MERWIN Planner: Mercury Enchanced With Novelty Heuristic

Abstract

Heuristic search with red-black planning heuristics is among the most effective approaches to satisficing planning and the driving power behind the state-of-the-art satisficing planner Mercury. Another recent success in satisficing planning is due to the introduction of novelty based heuristic guidance, in particular a guidance measuring the novelty of a heuristic estimate in a state. nn A satisficing planner that we baptize MERWIN empowers red-black planning heuristics with novelty based guidance, measuring the novelty of red-black planning heuristic estimates in explored states. MERWIN planner partitions the state space into novelty layers, expanding the most novel nodes first, and breaking ties within each layer by the redblack heuristic values.

Related