Abstract For autonomous driving, traversability analysis is one of the most basic and essential tasks. In this paper, we propose a novel LiDAR‐based terrain modeling approach, which could output stable, complete, accurate models results. As an inherent property environment that does not change with different view angles, our approach adopts multiframe information fusion strategy for modeling. Specifically, normal distributions transform mapping adopted to accurately model by fusing from...

LiDAR-based place recognition (LPR) is crucial for the navigation and localization of autonomous vehicles mobile robots in large-scale outdoor environments plays a critical role loop closure detection simultaneous mapping (SLAM). Existing LPR methods, which utilize 2D bird’s-eye view (BEV) projections 3D point clouds, achieve competitive performance efficiency accuracy. However, these methods often struggle with capturing global contextual information maintaining robustness to viewpoint...

For autonomous driving, it is important to obtain precise and high-frequency localization information. This paper proposes a novel method in which the Inertial Measurement Unit (IMU), wheel encoder, lidar odometry are utilized together estimate ego-motion of an unmanned ground vehicle. The IMU fused with encoder motion prior, involved three levels odometry: Firstly, we use information rectify intra-frame distortion scan, caused by vehicle’s own movement; secondly, provides better initial...

Range images are commonly used representations for 3D LiDAR point cloud in the field of autonomous driving. The approach generating a range image is generally regarded as standard approach. However, there do exist two different types approaches to image: In one approach, row defined laser ID, and other elevation angle. We named first Projection By Laser ID (PBID), second Elevation Angle (PBEA). Few previous works have paid attention difference these approaches. this work, we quantitatively...

For autonomous driving, drivable region detection is one of the most basic and essential tasks. In this paper, a novel LiDAR-based algorithm which could output complete, accurate stable result proposed. To promote completeness result, Bayesian generalized kernel inference bilateral filtering are utilized to estimate attribute those unobserved cells. ensure traversability, growing operator performed on normal vector map reflects slope terrain, thus closely related traversability vehicle....

Three-dimensional (3D) point cloud maps are widely used in autonomous driving scenarios. These usually generated by accumulating sequential LiDAR scans. When generating a map, moving objects (such as vehicles or pedestrians) will leave long trails on the assembled map. This is undesirable and reduces map quality. In this paper, we propose MapCleaner, an approach that can effectively remove from MapCleaner first estimates dense continuous terrain surface, based which then divided into noisy...

Abstract Robust localization is an essential capability for autonomous land vehicles. While a lot of work focused on structured environments, this article focuses navigation in off‐road environments. In the environment, due to lack salient features, scan matching algorithms tend degenerate. Therefore, first contribution paper propose reliable degeneracy indicator which can evaluate performance. The evaluated then integrated into factor graph optimization framework used both offline mapping...

High-precision 3D maps play an important role in autonomous driving. The current mapping system performs well most circumstances. However, it still encounters difficulties the case of Global Navigation Satellite System (GNSS) signal blockage, when surrounded by too many moving objects, or a featureless environment. In these challenging scenarios, either global navigation approach local will degenerate. With aim developing degeneracy-aware robust system, this paper analyzes possible...

Place recognition plays a crucial role in tasks such as loop closure detection and re-localization robotic navigation. As high-level representation within scenes, semantics enables models to effectively distinguish geometrically similar places, therefore enhancing their robustness environmental changes. Unlike most existing semantic-based LiDAR place (LPR) methods that adopt multi-stage relatively segregated data-processing storage pipeline, we propose novel end-to-end LPR model guided by...

3D occupancy grid map is an effective environment representation approach for autonomous vehicle. However, due to its heavy computational load, usage not as popular 2D counterpart. In this paper, we propose utilize range image accelerate the processing of map. We choose use stixels, in contrast commonly used voxels, primitives. By utilizing inherent relationship between stixel and image, expected depth each directly compared with value stored image. way, state can be updated without...

For autonomous driving, grid map is a popular tool to represent the environment, which can be used for obstacles crossing. To preserve more original information obtained from LIDAR and vertical structures in three dimensional (3D) this paper presents novel method generate 3D occupancy by integrating multi-frame point cloud. Firstly, we use of Inertial Measurement Unit (IMU) correct distortion cloud after delicate LIDAR/IMU hand-eye calibration procedure. Secondly, based on Gaussian process...

For autonomous driving, 3D mapping is an important task for accurate and robust navigation. This paper presents efficient method in urban scenarios. To improve the robustness, a degeneracy-aware factor graph constructed, which considers degradation of scan matching constraints prior observations. alleviate side effects caused by dynamic vehicles, likelihood-field-model-based multi-object detection tracking algorithm applied to filter objects. Extensive tests real-world datasets show that our...

Discriminating the traversability of terrains is a crucial task for autonomous driving in off-road environments. However, it challenging due to diverse, ambiguous, and platform-specific nature traversability. In this paper, we propose novel self-supervised terrain learning framework, utilizing contrastive label disambiguation mechanism. Firstly, weakly labeled training samples with pseudo labels are automatically generated by projecting actual experiences onto models constructed real time....

In autonomous driving scenarios, the point cloud generated by LiDAR is usually considered as an accurate but sparse representation. order to enrich cloud, this paper proposes a new technique that combines spatial adjacent frames and temporal frames. To eliminate “ghost” artifacts caused moving objects, identification algorithm introduced employs comparison between range images. Experiments are performed on publicly available Semantic KITTI dataset. Experimental results show proposed method...

For autonomous vehicle, ground segmentation is one of the most basic and essential capabilities. In this paper, to further improve performance LiDAR-based approaches, other than commonly used `oc-cupancy' information point clouds, we try utilize `free' naturally contained in any clouds. We choose use stixel as processing primitives. By utilizing intrinsic relationship between 3D voxel grid 2D range image, can efficiently calculate features that contain both occupancy free information. Using...