MoveIt Meeting Notes:
Attendance:
- Jens Petit - GSoC student, studying in Munich, integrating Bullet as a collision checker
- Omid Heidari - PhD student from Idaho State U, intern at Picknik, working with TrajOpt
- Jorge Nicho - SWRI in San Antonio, most internal projects at the moment
- Jon Binney - CTO at Iron Ox, has arms using MoveIt
- Bryce Willey - Realtime Robotics, Mentor for Jens, and OMPL refactoring
- Victor Vilches - port of MoveIt to ROS 2
- Simon Schmeisser
- Henning Kayser - Picknik
- Mike Lautman - Picknik, working on Cartesian interpolation planners, MoveIt Grasps
- Nathan Brooks - Picknik, Codeowners effort
- Gijs
- Andy McEvoy - PickNik
- Dave Coleman - PickNik
Moveit 2 Dashing Release
-
Acutronic provided 2 engineers originally, 2 more joined
-
Can successfully plan! Can reproduce the first example from the MoveIt tutorials!
-
a demo on how to estimate collisions without sensors?
- aiming to get some traction towards Dashing pre-release
-
Launching an alpha release for ROS2 for Dashing makes sense.
- Release it from the fork at Acturonics
- not enough time to get everything into the ros-planning main repo
- don’t want to impose on MoveIt maintainers before the complete port is finished.
- releasing a docker with everything
- support for 18.04 and OSX (definitely by the next 31st)
- still a lot of work to do
- Try it out here: any criticism is welcome
-
Acutronic is internally considering to contribute further (than just the initial ROS2 port), another announcement in 2-3 weeks
-
Qs:
- What does Window’s support look like:
- Not sure, no one present had extensive Windows experience
- OSX support involved a lot of CMakeLists.txt changes (linking and compilation)
- minor changes elsewhere
- Did ROS2 have everything there that we needed?
- Short answer: No.
- Acutronic is compiling a list of known issues
- Parameters weren’t working at the beginning, actions have a lot of issues
- FastRTS is okay with parameters, but memory leak in in ros_wmd makes some parameter fetching makes things buggy
- MoveIt is definitely a good battleground for testing ROS2
- Acutronic still had to port several external dependencies outside of MoveIt as well
- No debian files for release, just the docker hub image
- What does Window’s support look like:
ROSCon workshops
- Dave submitted an application for MoveIt workshops on the MoveIt Task Constructor and MoveIt 2.0
- Victor with be organizing the 2.0 workshop
- Volunteers will be needed to help run the workshop
GSoC/PickNik Intern Projects
- Jens Petit is integrating Bullet as a collision checker in MoveIt, for continuous collision detection in MoveIt.
- Current status, just beginning, but progressing
- Omid Heidari: doing TrajOpt as a new motion planner. Moving from TrajOpt_ros (tesseract planning effort)
- To use Jens’s work for collision detection and integrate into TrajOpt
- Has added an empty template planner, now starting to move TrajOpt in
MoveIt grasps: Mike Lautman
-
MoveIt grasps has had a lot of improvements recently, from Reputa (?) Robotics
-
Grasp planning (approach, retreat, lift planning)
-
Now being actively maintained
-
Differences between this an Intel?
-
Assumes that everything is cuboid, supports finger and suction graspers
- No machine learning, a simpler, more lightweight approach
-
Used the Schmalz multi-suction gripper, the Robotiq 2-finger gripper
-
Tutorial to enable for a new end effector?
Future Roadmap of MoveIt
-
Dave has some ideas, wants feedback
-
Having a strong roadmap can get us Funding/Engineering time, add new contributors
-
Different Phases:
- After MoveIt 2 in Beta, a refactor to look more like navigation 2
- Then a new focus on determinism
- TrajOpt, stopping OMPL from giving you random looking plans
- Making MoveIt realtime control ready
- What are the big needs of MoveIt?
-
Victor: everything realtime gets very complete
- We should consider if we want to do things fast, that gets in the way of determinism sometimes
- making layer 2 up to layer 7 of the OSI stack deterministic would be the first part of this
-
Jon: we should consider making plans that nicely run on realtime controllers
-
Could focus ideas by succinctly describing 3-6 well described use cases:
- Pick and place
- Research on a mobile robot
- planning to have a meeting outside of this one to brainstorm use cases for motivating the future roadmap
-
By realtime, we mean having certain planning components run faster.
- 1Khz controllers
- replanning 100Hz closed loop
-
Taking advantage of ROS 2 features
-
Biggest weakness of the project: OMPl is a little dissatisfying
-
A lot of the concepts in MoveIt are too complex for most industrial arm cases:
- MoveIt simple by Shaun Edwards
- In the MoveIt survey, a lot of people use Industrial arms
-
Victor: we need to improve/unify the software architecture diagrams to better understand how MoveIt works between all of the nodes/connections, etc.
- https://github.com/ros-planning/Moveit_docs/tree/kinetic-devel/diagram
- should work on moving some of these graphs over, they’re pretty great
Opening the Meeting to the general public
- Should we open the meeting to everyone who wants to be in it?
- originally, the purpose of this meeting was to keep everything alive, a small, once a quarter, intimate discussion on what to do
- Now has expanded to once monthly, more maintainers have been added recently
- Google Hangouts has a limit of 10, that we hit recently, but we’re now doing Google Meets, which could support up to 50 people
- we don’t want beginner users asking questions during this meeting, stick to the agenda
- usually about 1/2 agenda items are larger PRs under debates
- None opposed, will start announcing and linking the meeting to the public at large.
- How do we relate to the recently created manipulation working group?
- group all projects not necessarily in Moveit but are still relevant to manipulation.
- manipulation is focused soley on ROS2 however, so both meeting should be around.
- having an overall joined discussion would be nice, so consider stopping by the manipulation group
Agenda items not covered:
- Clang-tidy slow down (Andy Z and Robert H): neither were present
- Applications webpage update (Yu Yan): Yu was on call, but wasn’t responding, not much progress recently