Safety-Critical Learning for Long-Tail Events: The TUM Traffic Accident Dataset

Walter Zimmer ID, Ross Greer ID, Xingcheng Zhou ID, Rui Song ID, Marc Pavel ID, Daniel Lehmberg ID Ahmed Ghita ID Akshay Gopalkrishnan ID Mohan Trivedi ID Alois Knoll ID · 2025

Read Safety-Critical Learning for Long-Tail Events: The TUM Traffic Accident Dataset by Walter Zimmer ID, Ross Greer ID, Xingcheng Zhou ID, Rui Song ID, Marc Pavel ID, Daniel Lehmberg ID Ahmed Ghita ID Akshay Gopalkrishnan ID Mohan Trivedi ID Alois Knoll ID (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-14567.txt.

Description

Walter Zimmer and co-authors present a data set of real highway crashes in a 2025 paper on road safety. Rare, severe events are the hardest for safety systems to learn, since normal driving data contains almost none. Their TUM Traffic Accident set fills that gap with recordings of actual high-speed collisions.

The set holds ten crash sequences filmed by four roadside cameras and laser scanners. Nearly 300,000 two-dimensional boxes and 93,000 three-dimensional boxes are labelled across more than 48,000 frames, covering ten object classes. The team also built a detection model that mixes fixed rules with learned ones. Tests in the paper show the model is robust, and the data, model, and code are shared in public for other teams.

Catalog record

CREATORWalter Zimmer ID, Ross Greer ID, Xingcheng Zhou ID, Rui Song ID, Marc Pavel ID, Daniel Lehmberg ID Ahmed Ghita ID Akshay Gopalkrishnan ID Mohan Trivedi ID Alois Knoll ID
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE8 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-14567

Browse the full library · About the Tower of Babel