Description

The Arcade Learning Environment (ALE) wraps the Stella Atari 2600 emulator and exposes 100+ games (Breakout, Pong, Space Invaders, Ms. Pac-Man, …) as gymnasium envs. Importing stable_worldmodel registers every game under its standard ALE/<Game>-v5 id, so any title from the official catalogue is available via gym.make.

For the full game list, action sets, and per-game details, see the upstream docs: https://ale.farama.org/environments/complete_list/.

import stable_worldmodel as swm
from stable_worldmodel.policy import RandomPolicy

world = swm.World(
    'ALE/Breakout-v5',
    num_envs=1,
    image_shape=(224, 160),
    goal_conditioned=False,
    render_mode='rgb_array',
)
world.set_policy(RandomPolicy(seed=0))
world.evaluate(episodes=1, seed=0, video='./videos/breakout')

Install

ale-py ships with the env extra:

pip install 'stable-worldmodel[env]'

If it's missing at import time, a warning is emitted and ALE/* ids are not registered.

Notes

• Pass render_mode='rgb_array' so MegaWrapper can grab frames via env.render().
• ALE actions are discrete (Discrete(n)); n varies per game.
• Atari frames are 210×160. Pick an image_shape whose dims are divisible by 16 (e.g. (224, 160)) to avoid imageio's macro-block-size warning when writing mp4s.
• These envs are not goal-conditioned — pass goal_conditioned=False to swm.World.