1. Assisted Navigation
for an Autonomous Wheelchair
for the Severely Disabled
Xinyi Gong1
Mathieu Labbé2, Vivek Burhanpurkar3
Jonathan Kelly4
1Division of Engineering Science,
University of Toronto, Toronto;
2IntRoLab laboratory, Interdisciplinary Institute of
Technological Innovation (3IT),
Université de Sherbrooke, Sherbrooke;
3Cyberworks Robotics Inc., Orillia;
4Space & Terrestrial Autonomous Robotics Systems
(STARS) laboratory, Institute for Aerospace Studies,
University of Toronto, Toronto
2. Background
Electric-powered wheelchair by Pride Mobility
Visual Simultaneous Localization And Mapping (vSLAM)
implemented by Mathieu Labbé (PhD, Université de
Sherbrooke)
Path planning and autonomous navigation algorithms
available on Robotics Operating System (ROS)
Integrated with and designed the wheelchair’s driving and
sensing system by Charlie Guan (EngSci 1T6 + PEY) with
core values including:
o Use only low-cost consumer-grade hardware and sensors
o Reliability & Accuracy
4. Further work on…
Autonomous navigation + vSLAM
o Require pre-mapping the environments
o Maps saved in database
When map is used for autonomous navigation, user selects a
location & direction on map
Hence only useful for often visited places (e.g., home, office,
favourite coffee shop)
Time Consuming
Memory & Storage
5. My Approach
Alternative to Pre-mapping – Incremental Goal Generation
o Autonomous navigation + vSLAM still on
o Maps cleared with radius
Save memory and storage
Provide a buffer of information of previously unseen
spaces
Using the buffer, incremental goals are generated to
keep it moving with limited knowledge about the
environment
6. Assistive Mode using Incremental Goal Generation
Door Traversal
Desk Docking
Navigation through tight spaces requires sophisticated path
planning and accurate operation
Limitations in their own motor control
Corridor & Wall Following
Long period of joystick operation is difficult to sustain for the
severely disabled
7. Door Traversal
Most papers discussing door traversal only considered the
most basic scenario: a door in the middle a flat wall
8. Door Traversal
A simple Google Image search gives another common one:
http://www.chateauwaters.com/cw/apartment-homes/floorplans/
10. Corridor & Wall Following
When both walls within field of view (FOV) of camera, drive in-between.
When only a single wall is available, follow it with a reasonable distance.
At a corner by two walls, turn corresponding degrees and keep following.
Stop when no further information is available.
11. Desk Docking
• Work in progress
• Tested in real-world
3D sensor reading
• Detect desktop as a
surface
• Remove desktop to
analyze under-desk
space and calculate a
potential location and
orientation
12. Progress & Up Next
• So far, door traversal and corridor following have been tested in
real-world environment. Although successful experiments have
been repeated for considerable number of times, several
improvements are required and under investigation.
• Desk docking will be tested next.
• Will test in a variety of spaces on UofT campus by the end of
summer, 2016.
• End goal: demonstrate a 4-hour, failure-free supervised run of
the wheelchair without a passenger. If successful, further testing
will follow.