SmartWheeler

1
SmartWheeler
Vision
Project Members

• Amin Atrash
• Robert Kaplow
• Nan Lin
• Andrew Phan
• Joelle Pineau, Ass. Prof.
• Chris Prahacs
• Shane Saunderson

http//www.cs.mcgill.ca/smartwheeler
Previous Work Many techniques have been
proposed in recent years that are good at
avoiding obstacles or positive occlusions. Yet,
there is little or almost no literature on
avoiding pot holes or negative
occlusions. Proposed Research We propose a
passive (no lasers, just cameras) path planning
system that will build upon previous obstacle
avoidance algorithms and incorporate negative
occlusion detection based primarily on monocular
optical flow calculations. The goal is find the
fastest and safest path that will prevent our
wheelchair from getting stuck or, worse, toppling
over and thus injuring the user.
Project Sponsors
NSERC, Canadian Foundation for Innovation,
Sunrise Medical, Lippert Systems

• The goal of this project is to increase the
  autonomy and safety of
• individuals with severe mobility impairments by
  developing a robotic
• wheelchair that is adapted to their needs.
• SmartWheeler capabilities and objectives
• Automatic mapping and localization of
  indoor/outdoor environments.
• Detection and avoidance of positive and negative
  obstacles.
• Point-to-point planning and navigation.
• Shared control between autonomous controller and
  human user.
• Adapted interface for low-bandwidth
  communication.
• Testing and validation with target population.

Project Overview
Human Interface
The SmartWheeler's target users will be those
with limited motor skills patients suffering
from Multiple Sclerosis, or recovering from
spinal cord injuries. The focus is therefore on
compensating for those disabilities. Higher Level
Input Instead of the user driving the robot with
a joystick, the robot will have various levels of
autonomy. The user dictates the goal, the path,
and restrictions such as speed, and the robot
takes care of the driving. Touchscreen As an
alternative to the joystick, the touchscreen
offers the advantages of potentially reduced
fatigue levels and absolute positioning. To
further improve the ease-of-use of the
SmartWheeler, a gesture-based system will be
implemented
Sensing
SmartWheeler makes use of various sensors for
localization. One of the goals of the project
will be to determine the minimum sensing required
to complete specific higher level tasks. Each
type of sensor may be used for a combination of
navigation and obstacle avoidance. Long range
sensing is accomplished using two Sick LMS-200
laser range finders. The forward facing sensor is
mounted under the feet of the passenger, the
rearward sensor is mounted behind the seat. The
LMS-200 sensors have a range of 10m, a viewing
angle of 180deg., and a distance resolution of
15mm. Between four and eight Sharp GP2D12
infra-red short range (lt80cm) sensors will be
used for obstacle avoidance, as well as to cover
the gaps between the lasers. For wheel odometry,
two Hall-effect sensors are employed per main
wheel. The Allegro ATS651 sensors can detect both
speed and direction for each wheel.
Robust robot localization and map building
performed using Particle Filter methods
Partially Observable Markov Decisions Processes
(POMDPs) allow for high level decision making in
light of environmental uncertainty Integration
of wheelchair hardware into Carmen Robot
Navigation Toolkit Autonomous Navigation ideal
for disabled users Research will combine POMDPs
for high level planning with Particle Filter
methods, and explore methods to
learn POMDPs
Sensors will relay data to an HC12 based
microcontroller. This microcontroller will
manage all data before passing it along a serial
line at regular intervals to a PC/104 form factor
single board computer. The SBC (from Lippert
Systems) will be the workhorse of the navigation
and control of the SmartWheeler. Using a
customized version of the Carmen software, in a
linux environment, the SBC will interpret
information about the chairs surroundings, and
plot its course appropriately. Output signals
will be sent to another control module known as
the OMNI from PGDrives Tech. This step ensures
the universality of the control system as the
OMNI is available on almost all brands of
powered wheelchair and requires very basic input
parameters to drive. The OMNI will take
steering commands from the SBC and interpret them
into motor commands to drive the chair.
Navigation
Control