The Arcade Learning Environment (ALE) is a simple framework that allows researchers and hobbyists to develop AI agents for Atari 2600 games. It is built on top of the Atari 2600 emulator Stella and separates the details of emulation from agent design. This video depicts over 50 games currently supported in the ALE.

For an overview of our goals for the ALE read The Arcade Learning Environment: An Evaluation Platform for General Agents. If you use ALE in your research, we ask that you please cite this paper in reference to the environment. See the Citing section for BibTeX entries.

• Object-oriented framework with support to add agents and games.
• Emulation core uncoupled from rendering and sound generation modules for fast emulation with minimal library dependencies.
• Automatic extraction of game score and end-of-game signal for more than 100 Atari 2600 games.
• Multi-platform code (compiled and tested under macOS, Windows, and several Linux distributions).
• Python bindings through pybind11.
• Native support for Gymnasium, a maintained fork of OpenAI Gym.
• Visualization tools.
• Atari roms are packaged within the pip package

The ALE currently supports three different interfaces: C++, Python, and Gymnasium.

You simply need to install the ale-py package distributed via PyPI:

pip install ale-py

Note: Make sure you're using an up-to-date version of pip or the installation may fail.

You can now import the ALE in your Python projects with providing a direct interface to Stella for interacting with games

from ale_py import ALEInterface

ale = ALEInterface()
ale.loadROM("Breakout")
ale.reset_game()

reward = ale.act(0)  # noop
screen_obs = ale.getScreenRGB()

For simplicity for installing ale-py with Gymnasium, pip install "gymnasium[atari]" shall install all necessary modules and ROMs. See Gymnasium introductory page for description of the API to interface with the environment.

import gymnasium as gym

env = gym.make('ALE/Breakout-v5', render_mode="human")  # remove render_mode in training
obs, info = env.reset()
episode_over = False
while not episode_over:
    action = policy(obs)  # to implement - use `env.action_space.sample()` for a random policy
    obs, reward, terminated, truncated, info = env.step(action)

    episode_over = terminated or truncated
env.close()

For all the environments available and their description, see gymnasium atari page.

The following instructions will assume you have a valid C++17 compiler and vcpkg installed.

We use CMake as a first class citizen, and you can use the ALE directly with any CMake project. To compile and install the ALE you can run

mkdir build && cd build
cmake ../ -DCMAKE_BUILD_TYPE=Release
cmake --build . --target install

There are optional flags -DSDL_SUPPORT=ON/OFF to toggle SDL support (i.e., display_screen and sound support; OFF by default), -DBUILD_CPP_LIB=ON/OFF to build the ale-lib C++ target (ON by default), and -DBUILD_PYTHON_LIB=ON/OFF to build the pybind11 wrapper (ON by default).

Finally, you can link agaisnt the ALE in your own CMake project as follows

find_package(ale REQUIRED)
target_link_libraries(YourTarget ale::ale-lib)

If you use the ALE in your research, we ask that you please cite the following.

M. G. Bellemare, Y. Naddaf, J. Veness and M. Bowling. The Arcade Learning Environment: An Evaluation Platform for General Agents, Journal of Artificial Intelligence Research, Volume 47, pages 253-279, 2013.

In BibTeX format:

@Article{bellemare13arcade,
    author = {{Bellemare}, M.~G. and {Naddaf}, Y. and {Veness}, J. and {Bowling}, M.},
    title = {The Arcade Learning Environment: An Evaluation Platform for General Agents},
    journal = {Journal of Artificial Intelligence Research},
    year = "2013",
    month = "jun",
    volume = "47",
    pages = "253--279",
}

If you use the ALE with sticky actions (flag repeat_action_probability), or if you use the different game flavours (mode and difficulty switches), we ask you that you also cite the following:

M. C. Machado, M. G. Bellemare, E. Talvitie, J. Veness, M. J. Hausknecht, M. Bowling. Revisiting the Arcade Learning Environment: Evaluation Protocols and Open Problems for General Agents, Journal of Artificial Intelligence Research, Volume 61, pages 523-562, 2018.

In BibTex format:

@Article{machado18arcade,
    author = {Marlos C. Machado and Marc G. Bellemare and Erik Talvitie and Joel Veness and Matthew J. Hausknecht and Michael Bowling},
    title = {Revisiting the Arcade Learning Environment: Evaluation Protocols and Open Problems for General Agents},
    journal = {Journal of Artificial Intelligence Research},
    volume = {61},
    pages = {523--562},
    year = {2018}
}