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

World Model-Based End-to-End Scene Generation for Accident Anticipation in Autonomous Driving

arXiv 25.7 2025 53.5 method, application

TLDR

Proposes world model-based video generation and temporal reasoning to improve accident anticipation in autonomous driving, with a new benchmark dataset.

Reasoning

The paper addresses data scarcity and object-level cue absence by combining a world model for generative scene augmentation with adaptive temporal reasoning, and releases a new benchmark. Strengths include a clear problem motivation and empirical validation on public and new datasets; weaknesses are limited architectural details of the world model and lack of comparison to other world model approaches.

Read-first score

Read-first score 53.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 29.

Recency 8%
86.7

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

Methodology quality 25%
70

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

Reproducibility 25%
46

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset

Topical relevance 42%
41.4

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 397.

Keyword Scores

world model
9
video world model
8
generative world model
7
world dynamics prediction
2
world simulator
1
interactive world model
1
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • Generative scene augmentation using a world model guided by domain-informed prompts to create high-resolution, statistically consistent driving scenarios, enriching edge cases and complex interactions.
  • Dynamic prediction model with strengthened graph convolutions and dilated temporal operators to encode spatio-temporal relationships, addressing data incompleteness and transient visual noise.
  • Release of a new benchmark dataset designed to capture diverse real-world driving risks.

Methodology

The framework combines a video generation pipeline that leverages a world model with domain-informed prompts to produce high-resolution driving scenarios, and a dynamic prediction model that uses strengthened graph convolutions and dilated temporal operators for spatio-temporal encoding. Experiments are conducted on public and newly released datasets, evaluating accuracy and lead time of accident anticipation.

Key Results

Extensive experiments on public and newly released datasets demonstrate that the framework enhances both the accuracy and lead time of accident anticipation, offering a robust solution to current data and modeling limitations.

Resources

Paper

Tags