proportional to EMG signal intensity.
This work contributes, for the first time, a position-aware deep learning regression-based myoelectric control solution for use by persons with transradial amputation. The solution is an exciting new direction for prosthesis control—a departure from commonly used classification approaches. After future refinements to and testing of this solution, a long-term goal is to see it incorporated into prosthesis designs of today and tomorrow. Future extensions to this work could see regression-based control implemented in context-specific human movement therapies and technologies. Ultimately, the benefits of reliable control are far-reaching.
-
- Graduation date
- Fall 2024
-
- Type of Item
- Thesis
-
- Degree
- Doctor of Philosophy
-
- License
- This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.