Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DQO-MAP: Dual Quadrics Multi-Object mapping with Gaussian Splatting

Haoyuan Li, Ziqin Ye, Yue Hao, Weiyang Lin, Chao Ye

TL;DR

DQO-MAP presents an online, object-centric SLAM system that tightly integrates object pose estimation and reconstruction by combining 3D Gaussian Splatting for geometry with dual quadrics for pose. The architecture decouples object Gaussians on the CPU from GPU-accelerated optimization, enabling real-time performance and easy object extraction via unique IDs. Core innovations include an object-level association framework, an incremental Gaussian update strategy, and jointly trained losses for reconstruction and pose. Across synthetic and real datasets, the approach achieves improved reconstruction quality, competitive pose accuracy, and favorable runtime efficiency, illustrating its practical potential for object-aware navigation and manipulation.

Abstract

Accurate object perception is essential for robotic applications such as object navigation. In this paper, we propose DQO-MAP, a novel object-SLAM system that seamlessly integrates object pose estimation and reconstruction. We employ 3D Gaussian Splatting for high-fidelity object reconstruction and leverage quadrics for precise object pose estimation. Both of them management is handled on the CPU, while optimization is performed on the GPU, significantly improving system efficiency. By associating objects with unique IDs, our system enables rapid object extraction from the scene. Extensive experimental results on object reconstruction and pose estimation demonstrate that DQO-MAP achieves outstanding performance in terms of precision, reconstruction quality, and computational efficiency. The code and dataset are available at: https://github.com/LiHaoy-ux/DQO-MAP.

DQO-MAP: Dual Quadrics Multi-Object mapping with Gaussian Splatting

TL;DR

DQO-MAP presents an online, object-centric SLAM system that tightly integrates object pose estimation and reconstruction by combining 3D Gaussian Splatting for geometry with dual quadrics for pose. The architecture decouples object Gaussians on the CPU from GPU-accelerated optimization, enabling real-time performance and easy object extraction via unique IDs. Core innovations include an object-level association framework, an incremental Gaussian update strategy, and jointly trained losses for reconstruction and pose. Across synthetic and real datasets, the approach achieves improved reconstruction quality, competitive pose accuracy, and favorable runtime efficiency, illustrating its practical potential for object-aware navigation and manipulation.

Abstract

Accurate object perception is essential for robotic applications such as object navigation. In this paper, we propose DQO-MAP, a novel object-SLAM system that seamlessly integrates object pose estimation and reconstruction. We employ 3D Gaussian Splatting for high-fidelity object reconstruction and leverage quadrics for precise object pose estimation. Both of them management is handled on the CPU, while optimization is performed on the GPU, significantly improving system efficiency. By associating objects with unique IDs, our system enables rapid object extraction from the scene. Extensive experimental results on object reconstruction and pose estimation demonstrate that DQO-MAP achieves outstanding performance in terms of precision, reconstruction quality, and computational efficiency. The code and dataset are available at: https://github.com/LiHaoy-ux/DQO-MAP.

Paper Structure

This paper contains 20 sections, 14 equations, 8 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Object SLAM
  4. Gaussian Splatting in SLAM
  5. METHOD
  6. Objects’ Pose and Size Estimation
  7. Object-Level Association
  8. Gaussains Incremental Update
  9. Object Pose and Reconstruction Training Loss
  10. Object Pose Traing Loss
  11. Reconstruction Training Loss:
  12. EXPERIMENTS
  13. Evaluation of Online Object Reconstruction
  14. Cube-Diorama
  15. Replica
  16. ...and 5 more sections

Figures (8)

  • Figure 1: DQO-MAP simultaneously reconstructs objects using Gaussian Splatting and estimates their poses with quadrics.Each object is assigned a unique ID for association and extraction.
  • Figure 2: Overview of our proposed system. DQO-MAP tightly integrate object pose estimation and reconstruction, leveraging quadrics for object estimation and 3DGS for reconstruction.
  • Figure 3: Different dtypes of data association
  • Figure 4: Object reconstruction results in Room
  • Figure 5: Object reconstruction results in Replica
  • ...and 3 more figures