evo-pddl-airport-planner

Imported from Zhang-Henry/CoEvoSkills (artifacts/skills/virtualhome-agent-planning/evo-pddl-airport-planner/SKILL.md). Install upstream with npx skills add Zhang-Henry/CoEvoSkills --skill evo-pddl-airport-planner. Copyright stays with the author.

PDDL Airport Planner

Solves PDDL planning tasks for airport ground-traffic control domains from IPC competitions.

Overview

• Reads problem.json listing tasks with domain/problem PDDL file paths and output paths
• Uses pyperplan directly for parsing, grounding, and search (hFF + A*, fallback to BFS)
• Outputs plans in standard PDDL parenthesized format: (action_name arg1 arg2 ...)
• One action per line, lowercase (PDDL is case-insensitive)

Dependencies

• pyperplan (2.1+)
• Python 3.10+

Quick Start

import sys
sys.path.insert(0, '/app/environment/skills/evo-pddl-airport-planner/scripts')
from solve import solve_all_tasks, validate_plans

# Solve all tasks from problem.json
results = solve_all_tasks('/app/problem.json', base_dir='/app')

# Validate outputs
valid = validate_plans('/app/problem.json', base_dir='/app')
assert valid, "Some plans failed validation"
print("All tasks solved successfully!")

Utility Functions (scripts/utils.py)

load_tasks(problem_json_path)

Load task definitions from problem.json. Returns list of task dicts.

solve_pddl_task(domain_path, problem_path)

Solve a single PDDL task using pyperplan BFS. Returns list of action strings in standard PDDL format (action_name arg1 ...) or None.

solve_pddl_task_with_heuristic(domain_path, problem_path)

Solve using hFF heuristic with A* search, falling back to BFS. Returns list of action strings or None.

write_plan(actions, output_path)

Write action list to a plan file, one action per line.

solve_and_write(domain_path, problem_path, output_path)

Solve and write in one call. Returns True on success.

Entry Points (scripts/solve.py)

solve_all_tasks(problem_json_path, base_dir=None)

End-to-end: reads problem.json, solves all tasks, writes all plan files. Returns dict of task_id -> success boolean.

validate_plans(problem_json_path, base_dir=None)

Checks all output files exist, are non-empty, and have valid PDDL action syntax.

Key Design Decisions

1. Direct pyperplan: Uses pyperplan directly instead of through unified_planning to get native PDDL plan format without case conversion issues.

2. Standard PDDL format: Plans use (action_name arg1 arg2) parenthesized format with space-separated arguments, which is the standard IPC plan format.

3. Heuristic search with fallback: Tries hFF + A* first for efficiency on larger instances, falls back to BFS if heuristic search fails.