GGD-SLAM: Monocular 3DGS SLAM Powered by Generalizable Motion Model for Dynamic Environments

Yi Liu; Haoxuan Xu; Hongbo Duan; Keyu Fan; Zhengyang Zhang; Peiyu Zhuang; Pengting Luo; Houde Liu

cs.ROApr 14, 2026

GGD-SLAM: Monocular 3DGS SLAM Powered by Generalizable Motion Model for Dynamic Environments

Authors: Yi Liu, Haoxuan Xu, Hongbo Duan, Keyu Fan, Zhengyang Zhang, Peiyu Zhuang, Pengting Luo, Houde Liu

Organizations: Shenzhen International Graduate School, Tsinghua University, Shenzhen, China. · Thrust of Robotics and Autonomous Systems, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China. · the School of Cyber Science and Technology, Sun Yat-sen University, Shenzhen, China. · the Central Media Technology Institute, Huawei Incorporated Company, Shenzhen, China.

Abstract

Visual SLAM algorithms achieve significant improvements through the exploration of 3D Gaussian Splatting (3DGS) representations, particularly in generating high-fidelity dense maps. However, they depend on a static environment assumption and experience significant performance degradation in dynamic environments. This paper presents GGD-SLAM, a framework that employs a generalizable motion model to address the challenges of localization and dense mapping in dynamic environments - without predefined semantic annotations or depth input. Specifically, the proposed system employs a First-In-First-Out (FIFO) queue to manage incoming frames, facilitating dynamic semantic feature extraction through a sequential attention mechanism. This is integrated with a dynamic feature enhancer to separate static and dynamic components. Additionally, to minimize dynamic distractors' impact on the static components, we devise a method to fill occluded areas via static information sampling and design a distractor-adaptive Structure Similarity Index Measure (SSIM) loss tailored for dynamic environments, significantly enhancing the system's resilience. Experiments conducted on real-world dynamic datasets demonstrate that the proposed system achieves state-of-the-art performance in camera pose estimation and dense reconstruction in dynamic scenes.

Figures & tables

Fig. 1: Motivation : Left: DyPho-SLAM [ 5 ] requires specific semantic labels and depth input for dynamic removal; Right: The MLP in WildGS-SLAM [ 6 ] is constrained by per-scene rendering performance; Our GGD-SLAM introduces a generalizable motion model without semantic labels or depth input, reducing the need for per-scene 3DGS rendering loss supervision.

Fig. 2: Pipeline of our GGD-SLAM: the “Generalized Motion Model” ( III-A ) processes progressive frame $\bm{I}_{t}$ to extract prior $M_{t}$ , followed by encoder preprocessing, historical features $\mathcal{Q}_{t}$ maintenance, and the moving extractor. Subsequently, the “Tracking” module ( III-B ) passes through the frame graph $\mathcal{E}$ maintenance, and DBA optimization guided by prior to obtain the camera pose $\bm{T}_{t}$ . Finally, the “Mapping” module ( III-C ) utilizes priors to train the uncertainty model, and incrementally adds Gaussians as well as optimizes the Gaussian map $\mathcal{G}$ .

Fig. 3: Illustration of distractor-adaptive SSIM:(1) Computation of SSIM; (2) Conventional dynamic SSIM processing is polluted; (3) Our adapted kernel achieves clean SSIM under distractors.

Fig. 4: Qualitative results of the different dynamic extractors progressing in fr3/w/half.

Method	TUM								Bonn
Method	fr3/w/xyz		fr3/w/half		fr3/s/half		Avg.		balloon		balloon2		ps_track		crowd2		Avg.
RGBD Dense SLAM	ATE	Std.	ATE	Std.	ATE	Std.	ATE	Std.	ATE	Std.	ATE	Std.	ATE	Std.	ATE	Std.	ATE	Std.
RoDyn-SLAM [ 10 ]	8.3	5.5	5.6	2.8	4.4	2.2	4.1	2.3	7.9	2.7	11.5	6.1	14.5	4.6	6.5	3.9	12.3	4.4
GARAD-SLAM [ 23 ]	1.6	0.8	2.3	1.2	-	-	1.9	1.1	3.0	1.2	2.5	1.2	4.6	1.5	2.4	1.2	2.7	1.2
Gassidy [ 3 ]	3.5	1.6	3.7	1.9	2.4	1.4	2.6	1.3	2.6	0.8	7.6	3.4	10.3	4.4	-	-	7.8	3.1
DGS-SLAM [ 17 ]	4.1	2.2	5.5	2.8	4.1	1.6	3.0	1.5	2.9	0.8	6.0	2.8	9.8	4.1	-	-	7.3	3.0

TABLE I: Camera tracking results on dynamic scenes from the TUM and Bonn challenging datasets. The units for ATE

(\downarrow)

and Std.

(\downarrow)

are in cm. The best results among Mono Dense SLAM domains are in bold .

-

indicates that the relevant algorithms are not yet open-source, and the corresponding data is not provided in the original report.

Generalizable Prior	OTSU Binarize	Smoothness	ps_track	crowd2
✓	✓	$\times$	3.47	1.95
✓	$\times$	✓	3.44	1.86
$\times$	$\times$	✓	3.56	2.14
✓	✓	✓	3.41	1.79

TABLE II: Ablation study on Bonn RGB-D Dynamic Dataset. Generalizable Prior refers to prior information from GMM; OTSU Binarize is our solution for ambiguous edges; and Smoothness refers to the smoothness term in Tracking.

Fig. 5: Comparison of rendered results from state-of-the-art Gaussian Splatting SLAM methods.

Method	fr3/w/xyz			fr3/w/half			bonn/ps_tracking			bonn/crowd2			Avg.
Method	PSNR	SSIM	LPIPS	PSNR	SSIM	LPIPS	PSNR	SSIM	LPIPS	PSNR	SSIM	LPIPS	PSNR	SSIM	LPIPS
MonoGS	13.97	0.467	0.591	16.57	0.551	0.476	19.56	0.726	0.454	18.07	0.695	0.567	17.04	0.610	0.522
Splatam	14.29	0.527	0.486	15.53	0.478	0.493	17.81	0.596	0.344	15.79	0.517	0.592	15.81	0.526	0.479
WildGS-SLAM	21.53	0.821	0.141	20.89	0.829	0.136	23.35	0.904	0.148	23.15	0.885	0.264	22.23	0.855	0.172
Ours	22.68	0.751	0.117	22.25	0.848	0.134	22.92	0.902	0.162	24.27	0.918	0.219	23.03	0.859	0.158

TABLE III: Mapping results on dynamic scenes from the TUM and Bonn dynamic datasets. The unit for PSNR(

\uparrow

) is in dB. SSIM(

\uparrow

) and LPIPS(

\downarrow

) are both dimensionless metrics.

Dynamic SSIM	Static KD-Tree	fr3/w/xyz	bonn/cr2
✓	$\times$	21.96	23.75
$\times$	✓	21.59	23.47
✓	✓	22.68	24.27

TABLE IV: Ablation study on distractor-adaptive SSIM and static Gaussian KD-tree for our occlusion recovery method.

Fig. 6: Our GGD-SLAM performance on the Wild-SLAM Dataset.

Dynamic SLAM

See all 12 papers

RoSe-SLAM: Robust Semantic-Aware Gaussian Splatting SLAM from Dynamic Monocular Videos
Aug 29, 2026Wenting Wang, Jiaxin Guo, Wenzhen Dong +3Dynamic SLAM Visual SLAM
Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Hong Kong SAR. · Department of Mechanical, Aerospace and Civil Engineering, The University of Manchester, Manchester, UK. · Centre for Perceptual and Interactive Intelligence (CPII) Limited, Hong Kong SAR.
DL-SLAM: Enabling High-Fidelity Gaussian Splatting SLAM in Dynamic Environments based on Dual-Level Probability
Jul 2, 2026Ziheng Xu, Qingfeng Li, Xuefeng Liu +2Simultaneous Localization and Mapping Dynamic Scene Reconstruction
ACMMM 2026, , Rio de Janeiro, Brazil
Flow4DGS-SLAM: Optical Flow-Guided 4D Gaussian Splatting SLAM
Apr 24, 2026Yunsong Wang, Gim Hee LeeSimultaneous Localization and Mapping 3D Gaussian Splatting
School of Computing, National University of Singapore

Aug 29, 2026J/K move · Enter open · S save

RoSe-SLAM: Robust Semantic-Aware Gaussian Splatting SLAM from Dynamic Monocular Videos

Wenting Wang, Jiaxin Guo, Wenzhen Dong, Yun-Hui Liu+2

Department of Mechanical and Automation Engineering, The Chinese University of Hong Kong, Hong Kong SAR. · Department of Mechanical, Aerospace and Civil Engineering, The University of Manchester, Manchester, UK. · Centre for Perceptual and Interactive Intelligence (CPII) Limited, Hong Kong SAR.

Dynamic SLAM Visual SLAM 3DGS SLAM Dynamic Scene Reconstruction Camera Pose Estimation

Open paper AskPDF arXiv

In dynamic and unstructured environments, conventional SLAM systems generally suffer from significant accuracy degeneration due to their static assumptions. In this work, we propose Robust Semantic-aware Gaussian Splatting SLAM (RoSe-SLAM), to address the dynamic challenge by a holistic semantic scene understanding from uncalibrated monocular inputs, achieving accurate camera tracking and high-quality geometry reconstruction. Unlike conventional semantic SLAM using handcrafted semantic labels, our RoSe-SLAM exploits the semantic feature from 2D foundation model to enhance the dynamic tracking and mapping performance. By distilling the rich semantic features to our Gaussian fields, our method effectively identifies dynamic distractors and achieves semantic-aware multi-view consistency, significantly enhancing the geometric reconstruction and scene inpainting. Specifically, we propose a spatial-temporal motion mask generation module, enabling both long-term motion monitoring and short-term transient dynamics capturing, achieving robust and effective disentanglement of dynamic objects and static backgrounds. During global bundle adjustment, we propose an occlusion-aware keyframe selection mechanism to prioritize the occlusion as metric to pick the keyframes, and a multi-view semantic consistency module to improve the mapping quality in dynamic environments. By combining geometric motion cues with semantic priors, our system dynamically filters unreliable observations and reconstructs accurate static scene geometry. Extensive experiments conducted on benchmark datasets including dynamic TUM, Bonn and Wild-Mocap datasets, demonstrate that our method achieves superior performance in both trajectory estimation and static scene mapping, outperforming existing dynamic RGB SLAM baselines in long-term dynamic indoor environments.

Abstract

Dynamic SLAM

Explore similar work

RoSe-SLAM: Robust Semantic-Aware Gaussian Splatting SLAM from Dynamic Monocular Videos

DL-SLAM: Enabling High-Fidelity Gaussian Splatting SLAM in Dynamic Environments based on Dual-Level Probability

Flow4DGS-SLAM: Optical Flow-Guided 4D Gaussian Splatting SLAM