Skip to content
Bud's Wiki

Awesome World Models

TL;DR

Section titled “TL;DR”

A curated awesome-list of world-modeling papers and resources, organized by application domain: Game Simulation (pixel- and 3D-mesh-space), Autonomous Driving, Embodied AI, Science, plus sections on definitions, surveys, positions, theory/explainability, general approaches, and evaluation. Launched October 2025; passed 1k GitHub stars within 30 days. Aims to bridge the divergent “world model” usage across communities (Embodied AI / AV / NLP / video generation) by collecting them in one minimalist map. Useful as a navigation index when surveying a sub-area, not as a synthesized review.

Key claims

Section titled “Key claims”
  • The repo’s stated motivation is that “world model” is used with varying definitions across domains, and a single curated index is needed to bridge those communities [Aim of the Project].
  • The repo treats two papers as canonical definitional sources for world models: Ha & Schmidhuber 2018 (arXiv:1803.10122) and Yann LeCun’s “A Path Towards Autonomous Machine Intelligence” position paper [Definition of World Models].
  • Coverage is organized into 10 top-level sections: definition, surveys, game simulation, autonomous driving, embodied AI, science, positions, theory/explainability, general approaches, and evaluation [Overview].
  • The Surveys section enumerates 5 sub-clusters: (1) video generation, (2) 3D generation, (3) embodied AI, (4) autonomous driving, (5) other (safety, memory, child-like learning) [Surveys of World Models].
  • The Autonomous Driving section is explicitly maintained as a “more-than-concise” unsorted list and defers to the LMD0311/Awesome-World-Model repo for full coverage [§ World Models for Autonomous Driving Note].
  • The repo is CC0-1.0 licensed and accepts community PRs [README header].

Method

Section titled “Method”

A flat README-as-resource — each entry is Name, "Paper Title" plus badge links to arXiv, project website, and code. Entries within sections are unsorted (chronological-ish, depending on section). Notable curation choices: marking a subset of entries with ⭐️ as “must-reads”; splitting Game Simulation into Pixel Space vs 3D Mesh Space; embedding two prior awesome-lists (LMD0311/Awesome-World-Model for AV, leofan90/Awesome-World-Models for robotics) as upstream sources rather than re-listing.

Results

Section titled “Results”

The repo crossed 1k stars within 30 days of October 2025 launch. It is one of three active “awesome-world-models” lists that the wiki currently sees referenced (the others being LMD0311’s AV-focused list, leofan90’s robotics-focused list, and M-E-AGI-Lab’s general-purpose list cited under “From Masks to Worlds”). The Definition section pins the field to two canonical sources (Ha & Schmidhuber 2018; LeCun 2022), which matches how filed papers on world models in the wiki tend to cite their lineage.

Why it’s interesting

Section titled “Why it’s interesting”

Acts as a hub-page for the wiki’s World Foundation Models cluster — when a new paper lands and the question is “what was the prior art on physics-grounded video generation / driving world models / interactive game world models”, this list is the fastest map. The Embodied AI surveys section in particular (arXiv:2510.16732 “A Comprehensive Survey on World Models for Embodied AI” and arXiv:2507.00917) are worth fetching independently if/when the wiki’s WFM cluster needs synthesis beyond what filed papers cover. Note that the list does not synthesize — it’s an index, not a review — so claims about which sub-area “matters” or which approach dominates have to come from reading the underlying entries, not from this README.

See also

Section titled “See also”