Revision history [back]

So, I've got a few thoughts on this:

1. Have you seen the discussion about the navigation stack and nonholonomic robots? You can find it in the ros-users archive. If you read through it far enough, you'll get to some discussion about the spin-in-place scoring - basically the base_local_planner isn't the best at it so it does sometimes choose poor rotations. I don't know if this was ever solved, perhaps in the dwa_local_planner available in navigation_experimental.
2. Have you tried to visualize the potential fields that you are creating with your scoring parameters? See ticket #4620 for some examples. I believe that functionality is in the released version of base_local_planner, but you'll have to check out visualization trunk in order to get a version of rviz that has had the patch from ticket #4610 applied. Basically, use this as a sanity check to make sure that the final combination actually keeps the trajectories out of the "high cost" areas of the map.
3. Eitan wrote up a nice basic navigation stack tuning guide a bit ago. Have you guys looked through those hints?
4. Not sure if it would help you, but you can find the navigation stack settings we used last year at IGVC 2010 in the cwru-ros-pkg (pick the version of any file with "outdoors" in the name if there is a version with and without that in the name). Those were working reasonably well for us at IGVC until our motor controller blew a FET and we had to switch to one that could only do about half the current we needed :(. We didn't really have any issues with moving into "high cost" areas of the map or spinning 360 degrees when we were almost on the path and still had a good deal left to go. The only time it did a poor spin in place was at goal points, start points or any other place where the path made sense to include a spin-in-place.

Hope something in there helps you guys out.