1 of 13

"Moe" the Autonomous Lawnmower

Auburn University

William Woodall <w@auburn.edu> williamjwoodall.com

Michael Carroll <m@auburn.edu> mjcarroll.net

2 of 13

Overview

  • Competition Overview
  • Overview of “Moe”
  • Team History
  • Moving from robot_pose_ekf to custom EKF
  • Use of move_base
  • Architecture of path planning
  • Edge Cases
  • Conclusion

3 of 13

Competition Overview

  • Hosted by ION and AFRL
  • automow.com
  • Dayton, OH
  • June 2nd
  • Dynamic Competition
  • 2010 - Third
  • 2011 - Second
  • 2012 - First?

4 of 13

Overview�of "Moe"

ROS Features:

  • App Manager
  • ROSOSC Teleoperation
  • Full URDF
  • Gazebo

5 of 13

Team History

  • 2010:
    • 9 Mechanical
    • 7 Electrical
    • 2 Software
  • 2011:
    • 1 Electrical
    • 2 Software
  • 2012:
    • 1 Electrical
    • 1 Software

6 of 13

Team History

7 of 13

8 of 13

Reasons for Custom EKF

  • "Tighter" integration of raw messages
    • Wheel encoders used directly in model
  • Zero-velocity updates
  • Using lawnmower specific heuristics
    • Monitor GPS covariance and fix type
    • Monitor cutting blade status, as they affect the magnetometers in the AHRS
  • Remove the "black-box" effect
  • Learning Experience

9 of 13

Result of Custom EKF

  • Better Solution for the Lawnmower
  • Tighter control over resulting solution
  • ROS interfaces provide sufficient decoupling to easily drop-in replace the filter into the system

10 of 13

Move Base

  • Great General Solution
  • Point A to B without hitting things
  • Coverage Planning
  • Another "black-box"
  • Frustrations from tuning with noise from GPS

11 of 13

Coverage Path Planning with Move Base

  • A to B != Coverage
  • Custom global/local planner or feed move base�waypoints?
  • What's the �best way to �combine �coverage �and obstacle �avoidance?

12 of 13

Edge Cases

  • Most valuable grass is near obstacles
  • Move Base is difficult to predict in tight maneuvers
  • Scheduled�Controllers
    • Move Base for �general �navigation
    • Specialized local�controller for �edging obstacles
    • Simple "Maneuvers"

13 of 13

Conclusion

  • ROS provides great general solutions to common problems
  • Those solutions can sometimes be configured to be good enough for domain specific tasks, but can also struggle
  • ROS grew with the team
  • Autonomous Lawnmowers �are awesome

goo.gl/2ZPlD