The field of autonomous driving is rapidly progressing, with a focus on improving safety, efficiency, and accuracy. Recent developments have centered around enhancing the precision of trajectory data for vulnerable road users, such as pedestrians and cyclists, and advancing lane detection algorithms to achieve robust generalization across diverse environments. Another key area of research is the development of vision-centric HD mapping systems, which leverage vehicle-to-infrastructure communications and roadside cameras to expand the map perception range. Additionally, innovative methods for 3D object detection, traffic light recognition, and lane topology reasoning are being explored to address the complexities of autonomous driving. Notable papers in this area include:

• A high-resolution trajectory dataset for high-density vulnerable road users, which provides a comprehensive representation of VRU behavioral characteristics.
• A method that detects keypoints of lanes and predicts sequential connections between them to construct complete 3D lanes, demonstrating superior F1 scores and generalization capacity.
• A video-based end-to-end neural network for traffic light recognition, which achieves state-of-the-art performance while maintaining real-time processing capabilities.
• An intrinsic-feature-guided 3D object detection method, which extracts intrinsic features from templates and provides rich structural information for foreground objects.
• A novel 3D object detector that focuses on detecting objects' closer surfaces to the LiDAR sensor, enhancing robustness and cross-domain performance.