Integration of multiple sensor modalities and deep learning into Simultaneous Localization And Mapping (SLAM) systems are areas of significant interest in current research. Multi-modality is a stepping stone towards achieving robustness in challenging environments and interoperability of heterogeneous multi-robot systems with varying sensor setups. With maplab 2.0 , we provide a versatile open-source platform that facilitates developing, testing, and integrating new modules and features into a fully-fledged SLAM system. Through extensive experiments, we show that maplab 2.0 ’s accuracy is comparable to the state-of-the-art on the HILTI 2021 benchmark. Additionally, we showcase the flexibility of our system with three use cases: i) large-scale ( $\sim$ 10 $\text{km}$ ) multi-robot multi-session (23 missions) mapping, ii) integration of non-visual landmarks, and iii) incorporating a semantic object-based loop closure module into the mapping framework.
Robust and accurate pose estimation is crucial for many applications in mobile robotics. Extending visual Simultaneous Localization and Mapping (SLAM) with other modalities such as an inertial measurement unit (IMU) can boost robustness and accuracy. However, for a tight sensor fusion, accurate time synchronization of the sensors is often crucial. Changing exposure times, internal sensor filtering, multiple clock sources and unpredictable delays from operation system scheduling and data transfer can make sensor synchronization challenging. In this paper, we present VersaVIS, an Open Versatile Multi-Camera Visual-Inertial Sensor Suite aimed to be an efficient research platform for easy deployment, integration and extension for many mobile robotic applications. VersaVIS provides a complete, open-source hardware, firmware and software bundle to perform time synchronization of multiple cameras with an IMU featuring exposure compensation, host clock translation and independent and stereo camera triggering. The sensor suite supports a wide range of cameras and IMUs to match the requirements of the application. The synchronization accuracy of the framework is evaluated on multiple experiments achieving timing accuracy of less than . Furthermore, the applicability and versatility of the sensor suite is demonstrated in multiple applications including visual-inertial SLAM, multi-camera applications, multi-modal mapping, reconstruction and object based mapping.
23 February 2021 in IEEE Robotics and Automation Letters
To cope with the growing demand for transportation on the railway system, accurate, robust, and high-frequency positioning is required to enable a safe and efficient utilization of the existing railway infrastructure. As a basis for a localization system we propose a complete on-board mapping pipeline able to map robust meaningful landmarks, such as poles from power lines, in the vicinity of the vehicle. Such poles are good candidates for reliable and long term landmarks even through difficult weather conditions or seasonal changes. To address the challenges of motion blur and illumination changes in railway scenarios we employ a Dynamic Vision Sensor, a novel event-based camera. Using a sideways oriented on-board camera, poles appear as vertical lines. To map such lines in a real-time event stream, we introduce Hough 2 Map, a novel consecutive iterative event-based Hough transform framework capable of detecting, tracking, and triangulating close-by structures. We demonstrate the mapping reliability and accuracy of Hough 2 Map on real-world data in typical usage scenarios and evaluate using surveyed infrastructure ground truth maps. Hough 2 Map achieves a detection reliability of up to $92\,\%$ and a mapping root mean square error accuracy of 1.1518 m. 11 The code is available at https://github.com/ethz-asl/Hough2Map.
Increasing demand for rail transportation results in denser and more high-speed usage of the existing railway network, making new and more advanced vehicle safety systems necessary. Furthermore, high traveling speeds and the large weights of trains lead to long braking distances—all of which necessitates a Long-Range Obstacle Detection (LROD) system, capable of detecting humans and other objects more than 1000 m in advance. According to current research, only a few sensor modalities are capable of reaching this far and recording sufficiently accurate data to distinguish individual objects. The limitation of these sensors, such as a 1D-Light Detection and Ranging (LiDAR), is however a very narrow Field of View (FoV), making it necessary to use high-precision means of orienting to target them at possible areas of interest. To close this research gap, this paper presents a high-precision pointing mechanism, for the use in a future novel railway obstacle detection system, capable of targeting a 1D-LiDAR at humans or objects at the required distance. This approach addresses the challenges of a low target price, restricted access to high-precision machinery and equipment as well as unique requirements of our target application. By combining established elements from 3D printers and Computer Numerical Control (CNC) machines with a double-hinged lever system, simple and low-cost components are capable of precisely orienting an arbitrary sensor platform. The system’s actual pointing accuracy has been evaluated using a controlled, in-door, long-range experiment. The device was able to demonstrate a precision of 6.179 mdeg, which is at the limit of the measurable precision of the designed experiment.
Localization is an essential task for mobile autonomous robotic systems that want to use pre-existing maps or create new ones in the context of SLAM. Today, many robotic platforms are equipped with high-accuracy 3D LiDAR sensors, which allow a geometric mapping, and cameras able to provide semantic cues of the environment. Segment-based mapping and localization have been applied with great success to 3D point-cloud data, while semantic understanding has been shown to improve localization performance in vision based systems. In this paper we combine both modalities in SemSegMap, extending SegMap into a segment based mapping framework able to also leverage color and semantic data from the environment to improve localization accuracy and robustness. In particular, we present new segmentation and descriptor extraction processes. The segmentation process benefits from additional distance information from color and semantic class consistency resulting in more repeatable segments and more overlap after re-visiting a place. For the descriptor, a tight fusion approach in a deep-learned descriptor extraction network is performed leading to a higher descriptiveness for landmark matching. We demonstrate the advantages of this fusion on multiple simulated and real-world datasets and compare its performance to various baselines. We show that we are able to find 50.9 % more high-accuracy prior-less global localizations compared to SegMap on challenging datasets using very compact maps while also providing accurate full 6 DoF pose estimates in real-time.
Integration of multiple sensor modalities and deep learning into Simultaneous Localization And Mapping (SLAM) systems are areas of significant interest in current research. Multi-modality is a stepping stone towards achieving robustness in challenging environments and interoperability of heterogeneous multi-robot systems with varying sensor setups. With maplab 2.0, we provide a versatile open-source platform that facilitates developing, testing, and integrating new modules and features into a fully-fledged SLAM system. Through extensive experiments, we show that maplab 2.0's accuracy is comparable to the state-of-the-art on the HILTI 2021 benchmark. Additionally, we showcase the flexibility of our system with three use cases: i) large-scale (approx. 10 km) multi-robot multi-session (23 missions) mapping, ii) integration of non-visual landmarks, and iii) incorporating a semantic object-based loop closure module into the mapping framework. The code is available open-source at https://github.com/ethz-asl/maplab.