AmenablePDDL

AmenablePDDL is designed to simplify parsing and planning simple PDDL-defined domains and problems, providing an intuitive interface for implementing common planning algorithms. AmenablePDDL is a wrapper around the PDDL library by Marco Favorito, Francesco Fuggitti, and Christian Muise.

This interface was developed for Carnegie Mellon University's 16-280 Intelligent Robotic Systems.

Table of Contents

• Installation
• Overview
• Quick Start Example
• Public Methods
  • Constructor
  • Domain and Problem Access
  • State and Goal
  • Action Retrieval
  • Action Applicability and Application
• Implementing DFS (Example)
• Internal Methods (Optional)
• License

Installation

AmenablePDDL is available through pip. Install it on your machine by running:

pip install AmenablePDDL

Quick Start Example

from AmenablePDDL import AmenableP

# Initialize the interface with domain and problem files
interface = AmenableP("domain.pddl", "problem.pddl")

# Retrieve initial state and available actions
the_initial_state = interface.get_initial_state()
actions = interface.get_domain_actions()

print("Initial State:", the_initial_state)
print("Available Actions:", [action.name for action in actions])

This example initializes the AmenablePDDL interface, loads the domain and problem files, and prints out the initial state and available actions.

Public Methods

Constructor

interface = AmenableP(domain_file, problem_file)

• domain_file: Path to the PDDL domain file.
• problem_file: Path to the PDDL problem file.

Domain and Problem Access

• get_domain_actions() -> List[Action]: Returns a list of Action objects defined in the domain.

State and Goal

• get_initial_state() -> Set[Predicate]: Returns the set of positive predicates of the initial state.
• is_goal_state(state: Set[Predicate]) -> bool: Checks if a given state satisfies the goal expression.

Action Retrieval

• find_applicable_actions(state: Set[Predicate]) -> List[Tuple[Action, Dict[Variable, Constant]]]: Returns applicable actions with valid bindings in the given state.

Action Application

• apply_action(action: Action, state: Set[Predicate], binding: Dict[Variable, Constant]) -> Set[Predicate]: Applies the specified action with the given binding to the state, returning a new state.

Implementing DFS (Example)

Below is an example of how to implement a Depth-First Search (DFS) planner using AmenablePDDL.

from AmenablePDDL import AmenableP

# Initialize the interface
interface = AmenableP("domain.pddl", "problem.pddl")

# Define a simple DFS function

def dfs(state, plan, visited, depth_limit):
    if depth_limit <= 0:
        return None
    if interface.is_goal_state(state):
        return plan

    visited.add(frozenset(state))

    for action, binding in interface.find_applicable_actions(state):
        new_state = interface.apply_action(action, state, binding)
        state_key = frozenset(new_state)

        if state_key not in visited:
            new_plan = plan + [(action, binding)]
            result = dfs(new_state, new_plan, visited, depth_limit - 1)
            if result is not None:
                return result
    return None

# Run DFS starting from the initial state
visited_states = set()
plan = dfs(interface.get_initial_state(), [], visited_states, depth_limit=50)

if plan:
    print("Plan found:")
    for step, (action, binding) in enumerate(plan, start=1):
        bound_str = " ".join(str(binding[param]) for param in action.parameters)
        print(f"{step}: {action.name} {bound_str}")
else:
    print("No plan found.")

This script initializes the AmenablePDDL interface, defines a recursive DFS function leveraging the interface's methods to find applicable actions and apply them, and then searches for a plan with a given depth limit.

Internal Methods (Optional)

AmenablePDDL also provides internal methods for advanced usage:

• _evaluate_condition(expr, state, binding)
• _find_bindings_for_action(action, state)
• _apply_effects(action, state, binding)
• _ground_predicate(pred, binding)

These methods are used internally by the public methods, but users can extend the existing methods by forking the repository.

License

This project is licensed under the MIT License. See the LICENSE file for details.