Meeting Notes - 27 May 2021
Agenda Items
Release updates and blockers (Tyler)
MoveIt 2 updates (Henning)
Demos for new MoveIt 2 capabilities (David?)
Google Summer of Code student accepted (Mark)
broken/invalid CI setup
CI for Windows and macOS
Finishing up GHA migrations
Release Updates (Tyler)
Vatan’s been working on releases for Rolling and Galactic.
Backports have been merged in. If people are happy, we can release for Melodic and Noetic. There is still a circular dependency issue for the ROS1 versions with the resources package. Not a problem in ROS2 because the Pilz planner hasn’t been ported yet.
Robert: Pilz uses the IKFast solver for its robots and tests. To build IKFast generated solvers, we need access to moveit_core. However, the core packages depend on IKFast. Not a problem if everything is in the same workspace, but it surfaces if we’re building/testing an individual repository. Only causes problems if there are ABI changes in the core repos. Clean solution would be to put Pilz planner in a separate repo, but a downside would be that this introduces a separate release process.
Mark: Hesitant to start splitting planners out into separate repos. Lots of planners are also included in the core libraries.
Andy: One option could be to put Pilz planners in moveit_experimental again.
Robert: Pilz planner isn’t experimental anymore, so don’t think this is suitable.
Tyler: Move Pilz section of the resources repo into the main repo?
Robert: Pilz folks are not very responsive to this issue. We need to involve them again to get this resolved.
Henning: Could we move the Pilz tests into a downstream workspace?
Robert: Better to keep tests in the same repo as the code they’re testing. The Pilz packages have good test coverage and well-designed tests.
Felix: Ease of development and maintenance is also an important consideration. Prefer to not split things across different repos.
Robert: Will reach out to Pilz for help.
MoveIt2 (Henning)
New Foxy release completed.
Some upstream deps released to Rolling. Still blocked by some dependent packages.
Aiming to release for Galactic in mid-June.
Want main branch to target Foxy, Galactic, and Rolling at the same time. Expect to fork off Foxy eventually.
GSOC (Mark)
We picked a participant! Goes by CaptainYoshi on MoveIt Discord. Working on whole-system benchmarking for MoveIt, with realistic planning scenarios.
Broken CI
(same as Pilz planner discussion above, so skipping further conversation)
CI for Windows and MacOS (Wolf)
Wolf: Has been handling an ongoing project to release ROS packages through conda.
Related Discourse post for additional context (added by Joe).
MacOS support for ROS hasn’t been consistent.
Working to release ROS packages as conda packages, distributed through conda-forge. This benefits both Windows and MacOS users.
Patched some ROS packages to fix some issues related to releasing on conda. Working on merging patches into MoveIt libraries – fixes to CI to support releases. Starting with moveit_core for now.
Benefits of using conda: works on lots of Linux disributions, and not just a particular version of Ubuntu. Runs directly on the host system, so can benefit from hardware such as Cuda GPU acceleration. Better support for running different releases in parallel virtual environments, even weird combos like Kinetic/Noetic. Keep past builds for old versions as well.
Group discussion: How many people actually use ROS with MacOS?
Jorge: Beneficial to support a variety of OSes. Helps with some situations such as sending a demo to a client who’s not familiar with Linux.
Mark: Good to support MacOS, as long as it doesn’t clutter up the codebase with #ifdef statements to switch between different environments.
Github Actions migration
Vatan: Just about everything has been migrated. Some new repos will be added as part of Galactic release.
Other Topics
Fate of the fake_controller_manager
Henning: Is it OK to remove the fake_controller_manager from MoveIt2? The ros2_control version of this feature is much better, since it provides simulated hardware interface for all controller types. Downside is that we would totally rely on ros2_control for this functionality.
Robert: Another benefit is that we could remove the intermediate layer of the MoveIt controller manager – would simplify using controllers with MoveIt a lot.
Jorge: Currently have a use case for the MoveIt fake_controller_manager, but it’s in a situation where we will switch to MoveIt2 anyway.
ROS2 Demos
Mark: ROS2 Galactic just released. Planning on having a MoveIt2 demo ready in time. Polishing a demo with Hello Robot Stretch, so we (PickNik) will have something to show for the next meeting. We’ll also have a UR-5e demo.
ROS2 MoveIt Setup Assistant?
Jorge: Do we have a way to do MoveIt Setup Assistant things in ROS2?
Mark: Not really, currently. Launch file outputs are very different, so some careful thought is needed to make sure the MSA app produces the right files and puts the right info into them.
Henning: It will actually take a lot of work to get the MSA working well with ROS2. In the meantime, there’s also a standalone command to generate the allowed collision matrix for the SRDF, which is one important part of the MSA outputs. Need to document ongoing design work for ROS2 MSA.
Attendees
Abishalini Sivaraman
Adam Pettinger
Andy Zelenak
Bren Pierce
Cassidy Elliott
Dave Coleman
David Lu!!
Felix von Drigalski
Henning Kayser
Jafar Abdi
Joe Schornak
Jorge Nicho
Mark Moll
Marq Rasmussen
Robert Haschke
Simon Schmeisser
Tyler Weaver
Vatan Aksoy Tezer
Wolf Vollprecht
Wyatt Rees