Handling Numeric Criteria in Relaxed Planning Graphs

Nowadays, one of the main techniques used in heuristic planning is the generation of a relaxed planning graph, based on a Graphplan-like expansion. Planners like FF or MIPS use this type of graphs in order to compute distance-based heuristics during the planning process. This paper presents a new approach to extend the functionality of these graphs in order to manage numeric optimization criteria (problem metric), instead of only plan length optimization. This extension leads to more informed relaxed plans, without increasing significantly the computational cost. Planners that use the relaxed plans for further refinements can take advantage of this additional information to compute better quality plans.