Mapping 10 years of world model research (489 papers, 2012–2026)

Name: Mapping 10 years of world model research (489 papers, 2012–2026)
Availability: InStock
Author: Arthur12137

by Arthur12137·Mar 7, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●MidBig BrainRabbit Hole

Taxonomy of 489 papers, but GitHub Awesome lists + arXiv already fragment by paradigm.

Strengths

•Dual taxonomy (representation × application) provides genuine structure over flat lists
•Evolution tree visualizes paradigm shifts from latent RL to diffusion-based generators
•195 implementations tracked; venue + trend stats add curation value beyond aggregation

Weaknesses

•No novel insight or critical analysis—purely organizational
•Relies on community contributions; unmaintained lists become stale within months

Post Description

World models have evolved rapidly over the past decade — from latent model-based RL (PlaNet, Dreamer) to generative video simulators (Genie, DriveWorld) and embodied / multi-agent systems.

The literature is increasingly fragmented across RL, diffusion, 3D occupancy forecasting, safety-aware modeling, and autonomous driving.

I put together a structured repository organizing 489 papers (2012–2026) with:

– Dual taxonomy (representation × application) – A visual evolution tree – Venue & trend statistics – 195 open-source implementations

The goal is to organize the field into a coherent map rather than just an aggregated list.

Feedback and missing works are welcome.

https://github.com/Bowen12137/Awesome-World-Models