According to the paper's abstract, the Arcade Learning Environment (ALE) has been a cornerstone of deep reinforcement learning research, providing a standardized benchmark for evaluating general agents. However, the authors note that ALE lacks native support for object-extraction, which limits its usefulness for certain research directions.
To address this, the paper presents JAXtari, described in the title as a high-throughput and easy-to-modify Arcade Learning Environment. The use of JAX suggests a focus on hardware acceleration and programmability, though the abstract snippet does not provide implementation details.
The source is limited to the announcement and abstract, so no comparison with existing ALE variants or experimental results is available here. The paper is listed as a new arXiv submission.