[ompl-users] Strange Performance Issues

[daniel S. <tya...@ya...>]

Dear OMPL Users,

I implemented the ConnectRRT( or BiRRT ) based on the demos. (http://ompl.kavrakilab.org/RigidBodyPlanning_8cpp_source.html ,   95% same code)  The problem I have is the performance. I know the performance is dependent on the used collision detection but the performance of the CD is not the problem. The problem it is the amount of samples that are needed to solve the problem.


In first place I want to solve the easy alpha puzzle. The alpha 1.5. If using the RRT Connect the algorithm needs about 50.000-150.000 states and ~30-60sec time  to solve the alpha 1.5 puzzle. I think thats way over the top and I 'm pretty sure I'm doing something wrong.

Some details:
Geomerty of alpha Puzzle is from your OMPLapp/resources
CD is algorithm of our university Library.
The defined center of rotation = center of the Bounding Box of the robot
Planning Box area in the workspace is 3x Bounding Box of the robot.
Start point is almost (0/0/0) and goal is the nail outside the other obstacle nail (both points collision free)

Some things that are suspicious:
-Of course the amount of needed samples
-      The cout:
       Debug: Planner Range detected to be 0.660569
       Info: Staring with 1 states
       ...

I defined a start and goal, so why is the algorithm starting with one state?


I put the relevant code in the appendix although, as noted, it is 95% demo code.

Thanks in advance for any help! 

Daniel