Revision history [back]

DimitriProsser's answer is basically correct, I just wanted to add some additional information.

It seems the length of the portion of the global plan that move_base passes to the local planner depends on the size of the local costmap. After doing some more research on the way the TrajectoryPlannerROS is working I found out that the goal distance is not calculated through obstacles. It's more like grid cell steps in non-obstacle-cells. This leads to the conclusion, that my approach of shortening the plan is mostly useless.

However I was able to dynamically shorten the path inside the local planner, it is not necessary to modify move_base. In combination with an approach I call adaptive sim_time I think the navigation works better now. I'm still working on my modified planner, so I can't yet tell if it really improves the planner or it is just in my special case.