3D Simultaneous Localization And Mapping Research Articles

A Low-Cost 3D SLAM System Integration of Autonomous Exploration Based on Fast-ICP Enhanced LiDAR-Inertial Odometry

Advancements in robotics and mapping technology have spotlighted the development of Simultaneous Localization and Mapping (SLAM) systems as a key research area. However, the high cost of advanced SLAM systems poses a significant barrier to research and development in the field, while many low-cost SLAM systems, operating under resource constraints, fail to achieve high-precision real-time mapping and localization, rendering them unsuitable for practical applications. This paper introduces a cost-effective SLAM system design that maintains high performance while significantly reducing costs. Our approach utilizes economical components and efficient algorithms, addressing the high-cost barrier in the field. First, we developed a robust robotic platform based on a traditional four-wheeled vehicle structure, enhancing flexibility and load capacity. Then, we adapted the SLAM algorithm using the LiDAR-inertial Odometry framework coupled with the Fast Iterative Closest Point (ICP) algorithm to balance accuracy and real-time performance. Finally, we integrated the 3D multi-goal Rapidly exploring Random Tree (RRT) algorithm with Nonlinear Model Predictive Control (NMPC) for autonomous exploration in complex environments. Comprehensive experimental results confirm the system’s capability for real-time, autonomous navigation and mapping in intricate indoor settings, rivaling more expensive SLAM systems in accuracy and efficiency at a lower cost. Our research results are published as open access, facilitating greater accessibility and collaboration.

A new covariance intersection based integrated SLAM framework for 3D outdoor agricultural applications

AbstractThis letter introduces a novel integrated framework for simultaneous localization and mapping (SLAM) tailored for general agricultural applications. The framework combines a cutting‐edge SLAM method, LIO‐SAM, with covariance intersection for sensor fusion. Agricultural robots often operate in unstructured environments with sparse feature points and encounter repeated similar information, such as trees. Therefore, a fusion framework based on 3D SLAM augmented with additional information, such as feature‐independent GPS data, becomes essential. This study proposes an integrated SLAM framework by introducing a convergence strategy based on covariance analysis, incorporating a state‐of‐the‐art 3D SLAM technique. The convergence methods, namely “dynamic weight assignment” and “winner takes all”, are presented alternatively, tailored to seamlessly integrate with the proposed framework. Evaluations using a public dataset and an experiment demonstrate the effectiveness of the approach through numerical analysis and visual representation. The results illustrate that this method surpasses conventional approaches in accurately estimating the robot's position. In the future, this research will focus on automating crop cultivation and harvesting by integrating the proposed system with robot arm control.

Development of Autonomous Mobile Robot with 3DLidar Self-Localization Function Using Layout Map

In recent years, there has been growing interest in autonomous mobile robots equipped with Simultaneous Localization and Mapping (SLAM) technology as a solution to labour shortages in production and distribution settings. SLAM allows these robots to create maps of their environment using devices such as Lidar, radar, and sonar sensors, enabling them to navigate and track routes without prior knowledge of the environment. However, the manual operation of these robots for map construction can be labour-intensive. To address this issue, this research aims to develop a 3D SLAM autonomous mobile robot system that eliminates the need for manual map construction by utilizing existing layout maps. The system includes a PC for self-position estimation, 3DLidar, a camera for verification, a touch panel display, and the mobile robot itself. The proposed SLAM method extracts stable wall point cloud information from 3DLidar, matches it with the wall surface information in the layout map, and uses a particle filter to estimate the robot’s position. The system also includes features such as route creation, tracking, and obstacle detection for autonomous movement. Experiments were conducted to compare the proposed system with conventional 3D SLAM methods. The results showed that the proposed system significantly reduced errors in self-positioning and enabled accurate autonomous movement on specified routes, even in the presence of slight differences in layout maps and obstacles. Ultimately, this research demonstrates the effectiveness of a system that can transport goods without the need for manual environment mapping, addressing labour shortages in such environments.

IS-CAT: Intensity-Spatial Cross-Attention Transformer for LiDAR-Based Place Recognition.

LiDAR place recognition is a crucial component of autonomous navigation, essential for loop closure in simultaneous localization and mapping (SLAM) systems. Notably, while camera-based methods struggle in fluctuating environments, such as weather or light, LiDAR demonstrates robustness against such challenges. This study introduces the intensity and spatial cross-attention transformer, which is a novel approach that utilizes LiDAR to generate global descriptors by fusing spatial and intensity data for enhanced place recognition. The proposed model leveraged a cross attention to a concatenation mechanism to process and integrate multi-layered LiDAR projections. Consequently, the previously unexplored synergy between spatial and intensity data was addressed. We demonstrated the performance of IS-CAT through extensive validation on the NCLT dataset. Additionally, we performed indoor evaluations on our Sejong indoor-5F dataset and demonstrated successful application to a 3D LiDAR SLAM system. Our findings highlight descriptors that demonstrate superior performance in various environments. This performance enhancement is evident in both indoor and outdoor settings, underscoring the practical effectiveness and advancements of our approach.

A 3D Lidar SLAM System Based on Semantic Segmentation for Rubber-Tapping Robot

Simultaneous localization and mapping (SLAM) in rubber plantations is a challenging task for rubber-tapping robots. Due to the long-term stability of tree trunks in rubber plantations, a SLAM system based on semantic segmentation, called Se-LOAM, is proposed in this work. The 3D lidar point cloud datasets of trunks collected in rubber plantations of Hainan University are used to train the semantic model, and the model is used to extract features of trunk point clouds. After clustering the trunk point clouds, each single rubber tree instance is segmented based on the Viterbi algorithm. The point clouds of tree instances are fitted to the cylindrical trunk models for semantic cluster association and positional estimation, which are used for lidar odometry and mapping. The experimental results show that the present SLAM system is accurate in establishing online mapping, and the location of the trunk in the map is clearer. Specifically, the average relative pose error is 0.02 m, which is better than the positioning performance of LOAM and LeGO-LOAM. The average error of estimating the diameter at breast height (DBH) is 0.57 cm, and it only takes 401.4 kB to store a map of the area of approximately 500 m2, which is about 10% less than other classic methods. Therefore, Se-LOAM can meet the requirements of online mapping, providing a robust SLAM method for rubber-tapping robots.

A Robust Semi-Direct 3D SLAM for Mobile Robot Based on Dense Optical Flow in Dynamic Scenes.

Dynamic objects bring about a large number of error accumulations in pose estimation of mobile robots in dynamic scenes, and result in the failure to build a map that is consistent with the surrounding environment. Along these lines, this paper presents a robust semi-direct 3D simultaneous localization and mapping (SLAM) algorithm for mobile robots based on dense optical flow. First, a preliminary estimation of the robot's pose is conducted using the sparse direct method and the homography matrix is utilized to compensate for the current frame image to reduce the image deformation caused by rotation during the robot's motion. Then, by calculating the dense optical flow field of two adjacent frames and segmenting the dynamic region in the scene based on the dynamic threshold, the local map points projected within the dynamic regions are eliminated. On this basis, the robot's pose is optimized by minimizing the reprojection error. Moreover, a high-performance keyframe selection strategy is developed, and keyframes are inserted when the robot's pose is successfully tracked. Meanwhile, feature points are extracted and matched to the keyframes for subsequent optimization and mapping. Considering that the direct method is subject to tracking failure in practical application scenarios, the feature points and map points of keyframes are employed in robot relocation. Finally, all keyframes and map points are used as optimization variables for global bundle adjustment (BA) optimization, so as to construct a globally consistent 3D dense octree map. A series of simulations and experiments demonstrate the superior performance of the proposed algorithm.

Research on orchard navigation method based on fusion of 3D SLAM and point cloud positioning.

Accurate navigation is crucial in the construction of intelligent orchards, and the need for vehicle navigation accuracy becomes even more important as production is refined. However, traditional navigation methods based on global navigation satellite system (GNSS) and 2D light detection and ranging (LiDAR) can be unreliable in complex scenarios with little sensory information due to tree canopy occlusion. To solve these issues, this paper proposes a 3D LiDAR-based navigation method for trellis orchards. With the use of 3D LiDAR with a 3D simultaneous localization and mapping (SLAM) algorithm, orchard point cloud information is collected and filtered using the Point Cloud Library (PCL) to extract trellis point clouds as matching targets. In terms of positioning, the real-time position is determined through a reliable method of fusing multiple sensors for positioning, which involves transforming the real-time kinematics (RTK) information into the initial position and doing a normal distribution transformation between the current frame point cloud and the scaffold reference point cloud to match the point cloud position. For path planning, the required vector map is manually planned in the orchard point cloud to specify the path of the roadway, and finally, navigation is achieved through pure path tracking. Field tests have shown that the accuracy of the normal distributions transform (NDT) SLAM method can reach 5cm in each rank with a coefficient of variation that is less than 2%. Additionally, the navigation system has a high positioning heading accuracy with a deviation within 1° and a standard deviation of less than 0.6° when moving along the path point cloud at a speed of 1.0m/s in a Y-trellis pear orchard. The lateral positioning deviation was also controlled within 5cm with a standard deviation of less than 2cm. This navigation system has a high level of accuracy and can be customized to specific tasks, making it widely applicable in trellis orchards with autonomous navigation pesticide sprayers.

Accuracy Evaluation and Branch Detection Method of 3D Modeling Using Backpack 3D Lidar SLAM and UAV-SfM for Peach Trees during the Pruning Period in Winter

In the winter pruning operation of deciduous fruit trees, the number of pruning branches and the structure of the main branches greatly influence the future growth of the fruit trees and the final harvest volume. Terrestrial laser scanning (TLS) is considered a feasible method for the 3D modeling of trees, but it is not suitable for large-scale inspection. The simultaneous localization and mapping (SLAM) technique makes it possible to move the lidar on the ground and model quickly, but it is not useful enough for the accuracy of plant detection. Therefore, in this study, we used UAV-SfM and 3D lidar SLAM techniques to build 3D models for the winter pruning of peach trees. Then, we compared and analyzed these models and further proposed a method to distinguish branches from 3D point clouds by spatial point cloud density. The results showed that the 3D lidar SLAM technique had a shorter modeling time and higher accuracy than UAV-SfM for the winter pruning period of peach trees. The method had the smallest RMSE of 3084 g with an R2 = 0.93 compared to the fresh weight of the pruned branches. In the branch detection part, branches with diameters greater than 3 cm were differentiated successfully, regardless of whether before or after pruning.

Finding the best hardware configuration for 2D SLAM in indoor environments via simulation based on Google Cartographer

One of the most challenging topics in robotics is simultaneous localization and mapping (SLAM) in the indoor environments. Due to the fact that Global Navigation Satellite Systems cannot be successfully used in such environments, different data sources are used for this purpose, among others light detection and ranging (LiDARs ), which have advanced from numerous other technologies. Other embedded sensors can be used along with LiDARs to improve SLAM accuracy, e.g. the ones available in the Inertial Measurement Units and wheel odometry sensors. Evaluation of different SLAM algorithms and possible hardware configurations in real environments is time consuming and expensive. In our study, we evaluate the accuracy of mapping and localization (based on Absolute Trajectory Error and Relative Pose Error). Our use case is a robot used for room decontamination. The results for a small room show that for our robot the best hardware configuration consists of three LiDARs 2D, IMU and wheel odometry sensors. On the other hand, for long hallways, a configuration with one LiDAR 3D sensor and IMU works better and more stable. We also described a general approach together with tools and procedures that can be used to find the best sensor setup in simulation.

3D SLAM based on NDT matching and ground constraints for ground robots in complex environments

PurposeIn large-scale environments and unstructured scenarios, the accuracy and robustness of traditional light detection and ranging (LiDAR) simultaneous localization and mapping (SLAM) algorithms are reduced, and the algorithms might even be completely ineffective. To overcome these problems, this study aims to propose a 3D LiDAR SLAM method for ground-based mobile robots, which uses a 3D LiDAR fusion inertial measurement unit (IMU) to establish an environment map and realize real-time localization.Design/methodology/approachFirst, we use a normal distributions transform (NDT) algorithm based on a local map with a corresponding motion prediction model for point cloud registration in the front-end. Next, point cloud features are tightly coupled with IMU angle constraints, ground constraints and gravity constraints for graph-based optimization in the back-end. Subsequently, the cumulative error is reduced by adding loop closure detection.FindingsThe algorithm is tested using a public data set containing indoor and outdoor scenarios. The results confirm that the proposed algorithm has high accuracy and robustness.Originality/valueTo improve the accuracy and robustness of SLAM, this method proposed in the paper introduced the NDT algorithm in the front-end and designed ground constraints and gravity constraints in the back-end. The proposed method has a satisfactory performance when applied to ground-based mobile robots in complex environments experiments.

A NEW INDOOR LIDAR-BASED MMS CHALLENGING COMPLEX ARCHITECTURAL ENVIRONMENTS

Abstract. The use of moving devices equipped with range- and image-based sensor, generically defined Mobile Mapping systems (MMS), have been quite a disruptive innovation in the development of Geomatics techniques for 3D surveying large indoor-outdoor spaces and offer multiple solutions. The recent expansion of portable devices in the form of trolleys, backpacks, handheld tools largely implements SLAM (Simultaneous Localization and Mapping) algorithms and technology based on both Lidar and/or visual solutions for answering to the positioning and the 3D reconstruction problems. The research on MMS is directed to improve both multi-sensor integration implementation and usability of systems in diversified use contexts and application fields. The aim of the presented research is the evaluation of the potential of the Swift system recently developed by FARO Technologies, that has been fine-tuned for regular and large extent interiors mapping (such as factories, hospitals, airports, offices). The work tries to preliminary investigate the data delivery and usability of the integrated system. This is based on three elements mounted on a sliding trolley moved by the operator walking: the ScanPlan profilometer working for the 2D SLAM mapping, the static TLS Focus S-series, and the smartphone managing the sensors operation and the acquisition progress. The evaluation strategy undertaken will be based on the global and local performance analysis related to the trajectory, the data accuracy, the metric content and consistency. Two test studies belonging to the 20th century. architecture are presented in a preliminary framework of evaluation and validation: a Liberty-style cinema and the Torino Esposizioni Hall B designed in ferrocement by pier Luigi Nervi.

Canopy density estimation in perennial horticulture crops using 3D spinning lidar SLAM

AbstractWe propose a novel, canopy density estimation solution using a three‐dimensional (3D) ray cloud representation for perennial horticultural crops at the field scale. To attain high spatial and temporal fidelity in field conditions, we propose the application of continuous‐time 3D SLAM (simultaneous localization and mapping) to a spinning lidar payload (AgScan3D) mounted on a moving farm vehicle. The AgScan3D data are processed through a Continuous‐Time SLAM algorithm into a globally registered 3D ray cloud. The global ray cloud is a canonical data format (a digital twin) from which we can compare vineyard snapshots over multiple times within a season and across seasons. Then, the vineyard rows are automatically extracted from the ray cloud and a novel density calculation is performed to estimate the maximum likelihood canopy densities of the vineyard. This combination of digital twinning, together with the accurate extraction of canopy structure information, allows entire vineyards to be analyzed and compared, across the growing season and from year to year. The proposed method is evaluated both in simulation and field experiments. Field experiments were performed at four sites, which varied in vineyard structure and vine management, over two growing seasons and 64 data collection campaigns, resulting in a total traversal of 160 km, 42.4 scanned hectares of vines with a combined total of approximately 93,000 scanned vines. Our experiments show canopy density repeatability of 3.8% (relative root mean square error) per vineyard panel, for acquisition speeds of 5–6 km/h, and under half the standard deviation in estimated densities when compared with an industry standard gap‐fraction based solution. The code and field data sets are available at https://github.com/csiro-robotics/agscan3d.

Real-Time Lidar-based Localization of Mobile Ground Robot

Paper discusses modern methods for localization of mobile ground robot in outdoor environment. Special attention is payed to lidar-based methods which allow to do precise simultaneous localization and mapping (SLAM) independently to lighting conditions. Modular approach to lidar-based localization is proposed. It consists of three modules: map reconstructor, map merger and corrector, localization module. As map reconstructor we use 3D Lidar-based Graph SLAM, state-of-the-art method LOAM and its modification with point cloud segmentation techniques A-LeGO-LOAM. As map merger and corrector we apply the technique of interactive map correction for 3D Graph SLAM. For localization module we compare different real-time approaches: NDT scan matching and Localization with scan batch processing. They do not allow to localize the robot with an unknown initial position. To overcome this problem, a stage of global Monte Carlo localization is proposed, which allows to generate a switching condition and a rough initial pose. Matrix of localization convergence speed is introduced to study influence of possible inaccuracies of initial pose estimation to the convergence of real-time localization approach. An important emphasis in the paper is made to the evaluation of the localization quality with relative and absolute metrics. Experimental results are obtained for own data from mobile ground robot ClearPath Husky equipped with Velodyne HDL-32E lidar. Proposed approach to robot localization achieves real-time performance for different hardware platforms in particular single-board computer Nvidia Jetson Xavier. It demonstrates possibility for integration of studied modular localization approach to onboard control systems of autonomous vehicles.

Fusing Hand Postures and Speech Recognition for Tasks Performed by an Integrated Leg–Arm Hexapod Robot

Hand postures and speech are convenient means of communication for humans and can be used in human–robot interaction. Based on structural and functional characteristics of our integrated leg-arm hexapod robot, to perform reconnaissance and rescue tasks in public security application, a method of linkage of movement and manipulation of robots is proposed based on the visual and auditory channels, and a system based on hand postures and speech recognition is described. The developed system contains: a speech module, hand posture module, fusion module, mechanical structure module, control module, path planning module and a 3D SLAM (Simultaneous Localization and Mapping) module. In this system, three modes, i.e., the hand posture mode, speech mode, and a combination of the hand posture and speech modes, are used in different situations. The hand posture mode is used for reconnaissance tasks, and the speech mode is used to query the path and control the movement and manipulation of the robot. The combination of the two modes can be used to avoid ambiguity during interaction. A semantic understanding-based task slot structure is developed by using the visual and auditory channels. In addition, a method of task planning based on answer-set programming is developed, and a system of network-based data interaction is designed to control movements of the robot using Chinese instructions remotely based on a wide area network. Experiments were carried out to verify the performance of the proposed system.

Integration and evaluation of SLAM-based backpack mobile mapping system

Mobile mapping is an efficient technology to acquire spatial data of the environment. As a supplement of vehicle-borne and air-borne methods, Backpack mobile mapping system (MMS) has a wide application prospect in indoor and underground space. High-precision positioning and attitude determination are the key to MMS. Usually, GNSS/INS integrated navigation system provides reliable pose information. However, in the GNSS-denied environments, there is no effective long-term positioning method. With the development of simultaneous localization and mapping (SLAM) algorithm, it provides a new solution for indoor mobile mapping. This paper develops a portable backpack mobile mapping system, which integrates multi-sensor such as LiDAR, IMU, GNSS and panoramic camera. The 3D laser SLAM algorithm is applied to the mobile mapping to realize the acquisition of geographic information data in various complex environments. The experimental results in typical indoor and outdoor scenes show that the system can achieve high-precision and efficient acquisition of 3D information, and the relative precision of point cloud is 2~4cm, which meets the requirements of scene mapping and reconstruction.

A Robust 2D-SLAM Technology With Environmental Variation Adaptability

Simultaneous localization and mapping (SLAM) in complicated indoor/outdoor unknown environments is challenging. With a demand on high mobility and high integrity intelligent robotics, it is desired that the SLAM system should be portable and possibly standalone. To carry out the pose estimation as well as the mapping without relying on the information from other sensors, such as image, inertial measurement unit, rotary encoder of ground vehicle and so on, a single 2D light detection and ranging (LiDAR) is considered in this paper. In order to fulfill a robust 2D SLAM technology in unknown environments, the principal component analysis (PCA) is utilized to evaluate LiDAR scan contours and to carry out a corridor detector. The corridor detector is further extended to achieve adaptive unstable points removal, mapping probability adjustment as well as loop closure. Based on an adaptive grid map segmentation scheme, the cumulative mapping errors can obviously be reduced and a precise 2D map can be eventually carried out. Many experiments are conducted to verify the proposed method. Finally, for comparison, this paper utilizes the scan data and ground truth provided by the Computer Science and Artificial Intelligence Laboratory (CSAIL) at the Massachusetts Institute of Technology (MIT), to verify the localization precision of the proposed algorithm. Experiment shows that from the scan data in the route up to about 350 m, the maximum error can be as low as about 20 cm.

Reliable Graphs for SLAM

Estimation-over-graphs (EoG) is a class of estimation problems that admit a natural graphical representation. Several key problems in robotics and sensor networks, including sensor network localization, synchronization over a group, and simultaneous localization and mapping (SLAM) fall into this category. We pursue two main goals in this work. First, we aim to characterize the impact of the graphical structure of SLAM and related problems on estimation reliability. We draw connections between several notions of graph connectivity and various properties of the underlying estimation problem. In particular, we establish results on the impact of the weighted number of spanning trees on the D-optimality criterion in 2D SLAM. These results enable agents to evaluate estimation reliability based only on the graphical representation of the EoG problem. We then use our findings and study the problem of designing sparse SLAM problems that lead to reliable maximum likelihood estimates through the synthesis of sparse graphs with the maximum weighted tree connectivity. Characterizing graphs with the maximum number of spanning trees is an open problem in general. To tackle this problem, we establish several new theoretical results, including the monotone log-submodularity of the weighted number of spanning trees. We exploit these structures and design a complementary greedy–convex pair of efficient approximation algorithms with provable guarantees. The proposed synthesis framework is applied to various forms of the measurement selection problem in resource-constrained SLAM. Our algorithms and theoretical findings are validated using random graphs, existing and new synthetic SLAM benchmarks, and publicly available real pose-graph SLAM datasets.

Efficient Laser-Based 3D SLAM for Coal Mine Rescue Robots

An accurate description of laneway space with self-localization is a key issue when coal mine rescue robots (CMRRs) perform post-disaster exploration and rescue missions. The 3D simultaneous localization and mapping (SLAM) is an effective but time-critical and highly challenging task in complex laneway scenarios, especially after disasters. In this paper, we propose a novel real-time 3D SLAM based on normally distributed transform (NDT) that employs pose graph optimization and loop closure to further improve mapping consistency. We innovatively extract floors and walls in the laneway as plane nodes to construct landmark constraints, in addition to applying pose nodes from the lidar odometry via NDT. A lightweight and effective loop detection method is conducted using odometry with an appearance-based approach to building a globally consistent map. The proposed method was evaluated on a public dataset, and field tests in an underground coal mine were performed. Results indicate that our algorithm can achieve lower computational complexity and drift, which can provide pose estimation and environment description for CMRRs to realize remote control assistance and automatic navigation in coal mine rescue missions.

IMU-Assisted 2D SLAM Method for Low-Texture and Dynamic Environments

Generally, the key issues of 2D LiDAR-based simultaneous localization and mapping (SLAM) for indoor application include data association (DA) and closed-loop detection. Particularly, a low-texture environment, which refers to no obvious changes between two consecutive scanning outputs, with moving objects existing in the environment will bring great challenges on DA and the closed-loop detection, and the accuracy and consistency of SLAM may be badly affected. There is not much literature that addresses this issue. In this paper, a mapping strategy is firstly exploited to improve the performance of the 2D SLAM in dynamic environments. Secondly, a fusion method which combines the IMU sensor with a 2D LiDAR, based on framework of extended Kalman Filter (EKF), is proposed to enhance the performance under low-texture environments. In the front-end of the proposed SLAM method, initial motion estimation is obtained from the output of EKF, and it can be taken as the initial pose for the scan matching problem. Then the scan matching problem can be optimized by the Levenberg–Marquardt (LM) algorithm. For the back-end optimization, a sparse pose adjustment (SPA) method is employed. To improve the accuracy, the grid map is updated with the bicubic interpolation method for derivative computing. With the improvements both in the DA process and the back-end optimization stage, the accuracy and consistency of SLAM results in low-texture environments is enhanced. Qualitative and quantitative experiments with open-loop and closed-loop cases have been conducted and the results are analyzed, confirming that the proposed method is effective in low-texture and dynamic indoor environments.

Distributed stereo vision‐based 6D localization and mapping for multi‐robot teams

AbstractJoint simultaneous localization and mapping (SLAM) constitutes the basis for cooperative action in multi‐robot teams. We designed a stereo vision‐based 6D SLAM system combining local and global methods to benefit from their particular advantages: (1) Decoupled local reference filters on each robot for real‐time, long‐term stable state estimation required for stabilization, control and fast obstacle avoidance; (2) Online graph optimization with a novel graph topology and intra‐ as well as inter‐robot loop closures through an improved submap matching method to provide global multi‐robot pose and map estimates; (3) Distribution of the processing of high‐frequency and high‐bandwidth measurements enabling the exchange of aggregated and thus compacted map data. As a result, we gain robustness with respect to communication losses between robots. We evaluated our improved map matcher on simulated and real‐world datasets and present our full system in five real‐world multi‐robot experiments in areas of up 3,000 m2 (bounding box), including visual robot detections and submap matches as loop‐closure constraints. Further, we demonstrate its application to autonomous multi‐robot exploration in a challenging rough‐terrain environment at a Moon‐analogue site located on a volcano.