CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation

Max Fu, Justin Yu, Karim El-Refai, Ethan Kou, Haoru Xue, Huang Huang, Wenli Xiao, Guanzhi Wang, Fei-Fei Li, Guanya Shi, Jiajun Wu, Shankar Sastry, Yuke Zhu, Ken Goldberg, Linxi “Jim” Fan · 2026

Read CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation by Max Fu, Justin Yu, Karim El-Refai, Ethan Kou, Haoru Xue, Huang Huang, Wenli Xiao, Guanzhi Wang, Fei-Fei Li, Guanya Shi, Jiajun Wu, Shankar Sastry, Yuke Zhu, Ken Goldberg, Linxi “Jim” Fan (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-2603-22435.txt.

Description

Fu and co-authors present CaP-X, an open framework for studying coding agents that control robots. Inside it, CaP-Gym lets agents write and run programs that combine perception and control steps. Their CaP-Bench tests twelve frontier models at varied levels of abstraction and grounding. Scores rise with hand-crafted aids but fall as those aids are removed. The team shows that more test-time interaction and feedback can close much of the gap. Their CaP-Agent0 then matches human-level reliability on several tasks.

Catalog record

CREATORMax Fu, Justin Yu, Karim El-Refai, Ethan Kou, Haoru Xue, Huang Huang, Wenli Xiao, Guanzhi Wang, Fei-Fei Li, Guanya Shi, Jiajun Wu, Shankar Sastry, Yuke Zhu, Ken Goldberg, Linxi “Jim” Fan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE132069
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.22435

Browse the full library · About the Tower of Babel