Code and Data for the ECAI 2025 Paper "Merging Cartesian Abstractions for Classical Planning"

Code

File salerno-et-al-ecai2025-code.zip contains a copy of the Scorpion planning system, extended to include all methods described in the paper. To run the experiments as described in the paper, you will need to have a CPLEX installation that is compatible with the Scorpion planner.

Install uv:

curl -LsSf https://astral.sh/uv/install.sh | sh

To build the planner and run our configuration that creates Cartesian abstractions of up to two goals by merging single-goal abstractions, run:

uv run ./build.py
uv run ./fast-downward.py domain.pddl problem.pddl --search "astar(pho([\
cartesian([goals(size=1, order=hadd_up, random_seed=0)], random_seed=0, verbosity=silent, max_states=infinity,\
max_transitions=10M, max_time=infinity, memory_padding=500, pick_flawed_abstract_state=batch_min_h, pick_split=max_cover,\
tiebreak_split=max_refined, max_goals_per_abstraction=2, transition_representation=store, copy_merged_abstractions=true)],\
 saturated=true, orders=random_orders(), diversify=true, max_optimization_time=0, max_time=100))"

Scripts

A script to run the configurations found in the paper can be found in /experiments/multi-goal/paper-exp.py. For this, you need to set the env variable "DOWNWARD_BENCHMARKS" to a directory containing the IPC benchmarks.

Benchmarks

The file salerno-et-al-ecai2025-benchmarks.zip contains the PDDL benchmarks from the optimal tracks of IPC 1998-2018.

Data

The file salerno-et-al-ecai2025-data.zip contains the parsed values and basic reports for the experiments in the paper, in the folder data/reports-eval. It also includes a script reports.py to generate new reports using said data.