Learning Safe Numeric Action Models

Authors: Argaman Mordoch, Brendan Juba, Roni Stern

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, however, N-SAM can quickly learn a safe action model that can solve most problems in the domain. ... Experimental Results Although we do not obtain theoretical completeness guarantees for worst-case distributions, whether or not N-SAM learns applicable action models in practice is an empirical question. Thus, we implemented N-SAM and evaluated its performance on 12 benchmark domains...
Researcher Affiliation Academia Argaman Mordoch1, Brendan Juba2, Roni Stern1 1Ben Gurion University in Be er Sheva, Israel 2Washington University in St. Louis, USA
Pseudocode No The paper describes the N-SAM algorithm in prose and provides a graphical illustration (Figure 1), but it does not include formal pseudocode or a clearly labeled algorithm block.
Open Source Code Yes N-SAM s code is available in the link https://github.com/Search-BGU/numeric-sam.
Open Datasets Yes We implemented N-SAM and evaluated its performance on 12 benchmark domains, namely Depot, Driverlog, Zenotravel, Rovers, Satellite, Settlers, and UMT domains from the 3rd International Planning Competition (IPC3) (Long and Fox 2003), and Farmland, Counters, Plant-watering, and Sailing from (Scala et al. 2017).
Dataset Splits Yes These problems in each domain were split to train and test using a 5-fold cross-validation.
Hardware Specification Yes All experiments were run on a Linux machine with 8 cores and 16 GB of RAM.
Software Dependencies Yes To solve problems, we run two well-known planners with a timeout of 60 seconds: Metric-FF (version 2.1) (Hoffmann 2003) and ENHSP (Li et al. 2018).
Experiment Setup Yes To solve problems, we run two well-known planners with a timeout of 60 seconds: Metric-FF (version 2.1) (Hoffmann 2003) and ENHSP (Li et al. 2018). For Metric-FF, we used BFS with no cost minimization and running configuration EHC+H. For ENHSP, we used Greedy Best First Search with the MRP heuristic and helpful actions.