A compilation of Hard-To-Ground domains

• The three versions of the genome-edit-distance domain are taken from Patrik Haslum's website. The instances appearing in the IPC'14 seem to correspond to the easiest instances of the -split version here.

• The organic-synthesis benchmarks are different versions of the same problem. The instances appearing in the IPC'18 are not exactly these ones, but simpler instances.

• The pipesworld-tankage-nosplit folder contains the non-split version of the pipesworld-tankage, as it features e.g. in the downward-benchmark repos.

• The gnad-et-al-aaai19-large-instances folder contains larger instances of standard benchmarks, sourced from

  Gnad, D., Torralba, A., Domínguez, M., Areces, C., & Bustos, F. (2019). Learning how to ground a plan–partial grounding in classical planning. In Proceedings of the AAAI Conference on Artificial Intelligence (Vol. 33, pp. 7602-7609).

• The logistics folders just contains some handmade instances for testing purposes.

• The childsnack-contents folder contains a modification of the original Childsnack that adds contents to sandwiches (instead of just tracking the bread type).

• The visitall-multidimensional folder contains an extension of the original visitall domain that interchanges the 2D-grid with a n-dimensional hypercube.

• The *-large-simple folders contain larger instances of standard benchmarks with a low goal amount.

• The labyrinth folder contains the SAT and OPT set of the labyrinth domain from the IPC2023 dataset. The domain has been altered to include no negative preconditions

Scripts

• The flatten-structure.py script creates a copy of the benchmark using a flat directory structure. For the *-large-simple domains, it also needs to rename the PDDL files and readjust the name of some domains. This flat structure might be useful if you run experiments that do not expect nested directories in the benchmark.

  In more detail, if a domain X has several sub-directories with identical domain.pddl files, the script creates a single directory X containing all instance files. These instance files are renamed so you can identify from which sub-directory they are originally from. If a domain Y has several sub-directories using different domain.pddl files, the script creates a separate version of Y for each domain.pddl. Each version is named based on its original sub-directory.