A robust visual-centered SLAM for autonomous vehicles in outdoor degradation based on hierarchical affine consistency and two-stage detection

DWG-LIO: a robust LiDAR SLAM system with hard ground constraints and adaptive dynamic removal

Visual SLAM (VSLAM) is a key technology for intelligent unmanned systems to achieve environmental perception and autonomous localization. In response to the issues of unstable feature extraction and the decreased adaptability and accuracy of SLAM systems caused by dynamic illumination changes in practical application scenarios, this paper proposes a robust VSLAM system based on heterogeneous feature data association, which does not require pre-trained models and can operate stably on resource-constrained platforms. The system designs a boundary-aware adaptive feature detection framework that maintains a stable number of features under complex illumination conditions through mirror padding and dynamic threshold adjustment. It also incorporates an edge-feature-guided quadtree optimization mechanism, constructing a region of interest (ROI) spatial mask to guide feature distribution and improve feature repeatability. By combining the GMS algorithm to enhance matching robustness, high-quality feature points are provided for back-end pose estimation, comprehensively improving the accuracy and adaptability of the SLAM system. Experimental results show that the proposed method achieves more stable and abundant feature extraction under complex illumination conditions. Compared with ORB-SLAM2, the feature repeatability rate is improved by approximately 39.8%, and the RMSE in multiple indoor and outdoor scenarios is reduced by 21.9%, providing an effective solution for the stable deployment and reliable operation of SLAM in practical applications.

Enhanced spatial distribution for robust Gaussian SLAM with view-consistency optimization

ABSTRACT Efficient quaternion‐based Simultaneous Localization and Mapping framework designed for real‐time navigation of Unmanned Aerial Vehicles in GPS‐denied environments is presented here. Unlike conventional Euler angle or matrix‐based SLAM approaches, the proposed system employs quaternion‐based propagation in Inertial Measurement Unit (IMU) module, thereby eliminating singularities such as gimbal lock and ensuring numerically stable orientation estimation. It integrates Quaternion Error‐State Extended Kalman Filter for incremental fusion of visual and inertial measurements and incorporates an incremental pose‐graph backend with quaternion retraction to reduce drift during long‐term navigation. Novelty lies in unifying quaternion mathematics with lightweight incremental optimization, thereby achieving a balance between accuracy and computational feasibility for real‐time onboard deployment. Experimental validation on KAIST Visual‐Inertial Odometry dataset demonstrates significant performance gains. Raw IMU is −50 Hz (20–25 ms/frame), far exceeding latency constraints of UAV flight dynamics. Additional analysis confirmed robustness with an average UAV speed of 1261.98 m/s, velocity stability (std. dev. 689.43 m/s) and reliable feature tracking (> 1800 map points reconstructed). These findings establish the framework as a novel, lightweight and robust SLAM solution suitable not only for UAVs but also for space–air–ground autonomous systems requiring accurate real‐time navigation under computational constraints.

With the rapid evolution of Digital Twins and Embodied AI, achieving fast, dense, and high-precision 3D perception in unknown environments has become paramount. However, existing Visual SLAM paradigms face a critical dilemma: geometry-based methods often fail in texture-less areas due to feature scarcity, while learning-based approaches frequently suffer from scale drift and unphysical deformations. To bridge this gap, we propose VGGT-Geo, a novel SLAM system that synergizes generative priors from Large Foundation Models with multi-modal geometric optimization. Distinguishing itself from simple cascaded architectures, we construct a Probabilistic Geometric Fusion framework, consisting of (1) Generative Warm-start, leveraging the holistic scene understanding capabilities of the VGGT, (2) Confidence-Aware Optimization to extract dense features via DINOv3 and predict their confidence map, and (3) a Multi-Modal Constraint Closure that fuses point-line features and metric depth priors to constrain rotational Degrees of Freedom in Manhattan Worlds. We conducted systematic evaluations on TUM, Replica, Tanks and Temples, and a challenging self-collected dataset featuring extreme lighting and texture-less walls. Experimental results demonstrate that VGGT-Geo exhibits superior robustness and accuracy in unseen environments. On our most challenging dataset, it achieves an Absolute Trajectory Error of 4–5 cm and a Relative Rotation Error of 0.79°, outperforming current state-of-the-art methods by approximately 50% in trajectory accuracy. This study validates that synergizing the intuition of Large Foundation Models with geometric rigor is a viable path toward next-generation robust SLAM.

Abstract Simultaneous Localization and Mapping (SLAM) is a fundamental technology used for robotic perception and navigation. Conventional SLAM systems assume static environments, which often fails in real-world dynamic scenarios where moving objects cause significant errors or tracking failures. Current approaches to dynamic feature filtering primarily rely on either multi-view geometry or deep learning techniques. While geometry-based approaches are computationally intensive and environmentally sensitive, deep learning-based methods are constrained by fixed priors, making them less effective at handling pseudo-static objects and prone to missed detections under occlusion or motion blur. To overcome these limitations, we present DC-SLAM, a framework that synergistically combines geometric constraints with deep learning to enable both accurate dynamic prior updates and robust missed-detection compensation in real-time operation. Its key innovations include, first, accurately identifying pseudo-static objects using a dynamic reference frame comparison algorithm and decoupling their motion from the background via depth information and multi-view geometric constraints, enabling real-time prior updates. Second, to address missed detections from sudden viewpoint changes, motion blur, or occlusions, it predicts target positions through spatiotemporal motion consistency analysis and locates compensation boxes within expanded regions via a column-wise depth clustering algorithm, greatly improving robustness. To evaluate the proposed algorithm, comprehensive experiments are conducted on the TUM RGB-D benchmark against state-of-the-art SLAM systems, along with real-world tests in dynamic environments. On four TUM RGB-D Fr3 dynamic sequences, DC-SLAM attains the lowest ATE across all methods, with 93%–98% lower ATE RMSE than ORB-SLAM3 and up to 80% lower than recent dynamic-aware baselines including DS-SLAM, RDS-SLAM, and SG-SLAM. Real-world mobile robot experiments further demonstrate DC-SLAM’s capability to reliably filter dynamic objects and maintain precise localization, while exhibiting exceptional robustness against challenging conditions such as viewpoint variations and occlusions, ultimately delivering a highly dependable visual SLAM solution for dynamic environments.

Abstract. Tree volume estimation is fundamental to forest management and inventory, yet traditional methods rely on allometric equations that introduce significant uncertainties due to generalized relationships and measurement limitations. This study evaluates the accuracy of mobile personal laser scanning (PLS) technology for tree volume estimation in pedunculate oak (Quercus robur L.) forests through a controlled comparison framework. Field work was conducted in January 2025 under optimal leaf-off conditions in a lowland oak stand in central Croatia. Three morphologically typical mature oak trees were selected within a single plot to enable controlled comparison while minimizing environmental variability. Data was acquired using PLS Faro Orbis Scanner, emphasizing complete stem coverage from multiple azimuths to support robust SLAM trajectory estimation and minimize occlusion effects. Three principal volume estimation approaches were evaluated: (i) sectioning volume obtained after felling, serving as the operational reference; (ii) PLS Schumacher-Hall volume, computed from LiDAR derived DBH and total height using established allometric relationships; and (iii) PLS Trunk Volume, computed directly from point cloud data using LiDAR360's trunk slicing workflow. Following PLS data acquisition, target trees were felled and bucked into contiguous sections, with length and end diameters recorded for each section to compute reference volumes. The sectioning dataset was treated as an operational reference rather than absolute ground truth, acknowledging potential reconstruction errors due to field conditions and occasional stem breakage. The study reveals important trade offs between measurement accuracy, operational efficiency, and methodological complexity, with sectioning volume providing the most direct measurement approach by eliminating remote sensing processing uncertainties. The research establishes a robust methodological framework for evaluating PLS performance in oak forests while highlighting both significant potential and current limitations of mobile laser scanning for operational forest inventory applications.

HiRo-SLAM is a visual-inertial SLAM system developed to achieve high accuracy and enhanced robustness. To address critical limitations of conventional methods, including systematic biases from imperfect camera models, uneven spatial feature distribution, and the impact of outliers, we propose a unified optimization framework that integrates four key innovations. First, Precise Camera Projection Modeling (PCPM) embeds a fully differentiable camera model in nonlinear optimization, ensuring accurate handling of camera intrinsics and distortion to prevent error accumulation. Second, Visibility Pyramid-based Adaptive Non-Maximum Suppression (P-ANMS) quantifies feature point contribution through a multi-scale pyramid, providing uniform visual constraints in weakly textured or repetitive regions. Third, Robust Optimization Using Graduated Non-Convexity (GNC) suppresses outliers through dynamic weighting, preventing convergence to local minima. Finally, the Point-Line Feature Fusion Frontend combines XFeat point features with SOLD2 line features, leveraging multiple geometric primitives to improve perception in challenging environments, such as those with weak textures or repetitive structures. Comprehensive evaluations on the EuRoC MAV, TUM-VI, and OIVIO benchmarks show that HiRo-SLAM outperforms state-of-the-art visual-inertial SLAM methods. On the EuRoC MAV dataset, HiRo-SLAM achieves a 30.0% reduction in absolute trajectory error compared to strong baselines and attains millimeter-level accuracy on specific sequences under controlled conditions. However, while HiRo-SLAM demonstrates state-of-the-art performance in scenarios with moderate texture and minimal motion blur, its effectiveness may be reduced in highly dynamic environments with severe motion blur or extreme lighting conditions.

Multi-modal sensor fusion-based LiDAR SLAM is a key capability for reliable mobile robot operation in complex indoor environments. However, it remains susceptible to localization drift and global inconsistency in typical degenerate scenarios such as feature sparsity, repetitive structures, and dynamic disturbances. To address these challenges, we propose a LiDAR-inertial SLAM method enhanced with visual QR-code landmarks. The front-end employs a lightweight EKF-based LiDAR-IMU odometry to ensure real-time and robust motion estimation, while the back-end constructs a unified factor graph that tightly couples LiDAR, IMU, loop-closure, and QR-code landmark factors within a single state space to achieve globally consistent cross-modal constraints. QR codes are further incorporated as persistent artificial landmarks to provide strong global anchoring in long corridors and repetitive or feature-degraded environments. In addition, an adaptive covariance and hierarchical weighting mechanism dynamically adjusts factor influence based on residual statistics and observation quality, thereby improving robustness under occlusion, degradation, and sensor noise variations. Experimental results demonstrate that the proposed system significantly improves localization accuracy and mapping stability across various challenging indoor scenarios. These findings validate the effectiveness of deeply integrating visual landmarks with LiDAR-inertial information, offering new scientific evidence and practical value for robust multi-modal SLAM in indoor robotic perception—fully aligning with the research scope of Scientific Reports.

The present study addresses the issues of feature loss and map consistency in visual SLAM under low-texture, low-light, and unstructured scenes by proposing an improved system based on ORB-SLAM3. The following innovative features are worthy of note: Firstly, a combination of custom voxel mapping and sparse SLAM is proposed for the purpose of enhancing matching robustness and 3D reconstruction quality in low-texture regions. Secondly, the utilization of a depth map neural network for the fusion of point and line features is suggested. Tests on public datasets and other unstructured scenes demonstrate that this method significantly improves joint feature matching efficiency, validating the complementary advantages of point and line features. The results indicate a considerable boost in both localization accuracy and environmental adaptability in challenging scenarios, setting the foundation for more reliable SLAM applications in real-world conditions.

Abstract Despite the significant progress in dynamic SLAM research, achieving an optimal balance among accuracy, efficiency, and robustness remains an open challenge. To address this issue, this paper proposes GridClust-SLAM, an RGB-D SLAM method that integrates grid-based analysis with a lightweight semantic guidance mechanism. The method partitions the image into a grid, utilizing each grid cell as the fundamental unit for analysis. An edge-aware grid density clustering module is designed: initially, edge information is leveraged to isolate grid cells containing object boundaries; subsequently, the remaining non-edge grid cells are subjected to a weighted DBSCAN clustering based on their multi-dimensional attributes to obtain the grid clustering results. Furthermore, under the semantic guidance provided by a lightweight network, these clustering results are filtered and expanded in conjunction with a depth consistency check to accurately generate a dynamic grid mask. Experimental results on the TUM, BONN, and ICL-NUIM datasets demonstrate that the proposed method significantly outperforms existing state-of-the-art methods in terms of both absolute trajectory error and relative pose error, while maintaining high computational efficiency.

LoD-GS: Robust and Lightweight Gaussian Splatting SLAM for Real-Time Volumetric Scene Reconstruction

RIDR: Robust LiDAR SLAM with Invariant Distribution Registration and Flexible Constraint Modulation

In recent years, humanoid robots have gradually emerged as crucial “smart terminals” within the IoT, significantly expanding the application scenarios of IoT. This paper proposes a visual enhancement based robust 3D Gaussian SLAM (3DGS-SLAM) to improve the environmental perception ability of humanoid robots. First, the dynamic mask provided by YOLOv11 is optimized using composite morphological operations to obtain high-confidence static features, which are then employed for accurate estimation of the robot’s pose. Subsequently, a keyframe selection strategy combining scaled interval Kalman filtering (SIKF) with the motion model of a floating base humanoid robot is designed. It quantifies the uncertainty of pose by dynamically dividing the motion interval of the robot, and introduces a scaling factor to adaptively adjust the interval width, thereby filtering out high-quality keyframes to ensure stable pose tracking and efficient mapping. Further, a mapping strategy that combines Gaussian-Laplacian pyramid and Gaussian ellipsoid adaptive density control is developed by combining 3D Gaussian splashing. This method can not only identify Gaussian ellipsoids requiring segmentation by preserving high-frequency details in keyframes, but also control their splitting direction through gradient modulation, effectively reducing redundancy among Gaussian ellipsoids while enhancing mapping quality. The experimental results show that our method improves the average absolute trajectory error(ATE) on the BONN and TUM datasets by an average of 90.8% and 93.1% compared to ORB-SLAM3, significantly improving the localization accuracy and mapping consistency of humanoid robots in dynamic crowded scenes.

This paper presents a robust SLAM system designed for underwater environments, along with a simulated underwater localization dataset generated through a high-performance physical simulation platform. We investigate mainstream feature matching methods and incorporate improved versions into the SLAM framework to enhance its adaptability to challenging conditions such as poor illumination, low-texture regions, and significant camera motion. The proposed system supports multiple sensor configurations, including monocular, stereo, monocular-inertial, and stereo-inertial setups, providing flexibility for various real-world applications. Evaluations on our synthetic underwater dataset demonstrate that the system achieves reliable and robust performance in low-texture underwater environments, while maintaining high localization accuracy and strong tracking stability. The code will be released at https://github.com/Jiongmingkesi/ Underwater-Robust-SLAM after the manuscript is formally accepted.

Dynamic SLAM methods based on epipolar constraints provide a simple and efficient solution for distinguishing dynamic features. However, such constraints tend to fail under geometrically degenerate motion sce-narios, such as pure rotation, long-range linear motion, or coplanar scene structures, leading to misclassification of dynamic features, trajectory drift, and inaccurate map reconstruction. To address these challenges, a parallel SLAM framework is proposed, integrating semantic guidance, degeneracy-aware dynamic feature discrimination, and robust keyframe selection to handle degenerate motions. Without requiring inertial measurements, the proposed method relies solely on visual information to enable dynamic fea-ture removal, avoiding IMU drift. Specifically, a lightweight object detection module is introduced using the fast and compact YOLO-FASTEST model, enabling efficient semantic perception to provide prior information for dynamic point removal. Furthermore, a multi- frame Bayesian motion consistency criterion is proposed that jointly considers camera motion priors and observation residuals of feature points to enable dynamic feature discrimination in degenerate scenarios. In addition, an adaptive multi-metric keyframe insertion strategy is designed, jointly considering pose change magnitude, image entropy variation, and the ratio of constrained pixels, to enhance keyframe selection under motion-degenerate scenes. Experimental results demonstrate that the proposed method achieves superior trajectory accuracy and map completeness under various dynamic interference conditions, while maintaining real-time performance.

In dynamic and cluttered environments, traditional Simultaneous Localization and Mapping (SLAM) systems often suffer from degraded localization accuracy and unstable map construction due to the presence of moving objects and occlusions. To address these challenges, we propose DFusion-SLAM, a lightweight and robust SLAM framework that integrates an enhanced object detection module into ORB-SLAM3. The detection module is based on an improved D-Fine architecture, in which the original Transformer is replaced with a more efficient PolaLinear Attention mechanism. Furthermore, a MetaFormer-based semantic fusion structure is introduced to strengthen multi-scale feature representation. These architectural improvements jointly enhance detection accuracy while reducing model complexity, achieving a performance increase from 42.8% to 43.7% mean Average Precision (mAP). Experimental evaluations on dynamic RGB-D sequences from the TUM and Bonn datasets demonstrate that DFusion-SLAM significantly improves localization accuracy and mapping stability under dynamic conditions, while maintaining high computational efficiency. These results highlight the framework’s strong potential for real-time deployment in IoT-oriented mobile and robotic platforms operating in complex environments.

Traditional monocular and stereo visual SLAM systems often fail to operate stably in complex unstructured environments (e.g., weakly textured or repetitively textured scenes) due to feature scarcity from their limited fields of view. In contrast, multi-camera systems can effectively overcome the perceptual limitations of monocular or stereo setups by providing broader field-of-view coverage. However, most existing multi-camera visual SLAM systems are primarily feature-based and thus still constrained by the inherent limitations of feature extraction in such environments. To address this issue, a multi-camera visual SLAM framework based on the direct method is proposed. In the front-end, a detector-free matcher named Efficient LoFTR is incorporated, enabling pose estimation through dense pixel associations to improve localization accuracy and robustness. In the back-end, geometric constraints among multiple cameras are integrated, and system localization accuracy is further improved through a joint optimization process. Through extensive experiments on public datasets and a self-built simulation dataset, the proposed method achieves superior performance over state-of-the-art approaches regarding localization accuracy, trajectory completeness, and environmental adaptability, thereby validating its high robustness in complex unstructured environments.

Recent advancements in 3DGS-based explicit mapping have significantly improved SLAM performance, achieving more realistic environment reconstruction and faster processing. However, issues such as data loss caused by unstable data transmission, textureless and repetitive-texture often occur in real-world scenarios. These sensor degradation problems lead to tracking drift caused by incorrect feature or pixel matching, as well as artifacts due to rendering errors. To address these challenges, we propose the GaussR-SLAM, the first 3DGS-based SLAM system designed for sensor degradation scenarios. By initializing Gaussians using hybrid feature points and employing an adaptive tracking switch mechanism, we achieve efficient data association and pose correction. In the mapping thread, we propose fusion pruning based on the spatial distribution of Gaussians to eliminate artifacts and mapping errors, while also designing a hybrid descriptor loss for rendered images to achieve photorealistic rendering. Experimental results on standard datasets demonstrate that our system outperforms existing 3DGS-based SLAM systems under sensor degradation, particularly in scenarios involving data loss.