Laser-based Simultaneous Localization and Mapping (SLAM) is fundamental to autonomous navigation systems. However, conventional frameworks such as Lightweight and Ground-Optimized LiDAR Odometry and Mapping (LeGO-LOAM) face challenges in geometric segmentation robustness, feature extraction accuracy, and loop closure reliability, especially in complex and dynamic environments. To overcome these limitations, this paper proposes LeGO-LOAM-RAS, a semantic-aware, graph-based LiDAR SLAM framework that integrates RandLA-Net for multi-class semantic segmentation, a depth-guided AFE strategy, and a semantic-contextual loop closure mechanism. In the proposed system, RandLA-Net replaces traditional geometry-based segmentation with a deep learning approach, enabling fine-grained scene understanding and the effective discrimination of objects such as roads, vehicles, and buildings. This enhances both local contextual awareness and global structural representation. The AFE module dynamically adjusts neighborhood configurations and angular resolutions based on depth cues, employing depth-error-based noise suppression and curvature refinement to improve the reliability of planar and edge features. For loop closure, a semantic-contextual descriptor is constructed by fusing geometric features with semantic histograms in a polar grid representation, introducing joint geometric-semantic constraints. This design improves loop closure robustness by mitigating ambiguity in perceptually similar environments and suppressing the impact of dynamic elements. Extensive evaluations on publicly available benchmark datasets validate the effectiveness of LeGO-LOAM-RAS, demonstrating substantial improvements in localization accuracy and overall system robustness compared to state-of-the-art methods.

In recent years, semantic segmentation of large-scale point clouds has garnered significant attention due to its critical role in 3D scene understanding. However, the inherent complexity and uneven distribution of large-scale point clouds, coupled with substantial inter-class similarity, significantly hinder the discriminative power of existing segmentation approaches. RandLA-Net has shown strong capabilities in directly inferring semantic information. Building upon this foundation, we proposed three redesigned modules to improve the accuracy of point cloud segmentation: a Local Contextual Feature (LCF) module, a Global Contextual Feature (GCF) module and, a Contextual Feature Enhancement (CFE) module. The LCF module preserves the local spatial encoding unit and introduces an improved dual attention mechanism that independently computes geometric and feature-based attention scores. This facilitates more effective local feature aggregation and overcomes the segmentation artifacts caused by the difficulty in distinguishing similar classes. To complement local representations, the GCF module is integrated to capture scene-level semantics across all 3D points by using the spatial position volume ratio, thereby addressing feature extraction from both local and global perspectives. The CFE module is designed as a plug-and-play component, which enhances feature representations by integrating richer contextual cues from both explicit 3D geometry and implicit feature spaces, along with global bilinear interactions. Comprehensive experiments on the S3DIS (indoor) and Semantic3D (outdoor) datasets show that our method attains Overall Accuracy (OA) scores of 89.8% and 95.3%, and mean Intersection over Union (mIoU) scores of 73.3% and 78.0%, respectively, outperforming existing methods and providing new perspectives for large-scale point cloud semantic segmentation across diverse environments.

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.

To enhance the reliability and stability of simultaneous localization and mapping (SLAM) in dynamic environments, we propose a novel SLAM system integrating an advanced real-time object detection algorithm, the real-time detection transformer (RT-DETR). Our approach combines RT-DETR’s object detection capabilities with an optical flow-based dynamic thresholding method, effectively filtering out feature points associated with dynamic objects and thereby improving SLAM performance in such environments. We have optimized RT-DETR by substituting its original network backbone with lightweight modules, which reduces the number of parameters by 45% while only incurring a 5% reduction in accuracy. This optimization significantly lowers computational costs, making it feasible for deployment on mobile devices. Experiments conducted on the TUM and BONN dynamic datasets demonstrate that our system reduces the root mean square error (RMSE) of absolute trajectory and relative pose error (RPE) by approximately 28.82% compared to oriented fast and rotated brief-SLAM3 (ORB-SLAM3). Furthermore, experiments conducted on both a high-performance device and an embedded device demonstrate that, compared to Crowd-SLAM, which employs you only look once (YOLO) for dynamic object removal, our approach achieves an 8.52% improvement in absolute trajectory error (ATE), while the average frame per second (FPS) only decreases by 3.07%.