Online High-Definiton Map Construction with Transformer-based Model
I did this project as my Master's Thesis in RWTH-Aachen and BOSCH GmbH. This thesis explores the utilization of deep learning to generate High Definition (HD) maps for autonomous driving systems. HD maps are crucial for optimizing routes, improving environmental understanding, and enhancing vehicle safety. However, manual methods of generating HD maps encounter scalability issues, which can hinder their practical usage. The proposed methodologies in this thesis use data from onboard sensors (Cameras, LiDAR, Radar, Standard Definition Map) and integrate multiple modalities to enhance the generation of these maps. Additionally, these methodologies aim to fulfill the inference requirement, ensuring the map generation process is fast enough for deployment on a test vehicle. An optional extension of this thesis includes the integration of visibility for more detailed information in the HD map and adaptations of the network architecture. This research contributes to the development of more impactful and effective methods in generating HD maps, which are essential for the future of transportation.
