RGS-SLAM: Robust Gaussian Splatting SLAM with One-Shot Dense Initialization
Wei-Tse Cheng, Yen-Jen Chiou, Yuan-Fu Yang
2026-01-08
Summary
This paper introduces a new approach to building 3D maps called RGS-SLAM, which improves upon existing methods for creating these maps using a technology called Gaussian splatting.
What's the problem?
Current Gaussian splatting SLAM systems struggle with starting the map creation process and can be slow, especially in complex environments with lots of detail. They typically build the map bit by bit, reacting to errors, which isn't very efficient and can lead to inaccuracies, particularly when there's a lot of texture or clutter.
What's the solution?
RGS-SLAM takes a different approach. Instead of building the map gradually, it first creates a good initial 'seed' of 3D points represented as Gaussians. It does this by finding matching points in multiple images using a powerful image descriptor and then quickly converting those matches into a well-organized set of Gaussians. This initial seed is then refined, leading to faster map creation and better quality maps.
Why it matters?
This new method is important because it makes 3D mapping faster and more accurate, especially in challenging environments. It's also compatible with existing Gaussian splatting systems, meaning it can be easily integrated into current technology, and it can run very quickly – up to 925 frames per second – making it suitable for real-time applications like robotics and augmented reality.
Abstract
We introduce RGS-SLAM, a robust Gaussian-splatting SLAM framework that replaces the residual-driven densification stage of GS-SLAM with a training-free correspondence-to-Gaussian initialization. Instead of progressively adding Gaussians as residuals reveal missing geometry, RGS-SLAM performs a one-shot triangulation of dense multi-view correspondences derived from DINOv3 descriptors refined through a confidence-aware inlier classifier, generating a well-distributed and structure-aware Gaussian seed prior to optimization. This initialization stabilizes early mapping and accelerates convergence by roughly 20\%, yielding higher rendering fidelity in texture-rich and cluttered scenes while remaining fully compatible with existing GS-SLAM pipelines. Evaluated on the TUM RGB-D and Replica datasets, RGS-SLAM achieves competitive or superior localization and reconstruction accuracy compared with state-of-the-art Gaussian and point-based SLAM systems, sustaining real-time mapping performance at up to 925 FPS.