HPGS-SLAM: Hybrid Point-Guided Dense Visual SLAM With Online Mapping via Gaussian Splatting
Xin Su, Xiaoang Zhang, Rastin Pries, Eckehard G. Steinbach
Abstract
In this letter, we introduce HPGS-SLAM, a real-time RGB-D SLAM system guided by hybrid point features (combining traditional and learned point features), enabling high-precision tracking and online dense mapping with photorealistic reconstruction. HPGS-SLAM consists of two main components: (1) a lightweight feature-based frontend guided by hybrid points with adaptive learnable feature matching, aiming for accurate pose tracking and 3D landmarks generation; and (2) a backend that leverages 3D Gaussian Splatting for real-time dense mapping and photorealistic rendering, where the spawning of Gaussian primitives is guided by the 3D landmarks and hybrid keypoints shared from the frontend. HPGS-SLAM is designed in a distributed architecture to facilitate practical deployment. We evaluate HPGS-SLAM on the Replica, TUM-RGBD, and EuRoC MAV datasets. Both quantitative and qualitative results demonstrate that HPGS-SLAM outperforms existing systems in tracking accuracy and mapping efficiency, while achieving competitive visual quality for visual rendering.
BibTeX
@inproceedings{ral2026_hpgsslamhybridpo,
title = {HPGS-SLAM: Hybrid Point-Guided Dense Visual SLAM With Online Mapping via Gaussian Splatting},
author = {Xin Su and Xiaoang Zhang and Rastin Pries and Eckehard G. Steinbach},
booktitle = {RA-L 2026},
year = {2026}
}