ApproachFinder: Real-time Perception of Potential Docking Locations for Smart Wheelchairs

Abstract

According to the Canadian Survey on Disability 2017, 22% of the Canadian population suffers from some form of disability, and among older adults specifically, mobility impairment affects 24%.This thesis aims to assist the development of smart wheelchairs that can seamlessly collaborate with people to accomplish their mobility goals. To achieve this, we developed two algorithms that can find safe locations to park a wheelchair near indoor objects. First, we developed a vision pipeline using standard computer vision techniques and hand-tuned geometric properties. Next, we leveraged this vision pipeline to train a neural network that provides similar results with far fewer computational resource requirements. Finally, we present a way to integrate these safe locations into any shared controller. The larger goal of this work is to develop a semi-autonomous docking assistance system that provides a collision-free path for wheelchair navigation.