UniAda: Universal Adaptive Multi-objective Adversarial Attack for End-to-End Autonomous Driving Systems

Jingyu Zhang, Jacky Wai Keung, Yan Xiao, Yihan Liao, Yishu Li, Xiaoxue Ma · 2026

Read UniAda: Universal Adaptive Multi-objective Adversarial Attack for End-to-End Autonomous Driving Systems by Jingyu Zhang, Jacky Wai Keung, Yan Xiao, Yihan Liao, Yishu Li, Xiaoxue Ma (2026) 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-2604-23362.txt.

Description

Zhang and five colleagues introduce UniAda, a multi-objective white-box attack for autonomous driving systems that perturbs both steering and speed. Existing attacks focus on steering, missing speed controls and imperceptible perturbations. UniAda crafts an image-agnostic perturbation optimised jointly for steering and speed through an Adaptive Weighting Scheme. Tested on simulated and real driving data, it beats five benchmarks on two metrics. It induces steering deviations from 3.54 to 29 degrees and speed deviations from 11 to 22 km per hour on average.

Catalog record

CREATORJingyu Zhang, Jacky Wai Keung, Yan Xiao, Yihan Liao, Yishu Li, Xiaoxue Ma
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75714
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.23362

Browse the full library · About the Tower of Babel