Manipulation WG / MoveIt Maintainer Meeting - Jan 26

The next Manipulation WG Meeting / MoveIt Maintainer Meeting will be 2023-01-26T16:00:00Z.

Agenda

Meeting

Real-time meetings by Google. Using your browser, share your video, desktop, and presentations with teammates and customers.

Phone Numbers (US)

+1 347-486-5750

PIN: 644 991 049#

The Google Summer of Code (GSoC) application period for open source projects is from January 23 till February 5. It’d be good to start the discussion of possible MoveIt GSoC project ideas.

Attendence:

  • Bilal Gill
  • Paul Gesel
  • Mark Moll
  • Cassidy Elliott
  • Sebastian Castero
  • Henning Kayser
  • Simon
  • Chris Thrasher
  • Andy Zelenak
  • Nathan Brooks
  • Sebastian Jahr
  • Jafar Uruc
  • Peter Mitrano
  • Tyler Weaver

Notes

[Tyler] Refactor of CMake for ros2_control CMake to fix overriding. Would appreciate reviews.

[Henning] Rolling Binaries are broken, believe it is because of realtime_tools and we need a new release of realtime_tools

[Sebastian Jahr] Benchmarking to evaluate what should be default planning pipeline, consider parallel planner. Now it is RRT Connect, which is most likely to produce solutions, however the paths are crazy because it lacks optimization.

  • PILZ consistently gives shorter path lengths but fails to plan much more often.

  • In this specific case CHOMP reliably produces the best solutions. This is one of the case where using parallel planner gives you the advantage of having good plans from CHOMP when it works. CHOMP does not always produce the best plans.

  • Only RRT is able to solve this planning problem.
  • Considering solving a variety of problems, the parallel planner gives better default behavior. It won’t improve the path length of really hard problems where RRT is the only planner that succeeds but in cases where the planners that produce nicer plans succeed you can get the benefit of them with the fallback to RRT when they fail.
  • One downside to this implementation is that we always take as long as the longest planner and CHOMP consistently is the longest planner. Maybe we should consider not including it in the parallel set.
  • [Mark] We should consider different stopping criterias.
  • [Andy] Se have something like a library of stopping criteria already. Right now, what’s available is “run full time, take the shortest path” or “take the first solution and terminate early” - Add a default stopping criterion for parallel planning by AndyZe · Pull Request #1876 · ros-planning/moveit2 · GitHub
  • [Nathan] What does the logging look like for this? We need useful logging messages so users are not confused.
  • [Mark] What about a slider between Fast planning and Short path length in the RViz panel
  • [Sebastian Jahr] Not currently integrated into any ui, you have to change config params for this.
  • [Mark] Swept volume or an approximation of it would be a better heuristic than join angle path distance
  • [Nathan] Would like more tutorial content about this. Concerned it will be complex for users. It might be helpful to consider how this will look to users to evaluate what we want to make default.

[Tyler] Nightly debian builds: GitHub - moveit/moveit2_packages

[Henning] GSOC projects

  • [Sebastian Castro] pick_ik improvements like using NLOpt
  • [Nathan] servo pipeline refactor
  • [Sebastian Jahr] Support for different IK solvers in PILZ
  • [Henning] Logging system improvements
  • Sebastian Castro, Sebastian Jahr, and Andy Z would like to mentor