Best-First Width Search: Exploration and Exploitation in Classical Planning

Authors: Nir Lipovetzky, Hector Geffner

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

Reproducibility Variable Result LLM Response
Research Type Experimental We show first that width-based exploration in GBFS is more effective than GBFS with local GBFS search (GBFS-LS), and then proceed to formulate a simple and general computational framework where standard goal-oriented search (exploitation) and widthbased search (structural exploration) are combined to yield a search scheme, best-first width search, that is better than both and which results in classical planning algorithms that outperform the state-of-the-art planners. Table 1 shows the results of the baseline GBFS in comparison with GBFS-LS and GBFS-W.
Researcher Affiliation Academia Nir Lipovetzky The University of Melbourne Melbourne, Australia nir.lipovetzky@unimelb.edu.au Hector Geffner ICREA & Universitat Pompeu Fabra Barcelona, Spain hector.geffner@upf.edu
Pseudocode No The paper describes algorithms conceptually but does not include any explicit pseudocode blocks or figures.
Open Source Code No BFWS algorithms and BFS(f) implemented using LAPKT toolkit. (Refers to a toolkit used, not specific code release for this paper's methods). Ramirez, M.; Lipovetzky, N.; and Muise, C. 2015. Lightweight Automated Planning Tool Ki T. http://lapkt.org/. (This is a general toolkit website, not a direct link to the code for this paper's methods/experiments).
Open Datasets Yes The domains and instances are from the 2014 planning competition (Vallati et al. 2015).
Dataset Splits No The paper mentions using domains and instances from the 2014 planning competition but does not specify any training, validation, or test dataset splits.
Hardware Specification Yes Experiments performed on 2.40GHz Intel Processor; time and memory outs after 30 min or 8GB. All experiments performed on 2.40GHz Intel Processor, with time and memory outs after 30 min and 8GB resp. as in last IPC.
Software Dependencies No BFWS algorithms and BFS(f) implemented using LAPKT toolkit. (No version number for LAPKT or other software dependencies are provided).
Experiment Setup Yes We have evaluated this question experimentally using the same settings and parameters as those reported for GBFS-LS; namely, STALL-SIZE=1000, LS-SIZE=1000, and MAX-LOCAL-TRY=100. By default, novelty measures w(s) are computed with 2-level precision only; namely, w(s) is determined to be 1 or greater than 1. This includes their use in GBFS-W.