Navigation2 WG Meeting Notes 12/5 and 12/12/19

Next Generation ROS
1

12/5 Meeting Notes
Attendees:

  • Mohammad Haghighipanah
  • Carlos Orduno
  • Carl Delsey
  • Melih Erdogan
  • Matt Hansen
  • Mike Jeronimo
  • Brian Wilcox
  • Francisco Martin Rico
  • Geoffrey Biggs
  • Jeremy Adams
  • Shivang Patel

Use more complex world models in CI (#1394)

  • Turtlebot 3 world is very simple and is no longer a good test for navigation
  • It would be nice to have better worlds to be a more thorough system test
  • The models identified in the issue are big so we may have problems with with CI
  • We can maybe use github Filesystem https://git-lfs.github.com/ There are problems with storing big binary objects in Git, so this is an alternative
  • Add the files to the system test first and we’ll add it to bringup later if appropriate

Eloquent Release

  • Still 15 outstanding issues and PRs. Please close any issues that can’t make it in before the eloquent release.
  • eloquent-devel branch created. Any new PRs that go in will need to be backported to that branch. The dynamic params changes are broken on master due to upstream changes. They have been backed out of master but are still on the eloquent-devel branch.

12/12 Meeting notes
Attendees:

  • Mohammad Haghighipanah
  • Carl Delsey
  • Carlos Ordunu
  • Melih Erdogan
  • Yuthartha Tuladhar
  • Casimir Sowinski
  • Francisco Martin Rico
  • Geoffery Biggs
  • Steve Macenski

Topics:
Replan immediately on plan invalidation (#1395) and Rearchitect Behavior Tree / Nodes (#1387)

  • discussed the need to replan immediately in case of an error and also need to support the use case where the robot is not replanning at a regular interval. This could possibly be achieved with the existing ComputePathToPose node by creating another instance in the recovery behaviors.
  • Discussed the possibility of breaking up action BT nodes into separate nodes for requests and responses and getting feedback etc. This gives us more fine grained control on how we interact with actions within the behavior tree. This could give us the ability to cancel or otherwise interact with an existing request from different parts of the behavior tree. It was noted that we can cancel a current action by causing the behavior tree to halt the action node, however that is a less direct way of controller the action.

Potential Migration from QMCL to AMCL(#1391)

  • QMCL is a much cleaner implementation of a Monte Carlo localizer. It could be used to replace AMCL, however it needs some work - convert to ROS 2, add missing features.
  • The maintainer is open to using a more permissive license like MPL. It is currently GPL
  • QMCL needs testing to determine how it compares to AMCL in terms of localization quality.
  • We need testers for that.

Default SLAM for ROS2(#1389)

  • Deferred to next week

Rectify & describe map server tickets(#1280)

  • We need to remove and/or clean up redundant map related tickets.
  • The team at Intel is planning to tackle many of those tickets in 2020. We can clean up those tickets then.

Costmap Filters (#1263)

  • There is a lot of commonality between the implementations of traffic lanes, slow zones and keep out zones. Steve proposed they could all be implemented at a separate costmap layer.
  • Keep out zones could just be another static layer map.
  • Slow zones could be a special costmap layer plugin that publishes on a topic when the robot is in an area where is needs to go slow.
  • Traffic lanes - there was a mismatch in the understanding the requirements for this feature. One possibity is that traffic lanes are areas where the robot should preferentially travel - sort of the opposite of a keep out zone. In that case, there is no directionality to the lanes. The other thought was that traffic lanes are more like road lanes, where robots going in opposite directions should prefer to stay on opposite sides. We need to clarify the requirments on this one.

Thanks to @cdelsey for taking notes!

No further meetings in 2019

2 Likes