evo-tpp-solver

TPP PDDL Solver

Solves Travelling Purchase Problem (TPP) planning tasks encoded in PDDL using pyperplan via the unified_planning framework.

When to use this skill

• You have a problem.json file listing TPP planning tasks
• Each task specifies a PDDL domain file, problem file, and output path
• You need to generate valid plans and write them as formatted text files
• You need pickle-serialized plans for validation compatibility

Quick Start

Run the solver script directly:

cd /app
python -m environment.skills.evo-tpp-solver.scripts.solve

Or use the solver from Python:

from pathlib import Path
import sys
sys.path.insert(0, "/app/environment/skills/evo-tpp-solver")
from scripts.solve import solve_all_tasks

solve_all_tasks(Path("/app/problem.json"), Path("/app"))

Pipeline Overview

1. Load config — Read problem.json to get the list of tasks (each has domain path, problem path, output filename)
2. Parse PDDL — Use unified_planning.io.PDDLReader to parse domain + problem files into a Problem object
3. Solve — Use OneshotPlanner(name="pyperplan") with configurable search/heuristic. Default: gbf/hff for speed. Falls back to wastar/hadd if the fast config fails.
4. Format actions — Each action is formatted as action_name(arg1, arg2, ...) using str(action_instance) which already produces this format
5. Write plan — One action per line to the output text file
6. Serialize — Pickle the plan actions list for validation compatibility

Key Implementation Details

Action Format

The str() representation of a unified_planning ActionInstance already produces the required format:

drive(truck1, depot1, market1)
buy(truck1, goods1, market1, level0, level1, level0, level1)

All identifiers are lowercase (PDDL is case-insensitive; PDDLReader normalizes to lowercase).

Solver Configuration

pyperplan supports these search/heuristic combinations via OneshotPlanner params:

• Search: gbf, astar, wastar, bfs, ehs, ids
• Heuristic: hff, hadd, hmax, hsa, blind, lmcut, landmark