Menu
▾
▴
Re: [ompl-users] Workspace Planning
|
From: Mark M. <mm...@ri...> - 2021-08-04 15:40:11
|
Hi Adam, Expanding on Jonathan just wrote, you can combine his idea with the type of local optimization that is typically done in a post processing stage. There is a PathSimplifier::findBetterGoal <https://github.com/ompl/ompl/blob/main/src/ompl/geometric/PathSimplifier.h#L219> method that gets called by PathSimplifier::simplify to find a better IK solution that can be directed to the rest of the path. Here, “better” means the overall path cost (not just the cost of a state) as defined by the optimization objective (which is path length by default). For this to work, you need a goal of a type derived from GoalSampleableRegion <http://ompl.kavrakilab.org/classompl_1_1base_1_1GoalSampleableRegion.html>. The derived class GoalLazySamples might be a good start; all you need is a functor that returns IK solutions. Re. constrained planning, that often assumes the constraints apply to the whole path, not just the goal state, so this might not be what you want. (Side note: although it’s often convenient to reason lower-dimensional constraint manifolds, the ProjectedStateSpace <http://ompl.kavrakilab.org/classompl_1_1base_1_1ProjectedStateSpace.html> does not assume the set of constraint-satisfying configurations is a manifold or low-dimensional). Best, Mark > On Aug 4, 2021, at 7:31 AM, Jonathan Gammell <ga...@ro...> wrote: > > Hi Adam, > > This is not a complete solution, but an engineering solution to your problem would be to generate a sufficient number of goal arm configurations and then use a planning algorithm capable of solving single-start, multiple-goal problems to find the optimal path to that discrete set of goals. > > I don't know all the algorithms that would be appropriate, but I do know that BIT* (and similar algorithms, like AIT*) can handle multiple goals (and also multiple starts). > > Hope that helps, > Jon > > On 04/08/2021 14:32, omp...@li... wrote: >> Send ompl-users mailing list submissions to >> omp...@li... >> >> To subscribe or unsubscribe via the World Wide Web, visit >> https://lists.sourceforge.net/lists/listinfo/ompl-users >> or, via email, send a message with subject or body 'help' to >> omp...@li... >> >> You can reach the person managing the list at >> omp...@li... >> >> When replying, please edit your Subject line so it is more specific >> than "Re: Contents of ompl-users digest..." >> >> >> Today's Topics: >> >> 1. Workspace Planning (Adam Fishman) >> >> >> ---------------------------------------------------------------------- >> >> Message: 1 >> Date: Tue, 3 Aug 2021 11:44:10 -0700 >> From: Adam Fishman <afi...@cs...> >> To: omp...@li... >> Subject: [ompl-users] Workspace Planning >> Message-ID: <949...@cs...> >> Content-Type: text/plain; charset="utf-8" >> >> Hi all, >> >> I am trying to do optimal workspace planning with OMPL for a robotic manipulator, but I haven?t found many resources on how to do this. FWIW I am using the direct Python bindings for OMPL due to my system constraints (I can?t use ROS for my task). >> >> The simplest way to do it seems to be to use IK to get a target configuration and then to use something like BIT* to plan. The issue with this, however, is that the IK solution may not itself be optimal. That is, the optimal path to one IK solution may be shorter than the optimal path to another. And, finding the ideal IK solution seems to be a planning problem itself. >> >> I?ve been looking into Constrained Planning, but the workspace is not exactly a zero-volume constraint manifold, so I?m not sure how to fit this into the framework. >> >> I imagine this is a problem that?s been explored extensively, but I?m not sure how to proceed. Does anyone have suggestions? >> >> Thanks a lot, >> Adam Fishman >> He/Him >> fishbotics.com <http://fishbotics.com/> >> -------------- next part -------------- >> An HTML attachment was scrubbed... >> >> ------------------------------ >> >> >> >> ------------------------------ >> >> Subject: Digest Footer >> >> _______________________________________________ >> ompl-users mailing list >> omp...@li... >> https://lists.sourceforge.net/lists/listinfo/ompl-users >> >> >> ------------------------------ >> >> End of ompl-users Digest, Vol 108, Issue 1 >> ****************************************** > > -- > Jonathan Gammell, Ph.D. > > Departmental Lecturer in Robotics > Estimation, Search, and Planning (ESP) Research Group > Oxford Robotics Institute (ORI) > Department of Engineering Science > University of Oxford > office: +44 1865 613082 > skype: jdgammell > https://robotic-esp.com/gammell > https://twitter.com/robotic_esp > > > > _______________________________________________ > ompl-users mailing list > omp...@li... > https://lists.sourceforge.net/lists/listinfo/ompl-users |
