Solving QNP and FOND$^+$ with Generating, Testing and Forbidding

This project contains a GTF solver for qualitative numeric planning (QNP) and FOND$^+$ planning. The runnable code is under codes; the benchmark inputs are under domains.

Repository Layout

• codes/src/preOutput_plus: preprocessing wrapper. It converts QNP input to FOND$^+$ PDDL when needed, then runs the translator and preprocessor.
• codes/src/qnp2pddl.py and codes/src/genFONDInput_plus.py: QNP parser and QNP-to-PDDL conversion.
• codes/src/translate: modified Fast Downward translator with FOND$^+$ fairness support.
• codes/src/preprocess: Fast Downward-style preprocessor. It reads output.sas and writes output.
• codes/src/search: modified PRP/Fast Downward search code. search/downward is the built solver binary. search/fond_plus contains the generating, testing and forbidding loop.
• codes/src/numeric: numeric FF-based reducer used by the 3FN mode.
• codes/src/renderPolicy.py: optional Graphviz rendering for dumped policy graphs.
• domains/QNP: QNP instances and their PDDL translations.
• domains/FONDX: FOND$^+$ PDDL instances.

Requirements

• Linux or another Unix-like environment with bash.
• python3.
• make and a C++ compiler with C++11 support.
• Optional for 3FN: flex, bison, and a working build of codes/src/numeric/ff.
• Optional for policy rendering: Graphviz command line tools and the Python graphviz package.

Build

From the project root:

cd codes/src
./build_all

build_all builds src/preprocess/preprocess and src/search/downward.

For 3FN, also build the numeric FF:

cd codes/src/numeric
make -f makefile

Input Format

Set PROBLEM_PATH to a directory containing one of the following:

• FOND$^+$ PDDL: domain.pddl and problem.pddl.
• QNP: exactly one .qnp file.

The QNP text format used by qnp2pddl.py is:

1. problem name,
2. feature declaration line: N name type ..., where type 0 is Boolean and type 1 is numeric,
3. initial feature values,
4. goal feature values,
5. number of actions, followed by each action name, its precondition line, and its effect line.

Run

Use the wrapper script codes/GTF. It selects the solver preset, resolves --working_src_path to codes/src, checks that the planner binary exists, and then forwards the request to src/search/downward.

cd codes

./GTF 3FF ../domains/FONDX/BlocksColumns-Adv/p3-1-1

Examples:

./GTF BFF ../domains/FONDX/BlocksColumns-Adv/p3-1-1
./GTF 3FF ../domains/FONDX/BlocksColumns-Adv/p3-1-1 --show-policy 0
./GTF 3FN ../domains/QNP/BlocksColumns/InQNP/p4-1-3 --num-solutions 2

Extra options are appended after the wrapper defaults. For simple scalar options, pass the option again to override the default value.

Generated Files

The preprocessing and search pipeline writes intermediate and result files into PROBLEM_PATH, including:

• domain.pddl and problem.pddl when a .qnp instance is translated.
• output.sas, output, and elapsed.time from translation/preprocessing.
• result.info, policy.out, and policy.fasp from search.
• action.ind, state.ind, graph.out, pstate.ind, and pgraph.out when graph dumping is enabled.
• qnp_num/ and redundant_actions when 3FN runs the numeric reducer.
• S_*, NT, or NT_* when saving all solutions or non-terminating debug policies is enabled.

These products are ignored by .gitignore; the repository should keep source code and benchmark inputs, not local run outputs or binaries.

Policy Graph Rendering

Enable graph dumping in the solver command:

--dump-graph 1

Then render a policy graph:

python3 src/renderPolicy.py --main_path "$PROBLEM_PATH"

Useful flags are --p for compact partial-state graphs, --nt for non-terminating edges, and --ntc for compact non-terminating edges.

License

This repository is distributed under the GNU General Public License v3.0; see LICENSE. Third-party components keep their original copyright and license notices in the source files.