Delete-Relaxation Heuristics for Lifted Classical Planning

Abstract

Recent research in classical planning has shown the importance of search techniques that operate directly on the lifted representation of the problem, particularly in domains where the ground representation is prohibitively large. In this paper, we show how to compute the additive and maximum heuristics from the lifted representation of a problem. We do this by adapting well-known reachability analysis techniques based on a Datalog formulation of the delete relaxation of the problem. Our adaptation allows us to obtain not only the desired heuristic value, but also other useful heuristic information such as helpful actions. Our empirical evaluation shows that our lifted version of the additive heuristic is competitive with its ground counterpart on most of the standard international competition benchmarks, and significantly outperforms other state-of-the-art lifted heuristic methods in the literature.

Encoding Lifted Classical Planning in Propositional Logic

“…We evaluate our satisficing configuration (Alg. 2) against the PWL planner with (1) lazy best first search, helpful operators and the lifted h add heuristic (Corrêa et al 2021) , and (2) eager best first search with Goal Counting (GC). Further we use GC with the unary relaxation heuristic as tiebreaker, with (ur-d) and without (ur) disambiguation of static predicates (Lauer et al 2021).…”

“…Corrêa et al (2020) introduced an approach to generate successor states in this process efficiently. Heuristics to guide the search have been presented by Corrêa et al (2021) and Lauer et al (2021). The former computes the Add and the Max heuristic (Bonet and Geffner 2001) on the lifted model.…”

Best-First Width Search for Lifted Classical Planning

“…The alternative to grounding is to use planners that rely exclusively on the PDDL representation and avoid grounding altogether. Recent works show that such lifted planners outperform ground planners in hard-toground tasks (Corrêa et al 2021;Lauer et al 2021).…”

“…So far, two approaches for this challenge have been presented. First, we can compute the same heuristics used for ground planning, but this can be more expensive than the same computation on the ground task (Corrêa et al 2021). Second, we can simplify the lifted representation of the task before computing a heuristic with the price of reducing the informativeness of the resulting heuristic (Ridder and Fox 2014;Lauer et al 2021).…”

Classical Planning as QBF without Grounding

“…The simplified domains submitted to IPC-18 received an outstanding domain submission award. In addition, we also consider other Hard-to-ground domains (htg): Genomeedit-distance (without costs), Pipesworld from Corrêa et al (2021) and Childsnack, Visitall, Blocks and Logistics from Lauer et al (2021). For these domains, we consider a subset of instances which capture the hardness in grounding.…”