Using Swamps to Improve Optimal Pathfinding (Extended Abstract)

We address the problem of quickly finding shortest paths in known graphs. We propose a method that relies on identifying areas that tend to be searched needlessly (areas we call swamps), and exploits this knowledge to improve search. The method requires relatively little memory, and reduces search cost drastically, while still finding optimal paths. Our method is independent of the heuristics used in the search, and of the search algorithm. We present experimental results that support our claims, and provide an anytime algorithm for the pre-processing stage that identifies swamps.