Globally Consistent RGB-D SLAM with 2-D Gaussian Splatting

Abstract

Recently, 3-D Gaussian splatting (3DGS)-based RGB-D simultaneous localization and mapping (SLAM) displays remarkable performance of high-fidelity 3-D reconstruction. However, 3DGS suffers from a lack of depth rendering consistency, which leads to suboptimal geometric reconstruction in 3DGS-based SLAM. In addition, 3DGS-based SLAM methods typically lack efficient loop closure, limiting their ability to build globally consistent maps online. In this article, we present 2-D Gaussian splatting (2DGS-SLAM), an RGB-D SLAM system using 2-D Gaussian splatting as map representation. By leveraging the depth-consistent rendering property of the 2-D variant, we propose an accurate camera pose optimization method and achieve geometrically accurate 3-D reconstruction. In addition, we implement efficient loop detection and camera relocalization by leveraging MASt3R, a feedforward 3-D reconstruction model, and achieve efficient map updates by maintaining a local active map. Experiments show that our 2DGS-SLAM approach achieves superior tracking accuracy, higher surface reconstruction quality, and more consistent global map reconstruction compared to existing rendering-based SLAM methods, while maintaining high-fidelity image rendering and improved computational efficiency.