UniMapGen: A generative framework for large-scale map construction from multi-modal data

Yujian Yuan, Changjie Wu, Xinyuan Chang, Sijin Wang, Hang Zhang, Shiyi Liang, Shuang Zeng, Mu Xu

Abstract: Large-scale map construction is foundational for critical ap- plications such as autonomous driving and navigation sys- tems. Traditional large-scale map construction approaches mainly rely on costly and inefficient special data collection vehicles and labor-intensive annotation processes. While ex- isting satellite-based methods have demonstrated promising potential in enhancing the efficiency and coverage of map construction, they exhibit two major limitations: (1) inher- ent drawbacks of satellite data (e.g., occlusions, outdated- ness) and (2) inefficient vectorization from perception-based methods, resulting in discontinuous and rough roads that re- quire extensive post-processing. This paper presents a novel generative framework, UniMapGen, for large-scale map con- struction, offering three key innovations: (1) representing lane lines as discrete sequence and establishing an iterative strat- egy to generate more complete and smooth map vectors than traditional perception-based methods. (2) proposing a flexi- ble architecture that supports multi-modal inputs, enabling dynamic selection among BEV, PV, and text prompt, to over- come the drawbacks of satellite data. (3) developing a state update strategy for global continuity and consistency of the constructed large-scale map. UniMapGen achieves state-of- the-art performance on the OpenSatMap dataset. Further- more, UniMapGen can infer occluded roads and predict roads missing from dataset annotations.