PriorDrive: Enhancing Online HD Mapping with Unified Vector Priors

TL;DR

PriorDrive addresses online HD map construction by integrating diverse vector priors into existing mapping models through a Unified Vector Encoder (UVE) and a Hybrid Prior Representation (HPQuery). It introduces a vector-specific pre-training regime and tailored integration strategies to combine SD maps, outdated HD maps, and online local priors, achieving robust, long-range map predictions. Across nuScenes, Argoverse 2, and OpenLane-V2, PriorDrive consistently improves vector map accuracy and topology reasoning with minimal overhead, demonstrating plug-and-play compatibility. The results suggest a scalable path to continuously improve HD maps by leveraging historical priors and diverse sources for safer autonomous navigation.

Abstract

High-Definition Maps (HD maps) are essential for the precise navigation and decision-making of autonomous vehicles, yet their creation and upkeep present significant cost and timeliness challenges. The online construction of HD maps using on-board sensors has emerged as a promising solution; however, these methods can be impeded by incomplete data due to occlusions and inclement weather, while their performance in distant regions remains unsatisfying. This paper proposes PriorDrive to address these limitations by directly harnessing the power of various vectorized prior maps, significantly enhancing the robustness and accuracy of online HD map construction. Our approach integrates a variety of prior maps uniformly, such as OpenStreetMap's Standard Definition Maps (SD maps), outdated HD maps from vendors, and locally constructed maps from historical vehicle data. To effectively integrate such prior information into online mapping models, we introduce a Hybrid Prior Representation (HPQuery) that standardizes the representation of diverse map elements. We further propose a Unified Vector Encoder (UVE), which employs fused prior embedding and a dual encoding mechanism to encode vector data. To improve the UVE's generalizability and performance, we propose a segment-level and point-level pre-training strategy that enables the UVE to learn the prior distribution of vector data. Through extensive testing on the nuScenes, Argoverse 2 and OpenLane-V2, we demonstrate that PriorDrive is highly compatible with various online mapping models and substantially improves map prediction capabilities. The integration of prior maps through PriorDrive offers a robust solution to the challenges of single-perception data, paving the way for more reliable autonomous vehicle navigation. Code is available at https://github.com/MIV-XJTU/PriorDrive.

Figures (11)

• Figure 1: Overview of PriorDrive. PriorDrive seamlessly integrates diverse vectorized prior maps into existing online mapping frameworks, making final predictions more complete and accurate than those generated without priors. The optimized predictions can be uploaded to cloud servers for other vehicles to download and use as prior maps.
• Figure 2: Structure of UVE. Fused prior embedding enhances vector representation. Intra-instance and inter-instance attention refine local features and capture global context effectively.
• Figure 3: Pre-trained Pipeline. The noise & mask generator creates segment-level and point-level noise or mask, and then reconstructs the entire map via UVE and MLP.
• Figure 4: Our Integration Method. For query-based models, we enable interaction between the prior features and queries at both instance-level and point-level, ensuring a more nuanced integration. For the other models, we integrate prior features directly with the BEV features.
• Figure 5: Comparison of qualitative results between those with and without unified prior maps.