Classical Planning with Avoid Conditions
Authors: Marcel Steinmetz, Jörg Hoffmann, Alisa Kovtunova, Stefan Borgwardt9944-9952
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We run a large-scale experiment, comparing our techniques against compilation methods and against simple state pruning using ϕ. |
| Researcher Affiliation | Academia | 1 Saarland University, Saarland Informatics Campus, Saarbr ucken, Germany 2 Institute of Theoretical Computer Science, Technische Universit at Dresden, Germany |
| Pseudocode | Yes | Algorithm 1: k-trap computation. Adaptions to ϕtraps are highlighted in bold. |
| Open Source Code | Yes | Source code and benchmarks are available online1. 1https://doi.org/10.5281/zenodo.6338021 |
| Open Datasets | No | The paper discusses various benchmark domains like 'Cave Diving (IPC14)', 'Miconic', 'Nurikabe (IPC18)', and 'ROAD' domains, but it does not provide specific links, DOIs, or formal citations (with authors and year) for accessing these datasets or their modified versions. |
| Dataset Splits | No | The paper describes experimental scenarios with solvable and unsolvable instances but does not provide explicit details about dataset splits for training, validation, or testing. |
| Hardware Specification | Yes | The experiments were run on machines with Intel Xeon E5-2660 @ 2.20GHz CPUs, and 30 minutes time and 4 GB memory cutoffs. |
| Software Dependencies | No | The paper states 'We implemented all described methods in Fast Downward (FD) (Helmert 2006)', but does not provide a specific version number for Fast Downward or other software dependencies. |
| Experiment Setup | Yes | For each category, we chose a canonical base planner configuration: optimal planning via A search with LM-cut (Helmert and Domshlak 2009); satisficing planning via greedy best-first search with two open lists and preferred operators using h FF (Hoffmann and Nebel 2001); and proving unsolvability via depth-first search with hmax (Haslum and Geffner 2000) for dead-end detection. For the k-ϕ-traps, we experimented with k {2, 3, 4, 5}. To terminate CEGAR, we enforced an upper limit N on the number of abstract states, N {25k, 50k, 100k, 150k, 200k, 300k}. |