Code, Benchmarks and Experiment Data for the ICAPS 2024 Paper "Abstraction Heuristics for Factored Tasks"

We hereby provide all code, data, and benchmarks required to reproduce the experiments reported in the paper "Abstraction Heuristics for Factored Tasks" presented at ICAPS 2024.

buechner-et-al-icaps2024-code.zip contains our implementation of the concepts presented in the paper based on the Scorpion (Seipp, Keller and Helmert, JAIR 2020) planner.

buechner-et-al-icaps2024-scripts.zip contains experiment scripts compatible with Downward Lab 7.3 for reproducing the experiments of the paper as well as Apptainer images of external sources used in the experiments.

buechner-et-al-icaps2024-benchmarks.zip contains the benchmarks in finite domain representation (ending in .sas) used in the experiments. It consists of existing problems (Cave Diving from the IPC 2014 and Matrix Multiplication from Speck et al. (ICAPS 2023)) as well as completely new problems generated with our own problem generators to be found in buechner-et-al-icaps2024-generators.zip.

buechner-et-al-icaps2024-generators.zip contains the problem generators mentioned above. We implemented problem generators for four domains: Pancakes, Burnt Pancakes, Rubik's Cube, and TopSpin. Use the following commands to generate problems:
- Pancakes (directory "pancakes"): ./build.py -n N to obtain a .sas file for N pancakes.
- Burnt Pancakes (directory "pancakes"): ./build.py -n N -b to obtain a .sas file for N pancakes with orientation.
- Rubik's Cube (directory "rubiks-cube"): ./build.py -t T to obtain a .sas file for a 3x3x3 Rubik's Cube using the half-turn metric with T random turns from the solved state to initialize the problem. Use additionally parameters -s to obtain a 2x2x2 Rubik's Cube instead and/or -q to use the quarter-turn metric.
- TopSpin (directory "topspin"): ./build.py -n N -k K to obtain a .sas file for a TopSpin problem with N tiles and a flipping window of size K.

buechner-et-al-icaps2024-data.zip contains the experimental data. All directories except thse ending with "-eval" contain raw data of the experiments that were performed for the paper. Each of these contain a subdirectory tree structure "runs-" where each planner run has its own directory. For each run, there are symbolic links to the input files "problem.sas" (can be resolved by putting the benchmarks directory to the right place), the run log file "run.log" (stdout), possibly also a run error file "run.error" (stderr), the run script "run" used to start the experiment, and a "properties" file that contains data parsed from the log file(s). Directories with the "-eval" ending each contain a "properties" file which contains a JSON dictionary with combined data of all runs of the corresponding experiment as well as html and tex files which were used to generate the figures and tables in the paper.

Note on licence: we chose GPL v3.0 or later mainly because we consider our implementation based on Fast Downward the main contribution of this package, and Fast Downward comes with GPL v3.0.