TextWorld-to-PDDLGym

This is a simple library for converting TextWorld environment (an OpenAI gym) to PDDLGym's.

• TextWorld https://github.com/microsoft/TextWorld
• PDDLGym https://github.com/tomsilver/pddlgym

For installation

!pip install .

to go. It depends on (and automatically installs) TextWorld and PDDLGym.

First, you may generate your own TextWorld game.

import textworld
from textworld.generator import make_game, compile_game
from textworld.challenges import cooking

options = textworld.GameOptions()
options.seeds = 5611

settings = {
    'split': "test",
    'recipe_seed': 1234,
    'take': True,
    'open': True,
    'cook': True,
    'cut': True,
    'go': 6
}
game = cooking.make(settings, options)
gamefile = compile_game(game)

Or, your can use a given game file.

gamefile = "./demo_games/tw-cooking-test-take+cook+cut+open+go6-JrmLfNyMcErjF6LD.ulx"

Then, you should provide a TextWorld enviornment, specifying some EnvInfos options as follows.

import gym
import textworld.gym
from textworld import EnvInfos

request_infos = EnvInfos(description=True, entities=True, facts=True)
env_id = textworld.gym.register_game(gamefile, request_infos)
env = gym.make(env_id)
obs, infos = env.reset()

Now you are ready to go with TextWorld.

from pddlgym_textworld import TextWorldWrapper

env = TextWorldWrapper(tw_env)

ob, infos = env.reset()

Observation (ob) is in pddlgym.structs.State. You can consider the wrappeed environment as PDDLGym's logical environment.

from pddlgym.structs import Literal, Predicate, Type

#obs, reward, done, debug_info = env.step("go south")
action = Predicate('go/south', 3, [Type('P'), Type('r'), Type('r')])('P', 'bedroom', 'corridor')
obs, reward, done, debug_info = env.step(action)