ARGOS : Automated Functional Safety Requirement Synthesis for Embodied AI via Attribute-Guided Combinatorial Reasoning

Dongsheng Chen, Yuxuan Li, Yi Lin, Guanhua Chen, Jiaxin Zhang, Xiangyu Zhao, Lei Ma, Xin Yao, Xuetao Wei · 2026

Read ARGOS : Automated Functional Safety Requirement Synthesis for Embodied AI via Attribute-Guided Combinatorial Reasoning by Dongsheng Chen, Yuxuan Li, Yi Lin, Guanhua Chen, Jiaxin Zhang, Xiangyu Zhao, Lei Ma, Xin Yao, Xuetao Wei (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-2602-07007.txt.

Description

Dongsheng Chen and eight co-authors build a tool that writes safety requirements for embodied AI. Such robots act in open worlds on plain-language instructions. Old hazard analysis lists risks for fixed functions, but open instructions create risk combos that do not fit that mold. Language models can scale the work, yet their hazard write-ups often lack grounding in physics.

Their tool, ARGOS, splits each entity in an instruction into fine traits. That grounds the model's thinking in causal risk factors, so it can sketch hazard scenes that are physically plausible. It then maps broad safety standards, like ISO 13482, onto the robot's own skills to yield requirements fit to the context. In tests, ARGOS wrote high-quality requirements and beat rivals at spotting rare, long-tail risks.

Catalog record

CREATORDongsheng Chen, Yuxuan Li, Yi Lin, Guanhua Chen, Jiaxin Zhang, Xiangyu Zhao, Lei Ma, Xin Yao, Xuetao Wei
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE66 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07007

Browse the full library · About the Tower of Babel