Abstraction Heuristics for Classical Planning Tasks with Conditional Effects

Authors: Martín Pozo, Jendrik Seipp

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments show that these heuristics are competitive with and often surpass the state-of-the-art for conditional-effect tasks.
Researcher Affiliation Academia 1Universidad Carlos III de Madrid 2Link oping University EMAIL, EMAIL
Pseudocode Yes Algorithm 1: CEGAR loop for a task Π. Algorithm 2: Compute outgoing transitions from abstract state a via operator o in a projection to pattern P. Algorithm 3: Compute outgoing transitions from abstract state a via operator o in Cartesian abstraction α.
Open Source Code Yes All code, benchmarks, and experiment data is available online [Pozo and Seipp, 2025].
Open Datasets Yes As our benchmark set, we use the domains with conditional effects from the last two IPCs, the domains used by R oger et al. [2014], the matrix multiplication domain [Speck et al., 2023], and the domain for transforming quantum circuits into CNOT-only layouts [Shaik and van de Pol, 2024].
Dataset Splits No The paper uses established benchmark domains for planning tasks, which are typically defined by a set of problems rather than explicit training/validation/test splits in the machine learning sense. No specific dataset split information (percentages, sample counts, or explicit split files) is provided in the text for reproducibility.
Hardware Specification Yes We use Downward Lab [Seipp et al., 2017] and run experiments on Xeon Gold 6130 processors.
Software Dependencies No The paper mentions using the "Scorpion planning system [Seipp et al., 2020a], which is an extension of Fast Downward [Helmert, 2006]" and the "h2 preprocessor [Alc azar and Torralba, 2015]", but specific version numbers for these software components are not provided in the text.
Experiment Setup Yes For all configurations we use the h2 preprocessor [Alc azar and Torralba, 2015] and limit time and memory for each planner run to 30 minutes and 8 Gi B, respectively. ... For M&S, we use the recommended values for all parameters [Sievers, 2018] and up to 100 000 abstract states. For Cartesian abstractions, we use incremental search to find abstract plans [Seipp et al., 2020b] and a time limit of 900 seconds for the CEGAR loop.