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].

Encoding Domain Transitions for Constraint-Based Planning

Authors: Nina Ghanbari Ghooshchi, Majid Namazi, M.A.Hakim Newton, Abdul Sattar

JAIR 2017 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on a number of standard classical planning benchmark domains demonstrate TCPP s efficiency over the original Pa P2 running on SICStus Prolog and our reconstructed and enhanced versions of Pa P2 running on Minion. [...] 7. Experimental Results. We ran all experiments reported in this paper on the same high performance computing cluster Gowonda at Griffith University.
Researcher Affiliation Collaboration Nina Ghanbari Ghooshchi EMAIL Institute for Integrated and Intelligent Systems (IIIS) Griffith University, Brisbane, Australia Data61, CSIRO, Australia
Pseudocode Yes Algorithm 1 Construction of Redefined DTGs; Algorithm 2 A Non-Deterministic Search for Parallel Planning; Algorithm 3 DTGs To CSP Encoding; Algorithm 4 Transition constraint tc(vτ) using a table; Algorithm 5 Adding Parallelism Variables to Transition Constraints; Algorithm 6 Decoding CSP Solution to Plan; Algorithm 7 FSA To CSP Encoding; Algorithm 8 Action Succession Constraint asc(vτ, vτ+1) using a table; Algorithm 9 Synchronisation Constraint sc(vτ) using a table; Algorithm 10 Mutex Constraint mc(v, k, v, k, τ) using a negative table
Open Source Code No The paper does not contain any explicit statement about providing source code, a link to a repository, or code in supplementary materials for the methodology described.
Open Datasets Yes As our benchmark planning domains, we use 21 classical planning domains from international planning competitions. These domains are airport, blocks, driverlog, freecell, grid, gripper, logistics00, logistics98, miconic, mprime, mystery, pathway, psr-small, pipes-notankage, pipes-tankage, rovers, satellite, storage, tpp, and zenotravel. There are in total 786 problem instances in all 21 domains and we use them to evaluate the planners.
Dataset Splits No The paper uses '786 problem instances in all 21 domains' from international planning competitions for evaluation. These are individual planning problems, not a single dataset that is split into training, validation, and test sets in the traditional sense. The paper does not specify percentages or counts for such splits within a single dataset, but rather evaluates performance on a collection of distinct problem instances.
Hardware Specification Yes We ran all experiments reported in this paper on the same high performance computing cluster Gowonda at Griffith University. Each node of the cluster is equipped with Intel Xeon CPU E5-2670 processors @2.60 GHz, FDR 4x Infini Band Interconnect, having system peak performance 18949.2 Gflops. We ran experiments with 4GB memory limit and 60 minute timeout.
Software Dependencies No As mentioned in Section 3, we use Minion (Gent et al., 2006) as our constraint solver. [...] Our experiments on a number of standard classical planning benchmark domains demonstrate TCPP s efficiency over the original Pa P2 running on SICStus Prolog and our reconstructed and enhanced versions of Pa P2 running on Minion. [...] We use the translator used in Fast Downward (Helmert, 2006) to translate the PDDL description into SAS+ formalism. No specific version numbers are provided for Minion, SICStus Prolog, or Fast Downward.
Experiment Setup Yes We ran experiments with 4GB memory limit and 60 minute timeout. [...] Minion supports several options for variable ordering, but in our experiments we use the most well-known two: smallest domain first (sdf) and most conflict variable first (conflict). [...] For propagation, Minion supports generalised arc consistency (gac), singleton arc consistency (sac), and a few variants of sac. In our experiments, we use gac and the basic sac as our constraint propagation strategies. [...] In our experiments, we only use the ascending value ordering. [...] Each of TCPP and Pa PR variants running on Minion with the following configurations: 1. sdf-gac: smallest domain first variable ordering and generalised arc consistency; 2. sdf-sac: smallest domain first variable ordering and singleton arc consistency; 3. conf-gac: conflict variable ordering and generalised arc consistency; 4. conf-sac: conflict variable ordering and singleton arc consistency