MoveIt Maintainer Meeting - Sep 22

MoveIt
1

The next Manipulation WG Meeting / MoveIt Maintainer Meeting will be 2022-09-22T15:00:00Z.

Agenda

  • Old Business
  • MoveIt Updates
    • Parallel Planning & better defaults (Sebastian)
    • gd_ik (Tyler)

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#

2

Notes

  • New maintainer: Sebastian Jahr
  • [Sebastian] Better default plans
    • People are surprised by default planning behavior, “non-intuitive motions”
    • If we use KDL as default IK solver = non-deterministic IK solutions
      • leads to large joint jumps with small or no difference in end effector pose
      • No solution other than changing default IK from KDL
    • Parallel planning pipelines Parallel planning pipelines by sjahr · Pull Request #1420 · ros-planning/moveit2 · GitHub
      • plan with multiple algorithms in parallel and compare the output to give user the “best” plan
      • best is determined by cost functions run over the resulting trajectories from the planners
      • ompl/RRTConnect + ompl/PRMstarr + pilz/LIN
      • termination criteria - lambda that determines when to terminate planning, test for acceptable plan from any planner
      • can choose smoothest, shortest, or any
      • keep memory of which planner creates best solution for your problem
      • [Simon] doesn’t this exist in ompl
        • yes, but this allows to use different moveit planners in parallel, not just different ompl planners
      • [Andy] Can Carteasian planner be added as one of the parallel planners, a lot of times if Cartesian path would succeed that is what users would like. Pilz is similar to Cartesian planner and solves a similar problem.
      • [Andy] We should be careful about two pipelines using the same data file for the roadmap at the same time and warn users against doing that.
      • [Sebastian] We could make multiple ompl pipelines wait on each-other to solve that. Then encourage users to use parallel features of ompl with a single pipeline.
      • [Sebastian] Many people find shortest path to be the most intuitive. Working on benchmarking and testing this feature to figure out what good defaults are.
      • [Tyler] I like this because it is moving towards an interface where the user specifies what is important to them and we help them choose the right algorithm solutions for what they would like to achieve.
      • [Andy] It would be really useful to specify which planner you would like solutions from in order from pilz → RRT.
      • [Sebastian] That is possible with this.
      • [Henning] Did you look into having RRTConnect start with a straight line interpolation?
      • [Sebastian] To use this feature we need to add this to the ompl interface potentially as a parameter. I did look into this and sadly this is a larger API change. I have it on my todo list. I’m tracking my effort for this in issues on my fork of MoveIt.
  • [Tyler] New/Old IK solver - coming soon
  • [Andy] Enable splicing trajectories without re-prameterization
    • [Henning] Only works if you already have parameterized trajectories and you want to join them and only parameterize where they join and not the whole new trajectory. Useful when combining different types of trajectories.
  • [Henning] Simon, how is arm to arm collision checking done in the GSOC project
    • Done per trajectory point. If you have a sparse trajectory, good luck. Checks all points against all points of other trajectories.

Attendees

  • Vatan Aksoy Tezer
  • Sebastian Jahr
  • Nathan Brooks
  • Henning Kayser
  • Andy Zelenak
  • Simon Schmeisser
  • Tyler Weaver