Plan-Space Explanation via Plan-Property Dependencies: Faster Algorithms & More Powerful Properties
Authors: Rebecca Eifler, Marcel Steinmetz, Álvaro Torralba, Jörg Hoffmann
IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We empirically evaluate this method through experiments on a collection of IPC benchmarks extended with AS properties by Eif20. Formulating the same properties in LTLf, we find that our more general solution causes little overhead. Formulating new LTLf plan properties beyond AS properties, we get worse but still reasonable scaling behavior. |
| Researcher Affiliation | Academia | Rebecca Eifler , Marcel Steinmetz , Alvaro Torralba and J org Hoffmann Saarland University, Saarland Informatics Campus, Germany {eifler, steinmetz, torralba, hoffmann}@cs.uni-saarland.de |
| Pseudocode | No | The paper describes algorithms in text (e.g., Section 3 and 4) but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper states: "We implemented our All MUGS algorithms in Fast Downward [Helmert, 2006], on top of Eif20’s code base, using the Sym BA [Torralba et al., 2014] code for symbolic search." This indicates they built on existing open-source tools, but there is no explicit statement that their own implementation of the new algorithms or LTL framework is open-source or available. |
| Open Datasets | Yes | We use Eif20’s benchmarks, comprising two parts. First, the classical-planning IPC benchmarks, modified by enforcing a plan-cost bound. Second, four domains (resource-constrained planning by Nakhost et al. [2012], plus the Blocksworld) extended with Gsoft encoding action-set properties, setting the original goals as Ghard and choosing the cost bound so that some but not all of Gsoft can be satisfied. |
| Dataset Splits | No | The paper mentions using |
| Hardware Specification | Yes | All experiments were run on Intel E5-2660 machines running at 2.20 GHz, with a time (memory) limit of 30min (4GB). |
| Software Dependencies | No | The paper mentions using "Fast Downward [Helmert, 2006]", "Eif20’s code base", and "Sym BA [Torralba et al., 2014] code". While these are specific tools, no version numbers are provided for these or any other software components, which is required for reproducibility. |
| Experiment Setup | Yes | All experiments were run on Intel E5-2660 machines running at 2.20 GHz, with a time (memory) limit of 30min (4GB). Cost bounds set to x times the best known plan cost. |