The repository includes the code for generating planning tasks according to specified structure
@InProceedings{katz-et-al-socs2023,
title = "Generating SAS+ Planning Tasks of Specified Causal Structure",
author = "Michael Katz and Junkyu Lee and Shirin Sohrabi",
booktitle = "Proceedings of the 16th Annual Symposium on
Combinatorial Search (SoCS 2023)",
publisher = "{AAAI} Press",
year = "2023"
}
./generate.py --task --num-atoms 100 --num-variables 10 --num-goal-variables 4 --max-num-prevails 2 --max-num-effects 3 --max-atoms-per-layer 10 --domain-file-name domain.pddl --problem-file-name problem.pddl --sas-file-name output.sas --polytree-cg --edge-probability 0.2
./generate.py --domain --inverted-fork-cg --num-atoms 200
or, for a default set of possible number of atoms:
./generate.py --domain --inverted-fork-cg
./generate.py --task --input-json-file-name examples/sas_task.json --seed 2023
The json input file with task parameters can be obtained from pddl files by
building the code in this Fast Downward Fork, and
running the following (producing the file named sas_task.json):
./fast-downward.py domain.pddl problem.pddl --search 'lazy_greedy([cg(),const(infinity)])'