I have seen it used in automated story-telling (or game AI to control NPCs), and in NLG systems, where the generation of text is reinterpreted as a planning task.
What these systems have in common is that they're either off-line or in a simple environment (NPC control). I'm not sure they would be suitable for real-time applications, unless you can be sure that a feasible plan exists which can be found within certain time bounds. I wouldn't want to sit in a car going at high speed on the motorway and waiting for the driving unit to work out a plan how to avoid an obstacle that suddenly appears on the road.