Re: [ompl-users] Control planner with optimization objective?

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

If you google for “optimal kinodynamic planning”, you’ll find a number of relevant papers, including some by Frazzoli et al., Bekris et al., and Van den Berg et al. Bekris’ group also has some software available:

http://pracsyslab.org/sst_software

--
Mark Moll



> On Dec 14, 2014, at 7:00 PM, Eric Purdy <ep...@uc...> wrote:
> 
> Can you recommend any code/papers that implements a control planner which finds an optimal path?
> 
> On Sun, Dec 14, 2014 at 3:09 PM, Mark Moll <mm...@ri...> wrote:
> Hi Eric,
> 
> > On Dec 14, 2014, at 12:43 AM, Eric Purdy <ep...@uc...> wrote:
> > I want to solve a problem with controls, but also favor paths that use less fuel. Is it possible to use an optimization objective with a control planner? RRT (the control version) seems to ignore an optimization objective that I specified, while RRTstar (from geometric) ignores the control aspect of the problem and just finds a path through the state space.
> >
> > I am working with the Python bindings, if that makes a difference.
> 
> It would be very nice to have implementations of asymptotically optimal kinodynamic planning algorithms in OMPL, but we currently don’t. Such algorithms do exist, though, so you could implement one of them. Alternatively, you can find a kinematically feasible path first and then reparametrize the path such that the trajectory locally optimizes your objective. If your system has car-like dynamics, you could use the DubinsStateSpace or ReedsSheppStateSpace to get a path with a bounded radius of curvature.
> 
> --
> Mark Moll
> 
> 
> 
> 
> 
> --
> -Eric