World Models Survey Repository

This repository serves as a comprehensive, curated resource list for "World Models," stemming from an ACM CSUR 2025 survey paper. It addresses the need for a structured overview of research in this rapidly evolving field, targeting researchers, engineers, and practitioners by consolidating key papers, models, and applications across various domains. The benefit lies in providing a centralized, organized entry point to understand and navigate the landscape of world models.

How It Works

The repository functions as a structured catalog of research on world models, organized according to the comprehensive survey paper. It categorizes world models by their core functionalities: implicit representation of external worlds, future prediction (encompassing video generation and embodied environments), and their diverse applications. This classification aids users in navigating the multifaceted landscape of world model research, highlighting key approaches and their interrelations.

Quick Start & Requirements

This is an informational resource, not a software project requiring installation. The primary entry point is the survey paper "Understanding World or Predicting Future? A Comprehensive Survey of World Models," available on arXiv. Interested parties should consult the individual research papers cited within the repository for specific implementation details, dependencies, and setup instructions.

Highlighted Details

• Extensive coverage of world model research, including foundational areas like Model-based Reinforcement Learning, Self-supervised Learning (e.g., JEPA, DINO-WM), and the integration of Large Language Models (LLMs/MLLMs).
• Detailed sections on future prediction paradigms such as Video Generation (e.g., Sora, CogVideo) and the creation of Interactive 3D Environments.
• Exploration of diverse applications, notably Game Intelligence, Embodied Intelligence, and Urban Intelligence, with a significant focus on Autonomous Driving.
• The repository is linked to an academic survey paper accepted by ACM Computing Surveys, ensuring a rigorous and comprehensive overview.

Maintenance & Community

The project is associated with the Tsinghua University FIB Lab. Contact is available via email (dingjt15@tsinghua.org.cn). The core survey paper has been accepted by ACM Computing Surveys, indicating academic backing. No specific community channels (e.g., Discord, Slack) are listed.

Licensing & Compatibility

No explicit software license is provided for the repository itself. The content is primarily a curated list of research papers. Users should refer to the licenses of individual cited works for their respective terms of use and compatibility.

Limitations & Caveats

This repository serves as a curated bibliography and informational guide, not a deployable software system. It does not offer code implementations, benchmarks, or direct execution capabilities for the world models discussed. The scope is limited to the research cataloged in the associated survey paper, and the rapid pace of development in AI means the field is constantly evolving beyond the survey's publication date.