Operator Mutexes and Symmetries for Simplifying Planning Tasks

Authors: Daniel Fišer, Álvaro Torralba, Alexander Shleyfman7586-7593

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We propose four different methods for inference of operator mutexes and experimentally verify that they can be found in a sizable number of planning tasks. Experimental Results The proposed methods were implemented in C and the source code is publicly available1. Table 1 summarizes the results on the number of inferred op-mutex pairs (in thousands).
Researcher Affiliation Academia Daniel Fiˇser Czech Technical University in Prague, Faculty of Electrical Engineering, Prague, Czech Republic, Alvaro Torralba Saarland University, Saarland Informatics Campus, Saarbr ucken, Germany, Alexander Shleyfman Technion Haifa, Israel
Pseudocode Yes Algorithm 1: Fixpoint computation of a redundant set.
Open Source Code Yes The proposed methods were implemented in C and the source code is publicly available1. 1https://gitlab.com/danfis/cplan.git, branch aaai19
Open Datasets Yes We used all IPC benchmarks 2006 2018 from the optimal track.
Dataset Splits No The paper discusses 'train', 'validation', and 'test' as abstract components of a planning task (e.g., 'train' as a set of operators, 'validation' and 'test' as conceptual stages in a planning process like 'test set' for evaluation), but does not provide explicit details about data splits (e.g., percentages or counts) for experimental datasets.
Hardware Specification Yes The experiments were performed on a cluster with Intel E5-2670 2.6 GHz processor with 8 GB memory limit for each task.
Software Dependencies No The paper mentions implementation in C and using the BLISS library, but does not provide specific version numbers for these software components.
Experiment Setup Yes The experiments were performed on a cluster with Intel E5-2670 2.6 GHz processor with 8 GB memory limit for each task. We set the time limit to 30 minutes for the whole planning process. We used A* with the LM-Cut (lmc) heuristic, the merge-and-shrink (m&s) heuristic with SCCDFP merge strategy and non-greedy bisimulation shrink strategy, and the potential (pot) heuristic optimized for all syntactic states.