IRIS-SLAM: Unified Geo-Instance Representations for Robust Semantic Localization and Mapping

Tingyang Xiao 1 , Liu Liu 1 , Wei Feng 1 , Zhengyu Zou 1,3 , Xiaolin Zhou 1 , Wei Sui 2 , Hao Li 3 , Dingweng Zhang 3 , and Zhizhong Su 1 · 2026

Read IRIS-SLAM: Unified Geo-Instance Representations for Robust Semantic Localization and Mapping by Tingyang Xiao 1 , Liu Liu 1 , Wei Feng 1 , Zhengyu Zou 1,3 , Xiaolin Zhou 1 , Wei Sui 2 , Hao Li 3 , Dingweng Zhang 3 , and Zhizhong Su 1 (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-2602-18709.txt.

Description

Robots that map a room need both geometry and meaning. They must know where surfaces are, and which object is which, even seen from far apart. Tingyang Xiao, with eight co-authors, presents IRIS-SLAM in 2026. It is a semantic mapping system that unifies those two jobs in one representation. That representation builds on a geometry foundation model extended with instance information.

The same model predicts dense geometry and cross-view instance embeddings. Matching across views and recognising a place again, known as loop closure, both draw on viewpoint-agnostic semantic anchors. That shared basis bridges geometric reconstruction and open-vocabulary mapping. Tests show clear gains over leading methods. Gains are largest in map consistency and in reliable loop closure over wide baselines.

Catalog record

CREATORTingyang Xiao 1 , Liu Liu 1 , Wei Feng 1 , Zhengyu Zou 1,3 , Xiaolin Zhou 1 , Wei Sui 2 , Hao Li 3 , Dingweng Zhang 3 , and Zhizhong Su 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-18709

Browse the full library · About the Tower of Babel