⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

A Survey of World Models for Autonomous Driving

arXiv 25.01 2025 57.7 survey, application

TLDR

A systematic survey of world models for autonomous driving, proposing a three-tiered taxonomy covering generation, planning, and interaction.

Reasoning

The paper provides a comprehensive taxonomy and review of world models in autonomous driving, covering generation methods, behavior planning, and interaction. Strengths include clear categorization and coverage of training paradigms; weaknesses include lack of new empirical results and limited discussion of real-world deployment challenges.

Read-first score

Read-first score 57.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%
67.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%
60

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=analysis

Reproducibility 25%
30

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 351.

Keyword Scores

world model
10
world dynamics prediction
9
generative world model
8
interactive world model
7
model-based reinforcement learning world model
6
world simulator
4
video world model
3

Deep Analysis

Innovations

  • Proposes a three-tiered taxonomy for world models in autonomous driving: Generation of Future Physical World, Behavior Planning for Intelligent Agents, and Interaction between Prediction and Planning
  • Covers diverse generation methods including Image-, BEV-, OG-, and PC-based approaches enhanced by diffusion models and 4D occupancy forecasting
  • Integrates rule-driven and learning-based planning paradigms with cost map optimization and reinforcement learning
  • Introduces latent space diffusion and memory-augmented architectures for multi-agent collaborative decision-making
  • Analyzes training paradigms such as self-supervised learning, multimodal pretraining, and generative data augmentation

Methodology

This paper systematically reviews recent advances in world models for autonomous driving, proposing a three-tiered taxonomy that categorizes methods into generation, planning, and interaction. It further analyzes training paradigms and evaluates world models' performance in scene understanding and motion prediction tasks, providing a comprehensive technical roadmap.

Key Results

The survey provides a structured categorization of world model approaches and identifies key training paradigms, offering a technical roadmap for advancing safe and reliable autonomous driving solutions through world models.

Limitations

  • Self-supervised representation learning remains a key challenge for practical deployment
  • Multimodal fusion of diverse sensor data is not yet fully resolved
  • Advanced simulation environments are needed to bridge the gap between research and real-world complex urban driving

Resources

Paper

Tags