Modular Safety Guardrails Are Necessary for Foundation-Model-Enabled Robots in the Real World

Joonkyung Kim, Wenxi Chen, Davood Soleymanzadeh, Yi Ding, Xiangbo Gao, Zhengzhong Tu, Ruqi Zhang, Fan Fei, Sushant Veer, Yiwei Lyu, Minghui Zheng, Yan Gu · 2026

Read Modular Safety Guardrails Are Necessary for Foundation-Model-Enabled Robots in the Real World by Joonkyung Kim, Wenxi Chen, Davood Soleymanzadeh, Yi Ding, Xiangbo Gao, Zhengzhong Tu, Ruqi Zhang, Fan Fei, Sushant Veer, Yiwei Lyu, Minghui Zheng, Yan Gu (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-2602-04056.txt.

Description

Robots driven by foundation models need safety in layers, argue Joonkyung Kim and colleagues. Their 2026 paper divides robot safety into action safety, decision safety, and human-centered safety. A physically safe motion can still be semantically wrong or violate what a person actually wanted.

Static verification, single controllers, and end-to-end learned policies each miss parts of that stack, especially as tasks stay open-ended. The authors propose modular guardrails: monitoring layers that evaluate behavior and intervention layers that can step in. Modularity lets each layer adapt as tasks, environments, and human expectations shift over time.

Catalog record

CREATORJoonkyung Kim, Wenxi Chen, Davood Soleymanzadeh, Yi Ding, Xiangbo Gao, Zhengzhong Tu, Ruqi Zhang, Fan Fei, Sushant Veer, Yiwei Lyu, Minghui Zheng, Yan Gu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-04056

Browse the full library · About the Tower of Babel