USING VISION AND ELECTROMYOGRAPHY FOR AUTONOMOUS CONTROL OF ROBOTIC LEGS
Keywords: human-robotic legs, deep learning, computer vision, wearable robotics, prosthetics, exoskeletons
Overview of Research
Mobility impairments are abundant and can impact one’s ability to perform daily locomotor tasks; for example, over 40 million people in the United States alone have serious difficulties walking or climbing stairs. Robotic prostheses, exoskeletons, and other assistive devices can help address these limitations and allow users to regain mobility by increasing the safety and autonomy of walking through powered locomotor assistance.
Control is vital for these robotic-legged systems to provide safe and effective locomotion. However, the control of robotic legs remains a challenge, especially while navigating complex real-world environments. Systems published to date depend on proximate sensors (i.e., mechanical, inertial, and electromyography (EMG)), which are limited to current state information and struggle with the ability to transition between walking modes, analogous to walking blind. However, computer vision presents the opportunity to improve robotic leg control by classifying and adapting to environments prior to physical interaction.
This study focuses on the development and assessment of a high-level control system using large-scale computer vision and EMG data while leveraging modern deep-learning methods. The project aims to investigate if and when the use of vision and deep learning can improve the current performance of high-level control systems in terms of the speed and accuracy of the walking mode predictions.
Research Team
Alex Mihailidis, Ph.D. P.Eng. (University of Toronto)
Garrett Kurbis, (University of Toronto)
Brokoslaw Laschowski, Ph.D. (University of Toronto)
