A curated filter for the annual flood of RL papers

What it does This repository is a manually curated index of reinforcement learning papers spanning classics (DQN, PPO, SAC) through recent conference hits at ICLR, ICML, and NeurIPS through 2026. The maintainer organizes them into practical buckets—model-free and model-based methods, offline RL, meta-RL, adversarial settings, generalization, and the growing intersection with transformers and LLMs. Each entry typically includes a one-line description of the core idea, making it a quick reference rather than a deep review.

The interesting bit The value here is explicitly editorial: the README admits tens of thousands of RL papers appear yearly, and it only lists those the authors actually read and consider insightful. That honesty saves you from the trap of bibliography inflation where every paper is somehow “seminal.”

Key highlights

• Coverage runs from foundational work (Nature DQN, TRPO, DDPG) to fresh conference spotlights (NeurIPS 23, ICLR 25, ICML 25).
• Papers are grouped by technique—exploration, world models, diffusion-based offline RL—rather than just by year.
• Includes emerging areas like RL with Transformers/LLMs and unsupervised RL.
• Each paper gets a terse, technical description of its contribution rather than just a title and link.

Caveats

• The list is openly incomplete: the authors state they can only include papers they have personally read and found insightful.
• Some table rows contain empty description cells or stray HTML comments left in the markdown source.
• “Genaralisation” is misspelled in the table of contents, which hints that polish is not the primary goal.

Verdict Worth bookmarking if you are a graduate student or practitioner trying to trace the evolution of a specific RL subfield without reading every conference proceedings. Skip it if you need systematic meta-analyses or reproducible benchmarks; this is a reading list, not a survey with code.