Here is an interesting little tidbit on Microsoft Research about shortest path algorithms. I have found this entire problem space interesting ever since I was involved in building an application to support warehousing and distribution and saw that despite the software that we were building it was still going to be a very manual and error prone process.
In the future (when cars drive themselves) being able to optimise for shortest path based on a variety of conditions (traffic, cost, weather, other cars) is going to be important – at a macro level its probably more important than collision detection algorithms and obstacle avoidance algorithms.