Real-Time Evaluation of Autonomous Systems under Adversarial Attacks

Adithya Mohan, Xujun Xie, Venkatesh Thirugnana Sambandham, Torsten Schön · 2026

Read Real-Time Evaluation of Autonomous Systems under Adversarial Attacks by Adithya Mohan, Xujun Xie, Venkatesh Thirugnana Sambandham, Torsten Schön (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-2605-03491.txt.

Description

Adithya Mohan, Xujun Xie, Venkatesh Thirugnana Sambandham, and Torsten Schön evaluate self-driving policies under attack. They use real intersection data instead of pure simulation. Virtual testing misses structural and oversight effects in real recordings. Those effects shape how robust a policy really is.

Within a controlled data contract, they compare three approaches. The approaches are plain behavior cloning, transformer-based cloning with object tokens, and adversarial imitation learning. Nominal errors were similarly tiny. Yet robustness profiles split sharply by architecture and state design. Gradient attacks at inference pushed final position errors up to about eight meters. The work gives teams a clear way to test such systems before they go on the road.

Catalog record

CREATORAdithya Mohan, Xujun Xie, Venkatesh Thirugnana Sambandham, Torsten Schön
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.03491

Browse the full library · About the Tower of Babel