A Novel Technique for Avoiding Plateaus of Greedy Best-First Search in Satisficing Planning
Tatsuya Imai, Akihiro Kishimoto · Proceedings of the AAAI Conference on Artificial Intelligence · 2011
Greedy best-first search (GBFS) is a popular and effective algorithm in satisficing planning and is incorporated into high-performance planners. GBFS in planning decides its search direction with automatically generated heuristic functions. However, if the heuristic functions evaluate nodes inaccurately, GBFS may be misled into a valueless search direction, thus resulting in performance degradation. This paper presents a simple but effective algorithm considering a diversity of search directions to avoid the errors of heuristic information. Experimental results in solving a variety of planning problems show that our approach is successful.