1 September 2024 in arXiv
In autonomous driving, there is growing interest in end-to-end online vectorized map perception in bird's-eye-view (BEV) space, with an expectation that it could replace traditional high-cost offline high-definition (HD) maps. However, the accuracy and robustness of these methods can be easily compromised in challenging conditions, such as occlusion or adverse weather, when relying only on onboard sensors. In this paper, we propose HRMapNet, leveraging a low-cost Historical Rasterized Map to enhance online vectorized map perception. The historical rasterized map can be easily constructed from past predicted vectorized results and provides valuable complementary information. To fully exploit a historical map, we propose two novel modules to enhance BEV features and map element queries. For BEV features, we employ a feature aggregation module to encode features from both onboard images and the historical map. For map element queries, we design a query initialization module to endow queries with priors from the historical map. The two modules contribute to leveraging map information in online perception. Our HRMapNet can be integrated with most online vectorized map perception methods. We integrate it in two state-of-the-art methods, significantly improving their performance on both the nuScenes and Argoverse 2 datasets. The source code is released at https://github.com/HXMap/HRMapNet.
13 February 2026 in IEEE Robotics and Automation Letters
Lane-level maps are increasingly preferred over Standard-Definition (SD) and High-Definition (HD) maps, offering a better trade-off among detail richness, coverage breadth, and data freshness. However, constructing city-scale lane-level maps remains time-consuming and labor-intensive. To address these challenges, this letter presents an automated mapping framework that fuses crowdsourced trajectories with satellite imagery to enable scalable and accurate map generation. Our approach begins by mining billions of trajectories to extract the geometric and topological structure of road networks. To enrich feature representation, we introduce an effective multimodal fusion mechanism that integrates trajectory data with satellite images, leveraging the complementary strengths of both modalities. Furthermore, a spatiotemporal prior-fusion decoding strategy is proposed to enhance the accuracy and consistency of vectorized map element perception. Finally, a globally consistent, vectorized lane-level map is generated by synthesizing the geometric and semantic output of the perception pipeline. The proposed method achieves the state-of-the-art mean average precision (mAP) on both a large-scale self-curated dataset and Argoverse 2. Having processed over one million kilometers of road networks, the system demonstrates significant scalability and practicality for real-world lane-level mapping applications.