knightnemo/Awesome-World-Models: Curated Research Hub for World Modeling

knightnemo/Awesome-World-Models: Curated Research Hub for World Modeling

World modeling research is exploding across disciplines—embodied AI, autonomous driving, game simulation, NLP—but finding relevant papers means hunting through disconnected communities with conflicting definitions of what a "world model" even is. Researchers waste hours reconciling terminology from robotics, computer vision, and reinforcement learning literature. knightnemo/Awesome-World-Models addresses this fragmentation directly, offering a structured, community-maintained index that spans domains without diluting technical precision.

What is knightnemo/Awesome-World-Models?

knightnemo/Awesome-World-Models is a curated GitHub repository—not a software library or framework—that organizes academic and industry research on world models across multiple application domains. Maintained by Siqiao Huang and contributors, it reached 3,159 GitHub stars and 134 forks as of its last commit on July 14, 2026. The repository operates under a BSD 3-Clause license and explicitly positions itself as "a one-stop resource for researchers, practitioners, and enthusiasts interested in world modeling."

The project's core premise is that world models have become a "hot topic" attracting "unprecedented attention," yet the term is applied inconsistently across fields. Rather than imposing a single definition, the repository maps how different communities use world models—from Ha and Schmidhuber's original 2018 paper to Yann LeCun's autonomous machine intelligence framework—while tracking implementations in embodied AI, autonomous driving, game simulation, scientific computing, and NLP.

Notably, the repository builds upon two prior specialized lists: Awesome-World-Model-for-Autonomous-Driving and Awesome-World-Model-for-Robotics. Its rapid growth—exceeding 1,000 stars within 30 days of launch in late 2025—suggests strong demand for cross-domain synthesis in this research area.

Key Features

Cross-Domain Taxonomy — The repository organizes papers into twelve primary categories: tutorials and starter resources, surveys, game simulation, autonomous driving, embodied AI, scientific applications, position papers, theory and explainability, general approaches, evaluation methodologies, acknowledgements, and citations. This structure lets researchers trace how world model techniques transfer between domains.

Visual Navigation System — Since October 2025, all entries display standardized badges for arXiv papers, project websites, and code repositories. This eliminates the friction of hunting down implementations for papers that initially appear theoretical.

Curated Starter Resources — The repository prioritizes accessible entry points, highlighting implementations like Nano World Models (a minimalist future video prediction system), minWM (a full-stack real-time interactive video world model framework), StableWM (a reproducibility-focused platform), and WorldFoundry (unified inference and evaluation infrastructure).

Historical Anchoring — Rather than treating world models as a 2020s phenomenon, the repository explicitly traces the concept to two foundational sources: Ha and Schmidhuber's 2018 "World Models" paper and LeCun's 2022 OpenReview position piece on autonomous machine intelligence. This grounding prevents newer researchers from missing architectural precedents.

Community Contribution Workflow — The repository accepts pull requests for paper submissions and provides direct contact via siqiaohuang981@gmail.com, with explicit calls for domain experts to improve underdeveloped sections—notably autonomous driving, where the maintainer acknowledges non-expertise.

Use Cases

Literature Review Acceleration — Researchers entering world modeling can bypass weeks of scattered searching. The surveys section alone indexes fifteen major review papers from 2024-2025, including domain-specific treatments for embodied AI, autonomous driving, 3D/4D generation, and video generation.

Implementation Benchmarking — Practitioners evaluating world model frameworks can compare directly-linked codebases. The game simulation section, for instance, pairs papers like DIAMOND (diffusion-based Atari world modeling) and Matrix-Game 3.0 (real-time streaming interactive world models with long-horizon memory) with their GitHub repositories.

Cross-Domain Technique Transfer — A robotics researcher might discover that DriveDreamer-2's LLM-enhanced approach to diverse driving video generation shares architectural patterns with DreamGen's generalization strategy for robot learning—connections the repository's organization makes visible.

Course and Reading List Construction — Educators can build syllabi around the repository's tiered structure, using tutorials like iVideoGPT for foundational interactive video world models, then progressing through survey papers to current research frontiers.

Industry Technology Scouting — Engineering teams at autonomous driving or robotics companies can track which world model approaches have open implementations versus theoretical proposals, using the badge system to filter rapidly.

Installation & Setup

As a curated list rather than executable software, knightnemo/Awesome-World-Models requires no traditional installation. However, researchers typically interact with it through several workflows:

Clone for Local Reference

# Clone the repository for offline access and personal annotation
git clone https://github.com/knightnemo/Awesome-World-Models.git
cd Awesome-World-Models

Browse Structured Content

The repository uses standard Markdown^{↗ Smart Converter} with anchor-linked sections. Access specific domains directly via:

# Open the rendered README in your default browser
# (after cloning, or view directly on GitHub)
open README.md  # macOS
xdg-open README.md  # Linux

Contribute New Papers

# Fork the repository through GitHub UI, then clone your fork
git clone https://github.com/YOUR_USERNAME/Awesome-World-Models.git
cd Awesome-World-Models

# Create a feature branch for your additions
git checkout -b add-paper-[paper-name]

# Edit README.md following the existing badge format
# Submit pull request via GitHub

The repository includes CONTRIBUTING.md with submission guidelines. The maintainer emphasizes that community contributions are essential for keeping the autonomous driving section current, given acknowledged non-expertise in that domain.

Real Code Examples

The repository itself contains no executable code—it is a documentation resource. However, it surfaces implementations with direct code links. Below are representative examples of how the repository documents accessible implementations:

Example 1: Minimalist Video Prediction (Nano World Models)

The repository highlights Nano World Models as a pedagogical entry point:

Paper: "Nano World Models: A Minimalist Implementation of Future Video Prediction"
arXiv: 2605.23993
Website: https://simchowitzlabpublic.github.io/nano-world-model/
Code: https://github.com/simchowitzlabpublic/nano-world-model

This entry demonstrates the repository's value: a researcher seeking to understand world model fundamentals can immediately access a stripped-down implementation rather than parsing complex production systems.

Example 2: Real-Time Interactive Framework (minWM)

For practitioners needing full-stack capabilities:

Paper: "minWM: A Full-Stack Open-Source Framework for Real-Time Interactive Video World Models"
arXiv: 2605.30263
Code: https://github.com/shengshu-ai/minWM

The badge system makes the code link visually prominent, reducing the common friction where promising papers lack accessible implementations.

Example 3: Reproducibility Platform (StableWM)

Research groups prioritizing experimental validation:

Paper: "stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation"
arXiv: 2605.21800
Website: https://galilai-group.github.io/stable-worldmodel/
Code: https://github.com/galilai-group/stable-worldmodel

The repository explicitly flags such resources with [⭐️] markers, indicating recommended starting points. This curation layer—human judgment on which implementations merit attention—distinguishes it from automated paper aggregators.

Example 4: Unified Infrastructure (WorldFoundry)

For teams building evaluation pipelines:

Paper: "WorldFoundry: Unified World Model Inference & Evaluation Infrastructure"
Website: https://openenvision.github.io/WorldFoundry/
Code: https://github.com/OpenEnvision/WorldFoundry

Note that this entry lacks an arXiv badge, accurately reflecting that the repository documents resources regardless of publication venue—a pragmatic choice given the rapid preprint culture in world modeling research.

Advanced Usage & Best Practices

Domain-Specific Deep Dives — While the repository provides breadth, serious practitioners should cross-reference with the specialized lists it builds upon. The autonomous driving section explicitly defers to Awesome-World-Model-for-Autonomous-Driving for comprehensive coverage, acknowledging maintainer limitations.

Temporal Awareness — The "News & Updates" section tracks repository milestones but not individual paper advances. Researchers should verify paper versions against arXiv dates, as the repository may lag behind rapid revision cycles common in this field.

Badge Verification — The standardized badges (arXiv, Website, Code) are added by contributors, not automatically validated. Before relying on code links for reproduction, verify repository activity—some linked projects may be research prototypes without ongoing maintenance.

Citation Practice — The repository requests citation of the list itself for academic use. This supports community maintenance while acknowledging that individual paper credit belongs to original authors.

Cross-Reference with Theory Section — Practitioners implementing world models should consult the "Theory & World Models Explainability" section, which includes critical perspectives on whether current approaches genuinely model world dynamics or merely predict futures—a distinction with significant architectural implications.

Comparison with Alternatives

Trade-offs: The cross-domain scope of knightnemo/Awesome-World-Models sacrifices some depth in any single area—explicitly acknowledged in the autonomous driving section's "call for maintenance." Conversely, specialized lists may miss technique transfer opportunities visible only through cross-domain organization. Papers With Code offers superior benchmark tracking but lacks the curated narrative and historical framing this repository provides.

FAQ

What license covers the repository? BSD 3-Clause "New" or "Revised" License.

Can I use this for commercial research? Yes, BSD 3-Clause permits commercial use with attribution requirements.

How current is the paper list? Last commit July 14, 2026; community contributions accepted via PR.

Does the repository include code? No—it indexes papers and implementations with direct links to external repositories.

How do I submit a paper? Fork, edit README.md following existing badge format, and submit a PR per CONTRIBUTING.md.

Is the autonomous driving section complete? Explicitly noted as incomplete; the maintainer seeks domain expert contributions.

What's the difference from specialized lists? This repository bridges embodied AI, driving, NLP, and science, while prior lists focus on single domains.

Conclusion

knightnemo/Awesome-World-Models serves a specific, valuable function in the research ecosystem: it reduces the friction of navigating a terminologically fractured field without imposing artificial uniformity. For researchers whose work spans embodied AI and autonomous driving, or who need to trace how video generation techniques migrate to scientific simulation, this cross-domain organization saves substantial discovery time.

The repository is best suited for: graduate students entering world modeling research, interdisciplinary teams building on multiple world model traditions, and practitioners seeking to distinguish mature implementations from theoretical proposals. Its acknowledged limitations—particularly in autonomous driving depth—are transparently communicated rather than obscured.

The 3,159-star growth trajectory and explicit community contribution model suggest sustainable maintenance if domain experts respond to the maintainer's calls for participation. For current world modeling research, it belongs alongside domain-specific lists in any researcher's bookmark collection.

Explore the full curated collection at https://github.com/knightnemo/Awesome-World-Models.