Pedestrian Crossing, Paris, France
Agreesive Merge, Rossfeld, Germany
Fail-to-Yield, Ann Arbor, U.S.
Deny Merge, Chicago, U.S.
Highway Zigzag, Arizona, U.S.
Redlight Running, San Diego, U.S.
Safe and scalable deployment of end-to-end (E2E) autonomous driving requires extensive and diverse data, particularly safety-critical events. Existing data are mostly generated from simulators with a significant sim-to-real gap or collected from on-road testing that is costly and unsafe.
This paper presents TeraSim-World, an automated pipeline that synthesizes realistic and geographically diverse safety-critical data for E2E autonomous driving at anywhere in the world. Starting from an arbitrary location, TeraSim-World retrieves real-world maps and traffic demand from geospatial data sources. Then, it simulates agent behaviors from naturalistic driving datasets, and orchestrates diverse adversities to create corner cases. Inspired by street views of the same location, it achieves photorealistic, geographically grounded sensor rendering via the frontier video generation model Cosmos-Drive.
By bridging agent and sensor simulations, TeraSim-World provides a scalable and critical data synthesis framework for training and evaluation of E2E autonomous driving systems.
We showcase some representative safety-critical events from TeraSim-World. More synthesized data will be released.
Pedestrian Crossing
Test facilities like Mcity provide a controlled environment, but its miniature setup cannot fully reflect the real-world complexity. With TeraSim-World, existing facilities can now generate more realistic and diverse scenarios.
We also provide resources and related work for TeraSim-World.
We will continue to release supplementary materials and more synthesized data as they become available.
@article{wang2025terasim-world,
author = {Wang, Jiawei and Sun, Haowei and Yan, Xintao and Feng, Shuo and Gao, Jun and Liu, Henry},
title = {TeraSim-World: Worldwide Safety-Critical Data Synthesis for End-to-End Autonomous Driving},
journal = {arXiv preprint arXiv:2509.13164},
year = {2025},
}
Based on the template from this site