Parametrized Families of Hard Planning Problems from Phase Transitions
Authors: Eleanor Rieffel, Davide Venturelli, Minh Do, Itay Hen, Jeremy Frank
AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we present the results and analysis of the performance of several state-of-the-art planners on the different planning domains described above. |
| Researcher Affiliation | Academia | Eleanor G. Rieffel, Davide Venturelli, Minh Do, Itay Hen and Jeremy Frank NASA Ames Research Center Moffett Field, CA 94035 |
| Pseudocode | No | The paper describes the formulation of planning problems for UHP and GC in prose, detailing actions, preconditions, and effects, but does not present this information in a formal pseudocode or algorithm block. |
| Open Source Code | No | The paper mentions developing a C++ program ('We wrote a simple C++ program to generate these problems, and to produce the PDDL representation') and extending another ('we extended the graph generator program described in (Culberson, Beacham, and Papp 1995)'). It does not state that this code is open-source or provide a link. |
| Open Datasets | No | We randomly generate Erd os-R enyi graphs Gn,p (Erd os and R enyi 1960)... We then obtain a parametrized family of UHP-based planning problems... We wrote a simple C++ program to generate these problems, and to produce the PDDL representation (a domain and problem file pair) for each instance. The paper describes how the problems (datasets) are generated but does not provide concrete access information (link, repository, citation for pre-existing public dataset) to the generated problem instances themselves. |
| Dataset Splits | No | The paper discusses creating 'test sets' of problems for evaluation but does not specify a training, validation, or test split in the typical sense of model training, nor does it provide percentages or absolute counts for such splits. |
| Hardware Specification | Yes | These results were collected using a 64-bit Red Hat Linux machine with 8 Intel Core I7 cores running at 2.4 Ghz with 8 GB of RAM. For these problems, we used the Ivy Bridge nodes of NASA s Pleiades supercomputer, Intel Xeon E5-2680v2 10-core processors running at 2.8 Ghz with 32 GB of RAM. |
| Software Dependencies | Yes | Planners used: To get representative results, we sought to use a set of planners that: (1) use different planning algorithms; and (2) are considered state-of-the-art (have strong performance on existing benchmarks). Specifically, we tested the following planners1: (1) FF (Hoffmann and Nebel 2001) and LAMA-2011 (Richter and Westphal 2010), representing the current dominating forward-state-space search framework; (2) LPG (Gerevini, Saetti, and Serina 2003) representing local search approach; and (3) M and Mp (Rintanen 2012b) representing the compilation approach. |
| Experiment Setup | Yes | We used a complete planner, FF, in both cases, with each data point showing the median runtime of 50 random instances for the relevant parameters. We use a two-hour (7200 sec) cutoff. To compare different planners, we created test sets of 5000 problems at the phase transition, for each problem size from 16 to 40 at an increment of 2 (for a total of 65000 problems). |