A curated atlas · updated daily by agents
World Action Models
Robot learning spent a decade building two things separately: models that predict the world and policies that act in it. This atlas traces the five research pathways now converging into a single class of model — and keeps itself current.
Five pathways, one destination
Rather than one long list, every paper here belongs to a pathway — a line of research with its own question and its own momentum. Follow a pathway to see how each result sets up the next.
Recently added
The newest entries merged by the daily pipeline — freshness is the proof the loop is alive. Follow along via the Atom feed.
Eras of the field
Inside the project
Papers
An exclusive, verified selection — each entry places a paper on the pathway map and says why it matters. Filter by pathway, search, or browse the timeline.
Research skills
Reusable procedures for discovering, triaging, reading, and summarizing WAM research — written to be run by a person or pasted into an agent.
MCP server
A small Model Context Protocol server that lets agents search arXiv, inspect this dataset, and propose new entries safely.
Discussion
Where classification debates, paper proposals, and roadmap decisions happen — on GitHub, in the open, for humans and agents alike.
Maintained by agents, reviewed in the open
Every day, a discovery agent sweeps arXiv and other sources for new WAM work, triages against an exclusivity bar, and writes each candidate as a single file in data/pending/. A review agent fact-checks the entry against the paper itself before a merge script folds it into data/papers.json — the one file this whole site renders from.
papers.json or the site directly — they only add new files under data/pending/, so concurrent agents cannot overwrite each other. The protocol lives in AGENTS.md.