From: Gustavo G. <gus...@gm...> - 2012-03-02 16:51:10
|
I corresponded with the Marius, the author of FLANN. For now it doesn't support incrementally building the index, but that is a high-priority feature on the to-do list. I'd appreciate any guidance on adding bindings to specific instances of nearest neighbors -- namely NearestNeighborsGNAT<Motions*>, which is used by the RRT* implementation. On Mon, Feb 27, 2012 at 10:21 AM, Gustavo Goretkin <gus...@gm...> wrote: > I get the impression that FLANN builds the data structure in batch and > doesn't support adding points (as is required for sample-based motion > planning) . I've asked the author about this. Have you used FLANN for > motion planning? > > On Sun, Feb 26, 2012 at 5:43 AM, Gustavo Goretkin > <gus...@gm...> wrote: >> For on, I'd like to visualize the entire tree, not just the solution >> path. I would also like to test out the NearestNeighbors algorithms on >> my own distance metric (so yes, Euclidian metric is not what I need) >> >> Is there any reason to avoid exposing the data structures in OMPL? >> >> I am working on implementing this: >> http://people.csail.mit.edu/aperez/www/stuff/perez_icra12.pdf with the >> metric calculation happening in Python. Eventually, I'd be interested >> in trying to implement it all within the framework of OMPL. >> >> Thanks, >> Gustavo >> On Sat, Feb 25, 2012 at 11:12 PM, Mark Moll <mm...@ri...> wrote: >>> It should be possible.The NearestNeighbors classes are templated, though. You can only export fully instantiated types to python. That is, you can only export NearestNeighborsGNAT<int>, NearestNeighborts<State*>, etc. You could also look at the flann library, which already has python bindings. I could be wrong, but I believe that flann assumes a Euclidean distance metric, whereas GNAT doesn’t. It this is something you need (for rotations and rigid body movements, for instance), then GNAT could work. >>> >>> Can you describe how you’d plan to use this? If this is something that might conceivably be useful for others, we could add the python bindings to OMPL. >>> >>> On Feb 25, 2012, at 1:27 PM, Gustavo Goretkin wrote: >>> >>>> I'd like to build Python bindings to the different implementations of >>>> NearestNeighbors (specifically GNAT). I'd like to be able to perform >>>> queries on and additions to the NearestNeighbors data structure and >>>> also iterate through all the Motion objects in the data structure. Are >>>> there straightforward changes I could make to generate_bindings.py? >>>> >>>> I was peering in ompl/py-bindings/bindings/geometric and I see that >>>> there are files like NearestNeighbors.pypp.cpp . Does this mean that >>>> some of these bindings I want are already being generated? >>>> >>>> Thank you! >>>> Gustavo >>> >>> -- >>> Mark Moll >>> >>> >>> |