Incorporating Domain-Independent Planning Heuristics in Hierarchical Planning

Authors: Vikas Shivashankar, Ron Alford, David Aha

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical results with three benchmark domains demonstrate that simultaneously leveraging hierarchical knowledge and heuristic guidance substantially improves planning performance. Our experiments on three benchmark domains demonstrate that HOGL s ability to simultaneously leverage hierarchical knowledge and DIP heuristics allows it to outperform state-of-the-art optimal DIP algorithms and blind search HGN planning algorithms.
Researcher Affiliation Collaboration Vikas Shivashankar Knexus Research Corporation National Harbor, MD vikas.shivashankar@knexusresearch.com Ron Alford MITRE Mc Lean, VA ralford@mitre.org David W. Aha Navy Center for Applied Research in AI Naval Research Laboratory Washington, DC david.aha@nrl.navy.mil
Pseudocode No The paper describes procedures and mathematical formulations but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' block, nor structured code-like steps.
Open Source Code No The paper states 'we implemented HOGL' but does not provide any link or explicit statement about making the source code for HOGL publicly available. Footnote 1 refers to a third-party planning system (Fast-Downward) used for comparison, not the authors' own code.
Open Datasets Yes Our empirical results with three benchmark domains... Logistics. We ran the planners on 25 randomly generated problems for each problem size ranging from 4 . . . 10. The HGN domain models contained three methods... Blocks-World. We ran the planners on 25 randomly generated problems for each problem size ranging from 4, 6, . . . , 20. We used HGN methods from the GDP planner (Shivashankar et al. 2012)... Depots. Like in the first two domains, we randomly generated 25 problems for each problem size (num. packages) ranging from 4 . . . 10.
Dataset Splits No The paper describes generating '25 randomly generated problems for each problem size' in the Logistics, Blocks-World, and Depots domains. It does not mention explicit training, validation, or test dataset splits in the conventional sense of partitioning a fixed dataset.
Hardware Specification Yes We ran all problems on a Xeon E5-2639 with a per problem limit of 8 GB of RAM and 25 minutes of planning time.
Software Dependencies No The paper mentions using the 'Fast-Downward planning system' for a baseline comparison but does not provide specific version numbers for it or any other software dependencies, including those for their own implemented planner, HOGL.
Experiment Setup Yes For each domain, we randomly generated 25 problem instances per problem size. We ran all problems on a Xeon E5-2639 with a per problem limit of 8 GB of RAM and 25 minutes of planning time.