Best-first utility-guided search

• Authors: Wheeler Ruml;Minh B. Do
• Venue: IJCAI'07 Proceedings of the 20th international joint conference on Artifical intelligence
• Year: 2007

In many shortest-path problems of practical interest, insufficient time is available to find a provably optimal solution. One can only hope to achieve a balance between search time and solution cost that respects the user's preferences, expressed as a utility function over time and cost. Current stateof-the-art approaches to this problem rely on anytime algorithms such as Anytime A* or ARA*. These algorithms require the use of extensive training data to compute a termination policy that respects the user's utility function. We propose a more direct approach, called BUGSY, that incorporates the utility function directly into the search, obviating the need for a separate termination policy. Experiments in several challenging problem domains, including sequence alignment and temporal planning, demonstrate that this direct approach can surpass anytime algorithms without requiring expensive performance profiling.