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

TeraSim-World: Worldwide Safety-Critical Data Synthesis for End-to-End Autonomous Driving

arXiv 25.9 2025 47.6 system, application

TLDR

TeraSim-World synthesizes realistic, geographically diverse safety-critical data for end-to-end autonomous driving using real-world maps and video generation.

Reasoning

The paper presents a novel pipeline that combines real-world geospatial data with a video generation model to create safety-critical driving scenarios, addressing the sim-to-real gap. However, the abstract lacks empirical validation or comparison to existing methods, and the reliance on a video generation model may introduce artifacts.

Read-first score

Read-first score 47.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 15.

Recency 8%
86.7

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

Methodology quality 25%
80

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

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%
21.4

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: 495.

Keyword Scores

world simulator
4
generative world model
3
video world model
3
world model
2
interactive world model
1
world dynamics prediction
1
model-based reinforcement learning world model
1

Deep Analysis

Innovations

  • Automated pipeline for worldwide safety-critical data synthesis for end-to-end autonomous driving
  • Integration of real-world geospatial data (maps, traffic demand) with naturalistic driving datasets to simulate agent behaviors
  • Use of the frontier video generation model Cosmos-Drive for photorealistic, geographically grounded sensor rendering
  • Bridging agent and sensor simulations to reduce the sim-to-real gap

Methodology

TeraSim-World retrieves real-world maps and traffic demand from geospatial data sources for an arbitrary location. It simulates agent behaviors from naturalistic driving datasets and orchestrates diverse adversities to create corner cases. Finally, it uses the Cosmos-Drive video generation model to produce photorealistic sensor renderings informed by street views of the same location, bridging agent and sensor simulations.

Key Results

The abstract does not present specific experimental results; it describes the pipeline's design and claims it provides a scalable and critical data synthesis framework for training and evaluation of end-to-end autonomous driving systems.

Tags