As a variant of RRT (Rapidly-exploring Random Tree), RRT* is an important improvement of sampling-based algorithms. Although it can provide a feasible planning solution with higher quality, more resources on optimization are required, resulting a very slow convergence rate, which cannot satisfy the real-time requirements of most autonomous systems. In this paper, we propose a novel approach of RRT* in collaboration with a double-tree structure to separate the extension and optimization procedure. In our algorithm, the original RRT is employed to explore the unknown environment and to search feasible connecting areas, represented by piecewise lines. Different from the method of anytime RRT*, the RRT phase in our method is to find different homotopic paths during each iteration. Thereafter, a modified RRT* is used to obtain an optimal solution.Simulation results on two benchmarks demonstrate an improved performance of our approach in comparison with the original RRT* and its variants (e.g., DT-RRT). An additional evaluation on a real robotic vehicle further proves the eficiency of our approach.

Simple and efficient geometric controllers, like Pure-Pursuit, have been widely used in various types of autonomous vehicles to solve tracking problems. In this paper, we have developed a new pursuit method, named CF-Pursuit, which has been based on Pure-Pursuit but with certain differences. In CF-Pursuit, in order to reduce fitting errors, we used a clothoid C ¹ curve to replace the circle employed in Pure-Pursuit. This improvement to the fitting method helps the Pursuit method to decrease tracking errors. As regards the selection of look-ahead distance, we employed a fuzzy system to directly consider the path's curvature. There are three input variables in this fuzzy system, 6 mcurvature, 9 mcurvature and 12 mcurvature, calculated from the clothoid fit with the current position and the goal position on the defined path. A Sugeno fuzzy model was adapted to output a reasonable look-ahead distance using the experiences of human drivers as well as our own tests. Compared with some other geometric controllers, CF-Pursuit performs better in robustness, cross track errors and stability. The results from field tests have proven the CF-Pursuit is a practical and efficient geometric method for the path tracking problems of autonomous vehicles.

Autonomous driving is experiencing rapid development. A lane-level map is essential for autonomous driving, and a lane-level road network is a fundamental part of a lane-level map. A large amount of research has been performed on lane-level road network generation based on various on-board systems. However, there is a lack of analysis and summaries with regards to previous work. This paper presents an overview of lane-level road network generation techniques for the lane-level maps of autonomous vehicles with on-board systems, including the representation and generation of lane-level road networks. First, sensors for lane-level road network data collection are discussed. Then, an overview of the lane-level road geometry extraction methods and mathematical modeling of a lane-level road network is presented. The methodologies, advantages, limitations, and summaries of the two parts are analyzed individually. Next, the classic logic formats of a lane-level road network are discussed. Finally, the survey summarizes the results of the review.

Given the prominence of 3D sensors in recent years, 3D point clouds are worthy to be further investigated for environment perception and scene understanding. Learning accurate local and global contexts in point clouds is pivotal for semantic segmentation, and neighbor aggregation and Transformers have achieved notable success in local and global perception in point cloud analysis, respectively. Nevertheless, studying each independently is far from the optimal solution for comprehensive feature learning. To address this, we take a novel step towards investigating and integrating the structures of neighbor aggregation and Transformers. In this paper, we introduce Point Neighbor Aggregation with Transformer (PointNAT), a conceptually straightforward and effective approach aiming to enhance the performance of 3D point cloud semantic segmentation. PointNAT consists of a Neighbor Aggregation Block (NAB) for local perception, a Point Transformer Block (PTB) for global modeling, and a Hybrid Block to connect NABs and PTBs. NABs effectively learn complex local features at varying scales through an improved neighbor aggregation operation and a multi-head mechanism. PTBs efficiently perform global attention using a small set of learnable key points. Hybrid Blocks serve as high-and-low frequency signal hybridizers, merging the strengths of these two blocks by adaptively assigning hybrid weights to local and global contexts. We have evaluated the performance of PointNAT with state-of-the-art networks on several benchmarks, including S3DIS, Toronto3D, and SensatUrban. PointNAT achieves mIoU scores of 77.8%, 84.7%, and 65.2% in these three dataset, respectively. Furthermore, it outperforms the baseline approach PointNeXt by 3.0%, 1.3%, and 4.2%, respectively, while utilizing only 59.9% of the parameters and 15.2% of the FLOPs. The results demonstrate PointNAT’s superior ability in accurately segmenting large-scale 3D point cloud scenes, emphasizing its potential to advance environment perception and scene understanding. Our code is available at https://github.com/zeng-ziyin/PointNAT.

Real-time multichannel video analysis is significant for intelligent transportation. Considering that deep learning and correlation filter (CF) tracking are time-consuming, a vehicle tracking method for traffic scenes is presented based on a detection-based tracking (DBT) framework. To design the model of vehicle detection, the You Only Look Once (YOLO) model is used, and then, two constraints including object attribute information and intersection over union (IOU), are combined to modify the vehicle detection box. This approach improves vehicle detection precision. In the design of tracking model, a lightweight feature extraction network model for vehicle tracking is constructed. An inception module is used in this model to reduce the computational load and increase the adaptivity of the network scale. And a squeeze-and-excitation channel attention mechanism is adopted to enhance feature learning. Regarding the object tracking strategy, the method of combining a spatial constraint and filter template matching is adopted. The observation value and prediction value are matched and corrected to achieve stable tracking of the target. Based on the interference of occlusion in target tracking, the spatial position, moving direction, and historical feature correlation of the target are comprehensively employed to achieve continuous tracking of the target.

Vision-based lane-detection methods provide low-cost density information about roads for autonomous vehicles. In this paper, we propose a robust and efficient method to expand the application of these methods to cover low-speed environments. First, the reliable region near the vehicle is initialized and a series of rectangular detection regions are dynamically constructed along the road. Then, an improved symmetrical local threshold edge extraction is introduced to extract the edge points of the lane markings based on accurate marking width limitations. In order to meet real-time requirements, a novel Bresenham line voting space is proposed to improve the process of line segment detection. Combined with straight lines, polylines, and curves, the proposed geometric fitting method has the ability to adapt to various road shapes. Finally, different status vectors and Kalman filter transfer matrices are used to track the key points of the linear and nonlinear parts of the lane. The proposed method was tested on a public database and our autonomous platform. The experimental results show that the method is robust and efficient and can meet the real-time requirements of autonomous vehicles.