Path - based Heuristics ( Preliminary

A path in a classical planning problem is a sequence of causal links ai, pi, ai+1 that connects a given state with the goal. C3 is a recent based planner that computes paths from the initial state and uses them to recursively decompose the problem into subproblems (Lipovetzky and Geffner 2009). Paths are ranked with an heuristic that takes deletes into account and is built on top of a base heuristic, in this case, the additive heuristic. In this work, we use the resulting path-heuristic to define a state heuristic that we compare with the additive heuristic in the context of a standard greedy best first state planner. Interestingly, while the path-heuristic is an orderof-magnitude more expensive than the additive heuristic, it results in many less expanded nodes, and often appears to be cost-effective and produces better plans.