This repository is a curated index of research papers, open-source implementations, and resources related to offline reinforcement learning (offline RL). It serves as a comprehensive reference for researchers and practitioners in the field, aiming to consolidate knowledge and facilitate advancements in learning from static datasets.

How It Works

The project functions as a structured collection of academic papers, categorized by sub-topics within offline RL, such as theory, methods, benchmarks, and applications. It also lists relevant open-source libraries, datasets, blogs, podcasts, workshops, and tutorials, providing a holistic view of the offline RL landscape.

Quick Start & Requirements

This repository is a collection of links and does not require installation or specific software.

Highlighted Details

• Comprehensive categorization of over 500 research papers on offline RL.
• Links to key open-source libraries like d3rlpy, CORL, and D4RL.
• Extensive coverage of benchmarks and datasets for evaluating offline RL algorithms.
• Includes resources on off-policy evaluation (OPE) and its theoretical underpinnings.

Maintenance & Community

The project is maintained by Haruka Kiyohara and Yuta Saito, and actively seeks contributions and maintainers.

Licensing & Compatibility

The repository itself is not software and thus does not have a license. The linked resources may have various licenses.

Limitations & Caveats

As a curated list, the repository's content is dependent on the maintainers' and contributors' efforts to keep it updated with the latest research and resources.