Robust 3DGS-based SLAM via Adaptive Kernel Smoothing

TL;DR

This work reframes 3DGS-SLAM by prioritizing robustness of the rasterization process to Gaussian parameter noise over perfect scene fidelity. It introduces Corrective Blurry KNN (CB-KNN), an adaptive, local smoothing mechanism that temporarily perturbs nearby Gaussians during rendering to regularize pose optimization without altering the underlying map, and applies this mainly to keyframes. The method preserves differentiable rendering and integrates with existing 3DGS pipelines, including adaptive selection of the kernel size and selective keyframe processing. Experimental results across synthetic and real datasets show improved pose tracking accuracy and stability while maintaining high-quality mapping and rendering metrics. The approach offers a practical route to more robust, real-world 3DGS-SLAM on devices with limited computational resources.

Abstract

In this paper, we challenge the conventional notion in 3DGS-SLAM that rendering quality is the primary determinant of tracking accuracy. We argue that, compared to solely pursuing a perfect scene representation, it is more critical to enhance the robustness of the rasterization process against parameter errors to ensure stable camera pose tracking. To address this challenge, we propose a novel approach that leverages a smooth kernel strategy to enhance the robustness of 3DGS-based SLAM. Unlike conventional methods that focus solely on minimizing rendering error, our core insight is to make the rasterization process more resilient to imperfections in the 3DGS parameters. We hypothesize that by allowing each Gaussian to influence a smoother, wider distribution of pixels during rendering, we can mitigate the detrimental effects of parameter noise from outlier Gaussians. This approach intentionally introduces a controlled blur to the rendered image, which acts as a regularization term, stabilizing the subsequent pose optimization. While a complete redesign of the rasterization pipeline is an ideal solution, we propose a practical and effective alternative that is readily integrated into existing 3DGS frameworks. Our method, termed Corrective Blurry KNN (CB-KNN), adaptively modifies the RGB values and locations of the K-nearest neighboring Gaussians within a local region. This dynamic adjustment generates a smoother local rendering, reducing the impact of erroneous GS parameters on the overall image. Experimental results demonstrate that our approach, while maintaining the overall quality of the scene reconstruction (mapping), significantly improves the robustness and accuracy of camera pose tracking.

Figures (4)

• Figure 1: In the fr1/desk scene of the TUM-RGBDTUM-RGBD dataset, compared with the baseline(SplaTAMsplatam) method, our method results in a smoother trajectory that is closer to the ground-truth trajectory.
• Figure 2: The flowchart illustrates the workflow of a 3DGS - SLAM method based on CB - KNN. It starts with initial Gaussian map construction, proceeds through camera tracking and Gaussian densification, then keyframing (where keyframes undergo CB - KNN - based position correction and color smoothing), and finally map update, forming a complete SLAM cycle.
• Figure 3: Rendering results on the Room0 from Replicareplica-dataset. It can be observed from the rendering results that our CB-KNN method does not exhibit image blurring in both novel and training views. Compared with SplaTAMsplatam, our method achieves comparable fidelity to the ground truth in both color maps and depth maps rendered for novel and training views.
• Figure 4: Trajectory maps on Replicareplica-dataset, TUM-RGBDTUM-RGBD, and ScanNetscannet.