Skip to content


Repository files navigation


RLE has been succeeded by the wonderful work of the folks at OpenAI under their repository Gym-retro. Since they now support all the consoles and games that we supported and most of our features, there's no reason to maintain this one.


A learning framework based on the Arcade Learning Environment (ALE) and Libretro (Stella for Atari and SNES9X for the Super Nintendo Entertainment System).

The environment provides an interface to training and evaluating AI algorithms against different console games using its screen as input.

The currently supported games can be found in the src/games/supported directory . Some popular games include: Mortal Kombat, Super Mario All Stars, F-Zero, Castle Wolfenstein and Gradius III.

A paper is available for RLE at If you use RLE in your publication, please use the following BibTex entry:

  title = {Playing SNES in the Retro Learning Environment},
  author = {Bhonker, Nadav and Rozenberg, Shai and Hubara, Itay},
  journal = {arXiv preprint arXiv:1611.02205},
  year = {2016}

Quick Start

Install main dependencies:

sudo apt-get install libsdl1.2-dev libsdl-gfx1.2-dev libsdl-image1.2-dev cmake

To install as a Gym environment:
Go to the gym-rle repository and follow the instructions there.

To install the python interface:
Either install via PyPi:

$ pip install rle-python-interface

or by cloning the repository and running the following:

$ pip install .
$ pip install --user .

To use the shared_library interface:

$ mkdir build && cd build
$ make -j4

To install the lua (Torch) interface, the additional alewrap module is required:

luarocks install
luarocks install

DQN Implementations Using RLE


  • @mgbellemare for his work on ALE and his useful advice.
  • @Alcaro and the @libretro community for their assistance in incorporating their work into our framework.