DynaGSLAM: Real-Time Gaussian-Splatting SLAM for Online Rendering, Tracking, Motion Predictions of Moving Objects in Dynamic Scenes

Runfa Blark Li*†, Mahdi Shaghaghi†, Keito Suzuki*, Xinshuang Liu*, Varun Moparthi*, Bang Du*, Walker Curtis†, Martin Renschler†, Ki Myung Brian Lee*, Nikolay Atanasov*, Truong Nguyen*
*UC San Diego, †Qualcomm XR Advanced Technology
Paper arXiv Code

Abstract

Simultaneous Localization and Mapping (SLAM) is one of the most important environment-perception and navigation algorithms for computer vision, robotics, and autonomous cars/drones. Hence, high quality and fast mapping becomes a fundamental problem. With the advent of 3D Gaussian Splatting (3DGS) as an explicit representation with excellent rendering quality and speed, state-of-the-art (SOTA) works introduce GS to SLAM. Compared to classical pointcloud-SLAM, GS-SLAM generates photometric information by learning from input camera views and synthesize unseen views with high-quality textures. However, these GS-SLAM fail when moving objects occupy the scene that violate the static assumption of bundle adjustment. The failed updates of moving GS affects the static GS and contaminates the full map over long frames. Although some efforts have been made by concurrent works to consider moving objects for GS-SLAM, they simply detect and remove the moving regions from GS rendering ("anti" dynamic GS-SLAM), where only the static background could benefit from GS. To this end, we propose the first real-time GS-SLAM, "DynaGSLAM", that achieves high-quality online GS rendering, tracking, motion predictions of moving objects in dynamic scenes while jointly estimating accurate ego motion. Our DynaGSLAM outperforms SOTA static & "Anti" dynamic GS-SLAM on three dynamic real datasets, while keeping speed and memory efficiency in practice.

Overview

We focus on three modules - Dynamic GS (a) Segmentation & Flow, (b) Management and (c) Tracking & Prediction. DynaGSLAM takes RGBD sequence as input to construct map with GS, (a) segment dynamic GS from static GS in 3D, and estimate dynamic GS 3D motion flow between frames. (b) Dynamic GS are managed separately from static GS with GS flow, but combined to jointly optimize. Case 1&2 are the rules for dynamic GS adding; "Cond. 1&2" denotes the conditions for dynamic GS deletion. (c) The optimized dynamic GS at current and past frames are used to interpolate/extrapolate dynamic GS in the continuous timeline from past to future. "CHS" refers to "cubic Hermite spline" and "LF" refers to "linear function". Please refer to the paper for details.

Results

TUM Dataset

Bonn Dataset

BibTeX

@misc{dynagslam,
      title={DynaGSLAM: Real-Time Gaussian-Splatting SLAM for Online Rendering, Tracking, Motion Predictions of Moving Objects in Dynamic Scenes}, 
      author={Runfa Blark Li and Mahdi Shaghaghi and Keito Suzuki and Xinshuang Liu and Varun Moparthi and Bang Du and Walker Curtis and Martin Renschler and Ki Myung Brian Lee and Nikolay Atanasov and Truong Nguyen},
      year={2025},
      eprint={2503.11979},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.11979}, 
}