Re: [ompl-users] MATLAB interfacing

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

Hi Ros,

What you’re asking is possible but challenging. It is difficult to set up a fair comparison. A naive Matlab-OMPL interface would likely involve copying/converting data from one representation to another. If this is done via callback functions that are called often (e.g., for state validation or collision checking), then a performance comparison between Matlab code and OMPL would not be very meaningful. If your Matlab reads the environment from a file in a standard mesh format, then the ompl_app library could be used to do the same on the OMPL side.

It is typically difficult to generalize results for a 2D point robot to more realistic scenarios. You may want to look at Matlab interfaces for, e.g., ROS and V-REP. These interfaces provide only high-level access (I think), so it still wouldn’t be easy to compare your planning methods against OMPL if your methods are implemented in Matlab.

Best,

Mark



> On Aug 1, 2019, at 11:31 AM, R Shinkle <ros...@gm...> wrote:
> 
> Greetings OMPL geniuses, 
> 
> I’m hoping to use OMPL’s RRT (and variants) as a baseline against which I can test my planning methods. All of my testing is done in MATLAB right now, so I’d like to use mex files to link to the C++ files. This is all in 2D. 
> 
> The plan right now is to set up the planner with simple setup in a C++ file, and interface with that C++ file through MATLAB’s mex files. 
> 
> The tricky part is that I would like to be able to send the information about the environment (size and position of obstacles) from MATLAB to the planner. 
> 
> This will be my first time interfacing between C++ and MATLAB (although I’ve written both individually) and also my first time using OMPL. Has anyone done any information passing for obstacles like this in the past? Will it be possible? Any suggestions on where to begin? 
> 
> Thank you!
> 
> Ros
>