Re: [ompl-users] Edge checking in control based planners

[Peter M. <mit...@gm...>]

Hi mark, my apologies for not responding to this. In fact that does not do
what I want, I have a quite unusual planning problem where the validity of
an edge cannot be determined by discretizing and checking individual
states. But it was not a problem to implement this within OMPL.

On Sun, Dec 8, 2019 at 6:26 PM Mark Moll <mm...@ri...> wrote:

> Hi Peter,
>
> There is no notion of a motion validator for kinodynamic planning on OMPL.
> Instead, the state validator is called after each call to the state
> propagator. The propagation step size should thus be sufficiently small so
> that the max. possible change in position is roughly equal to the collision
> checking resolution. Don’t confuse propagation step size with integration
> step size, though. Integration step size is often on the order of magnitude
> of .01 or smaller. The propagation step size can often be safely set to be
> an order of magnitude larger. The SpaceInformation::propagateWhileValid
> methods take as one of their arguments the max. number of steps (i.e.,
> calls to the state propagator) that will be taken. It seems like calling
> that method should give you what you want, no?
>
> Best,
>
> Mark
>
>
>
> On Nov 22, 2019, at 7:47 AM, Peter Mitrano <mit...@gm...> wrote:
>
> Hi Mark & others, I'm wondering if there is a hook for implementing
> continuous edge checking in the control based planners. Specifically, I
> want to validate a (x, u, x') tuple. A motion validator only validates (x,
> x') and state validity is of course just (x). At the moment, I am
> implementing this by setting the state to be infinity inside my
> propagate function, and using a state validity checker to check that
> result. I feel a much cleaner method would be to make propagate return bool
> instead of void, and then propagateWhileValid would check that in addition
> to a stateValidity checker. Is there a better way to do it?
>
> Thank you,
> -Peter Mitrano
> _______________________________________________
> ompl-users mailing list
> omp...@li...
> https://lists.sourceforge.net/lists/listinfo/ompl-users
>
>
>