Driving · 2026-10-19

JAXtari: High-Throughput and Easy-to-Modify Arcade Learning Environment

Quentin Delfosse, Raban Emunds, Paul Seitz, Sebastian Wette, Jannis Bl\"uml, Daniel Kirn, Dominik Mandok, Kristian Kersting

arXiv:2610.23650PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

JAXtari is an open-source, JAX-native reimplementation of Atari 2600 (ALE) that removes CPU-bound emulation bottlenecks and enables accelerator-native, high-throughput PPO training. The environment also provides object-centric observations and a flexible modification plugin system for direct semantic edits, enabling zero-shot evaluation of robustness to controlled task variations.

Why it matters

ALE is widely used but lacks native semantic editability and object-centric state access, and its CPU-based emulation bottlenecks modern GPU training pipelines. JAXtari targets behavioral/learning-task parity with ALE while enabling (1) faster, GPU-parallel training/evaluation, and (2) source-level modifications and object-centric observations for studying shortcut learning and robustness under controlled distribution shifts.

Method

  • Reimplement Atari environments as purely functional, fully JIT-compiled JAX transition functions (step(state, action) → observation, next_state, reward, done, info) instead of wrapping ROM-based ALE emulation.
  • Provide object-centric state observations (ObjectCentricWrapper) and pixel observations via a JAXGameRenderer; enable vectorized accelerator execution with jax.vmap to run thousands of instances on a single GPU.
  • Use a modular plugin modification system with internal (during construction) and post-step (after each step) interventions, enabling zero-shot evaluations on controlled semantic variations.
Abstract (from arXiv)

The Arcade Learning Environment (ALE) has been a cornerstone of Deep Reinforcement Learning research, providing a standardized benchmark for evaluating general agents. However, its lack of native support for object-extraction and environment modification restricts research into evaluating whether agents learn robust causal models or rely on brittle visual heuristics. Furthermore, the ALE's dependence on CPU-based emulation creates a significant bottleneck in modern GPU-accelerated training pipelines, limiting both experimental scale and iteration frequency. In this work, we introduce JAXtari, an open-source, high-performance reimplementation of Atari 2600 environments. By leveraging massive GPU parallelization, JAXtari allows PPO agents to finish training on 100M environment steps on a single A100-GPU in under an hour. Beyond speed, it provides object-centric observations and a flexible modification system, allowing researchers to, e.g. develop neuro-symbolic methods or discover novel shortcuts and subsequently train and adapt their agents rapidly, sometimes in only minutes.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-10-16 · Indexed by Pub-AI: 2026-10-19 UTC

preprint

JAXtari: High-Throughput and Easy-to-Modify Arcade Learning Environment

Quentin Delfosse, Raban Emunds, Paul Seitz, Sebastian Wette, Jannis Blüml, Daniel Kirn, Dominik Mandok, Kristian Kersting

Preprint posted: 2026-10-16 · Original record ↗

arxiv
2610.23650
Explicit source versions (1)
  • v1 · 2026-10-16
  1. Preprint available

    Source date: 2026-10-16 · First detected: 2026-10-25 UTC · Historical baseline

    Source evidence (1)
Abstract · arxiv

The Arcade Learning Environment (ALE) has been a cornerstone of Deep Reinforcement Learning research, providing a standardized benchmark for evaluating general agents. However, its lack of native support for object-extraction and environment modification restricts research into evaluating whether agents learn robust causal models or rely on brittle visual heuristics. Furthermore, the ALE's dependence on CPU-based emulation creates a significant bottleneck in modern GPU-accelerated training pipelines, limiting both experimental scale and iteration frequency. In this work, we introduce JAXtari, an open-source, high-performance reimplementation of Atari 2600 environments. By leveraging massive GPU parallelization, JAXtari allows PPO agents to finish training on 100M environment steps on a single A100-GPU in under an hour. Beyond speed, it provides object-centric observations and a flexible modification system, allowing researchers to, e.g. develop neuro-symbolic methods or discover novel shortcuts and subsequently train and adapt their agents rapidly, sometimes in only minutes.