All,
If you’re interested in our ROS2 Navigation plans and status, we’re going to start a regular working group to discuss.
Our first WG will be this Thursday, 8/2/18, at 8am Pacific time. If you are interested in joining, message me with your email and I’ll send you an invite.
Thanks,
Matt
2 Likes
Update - due to a conflict, I have to move this meeting to 7am Pacific time, still on Thursday 8/2. If you’re interested, message me with your email address.
Matt
Roser
August 3, 2018, 7:18am
3
Will you guys publish your meeting minutes each time after the meeting ? It benefits the community to get aware of what’s going on, thanks !
Thanks to everyone who attended the first WG, overall it was a success. There is obviously a lot of work to be done, and we’ll meet again in a few weeks.
If you replied to me and didn’t get an invite, I apologize, I didn’t see some replies until after the meeting. Next time I’ll try to create an event with a link that is public so anyone can join.
The presentation slides - Navigation 2 Overview, are being pushed to our Github repo: https://github.com/ros-planning/navigation2/blob/master/doc/README.md
Feel free to file issues there for anything you want addressed.
Thanks again,
Matt
All, here were the notes captured from the first meeting (thanks to @mjeronimo ):
Next time I’ll try to be more organized.
Michael Ferguson
Comment: Task dispatch engine: make this a separate library?
Possibly, if it makes sense. Our first release will probably have it integrated
Should implement continuous re-planning in the navigation node
ROS2: compress these modules into same process?
Warning: Environmental model across many processes could be performance issue
Huge state machine graphs floating around
Could we solve with local caching?
Questions
Porting A* and DWA or re-implementing?
Mostly porting, with some refactoring
Add some unit tests?
Yes, we plan to have unit tests, integration tests and system tests
Working on a branch, review upon integration into master?
Yes we will integrate into master, branch is for temporary / work in progress code
Timeline?
December 14th (Dirk mentioned Crystal release)
Fully working system?
At least basic move to pose functionality, with collaterals
Default platform?
David Lu
Sharing of resources - reason for putting everything in one process, one node for move base
Have to pass entire cost map - could be performance limiter
Part of the reason to combine it was performance-related
Overall trade-off between ability control things vs. ability to be modular
Would we be better off (nav core) not having the cost map object not managed at the high level
Planner tracking its own cost map
What you lose is ability to call specific recoveries
3D version of DWA
Adam Duncan (Amazon)
Anything in ROS2 design; fundamentally aren’t achievable in ROS2?
Answer - no, we believe this all is possible when Actions are implemented
Any core components missing on ROS2
Tf2 message filter
Actions
Ray Cole (Amazon, lead engineer on internal team, ROS)
Performance: areas to try to test? Use cases, data flow.
Something to do early on
Introspection into what’s going on? Management console.
Errors, invalid states
Productized
Tom Moore
Wei Zhi Liu
Dirk Thomas
Overhead between processes
Could put upper cap on cross-language capability
Could incur high run-time cost
Could stay within same language, build into shared libraries
Shivang Tripathi
Steve
Doug (Amazon)