Chatterjee

The efficiency of heuristic search depends dramatically on the quality of the heuristic function. For an optimal heuristic search, heuristics that estimate cost-to-goal better typically lead to faster searches. For a sub-optimal heuristic search such as weighted A*, the search speed depends more on the correlation between the heuristic and the true cost-to-goal. In this extended abstract, we discuss our preliminary work on computing heuristic functions that exploit this fact. In particular, we introduce a many-to-one mapping from an original search space to a conservative abstract space.