Performing Loop Closure Research Articles

Neural Surfel Reconstruction: Addressing Loop Closure Challenges in Large-Scale 3D Neural Scene Mapping.

Efficiently reconstructing complex and intricate surfaces at scale remains a significant challenge in 3D surface reconstruction. Recently, implicit neural representations have become a popular topic in 3D surface reconstruction. However, how to handle loop closure and bundle adjustment is a tricky problem for neural methods, because they learn the neural parameters globally. We present an algorithm that leverages the concept of surfels and expands relevant definitions to address such challenges. By integrating neural descriptors with surfels and framing surfel association as a deformation graph optimization problem, our method is able to effectively perform loop closure detection and loop correction in challenging scenarios. Furthermore, the surfel-level representation simplifies the complexity of 3D neural reconstruction. Meanwhile, the binding of neural descriptors to corresponding surfels produces a dense volumetric signed distance function (SDF), enabling the mesh reconstruction. Our approach demonstrates a significant improvement in reconstruction accuracy, reducing the average error by 16.9% compared to previous methods, while also generating modeling files that are up to 90% smaller than those produced by traditional methods.

H-SLAM: Hybrid direct–indirect visual SLAM

The recent success of hybrid methods in monocular odometry has led to many attempts to generalize the performance gains to hybrid monocular SLAM. However, most attempts fall short in several respects, with the most prominent issue being the need for two different map representations (local and global maps), with each requiring different, computationally expensive, and often redundant processes to maintain. Moreover, these maps tend to drift with respect to each other, resulting in contradicting pose and scene estimates, and leading to catastrophic failure. In this paper, we propose a novel approach that makes use of descriptor sharing to generate a single inverse depth scene representation. This representation can be used locally, queried globally to perform loop closure, and has the ability to re-activate previously observed map points after redundant points are marginalized from the local map, eliminating the need for separate map maintenance processes. The maps generated by our method exhibit no drift between each other, and can be computed at a fraction of the computational cost and memory footprint required by other monocular SLAM systems. Despite the reduced resource requirements, the proposed approach maintains its robustness and accuracy, delivering performance comparable to state-of-the-art SLAM methods (e.g., LDSO, ORB-SLAM3) on the majority of sequences from well-known datasets like EuRoC, KITTI, and TUM VI. The source code is available at: https://github.com/AUBVRL/fslam_ros_docker.

NDT RC: Normal Distribution Transform Occupancy 3D Mapping With Recentering

The Normal Distribution Transform Occupancy Map (NDT OM) is a mapping algorithm able to represent a dynamic 3D environment. The resulting map has fixed boundaries, thus a robot with unbounded displacement might fall outside of the map due to memory limitation. In this paper, a recentering algorithm called NDT RC is proposed to avoid this issue. NDT RC extends the use of NDT OM for vehicles with unbounded displacements. NDT RC provides a seamless translation of the map as the robot gets far from the center of the previous map. The influence of NDT RC on the precision of the estimated trajectory of the robot, or odometry, is examined on two publicly available datasets, the KITTI and Ford datasets. An analysis of the sensitivity of the NDT RC to its tuning parameters is carried out using the Ford dataset, while the KITTI dataset is used to measure the influence of the density of the input point cloud. The results show that the proposed recentering strategy improves the accuracy of the odometry calculated by registering the latest lidar scan on the generated map compared to other NDT based approaches (NDT OM, NDT OM Fusion, SE-NDT). In particular, the proposed method, which does not perform loop closure, reduces the mean absolute translation error by 16% and the runtime by 88% compared to the NDT OM Fusion on the Ford dataset.

LFVB-BioSLAM: A Bionic SLAM System with a Light-Weight LiDAR Front End and a Bio-Inspired Visual Back End.

Simultaneous localization and mapping (SLAM) is one of the crucial techniques applied in autonomous robot navigation. The majority of present popular SLAM algorithms are built within probabilistic optimization frameworks, achieving high accuracy performance at the expense of high power consumption and latency. In contrast to robots, animals are born with the capability to efficiently and robustly navigate in nature, and bionic SLAM algorithms have received increasing attention recently. Current bionic SLAM algorithms, including RatSLAM, with relatively low accuracy and robustness, tend to fail in certain challenging environments. In order to design a bionic SLAM system with a novel framework and relatively high practicality, and to facilitate the development of bionic SLAM research, in this paper we present LFVB-BioSLAM, a bionic SLAM system with a light-weight LiDAR-based front end and a bio-inspired vision-based back end. We adopt a range flow-based LiDAR odometry as the front end of the SLAM system, providing the odometry estimation for the back end, and we propose a biologically-inspired back end processing algorithm based on the monocular RGB camera, performing loop closure detection and path integration. Our method is verified through real-world experiments, and the results show that LFVB-BioSLAM outperforms RatSLAM, a vision-based bionic SLAM algorithm, and RF2O, a laser-based horizontal planar odometry algorithm, in terms of accuracy and robustness.

DE‐SLAM: SLAM for highly dynamic environment

AbstractSimultaneous localization and mapping (SLAM) is crucial for autonomous mobile robots. Most of the current SLAM systems are based on an assumption: the environment is static. However, the real environment is full of dynamic elements, such as pedestrians or vehicles, as well as changes in illumination and appearance over time. In this paper, DE‐SLAM, a visual SLAM system that can deal with short‐term and long‐term dynamic elements at the same time is proposed. A novel dynamic detection and tracking module that utilizes both semantic and metric information is proposed, and the localization accuracy is highly improved by eliminating features falling on the dynamic objects. A unified loop detection, loop check and global optimization module is used to perform loop closure. Experimental results on datasets and real environments show that DE‐SLAM outperforms other state‐of‐the‐art SLAM systems in dynamic environments.

Learning of Holism-Landmark Graph Embedding for Place Recognition in Long-Term Autonomy

Place recognition plays an important role to perform loop closure detection of large-scale, long-term simultaneous localization and mapping in loopy environments. The long-term place recognition problem is challenging because the environment appearance exhibits significant long-term variations across various times of the day, months, and seasons. In this letter, we introduce a novel place representation approach that simultaneously integrates semantic landmarks and holistic information to achieve place recognition in long-term autonomy. First, a graph is constructed for each place. The graph nodes encode all landmarks and the holistic image of the place scene recorded in different scenarios. The edges connecting the nodes indicate that these nodes represent the same landmark or place, even though places and landmarks encoded by the nodes may exhibit different appearances in the long-term periods. Then, a graph embedding is learned to preserve the locality in the feature descriptor space, i.e., finding a projection such that the same landmark and place have the identical representation in the new projected descriptor space, no matter in what scenarios they are recorded. We formulate the embedding learning as an optimization problem and implement a new solver that provides a theoretical convergence guarantee. Extensive evaluations are conducted using large-scale benchmark datasets of place recognition in long-term autonomy, which has shown our approach's promising performance.

Integrating SLAM and gas distribution mapping (SLAM-GDM) for real-time gas source localization

ABSTRACTGas distribution mapping (GDM) learns models of the spatial distribution of gas concentrations across 2D/3D environments, among others, for the purpose of localizing gas sources. GDM requires run-time robot positioning in order to associate measurements with locations in a global coordinate frame. Most approaches assume that the robot has perfect knowledge about its position, which does not necessarily hold in realistic scenarios. We argue that the simultaneous localization and mapping (SLAM) algorithm should be used together with GDM to allow operation in an unknown environment. This paper proposes an SLAM-GDM approach that combines Hector SLAM and Kernel DM + V through a map merging technique. We argue that Hector SLAM is suitable for the SLAM-GDM approach since it does not perform loop closure or global corrections, which in turn would require to re-compute the gas distribution map. Real-time experiments were conducted in an environment with single and multiple gas sources. The results showed that the predictions of gas source location in all trials were often correct to around 0.5–1.5 m for the large indoor area being tested. The results also verified that the proposed SLAM-GDM approach and the designed system were able to achieve real-time operation.

Dense RGB-D visual odometry using inverse depth

In this paper we present a dense visual odometry system for RGB-D cameras performing both photometric and geometric error minimisation to estimate the camera motion between frames. Contrary to most works in the literature, we parametrise the geometric error by the inverse depth instead of the depth, which translates into a better fit of the distribution of the geometric error to the used robust cost functions. To improve the accuracy we propose to use a keyframe switching strategy based on a visibility criteria between frames. For the comparison of our approach with state-of-the-art approaches we use the popular datasets from the TUM for RGB-D benchmarking as well as two synthetic datasets. Our approach shows to be competitive with state-of-the-art methods in terms of drift in metres per second, even compared to methods performing loop closure too. When comparing to approaches performing pure odometry like ours, our method outperforms them in the majority of the tested datasets. Additionally we show that our approach is able to work in real time and we provide a qualitative evaluation on our own sequences showing a low drift in the 3D reconstructions. We have implemented this method within the scope of PCL (Point Cloud Library) as a branch of the code for large scale KinectFusion, where the original ICP system for odometry estimation has been completely substituted by our method. A PCL fork including the modified method is available for download.

Vision-based sparse topological mapping

Most of the existing appearance-based topological mapping algorithms produce dense topological maps in which each image stands as a node in the topological graph. Sparser maps can be built by representing groups of visually similar images of a sequence as nodes of a topological graph. In this paper, we present a sparse/hierarchical topological mapping framework which uses Image Sequence Partitioning (ISP) to group visually similar images of a sequence as nodes which are then connected on the occurrence of loop closures to form a topological graph. An indexing data structure called Hierarchical Inverted File (HIF) is proposed to store the sparse maps so as to perform loop closure at the two different resolutions of the map namely the node level and image level. TFIDF weighting is combined with spatial and frequency constraints on the detected features for improved loop closure robustness. Our approach is compared with two other existing sparse mapping approaches which use ISP. Sparsity, efficiency and accuracy of the resulting maps are evaluated and compared to that of the other two techniques on publicly available outdoor omni-directional image sequences.