Revision history [back]

You are definitely going to need more elaborate motion planners. I suggest you take a look at the SBPL and OMPL packages and also the navigation_experimental stack. The first two packages are able to implement sampling based planners, which are well suited for navigation in complex, changing environments. You can then find a way to implement different metrics for path evaluation, incorporating probabilistic and dynamic information of the environment. The navigation_experimantal stack has the eband_local_planner, which can locally change the robot's path to avoid moving objects.

As far as I know, ROS does not have a ready to use motion planner for dynamic environment. If you want to stick with ROS architecture, my suggestion is that you should think about a global planner that is able to plan considering predictions on the motion of moving objects and a local planner that is able to escape from unpredicted situations.