Welcome to JAXAtari’s Documentation!

JAXAtari is a GPU-accelerated, object-centric Atari environment framework built with JAX. Inspired by OCAtari, it enables massively parallelized training for reinforcement learning research.

Built and maintained by students from TU Darmstadt.

Note

If you’re looking for a quick start, head to the usage section below or browse the API reference.

---

Features

• Object-centric extraction of Atari game states.

• JAX-based vectorized execution with GPU support.

• Compatible API with ALE (Arcade Learning Environment).

• Built-in benchmarking tools.

• Modular wrappers and utilities.

---

Getting Started

You can install and use JAXAtari as follows:

python3 -m venv .venv
source .venv/bin/activate
pip install -e .

To run a game manually:

python -m jaxatari.games.jax_seaquest

---