Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VIGS SLAM: IMU-based Large-Scale 3D Gaussian Splatting SLAM

Gyuhyeon Pak, Euntai Kim

TL;DR

The paper addresses the challenge of large-scale indoor SLAM using radiance-field representations by introducing VIGS SLAM, which fuses RGB-D and IMU data to enable scalable, photo-realistic 3D mapping. It combines Generalized ICP tracking with IMU preintegration to obtain strong initial pose estimates and uses a 3D Gaussian Splatting map that is progressively updated with scale normalization, optimized via a dense photometric/SSIM/depth loss. The key contributions are (1) integrating IMU measurements into Gaussian Splatting SLAM to boost tracking over large environments, (2) an ICP-based tracking framework driven by IMU preintegration for robust pose estimation, and (3) comprehensive experiments on the uHumansV1/V2 datasets showing improved tracking accuracy and rendering quality compared to state-of-the-art methods. The work demonstrates that IMU-informed Gaussian Splatting SLAM can achieve both high accuracy and scalability in large-scale indoor spaces, with practical implications for robotics and AR applications.

Abstract

Recently, map representations based on radiance fields such as 3D Gaussian Splatting and NeRF, which excellent for realistic depiction, have attracted considerable attention, leading to attempts to combine them with SLAM. While these approaches can build highly realistic maps, large-scale SLAM still remains a challenge because they require a large number of Gaussian images for mapping and adjacent images as keyframes for tracking. We propose a novel 3D Gaussian Splatting SLAM method, VIGS SLAM, that utilizes sensor fusion of RGB-D and IMU sensors for large-scale indoor environments. To reduce the computational load of 3DGS-based tracking, we adopt an ICP-based tracking framework that combines IMU preintegration to provide a good initial guess for accurate pose estimation. Our proposed method is the first to propose that Gaussian Splatting-based SLAM can be effectively performed in large-scale environments by integrating IMU sensor measurements. This proposal not only enhances the performance of Gaussian Splatting SLAM beyond room-scale scenarios but also achieves SLAM performance comparable to state-of-the-art methods in large-scale indoor environments.

VIGS SLAM: IMU-based Large-Scale 3D Gaussian Splatting SLAM

TL;DR

The paper addresses the challenge of large-scale indoor SLAM using radiance-field representations by introducing VIGS SLAM, which fuses RGB-D and IMU data to enable scalable, photo-realistic 3D mapping. It combines Generalized ICP tracking with IMU preintegration to obtain strong initial pose estimates and uses a 3D Gaussian Splatting map that is progressively updated with scale normalization, optimized via a dense photometric/SSIM/depth loss. The key contributions are (1) integrating IMU measurements into Gaussian Splatting SLAM to boost tracking over large environments, (2) an ICP-based tracking framework driven by IMU preintegration for robust pose estimation, and (3) comprehensive experiments on the uHumansV1/V2 datasets showing improved tracking accuracy and rendering quality compared to state-of-the-art methods. The work demonstrates that IMU-informed Gaussian Splatting SLAM can achieve both high accuracy and scalability in large-scale indoor spaces, with practical implications for robotics and AR applications.

Abstract

Recently, map representations based on radiance fields such as 3D Gaussian Splatting and NeRF, which excellent for realistic depiction, have attracted considerable attention, leading to attempts to combine them with SLAM. While these approaches can build highly realistic maps, large-scale SLAM still remains a challenge because they require a large number of Gaussian images for mapping and adjacent images as keyframes for tracking. We propose a novel 3D Gaussian Splatting SLAM method, VIGS SLAM, that utilizes sensor fusion of RGB-D and IMU sensors for large-scale indoor environments. To reduce the computational load of 3DGS-based tracking, we adopt an ICP-based tracking framework that combines IMU preintegration to provide a good initial guess for accurate pose estimation. Our proposed method is the first to propose that Gaussian Splatting-based SLAM can be effectively performed in large-scale environments by integrating IMU sensor measurements. This proposal not only enhances the performance of Gaussian Splatting SLAM beyond room-scale scenarios but also achieves SLAM performance comparable to state-of-the-art methods in large-scale indoor environments.
Paper Structure (14 sections, 12 equations, 5 figures, 3 tables)

This paper contains 14 sections, 12 equations, 5 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. Sensor Fusion for SLAM
  4. 3DGS SLAM
  5. METHOD
  6. Generalized ICP Tracking
  7. IMU Preintegration
  8. Update Preintegration
  9. 3D Gaussian Splatting Mapping
  10. EXPERIMENTS
  11. Experiments Setup
  12. Performance on the uHumansV1
  13. Performance on the uHumansV2
  14. CONCLUSIONS & FUTURE WORK

Figures (5)

  • Figure 1: Comparison of rendering image quality according to initial estimates on Humans12 (upper 2 rows) and Humans24 (lower 2 rows) of uHumansV1 dataset
  • Figure 2: Overview of VIGS SLAM. The input of VIGS SLAM is RGB-D image and IMU meausrements. The system generates a point cloud from RGB and depth inputs, followed by GICP tracking. The IMU preintegration values are used as a good initial guess to enhance tracking performance, and these values are updated after tracking each frame. Keyframes are identified for 3D Gaussian Splatting-based mapping, which efficiently updates the 3DGS map with detailed environmental representations.
  • Figure 3: Comparison of map reconstruction on uHumansV1. GS-ICP SLAM (left), VIGS SLAM (right)
  • Figure 4: Trajectory in Humans12(left), Humans24(middle), Humans60(right) of uHumansV1, compared with ORB-SLAM3, VINS-Mono, GS-ICP SLAM, and VIGS SLAM(ours).
  • Figure 5: Qualitative rendering results on the apartment scene of uHumansV2 dataset, compared with VIGS SLAM(ours), GS-ICP SLAM, MonoGS, and PhotoSLAM.