Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Property Directed Reachability for Automated Planning

Authors: M. Suda

JAIR 2014 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental An experimental comparison to the state of the art planners finds it highly competitive, solving most problems on several domains. 5. Experiments
Researcher Affiliation Academia Martin Suda EMAIL Max-Planck-Institut f ur Informatik, Saarbr ucken, Germany Charles University, Prague, Czech Republic
Pseudocode Yes Pseudocode 1 Algorithm PDR(Ξ£, I, G, T): Input: A symbolic transition system S = (Ξ£, I, G, T) Output: A witnessing path for S or a guarantee that no path exists
Open Source Code Yes The source code of PDRplan is publicly available on our web page (Suda, 2014), which also contains all the other material relevant for reproducing the experiments.
Open Datasets Yes We tested the planners on the STRIPS15 benchmarks of the International Planning Competition (IPC, 2014).
Dataset Splits No We tested the planners on the STRIPS15 benchmarks of the International Planning Competition (IPC, 2014). We used all the available STRIPS domains except the following: ... Altogether, we collected 1561 problems in 49 domains (see Table 3 on page 304 for a detailed list). While the paper uses established competition benchmarks, it does not specify any explicit training/test/validation dataset splits used by the authors themselves for their experiments.
Hardware Specification Yes We performed the experiments on machines with 3.16 GHz Intel Xeon CPU, 16 GB RAM, running Debian 6.0.
Software Dependencies Yes It internally relies on the SAT-solver Minisat (E en & S orensson, 2003) version 2.2.
Experiment Setup Yes We used a time limit of 180 seconds per problem instance for most of the runs, but increased it to 1800 seconds for the main comparison.