arXiv Robotics — research abstracts· Maolin Ye, Dongqing Zhao, Linyang Li, Shuai Xue, Yuan Liu, Yiyuan Wang, Wen Li·· 2 days agoSignalEditorial score75
V3DSAM: Semantic-Geometric Loop Verification and Geometry-Anchored Persistent Instance Mapping for Feed-Forward Dense SLAM
V3DSAM: Semantic-Geometric Loop Verification and Geometry-Anchored Persistent Instance Mapping for Feed-Forward Dense SLAM
Summary
This research presents V3DSAM, an open-vocabulary semantic dense SLAM system designed to address unreliable cross-view association in long sequences.
Editorial context
The paper introduces V3DSAM, a feed-forward dense SLAM system that combines geometric estimates from VGGT-SLAM 2.0 with open-vocabulary observations from SAM3 to improve long-sequence cross-view association without relying on internal backbone features or temporal ID propagation.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org