VarSplat: Uncertainty-aware 3D Gaussian Splatting for Robust RGB-D SLAM

Anh Thuan Tran; Jana Kosecka

cs.CVMar 10, 2026

VarSplat: Uncertainty-aware 3D Gaussian Splatting for Robust RGB-D SLAM

Authors: Anh Thuan Tran, Jana Kosecka

Organizations: Department of Computer Science, George Mason University

Abstract

Simultaneous Localization and Mapping (SLAM) with 3D Gaussian Splatting (3DGS) enables fast, differentiable rendering and high-fidelity reconstruction across diverse real-world scenes. However, existing 3DGS-SLAM approaches handle measurement reliability implicitly, making pose estimation and global alignment susceptible to drift in low-texture regions, transparent surfaces, or areas with complex reflectance properties. To this end, we introduce VarSplat, an uncertainty-aware 3DGS-SLAM system that explicitly learns per-splat appearance variance. By using the law of total variance with alpha compositing, we then render differentiable per-pixel uncertainty map via efficient, single-pass rasterization. This map guides tracking, submap registration, and loop detection toward focusing on reliable regions and contributes to more stable optimization. Experimental results on Replica (synthetic) and TUM-RGBD, ScanNet, and ScanNet++ (real-world) show that VarSplat improves robustness and achieves competitive or superior tracking, mapping, and novel view synthesis rendering compared to existing studies for dense RGB-D SLAM.

Figures & tables

Figure 2: VarSplat architecture. During mapping, each 3D Gaussian jointly learns position, appearance, and variance $\sigma^{2}$ . The per-splat variances are composited into per-pixel uncertainty $V$ through the differentiable rasterizer in mapping . This uncertainty map is then served as confidence in tracking and registration , while $\sigma^{2}$ is used for uncertainty-aware loop detection .

Neural Implicit Fields
Method	UC	R0	R1	R2	Off0	Off1	Off2	Off3	Off4	Avg.
NICE-SLAM [ 55 ]	✗	0.97	1.31	1.07	0.88	1.00	1.06	1.10	1.13	1.06
ESLAM [ 15 ]	✗	0.71	0.70	0.52	0.57	0.55	0.58	0.72	0.63	0.63
Point-SLAM [ 30 ]	✗	0.61	0.41	0.37	0.38	0.48	0.54	0.69	0.72	0.52
GO-SLAM [ 52 ]	✗	0.34	0.29	0.29	0.32	0.30	0.39	0.39	0.46	0.35
Loopy-SLAM [ 23 ]	✗	0.24	0.24	0.28	0.26	0.40	0.29	0.22	0.35	0.29

Table 1: Tracking Performance on Replica [ 36 ] (ATE RMSE

\downarrow

[cm]). UC indicates uncertainty. The results are highlighted as best , second and third . VarSplat achieves the best average performance while remaining competitive across scenes. *Photo-SLAM [ 12 ] use ORB-SLAM3 features [ 3 ] for tracking and loop closure.

Neural Implicit Fields
Method	a	b	c	d	e	Avg.
Point-SLAM [ 30 ]	246.16	632.99	830.79	271.42	574.86	511.24
ESLAM [ 15 ]	25.15	2.15	27.02	20.89	35.47	22.14
GO-SLAM [ 52 ]	176.28	145.45	38.74	85.48	106.47	110.49
Loopy-SLAM [ 23 ]	N/A	N/A	25.16	234.25	81.48	113.63
3D Gaussian Splatting

Table 2: Tracking Performance on ScanNet++ (ATE RMSE

\downarrow

[cm]). VarSplat achieves the highest accuracy with robustness on large motion camera.

Neural Implicit Fields
Method	UC	fr1/desk	fr1/desk2	fr1/room	fr2/xyz	fr3/off.	Avg.
NICE-SLAM [ 55 ]	✗	4.26	4.99	34.49	6.19	3.87	10.76
MIPS-Fusion [ 40 ]	✗	3.00	N/A	N/A	1.4	4.6	N/A
Point-SLAM [ 30 ]	✗	4.34	4.54	30.92	1.31	4.8	8.92
ESLAM [ 15 ]	✗	2.47	3.69	29.73	1.11	2.42	7.89
Co-SLAM [ 41 ]	✗	2.40	N/A	N/A	1.70	2.40	N/A

Table 3: Tracking Performance on TUM-RGBD (ATE RMSE

\downarrow

[cm]). UC indicates uncertainty. VarSplat outperforms both 3DGS and NeRF baselines. *Photo-SLAM [ 12 ] use ORB-SLAM3 features [ 3 ] for tracking and loop closure.

Neural Implicit Fields
Method	UC	00	59	106	169	181	207	Avg.
Co-SLAM [ 41 ]	✗	7.1	11.1	9.4	5.9	11.8	7.1	8.7
NICE-SLAM [ 55 ]	✗	12.0	14.0	7.9	10.9	13.4	6.2	10.7
ESLAM [ 15 ]	✗	7.3	8.5	7.5	6.5	9.0	5.7	7.4
Point-SLAM [ 30 ]	✗	10.2	10.8	8.7	7.2	22.2	14.8	12.3
GO-SLAM [ 52 ]	✗	5.4	7.5	7.0	7.0	6.8	6.9	6.8

Table 4: Tracking Performance on ScanNet. UC indicates uncertainty. VarSplat achieves the best overall, showing robustness to noisy indoor scenes.

Neural Implicit Fields
Method	Metric	R0	R1	R2	Off0	Off1	Off2	Off3	Off4	Avg.
NICE-SLAM [ 55 ]	Depth L1	1.81	1.44	2.04	1.39	1.76	8.33	4.99	2.01	2.97
	F1 [%]	45.0	44.8	43.6	50.0	51.9	39.2	39.9	36.5	43.9
ESLAM [ 15 ]	Depth L1	0.97	1.07	1.28	0.86	1.26	1.71	1.43	1.06	1.18
	F1 [%]	81.0	82.2	83.9	78.4	75.5	77.1	75.9	79.1	79.1
Point-SLAM [ 30 ]	Depth L1	0.53	0.22	0.46	0.30	0.57	0.49	0.51	0.64	0.46

Table 5: Reconstruction Performance on Replica. VarSplat achieve third-best result after Loopy-SLAM and LoopSplat, showing that variance regularization preserves mesh quality.

Dataset	Replica [ 36 ]			TUM [ 38 ]			ScanNet [ 5 ]
Method	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$	PSNR $\uparrow$	SSIM $\uparrow$	LPIPS $\downarrow$
NICE-SLAM [ 55 ]	24.42	0.892	0.233	14.86	0.614	0.441	17.54	0.621	0.548
ESLAM [ 15 ]	28.06	0.923	0.245	15.26	0.478	0.569	15.29	0.658	0.488
Point-SLAM [ 30 ]	35.17	0.975	0.124	16.62	0.696	0.526	19.82	0.751	0.514
Loopy-SLAM [ 23 ]	35.47	0.981	0.109	12.94	0.489	0.645	15.23	0.629	0.671
SplaTAM [ 16 ]	34.11	0.970	0.100	22.80	0.893	0.178	19.14	0.716	0.358

Table 6: Rendering performance on 3 datasets. VarSplat achieves competitive results on both synthetic and real-world datasets. Gray indicates evaluation on submaps rather than global map.

Method	a	b	c	d	e	Avg.
ESLAM [ 15 ]	13.63	11.86	11.83	10.59	10.64	11.71
SplaTAM [ 16 ]	23.95	22.66	13.95	8.47	20.06	17.82
Gaussian-SLAM [ 51 ]	26.66	24.42	15.01	18.35	21.91	21.27
LoopSplat [ 54 ]	25.60	23.65	15.87	18.86	22.51	21.30
VarSplat (Ours)	26.97	24.52	15.19	18.07	21.92	21.33

Table 7: Novel View Synthesis on ScanNet++ (PSNR

\uparrow

). VarSplat achives the best result.

Model	Tracking	Loop	Registration	ATE RMSE $\downarrow$
I	✗	✗	✗	8.20
II	✓	✗	✗	7.63
III	✓	✓	✗	7.49
IV	✗	✓	✓	7.51
V	✓	✓	✓	6.53

Table 8: Effect of uncertainty on pose estimation.

Model	Track Frozen	Depth	Square L2	ATE RMSE $\downarrow$
I	✗	✓	✓	7.55
II	✓	✗	✓	7.17
II	✓	✓	✗	7.38
IV	✓	✓	✓	6.53

Table 9: Ablation on Variance Training.

Method	Mapping	Mapping	Tracking	Tracking	ATE
	/fr(s)	/iter(ms)	/fr(s)	/iter(ms)	RMSE
NICE-SLAM [ 55 ]	5.8	90.6	8.1	20.8	0.97
Point-SLAM [ 30 ]	28.7	93.1	7.1	29.0	0.61
SplaTAM [ 16 ]	5.8	96.8	3.5	86.8	0.31
LoopSplat [ 54 ]	1.2	30.1	1.8	29.3	0.28
VarSplat (Ours)	1.9	50.3	2.0	34.1	0.20

Table 10: Runtime on Replica/Room0 using A100 80GB. Per-frame runtime is computed as total optimization time divided by the sequence length.

Figure 3: Uncertainty ablation on ScanNet (scene0181). Without uncertainty, tracking jitters, loop detection has long-range drift, and registration ghosts submaps. With VarSplat enabled, the trajectory is smooth and overlaps align.

Figure 4: Visualization of challenging conditions (scene0169 ScanNet). In the top example, uncertainty is high on the texture-poor wall region (white box), while the map continues to grow (blue box). In the bottom example, the uncertainty map can distinguish reliable regions on transparent surface.

Figure 5: Per-pixel uncertainty with vs. without depth on TUM-RGBD (fr1/desk2). With depth, uncertainty focuses on textureless areas with depth holes and stays low on well constrained surfaces, avoiding overconfidence on glossy areas.

Params	Replica [ 36 ]	TUM-RGBD [ 38 ]	ScanNet [ 5 ]	ScanNet++ [ 50 ]
$\lambda_{c}$	0.95	0.6	0.6	0.5
$l_{r}$	0.0002	0.002	0.002	0.002
$l_{t}$	0.002	0.01	0.01	0.01
$\text{iter}_{t}$	60	200	200	300
$\text{iter}_{m}$	100	100	100	500
$\tau$	10	50	5	10

Table S1 : Per-Dataset Hyperparameters.

Dataset	Replica	TUM	ScanNet	ScanNet++
Translation (cm)	1.07	1.39	1.34	14.77
Rotation (°)	0.50	1.37	0.69	13.43

Table S2 : Frame Motion on Replica [ 36 ] , TUM-RGBD [ 38 ] , ScanNet [ 5 ] , and ScanNet++ [ 50 ] .

Method	Metric	Rm0	Rm1	Rm2	Off0	Off1	Off2	Off3	Off4	Avg.
NICE-SLAM [ 55 ]	PSNR $\uparrow$	22.12	22.47	24.52	29.07	30.34	19.66	22.23	24.94	24.42
	SSIM $\uparrow$	0.689	0.757	0.814	0.874	0.886	0.797	0.801	0.856	0.809
	LPIPS $\downarrow$	0.330	0.271	0.208	0.229	0.181	0.235	0.209	0.198	0.233
ESLAM [ 15 ]	PSNR $\uparrow$	25.25	27.39	28.09	30.33	27.04	27.99	29.27	29.15	28.06
	SSIM $\uparrow$	0.874	0.890	0.935	0.934	0.910	0.942	0.953	0.948	0.923
	LPIPS $\downarrow$	0.315	0.296	0.245	0.213	0.254	0.238	0.186	0.210	0.245

Table S3 : Rendering Performance on Replica. * denotes evaluating on submaps instead of a global one.

Method	Metric	fr1/desk	fr2/xyz	fr3/office	Avg.
NICE-SLAM [ 55 ]	PSNR $\uparrow$	13.83	17.87	12.89	14.86
	SSIM $\uparrow$	0.569	0.718	0.554	0.614
	LPIPS $\downarrow$	0.482	0.344	0.498	0.441
ESLAM [ 15 ]	PSNR $\uparrow$	11.29	17.46	17.02	15.26
	SSIM $\uparrow$	0.666	0.310	0.457	0.478
	LPIPS $\downarrow$	0.358	0.698	0.652	0.569

Table S4 : Rendering Performance on TUM RGB-D. * denotes evaluating on submaps instead of a global one.

Method	Metric	0000	0059	0106	0169	0181	0207	Avg.
NICE-SLAM [ 55 ]	PSNR $\uparrow$	18.71	16.55	17.29	18.75	15.56	18.38	17.54
	SSIM $\uparrow$	0.641	0.605	0.646	0.629	0.562	0.646	0.621
	LPIPS $\downarrow$	0.561	0.534	0.510	0.534	0.602	0.552	0.548
ESLAM [ 15 ]	PSNR $\uparrow$	15.70	14.48	15.44	14.56	14.22	17.32	15.29
	SSIM $\uparrow$	0.687	0.632	0.628	0.656	0.696	0.653	0.658
	LPIPS $\downarrow$	0.449	0.450	0.529	0.486	0.482	0.534	0.488

Table S5 : Rendering Performance on ScanNet. * denotes evaluating on submaps instead of a global one.

Simultaneous Localization and Mapping

Simultaneous localization and mapping estimates a robot’s motion while building a map of an unknown environment.

See all 197 papers

VBGS-SLAM: Variational Bayesian Gaussian Splatting Simultaneous Localization and Mapping
Apr 3, 2026Yuhan Zhu, Yanyu Zhang, Jie Xu +13DGS SLAM Simultaneous Localization and Mapping
Department of Electrical and Computer Engineering, University of California, Riverside, CA, 92521, USA.
LightSplat: Real-Time High-Fidelity 3D Gaussian SLAM with Loop Closure
Sep 7, 2026Junze Bao, Ye Gao, Yiming Huang +5Simultaneous Localization and Mapping 3D Gaussian Splatting
School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, P. R. China, and also with Zhejiang Key Laboratory of Industrial Big Data and Robot Intelligent Systems, Hangzhou Innovation Institute, Beihang University, Hangzhou 310051, P. R. China. · School of Mechanical Engineering, Zhejiang Sci-Tech University, Hangzhou 310018, P. R. China.
Geometry-Aware Online Mapping for 3D Gaussian Splatting SLAM
Aug 14, 2026Thai Luu, Quan Tran, Hieu Phan +13DGS SLAM 3D Gaussian Splatting
Cognitive Robotics Lab, Department of Electrical Engineering and Computer Science, University of Arkansas, USA. · Center for AI Research, VinUniversity, Ha Noi, Vietnam.

Apr 3, 2026J/K move · Enter open · S save

VBGS-SLAM: Variational Bayesian Gaussian Splatting Simultaneous Localization and Mapping

Yuhan Zhu, Yanyu Zhang, Jie Xu, Wei Ren

Department of Electrical and Computer Engineering, University of California, Riverside, CA, 92521, USA.

3DGS SLAM Simultaneous Localization and Mapping 3D Gaussian Splatting

Open paper AskPDF arXiv

3D Gaussian Splatting (3DGS) has shown promising results for 3D scene modeling using mixtures of Gaussians, yet its existing simultaneous localization and mapping (SLAM) variants typically rely on direct, deterministic pose optimization against the splat map, making them sensitive to initialization and susceptible to catastrophic forgetting as map evolves. We propose Variational Bayesian Gaussian Splatting SLAM (VBGS-SLAM), a novel framework that couples the splat map refinement and camera pose tracking in a generative probabilistic form. By leveraging conjugate properties of multivariate Gaussians and variational inference, our method admits efficient closed-form updates and explicitly maintains posterior uncertainty over both poses and scene parameters. This uncertainty-aware method mitigates drift and enhances robustness in challenging conditions, while preserving the efficiency and rendering quality of existing 3DGS. Our experiments demonstrate superior tracking performance and robustness in long sequence prediction, alongside efficient, high-quality novel view synthesis across diverse synthetic and real-world scenes.

Abstract

Simultaneous Localization and Mapping

Explore similar work

VBGS-SLAM: Variational Bayesian Gaussian Splatting Simultaneous Localization and Mapping

LightSplat: Real-Time High-Fidelity 3D Gaussian SLAM with Loop Closure

Geometry-Aware Online Mapping for 3D Gaussian Splatting SLAM