UNIC: Learning Unified Multimodal Extrinsic Contact Estimation

Zhengtong Xu, Yuki Shirai · 2026

Read UNIC: Learning Unified Multimodal Extrinsic Contact Estimation by Zhengtong Xu, Yuki Shirai (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-2601-04356.txt.

Description

Zhengtong Xu and Yuki Shirai teach robots to feel where a held object touches the world, in a 2026 arXiv paper. Such extrinsic contacts — between the grasped object and its surroundings, not the gripper — carry the context a robot needs to plan and learn. earlier estimators assumed fixed grasps, known contact types, or calibrated cameras, and broke in messy settings.

Their framework, UNIC, needs no prior knowledge and no camera calibration. It reads the scene directly in camera coordinates, represents possible contacts as affordance maps. fuses vision with the robot's own body sensing and touch, using random masking during training so missing senses do not break it. In experiments it locates unseen contacts with average error under a centimetre, handles new objects and moving viewpoints, and stays robust when a modality drops out.

Catalog record

CREATORZhengtong Xu, Yuki Shirai
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-04356

Browse the full library · About the Tower of Babel