guidance-for-wheelchair

EE887 Special Topics in Robotics
Paper Review
Initial Results in the Development
of a Guidance System
for a Powered Wheelchair
2000. 6. 7.
BSCL
Lee Hyong Euk
General Idea of this paper
The autonomous navigation and the control of wheelchair
What are the issues in the wheelchair application?
 User Interface
 Obstacle avoidance
 Battery usage
 Seating Comport
 User demographics
…
Focus :
The navigation(guidance) of the wheelchair,
Particularly the estimation and control of the system
Requirement for Wheelchair Application
1.
2.
3.
4.
5.
System must be more accurate
System must be robust and repeatable
Smooth ride to ensure user comport
System should be simple and inexpensive
Remember that the passenger is a human being
Requirement for Wheelchair Application
1.
2.
3.
4.
5.
System must be more accurate
System must be robust and repeatable
Smooth ride to ensure user comport
System should be simple and inexpensive
Remember that the passenger is a human being
 This requirements need ‘exact position estimation’
Experimental Wheelchair System
1000 count-per revolution
Optical encoders
<Fig1. Experimental Wheelchair System>
Everest & Jennings Powered Wheelchair
512 by 480 pixel CCD
Black and white
26 inches
The camera view is not blocked by
the user’s lower extremities.
<Fig2. Wheelchair Schematic>
Approach
(1)
Odometry information(wheel rotation)
External vision-based observation of surrounding envrionment
Combining and Applying
Extended Kalman Filter algorithm
Optimal estimate of the wheelchair’s pose
Observation or measurement noise are modeled by Gaussian Distributed white process
A set of diff. Equation which relate wheel motion to the position and orientation of the
wheel chair are numerically integrated to produce the so-called “dead-reckoned”
Approach
(2)
Automatic Guidance of the wheelchair
 “Teach-repeat” Procedure
The role of two video camera
 Detect the cues in the surroundings
Approach
(2)
Automatic Guidance of the wheelchair
 “Teach-repeat” Procedure
The role of two video camera
 Detect the cues in the surroundings
: Reference hints for pos. estimation
(ex. Desk, wall, or any fixed one in the workspace)
16 cues were used for this system
(some objects in the experimental workspace)
‘cue’ is a priori information
Approach
(3)
<Fig3. Wheelchair system guidance
Flowchart>
Experiments
Test environment : home, office, classroom, laboratory, …
Load : 200-lb human passenger and other equipments(PC, …)
Floor surface : smooth poured concrete, tile, various carpet type
Ref. Path was taught in the
Mechanical Systems and Robotics Lab.
At the University of Notredam
Room layout with nominal path
Experiment Results
(1)
Tracking Ref. Path Result
Total Time : 175s
Avr. Speed : 0.5 ft/s
16 cues
Experiment Results
(2)
Tracking Ref. Path Result with obstacle avoidance
Manual Control for
obstacle avoidance
Experiment Results
(3)
Speed Control : error between actual and estimated position
Avr. X
error (in)
Avr. Y
error (in)
Max. X
error (in)
Max. Y
error (in)
Nominal speed
1.006
1.167
2.884
2.393
Low speed
0.175
0.146
0.369
0.393
For 10 consecutive run case
< Nominal Speed : 0.5ft/s >
< Lower Speed : 0.3ft/s >
Discussion
1. The extended Kalman filter accurately estimate
the system’s pose(position & orientation)
2. The limitation of evaluating accuracy depend on
the position of the ‘cue’s.
 a total of four cues were available for the last 8 ft of the path.
3. The obstacle avoidance strategy must be
developed
4. User interface and other considerable factor is
remained jobs.