Jiachen Zhang

Dynamic object detection from point clouds has been widely studied in recent years to achieve accurate and robust LiDAR odometry for autonomous driving. Satisfactory accuracy can be achieved by Dynamic object detection from point clouds has been widely studied in recent years to achieve accurate and robust LiDAR odometry for autonomous driving. Satisfactory accuracy can be achieved by detecting and removing the object points in the urban environment. However, it is still not clear how dynamic objects numerically affect the performance of LiDAR odometry. In addition, the existing solutions tended to directly remove the LiDAR features belonging to the dynamic object, which can lead to the degradation of the geometry constraints of the surrounding features. This paper aims to give answers to these problems by evaluating the effects of dynamic objects as well as reweighting both dynamic objects and static objects. Three factors affecting the performance of LiDAR odometry in highly dynamic scenarios, including the number , geometry distribution , and velocity of the dynamic objects , are first extensively studied using generated scenarios by leveraging real data. Instead of brutely removing the dynamic features, this paper proposes to adaptively assign weightings to the dynamic features. Then both the dynamic and static features are employed to estimate the LiDAR odometry. The effectiveness of the proposed method is verified using UrbanNav and nuScenes datasets that include numerous dynamic and static objects. To benefit the community, the implementation of the dynamic vehicle simulator and the code for the proposed method are both open-sourced.

Global Navigation Satellite System Real-time Kinematic (GNSS-RTK) is an indispensable source for the absolute positioning of autonomous systems. Unfortunately, the performance of the GNSS-RTK is significantly degraded in urban canyons, due to the notorious multipath and Non-Line-of-Sight (NLOS). On the contrary, LiDAR/inertial odometry (LIO) can provide locally accurate pose estimation in structured urban scenarios but is subjected to drift over time. Considering their complementarities, GNSS-RTK, adaptively integrated with LIO was proposed in this paper, aiming to realize continuous and accurate global positioning for autonomous systems in urban scenarios. As one of the main contributions, this paper proposes to identify the quality of the GNSS-RTK solution based on the point cloud map incrementally generated by LIO. A smaller mean elevation angle mask of the surrounding point cloud indicates a relatively open area thus the correspondent GNSS-RTK would be reliable. Global factor graph optimization is performed to fuse reliable GNSS-RTK and LIO. Evaluations are performed on datasets collected in typical urban canyons of Hong Kong. With the help of the proposed GNSS-RTK selection strategy, the performance of the GNSS-RTK/LIO integration was significantly improved with the absolute translation error reduced by more than 50%, compared with the conventional integration method where all the GNSS-RTK solutions are used.

Accurate positioning and mapping are significant for autonomous systems with navigation requirements. In this paper, a coarse-to-fine loosely-coupled (LC) LiDAR-inertial odometry (LC-LIO) that could explore the complementariness of LiDAR and inertial measurement unit (IMU) was proposed for the real-time and accurate pose estimation of a ground vehicle in urban environments. Different from the existing tightly-coupled (TC) LiDAR-inertial fusion schemes which directly use all the considered ranges and inertial measurements to optimize the vehicle pose, the method proposed in this paper performs loosely-couped integrated optimization with the high-frequency motion prediction, which was produced by IMU integration based on optimized results, employed as the initial guess of LiDAR odometry to approach the optimality of LiDAR scan-to-map registration. As one of the prominent contributions, thorough studies were conducted on the performance upper bound of the TC LiDAR-inertial fusion schemes and LC ones, respectively. Furthermore, the experimental verification was performed on the proposition that the proposed pipeline can fully relax the potential of the LiDAR measurements (centimeter-level ranging accuracy) in a coarse-to-fine way without being disturbed by the unexpected IMU bias. Moreover, an adaptive covariance estimation method employed during LC optimization was proposed to explain the uncertainty of LiDAR scan-to-map registration in dynamic scenarios. Furthermore, the effectiveness of the proposed system was validated on challenging real-world datasets. Meanwhile, the process that the proposed pipelines realized the coarse-to-fine LiDAR scan-to-map registration was presented in detail. Comparing with the existing state-of-the-art TC-LIO, the focus of this study would be placed on that the proposed LC-LIO work could achieve similar or better accuracy with a reduced computational expense.