Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Embracing Dynamics: Dynamics-aware 4D Gaussian Splatting SLAM

Zhicong Sun, Jacqueline Lo, Jinxing Hu

TL;DR

This work tackles pose drift and incomplete maps in dynamic environments by replacing static-geometry SLAM with a 4D Gaussian Map that captures spatio-temporal scene structure. It introduces D4DGS-SLAM, integrating LEAP-based dynamics-aware tracking and a dynamics-aware 4DGS mapping pipeline, including isotropic regularization that allocates Gaussians to dynamic regions. The approach yields state-of-the-art tracking and high-fidelity dynamic maps on real datasets (BONN and TartanAir-Shibuya), outperforming 3DGS and NeRF-based baselines in both pose accuracy and rendering quality. The results demonstrate the practicality of dynamic scene modeling for robust SLAM, while highlighting future work on occlusion handling and efficiency for resource-constrained platforms.

Abstract

Simultaneous localization and mapping (SLAM) technology has recently achieved photorealistic mapping capabilities thanks to the real-time, high-fidelity rendering enabled by 3D Gaussian Splatting (3DGS). However, due to the static representation of scenes, current 3DGS-based SLAM encounters issues with pose drift and failure to reconstruct accurate maps in dynamic environments. To address this problem, we present D4DGS-SLAM, the first SLAM method based on 4DGS map representation for dynamic environments. By incorporating the temporal dimension into scene representation, D4DGS-SLAM enables high-quality reconstruction of dynamic scenes. Utilizing the dynamics-aware InfoModule, we can obtain the dynamics, visibility, and reliability of scene points, and filter out unstable dynamic points for tracking accordingly. When optimizing Gaussian points, we apply different isotropic regularization terms to Gaussians with varying dynamic characteristics. Experimental results on real-world dynamic scene datasets demonstrate that our method outperforms state-of-the-art approaches in both camera pose tracking and map quality.

Embracing Dynamics: Dynamics-aware 4D Gaussian Splatting SLAM

TL;DR

This work tackles pose drift and incomplete maps in dynamic environments by replacing static-geometry SLAM with a 4D Gaussian Map that captures spatio-temporal scene structure. It introduces D4DGS-SLAM, integrating LEAP-based dynamics-aware tracking and a dynamics-aware 4DGS mapping pipeline, including isotropic regularization that allocates Gaussians to dynamic regions. The approach yields state-of-the-art tracking and high-fidelity dynamic maps on real datasets (BONN and TartanAir-Shibuya), outperforming 3DGS and NeRF-based baselines in both pose accuracy and rendering quality. The results demonstrate the practicality of dynamic scene modeling for robust SLAM, while highlighting future work on occlusion handling and efficiency for resource-constrained platforms.

Abstract

Simultaneous localization and mapping (SLAM) technology has recently achieved photorealistic mapping capabilities thanks to the real-time, high-fidelity rendering enabled by 3D Gaussian Splatting (3DGS). However, due to the static representation of scenes, current 3DGS-based SLAM encounters issues with pose drift and failure to reconstruct accurate maps in dynamic environments. To address this problem, we present D4DGS-SLAM, the first SLAM method based on 4DGS map representation for dynamic environments. By incorporating the temporal dimension into scene representation, D4DGS-SLAM enables high-quality reconstruction of dynamic scenes. Utilizing the dynamics-aware InfoModule, we can obtain the dynamics, visibility, and reliability of scene points, and filter out unstable dynamic points for tracking accordingly. When optimizing Gaussian points, we apply different isotropic regularization terms to Gaussians with varying dynamic characteristics. Experimental results on real-world dynamic scene datasets demonstrate that our method outperforms state-of-the-art approaches in both camera pose tracking and map quality.

Paper Structure

This paper contains 15 sections, 13 equations, 7 figures, 4 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Work
  3. METHOD
  4. 4D Gaussian Map Representation
  5. Long-term Effective Any Point Tracking
  6. SLAM System
  7. Dynamics-aware Tracking
  8. Keyframing
  9. Dynamics-aware 4DGS Mapping
  10. EXPERIMENT
  11. Experimental Setup.
  12. Evaluation of Camera Tracking Performance
  13. Evaluation of Mapping Performance
  14. Ablation study
  15. CONCLUSIONS

Figures (7)

  • Figure 1: Comparison of rendering results between 3DGS-based MonoGS and our 4DGS-based D4DGS-SLAM on Bonn and TartanAir-Shibuya Datasets.
  • Figure 2: Overview of D4DGS-SLAM
  • Figure 3: Visualization of anchors.
  • Figure 4: Anchors with high dynamics (in red) and low reliability (in yellow)
  • Figure 5: Comparison of Gaussian distributions with and without dynamics-aware InfoModule
  • ...and 2 more figures