Geometrically Constrained Gaussian Splatting SLAM

Guanghao Li, Qi Chen, Sijia Hu, Yuxiang Yan, Jian Pu

Keywords: 3DGS, SLAM, Robotics

TL;DR: We propose a dense RGB-D SLAM system that leverages an implicit TSDF hash grid to constrain the distribution of Gaussian ellipsoids.

Abstract: 3D Gaussian Splatting (3DGS) has emerged as a promising technique in SLAM due to its rapid and high-quality rendering capabilities. However, its reliance on discrete Gaussian ellipsoid primitives limits its effectiveness in capturing essential geometric features crucial for accurate pose estimation. To overcome this limitation, we propose a novel dense RGB-D SLAM system that integrates an implicit Truncated Signed Distance Function (TSDF) hash grid to constrain the distribution of Gaussian ellipsoids. This innovative approach enables precise estimation of the scene's geometric structure by smoothing the discrete Gaussian ellipsoids and anchoring them to the scene's surface. Acting as a low-pass filter, the implicit TSDF hash grid mitigates the inductive biases inherent in traditional 3DGS methods while preserving rendering quality. Our geometrically constrained map also significantly enhances generalization capabilities for depth estimation in novel views. Extensive experiments on the Replica, ScanNet, and TUM datasets demonstrate that our system achieves state-of-the-art tracking and mapping accuracy at speeds up to 30 times faster than existing 3DGS-based systems.

Primary Area: applications to robotics, autonomy, planning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9584