Simultaneous localization and mapping (SLAM) technology based on light detection and ranging (LiDAR) sensors has been widely used in various environmental sensing tasks indoors and outdoors. However, it still lacks effective constraints in structured environments such as corridors and parking lots, and its accuracy needs improvement. Based on this, a planar constraint-assisted LiDAR SLAM algorithm based on the Manhattan World (MW) assumption is proposed in this paper. The algorithm extracts planes from the environment point cloud submap, classifies the planes according to the ground and vertical planes, and calculates the main direction angles of the ground and vertical plane, respectively, to construct constraints. To enhance the stability and robustness of the system, a two-step main direction angle calculation and update strategy are designed, and a hysteresis update is used to avoid the introduction of errors by unoptimized planes. This paper uses a backpack laser scanning system to collect experimental data in various scenes. These data are used to compare our method with three open-source LiDAR SLAM algorithms, that are currently more widely used and perform better. Qualitative and quantitative experiments are conducted to verify the effectiveness of our method. The experimental results show that the absolute accuracy of the point clouds obtained by our method is improved by 77.46% on average compared with the other three algorithms in the environment, conforming to the MW assumption, which verifies the effectiveness of the algorithm.

Loop closure detection is a fundamental technique in SLAM (Simultaneous Localization and Mapping), playing a crucial role in the map-building process. Inspired by the Mercator projection and depth maps, this paper presents a novel loop closure detection method for LiDAR SLAM, termed the Mercator Descriptor. Our approach applies the Mercator projection principles to the point cloud data captured by the sensor in each frame and extracts depth information to generate descriptors. Additionally, we utilize a two-step identification strategy, from coarse to fine, to improve the accuracy and recall of loop closure detection. Testing on public datasets demonstrates that our algorithm achieves an average maximum F1 score of 94.57%, performing well across various datasets. Notably, we have integrated our algorithm into existing SLAM frameworks, and results from multiple datasets show that our algorithm can perform real-time and efficient map optimization. To share our findings and contribute to the community, we have opensourced our code on GitHub: https://github.com/wangzika/Mercator-Descriptor.git.

Accurate LiDAR-inertial odometry (LIO) highly depends on the geometric fidelity of the underlying environment representation. We explore the new and interesting research direction of integrating semantic segmentation models into metric odometry algorithms to enrich their representational capacity. Specifically, this letter proposes a semantic-driven hybrid voxel representation in which an off-the-shelf 3D segmentation network assigns every voxel to either a planar or nonplanar class, using planar and Gaussian representations, respectively. Consequently, a hybrid scan matching strategy is presented using class-specific residual models that are tailored to the distinct error statistics of each surface category. The scan matcher is embedded within an Iterated Extended Kalman Filter (IEKF) for odometry and mapping. We evaluate our method on diverse platforms and environments, and show improved localization accuracy across various indoor and outdoor scenarios, while maintaining real-time performance.

Frame-wise semantic segmentation of indoor lidar scans is a fundamental step toward higher-level 3D scene understanding and mapping applications. However, acquiring frame-wise ground truth for training deep learning models is costly and time-consuming. This challenge is largely addressed, for imagery, by Visual Foundation Models (VFMs) which segment image frames. The same VFMs may be used to train a lidar scan frame segmentation model via a 2D-to-3D distillation pipeline. The success of such distillation has been shown for autonomous driving scenes, but not yet for indoor scenes. Here, we study the feasibility of repeating this success for indoor scenes, in a frame-wise distillation manner by coupling each lidar scan with a VFM-processed camera image. The evaluation is done using indoor SLAM datasets, where pseudo-labels are used for downstream evaluation. Also, a small manually annotated lidar dataset is provided for validation, as there are no other lidar frame-wise indoor datasets with semantics. Results show that the distilled model achieves up to 56% mIoU under pseudo-label evaluation and around 36% mIoU with real-label, demonstrating the feasibility of cross-modal distillation for indoor lidar semantic segmentation without manual annotations.

Simultaneous Localization and Mapping (SLAM) technology, utilizing Light Detection and Ranging (LiDAR) sensors, is crucial for 3D environment perception and mapping. However, the absence of absolute observations and the inefficiency of single-robot perception present challenges for LiDAR SLAM in indoor environments. In this paper, we propose a multi-robot (MR) collaborative mapping method based on the Manhattan descriptor (MD) named MR-MD to overcome these limitations and improve the perception accuracy of LiDAR SLAM in indoor environments. The proposed method consists of two modules: MD generation and MD optimization. In the first module, each robot builds a local submap and constructs MD by parameterizing the planes in the submap. In the second module, the global main direction is updated using the historical MD of each robot, and constraints are built for each robot's horizontal and vertical directions according to their current MD and optimized. We conducted extensive comparisons with other multi-robot and single-robot LiDAR SLAM methods using real indoor data, and the results show that our method achieved higher mapping accuracy.