Communicating about Space: Language-Mediated Spatial Integration Across Partial Views

Ankur Sikarwar, Debangan Mishra, Sudarshan Nikhil, [0.3em] Ponnurangam Kumaraguru, Aishwarya Agrawal, [0.5em] Mila -- Quebec AI Institute · 2026

Read Communicating about Space: Language-Mediated Spatial Integration Across Partial Views by Ankur Sikarwar, Debangan Mishra, Sudarshan Nikhil, [0.3em] Ponnurangam Kumaraguru, Aishwarya Agrawal, [0.5em] Mila -- Quebec AI Institute (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-2603-27183.txt.

Description

Sikarwar and co-authors introduce COSMIC, a benchmark for spatial communication between multimodal models. Two agents see one indoor scene from different viewpoints and chat to answer spatial questions. The set holds 899 scenes and 1,250 question-answer pairs across five tasks. Models proved best at spotting shared anchor objects across views. Relational reasoning was weaker, and global map building sat near chance. Extra thinking helped anchor grounding but not the higher skills. Humans reached 95 percent accuracy in collected dialogues. The best model managed only 72 percent.

Catalog record

CREATORAnkur Sikarwar, Debangan Mishra, Sudarshan Nikhil, [0.3em] Ponnurangam Kumaraguru, Aishwarya Agrawal, [0.5em] Mila -- Quebec AI Institute
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72602
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.27183

Browse the full library · About the Tower of Babel