Current perception models in autonomous driving heavily rely on large-scale labelled 3D data, which is both costly and time-consuming to annotate. This work proposes a solution to reduce the dependence on labelled 3D training data by leveraging pre-training on large-scale unlabeled outdoor LiDAR point clouds using masked autoencoders (MAE). While existing masked point autoencoding methods mainly focus on small-scale indoor point clouds or pillar-based large-scale outdoor LiDAR data, our approach introduces a new self-supervised masked occupancy pre-training method called Occupancy-MAE, specifically designed for voxel-based large-scale outdoor LiDAR point clouds. Occupancy-MAE takes advantage of the gradually sparse voxel occupancy structure of outdoor LiDAR point clouds and incorporates a range-aware random masking strategy and a pretext task of occupancy prediction. By randomly masking voxels based on their distance to the LiDAR and predicting the masked occupancy structure of the entire 3D surrounding scene, Occupancy-MAE encourages the extraction of high-level semantic information to reconstruct the masked voxel using only a small number of visible voxels. Extensive experiments demonstrate the effectiveness of Occupancy-MAE across several downstream tasks. For 3D object detection, Occupancy-MAE reduces the labelled data required for car detection on the KITTI dataset by half and improves small object detection by approximately 2% in AP on the Waymo dataset. For 3D semantic segmentation, Occupancy-MAE outperforms training from scratch by around 2% in mIoU. For multi-object tracking, Occupancy-MAE enhances training from scratch by approximately 1% in terms of AMOTA and AMOTP. Codes are publicly available at https://github.com/chaytonmin/Occupancy-MAE .

Multi-camera 3D perception has emerged as a prominent research field in autonomous driving, offering a viable and cost-effective alternative to LiDAR-based solutions. The existing multi-camera algorithms primarily rely on monocular 2D pre-training. However, the monocular 2D pre-training overlooks the spatial and temporal correlations among the multi-camera system. To address this limitation, we propose the first multi-camera unified pre-training framework, called UniScene, which involves initially reconstructing the 3D scene as the foundational stage and subsequently fine-tuning the model on downstream tasks. Specifically, we employ Occupancy as the general representation for the 3D scene, enabling the model to grasp geometric priors of the surrounding world through pre-training. A significant benefit of UniScene is its capability to utilize a considerable volume of unlabeled image-LiDAR pairs for pre-training purposes. The proposed multi-camera unified pre-training framework demonstrates promising results in key tasks such as multi-camera 3D object detection and surrounding semantic scene completion. When compared to monocular pre-training methods, UniScene shows a significant improvement of about 2.0% in mAP and 2.0% in NDS for multi-camera 3D object detection, as well as a 3% increase in mIoU for surrounding semantic scene completion. By adopting our unified pre-training method, a 25% reduction in 3D training annotation costs can be achieved, offering significant practical value for the implementation of real-world autonomous driving.

Autonomous vehicles (AVs) heavily depend on machine learning-based algorithms for the purpose of environmental perception. However, extensively utilized deep learning-based visual perception methods are susceptible to adversarial visual sensor attacks, thereby introducing latent risks to vehicle driving safety. In this paper, we first investigate a novel adversarial patch attack method for the DNN-based visual object detection method widely used in AVs. The proposed adversarial attack strategy contains an evasion attack mode and a misclassified attack mode. For each mode, the corresponding adversarial patch generation method is designed by a dedicated optimization process. In addition, we propose an adversarial attack defense method based on texture feature and local denoising. According to the texture features, the method can detect the locations of the possible adversarial patches in the image, and then the image is denoised locally from the detection results. The attack effects of the generated adversarial patch are validated on the KITTI dataset and in real driving scenes, respectively. The results show that the average accuracy of car and pedestrian detection under evasion attacks decreases by 56% and 68%, respectively, while under misclassification attacks, the average accuracy decreases by 35% and 85% for car and pedestrian, respectively. The proposed method can effectively mislead the target detection model which seriously affects the perception module. This paper conduct a defense experiment using data generated by the adversarial patch, resulting in a 92% increase in mean average precision for detecting cars and pedestrians, with detection time reduced to 17% of its original value. The experimental results demonstrate the defensive effectiveness and efficiency of the proposed defense method.

This paper addresses the lateral and longitudinal control problem for vehicle platoons subject to multiple communication delays. The essential lateral control problem and different delays in search-based routes of the vehicular ad-hoc networks (VANET) are introduced to vehicle platoon control. We first design an objective-based decoupled lateral and longitudinal control framework for vehicle platoons. With the Frenet coordinate system, a robust controller for lateral control is established to ensure each follower vehicle tracks the desired path and remain within the platoon. For the longitudinal control, the longitudinal error dynamic of vehicle- $ i$ with a specific interacted vehicle- $ j$ is modeled. Furthermore, with an unknown interaction topology, the error dynamic is augmented as multiple low-dimensional subsystems instead of an high-dimensional platoon system. To accommodate the communication delays, we design a distributed robust controller based on the Lyapunov-Krasovskii function. In addition, a delay-tolerant controller design theorem is obtained with linear matrix inequalities (LMIs). Simulations are conducted to demonstrate the effectiveness and improvement of the proposed controllers.

One of the most important tasks in the advanced transportation systems is road extraction. Unlike single geospatial objects extraction, extracting road region from high-resolution remote sensing imagery is challenging due to complicated background and rural road networks that have heterogeneous forms with low interclass and high intraclass differences. In the past few years, convolutional neural networks (CNNs) have achieved milestones in road extraction. In particular, the networks based on U -shaped architecture and skip-connections have been widely used in a variety of road extraction tasks. However, although CNN has achieved excellent performance, global and long-range semantic information interaction cannot be learned well due to the locality of convolution operation. In this paper, we propose Swin Transformer Unet which combined U-shaped architecture with hierarchical Swin Transformer with shifted windows. Moreover, we improve the loss function to make it more suitable for our road extraction task. The techniques of data augmentation and data preprocessing are used in order to get better performance. The Massachusetts roads dataset is chosen as the dataset to carry out the experiment of road extraction, and the result shows that this model outperforms other U-shaped networks of road extraction from remote sensing images.

The localization of autonomous vehicles mainly depends on GPS-based methods in large-scale outdoor scenarios. Unfortunately, GPS is vulnerable and not reliable in some regions, such as urban canyons. Odometry is taken as a complementary method for positioning in GPS-limited areas, but it suffers from accumulated errors, which limits its application. Topological map is a concise and low-cost representation of real world environment, which can be deemed as one of the information sources for localization. To realize accurate localization in the GPS-limited and large-scale scenes for autonomous vehicles, a generalized localization framework based on the topological map and odometry is proposed in this paper. The odometry trajectory can be automatically aligned with the correct paths of the map through matching without the initial global position. Under this framework, odometry is deemed as the kinematic model while topological map is regarded as a global sensor that is utilized to correct the drift errors of odometry. As an example, the integration of visual odometry and topological map is used to validate the effectiveness of the proposed method on KITTI benchmark. Experimental results demonstrate that the proposed method can effectively constrain the drift errors of odometry, and it achieves state-of-the-art localization performance compared with existing map-aided methods.