Meeting Notes
PILZ industrial planning PR
Joachim -
Pilz PR still not merged (#1893), test time is still long
v4hn -
Pilz robot has not been added to moveit_resources pkg
Convert some ros tests to unit tests to avoid roscore overhead?
Joachim -
Will not make further changes to tests (as discussed last mtg) because we might never make progress
moveit_resources refactoring
v4hn -
has made a large PR (#2199), many paths have been changed
could use some help maintaining moveit_resources
there’s almost no reason not to use the original packages, they are mostly copy/pasted into moveit_resources
moveit needs its own copy of some robot configs for testing, does not want to depend on upstream repos
will need to sync with upstream every so often
MoveIt2
Henning -
synced with MoveIt1 again
moveit_ros_benchmarks is done (Yu Yen)
Foxy branch is finally functional, feedback welcomed. It is the next release candidate
Some conflicts with urdf_dom, waiting for new release
move_group is ported, PR being reviewed
just a few packages left to port
still working on porting tests, some deprecated headers etc to be replaced
Mark/Tyler -
OpenCV version might need updating
MoveIt2 clang format version
Henning -
still discussing clang format version (6 or 10?)
Andy -
prefer the newer one for better forward compatability
Mark -
who is in favor of 10?
(no dissent from anybody)
Student updates
Jeroen -
Google Summer of Code, Belgian PhD student
working with OMPL interface
implementing a constraint interface
showed some cool videos
goal: replace a hacky patch to prevent joint flips when constraints are applied
the new solution works great in some cases
tolerances of 0 are possible (exact solution)
did some code cleanup too
starting a tutorial
Adam -
Google Summer of Code, UTexas PhD student
made many updates to the ROS1 realtime servo node
has now ported it to ROS2
showed video of realtime servo with a Panda robot, rviz, and fake_joint_driver
now working on porting and adding tests
upcoming features:
split out the jacobian to allow servoing for non-serial manipulators, too (e.g. Stewart platforms)
damped least squares method
v4hn -
a GUI to easily test servoing would be nice
Boston -
a PickNik summer intern
doing grasp planning with deep learning from point clouds
started with DexNet
showed a cool video with a real UR5 grasping a bar of soap
Considering whether to add this to MoveIt Task Constructor [MTC] or an action server / service interface
v4hn -
does not see the need for an action/service interface
prefers an MTC grasp generator stage
MTC has the advantage of running in parallel
Mark -
MTC may make it more accessible
MTC documentation needs work
Melodic release
Tyler -
MoveIt Melodic release coming soon