Automated Synthesis of Social Laws in STRIPS

Authors: Ronen Nir, Alexander Shleyfman, Erez Karpas9941-9948

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical Evaluation We now present an empirical evaluation of the techniques described above.
Researcher Affiliation Academia Ronen Nir, Alexander Shleyfman, Erez Karpas Technion Israel Institute of Technology Haifa, Israel {ronenn, karpase}@technion.ac.il, shleyfman.alexander@gmail.com
Pseudocode No The paper describes various algorithms and functions (e.g., IS ROBUST, FIND CONFLICT) in prose but does not include any structured pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not include any statement about releasing the source code for the described methodology or provide a link to a code repository.
Open Datasets Yes Following the work on social law robustness verification (Karpas, Shleyfman, and Tennenholtz 2017), we evaluate these configurations on benchmarks from the first Competition of Distributed and Multiagent Planners (Komenda, Stolba, and Kovacs 2016)
Dataset Splits No The paper evaluates its approach on 'benchmarks from the first Competition of Distributed and Multiagent Planners', but it does not specify any training, validation, or test dataset splits in terms of percentages, sample counts, or explicit splitting methodology.
Hardware Specification Yes All planners had a memory limit of 8GB on an Intel Xeon E5-2695 CPU running at 2.10GHz (limited to one core only).
Software Dependencies Yes we first run Fast Downward Stone Soup 2014 (R oger, Pommerening, and Seipp 2014) with a timeout of 1800 seconds... we then switch to running Sym PA (Torralba 2016), a planner from the 2016 unsolvability International Planning Competition (IPC)... We ran 16 instances at a time with the help of GNU Parallel (Ferrer-Mestres, Franc es, and Geffner 2017).
Experiment Setup Yes We set Sym PA with a similar timeout of 1800 seconds. The search stopped exploring nodes after 1800 seconds. All planners had a memory limit of 8GB on an Intel Xeon E5-2695 CPU running at 2.10GHz (limited to one core only). In order to exploit these preferred operator criteria, we use a multi-queue mechanism. Since we have 2 criteria we use a 4 queue mechanism, where one queue contains operators that are preferred according to both criteria, another contains only operators that are preferred according to early, a third contains only operators that are preferred according to public, and a fourth queue which contains all operators.