Re: [ompl-users] Integrating geometric rrt with another planner

[Mark M. <mm...@ri...>]

Hi Hemang,

> On May 26, 2017, at 12:42 PM, Hemang Chawla <hem...@gm...> wrote:
> I am trying to plan for a non-holonomic wheeled mobile manipulator.
> 
> For planning the path of the base, the control based planners ma be used. However they cannot be extended directly to mobile manipulators due to higher dimensionality.

It *can*, but it will likely be very slow. If the type of problems you want to solve can be solved quasi statically, then the approach you suggest is likely the better way to go.

> 
> On the contrary, the planning with reed shepps planner uses geometric planning. I wish to plan for the base and the arm simultaneously. Is there a way to plan for a robot with base in SE(2) planned with the rest of the joints in R^6 ? Perhaps a new state space will have to be created for the same?

You can create a CompoundStateSpace <http://ompl.kavrakilab.org/classompl_1_1base_1_1CompoundStateSpace.html> with DubinsStateSpace and R^6 as the subspaces. See the tutorial at http://ompl.kavrakilab.org/workingWithStates.html <http://ompl.kavrakilab.org/workingWithStates.html>.


> I recognize the drawback here is the large amount of time it may take as the planning with reed shepps is not optimized. However I wish to plan only in a small area of a couple meters^2.

I’m not sure what you mean with "planning with reed shepps is not optimized.” You should be able to use RRT* or other optimizing planners with DubinsStateSpace (or a CompoundStateSpace with a DubinsStateSpace as a subspace).

Mark