Towards an Intelligent Vision-Guided Wheelchair
There are a number of active investigations into various aspects of autonomous wheelchairs, which are briefly summarized:
Object Detection in Nursing Homes for Autonomous Wheelchair Software
When creating autonomous robots for indoor health-care environments, common techniques such as LIDAR, GPS, and RFID chips, can be unreliable, and interfere with medical technology. Thus, computer vision is proposed as a less invasive and more reliable technique. This project, which is divided into two parts, aims to develop software for an autonomous wheelchair, which will use vision to navigate around a nursing home.
The first part of this project is building a data set of items encountered in a nursing home (e.g., wheelchairs, walkers, canes, etc.). This dataset will be trained on an existing object detection algorithm (e.g., YOLO9000) and tested in a nursing home setting. The second part of the project is to determine what features of the nursing home can be used to localize the wheelchair, allowing it to autonomously navigate within a pre-mapped space reliably. The strategy for localization is a combination of an existing global pose refinement framework, as well as object classification using the dataset put together in the first part of the project. Successful classification of objects will inform the current location of the wheelchair (i.e., answer “where am I right now?”); global pose refinement, which is guided by wheel odometry and feature detection of wall edges, will allow the wheelchair to localize as it moves through nursing home corridors.
Wheelchair Interface Design
The interface for how a user may access an autonomous wheelchair’s functionality is critical. Our efforts towards a new design can be found here.
Wheelchair Hardware Design
We are developing a new hardware platform and more detail can be found here.