Uni-Map: Unified Camera-LiDAR Perception for Robust HD Map Construction
Keywords: HD Map Construction, Sensor Failures; Out-of-Distribution Robustness
TL;DR: Unified Camera-LiDAR Perception for Robust HD Map Construction
Abstract: High-definition (HD) map construction methods play a vital role in providing precise and comprehensive static environmental information essential for autonomous driving systems. The primary sensors used are cameras and LiDAR, with input configurations varying among camera-only, LiDAR-only, or camera-LiDAR fusion based on cost-performance considerations, while fusion-based methods typically perform the best. However, current methods face two major issues: high costs due to separate training and deployment for each input configuration, and low robustness when sensors are missing or corrupted. To address these challenges, we propose the Unified Robust HD Map Construction Network (Uni-Map), a single model designed to perform well across all input configurations. Our approach designs a novel Mixture Stack Modality (MSM) training scheme, allowing the map decoder to learn effectively from camera, LiDAR, and fused features. We also introduce a projector module to align Bird's Eye View features from different modalities into a shared space, enhancing representation learning and overall model performance. During inference, our model utilizes a switching modality strategy to adapt seamlessly to any input configuration, ensuring compatibility across various modalities. To evaluate the robustness of HD map construction methods, we designed 13 different sensor corruption scenarios and conducted extensive experiments comparing Uni-Map with state-of-the-art methods. Experimental results show that Uni-Map outperforms previous methods by a significant margin across both normal and corrupted modalities, demonstrating superior performance and robustness. Notably, our unified model surpasses independently trained camera-only, LiDAR-only, and camera-LiDAR MapTR models with a gain of 4.6, 5.6, and 5.6 mAP on the nuScenes dataset, respectively. The source code will be released.
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3091
Loading