A Field Guide to the RL Exploration Wilderness

What it does This repository is a curated bibliography that catalogs research papers on exploration methods in reinforcement learning. It sorts works from venues like NeurIPS, ICML, and ICLR into chronological lists, then layers a custom taxonomy on top that splits methods into collect-phase strategies—such as action perturbation and state guidance—and train-phase strategies like count-based and prediction-based incentives. A visual diagram maps landmark algorithms including Go-Explore, RND, and ICM onto these buckets so you can orient yourself without reading every abstract.

The interesting bit Most awesome-lists are flat chronological dumps; this one tries to impose order on the chaos by classifying papers according to when they intervene in the RL loop. The taxonomy openly admits that categories overlap, but it still gives the field a rough shared vocabulary that a simple tag cloud cannot.

Key highlights

• Tracks papers from NeurIPS 2021 through ICML 2026, plus a dedicated section for classic pre-2021 works.
• Divides exploration methods into Augmented Collecting Strategy and Augmented Training Strategy, with subcategories like Parameter Space Perturbation and Entropy Augmented.
• Provides a diagram mapping specific landmark papers to their taxonomic buckets.
• Uses a MiniGrid hard-exploration screenshot to illustrate why sparse-reward navigation is worth studying.

Caveats

• This is strictly a reading list: no code, no implementations, and no reproduced benchmarks.
• The taxonomy notes that algorithm categories overlap, so the boundaries are fuzzy rather than absolute.

Verdict Bookmark it if you are writing a literature review or designing a new intrinsic reward. Skip it if you need copy-pasteable training scripts.