Path Tracking for CarLike Mobile Robots Using Computer...
Transcript of Path Tracking for CarLike Mobile Robots Using Computer...
![Page 1: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/1.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Path Tracking for CarLike Mobile Robots Using Computer Vision
Jacky BaltesOffice 529 Machray Hall
Department of Computer ScienceUniversity of Manitoba
Andrew ThomsonCITR
University of AucklandNew Zealand
![Page 2: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/2.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Introduction
● Following a path marked on the floor with a mobile robot
● Simple, quick algorithm● Path is unknown● Few assumptions about the kinematics of the
robot● Visual servoing: path marked on the floor with
white tape● Evaluation using a global vision system
![Page 3: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/3.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Robot Hardware
Base: Tamiya Toy Car, proportional speed and direction. No odometry.
Processing: Eyebot controller, 25MHz 68332robotics controller, 1MB Ram
Sensors: CMOS camera80x60 resolution in 24 bit color. 15 FPS
![Page 4: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/4.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Global View of the Path
![Page 5: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/5.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Path Geometry
![Page 6: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/6.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Camera Geometry
● Camera slanted towards the ground● Robot can not see directly in front
– Minimum distance is 8.8cm
● Lookahead distance
![Page 7: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/7.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Camera Views
● Simple RGB threshold to extract path from image● Sufficient in robotics application● Green tint because the CMOS sensor is
uncalibrated for R,G, and B gains
![Page 8: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/8.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Path Tracking Control
● Most path tracking controllers for mobile robots use offset and gradient
● Balluchi uses offset, gradient and curvature
● Egerstedt uses look ahead
Path
~yØ
Ø
p
Ø~
xc,y
cx
p,y
p
l
![Page 9: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/9.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Image Processing
● Maintain 8 fps
● Process rectangle
● P1 bottom and right edge
● P2 top and left edge
● Worst case 210 Pixels
● Offset: distance to path
– Avg. x of P1 and P2● Gradient of P2P1
![Page 10: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/10.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Image Processing Example
● Scan for P1: Bottom and left● Scan for P2: Top and right
![Page 11: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/11.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Control
● Gradient and offset are normalized to 1 .. +1● Weighted sum of gradient and offset is computed
and a steering angle is extracted from a lookup table (9 entries)
● Steering angle determines speed (Slow, medium, fast)
● Specific to robots– Linear interpolation between the extreme steering
angles
![Page 12: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/12.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Evaluation
● Influence of different weightings for offset and gradient.
● What information is more important?● Avg. error vs speed and avg. error vs. commands● 3 Lookahead distances
– 10 Pixels = 11.3cm
– 20 Pixels = 13.5cm
– 30 Pixels = 17.1cm
● 5 laps around the racetrack as quickly as possible
![Page 13: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/13.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs Offset Weighting10 Pixel Lookahead
● Offset clearly needed● Small influence of gradient
![Page 14: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/14.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Path Offset 5%10 Pixel Lookahead
![Page 15: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/15.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Evaluation Offset 50%10 Pixel Lookahead
![Page 16: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/16.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs Offset20 Pixel Lookahead
● Similar result to 10 pixel lookahead● More variety in results● Worse at 100% offset weighting
![Page 17: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/17.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Evaluation Offset 75%20 Pixel Lookahead
● Smoother curve than 10 pixel lookahead
![Page 18: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/18.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs Offset Weighting30 Pixel Lookahead
● Poor performance on 100% offset● Average X coordinate of P1 and P2
![Page 19: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/19.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Evaluation Offset 85%30 Pixel Lookahead
● Lost sight of track sometime● Follow last known position
![Page 20: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/20.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs. SpeedAverage Error vs. #Commands
● What is the influence of the error on the average and maximum speed of the robot
● No correlation● Strong correlation between number of commands
and error– Slow motion of the robot
– More commands => closer to the path
– No oversteering (Balluchi)
![Page 21: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/21.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs Speed10 Pixel Lookahead
![Page 22: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/22.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs Speed30 Pixel Lookahead
![Page 23: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/23.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Average Error vs Control Work10 Pixel Lookahead
![Page 24: Path Tracking for CarLike Mobile Robots Using Computer Visionjacky/Publications/pdf/thomson02:_mobil_ro… · Author: Jacky Baltes University of Manitoba 1/19/06 Introduction Following](https://reader033.fdocuments.net/reader033/viewer/2022042321/5f0acff67e708231d42d76a3/html5/thumbnails/24.jpg)
Author: Jacky Baltes University of Manitoba 1/19/06
Conclusion
● Simple control algorithm for visual servoing of mobile robots
● Quick approximation of offset and gradient● Short sighted robots: offset much more important● New controller using Fuzzy Logic● Better extraction of the path