LDMapNet-U: An End-to-End System for City-Scale Lane-Level Map Updating

TL;DR

LDMapNet-U introduces an end-to-end system for city-scale lane-level map updating that unifies vectorized map generation and change detection by leveraging historical maps as priors and current BEV observations. The key innovations are the Prior-Map Encoding (PME) module, which encodes historical map priors, and the Instance Change Prediction (ICP) module, which learns associations between predicted and historical lane instances to generate change labels within a single framework. Extensive experiments on the LD-U and LD-U-L datasets show consistent gains over strong baselines in both map construction and change detection, and the approach has been deployed in production at Baidu Maps to support weekly updates for hundreds of cities. The work demonstrates a practical path toward automated, scalable, and high-quality lane-level map maintenance with substantial cost and time savings for autonomous driving applications.

Abstract

An up-to-date city-scale lane-level map is an indispensable infrastructure and a key enabling technology for ensuring the safety and user experience of autonomous driving systems. In industrial scenarios, reliance on manual annotation for map updates creates a critical bottleneck. Lane-level updates require precise change information and must ensure consistency with adjacent data while adhering to strict standards. Traditional methods utilize a three-stage approach-construction, change detection, and updating-which often necessitates manual verification due to accuracy limitations. This results in labor-intensive processes and hampers timely updates. To address these challenges, we propose LDMapNet-U, which implements a new end-to-end paradigm for city-scale lane-level map updating. By reconceptualizing the update task as an end-to-end map generation process grounded in historical map data, we introduce a paradigm shift in map updating that simultaneously generates vectorized maps and change information. To achieve this, a Prior-Map Encoding (PME) module is introduced to effectively encode historical maps, serving as a critical reference for detecting changes. Additionally, we incorporate a novel Instance Change Prediction (ICP) module that learns to predict associations with historical maps. Consequently, LDMapNet-U simultaneously achieves vectorized map element generation and change detection. To demonstrate the superiority and effectiveness of LDMapNet-U, extensive experiments are conducted using large-scale real-world datasets. In addition, LDMapNet-U has been successfully deployed in production at Baidu Maps since April 2024, supporting map updating for over 360 cities and significantly shortening the update cycle from quarterly to weekly. The updated maps serve hundreds of millions of users and are integrated into the autonomous driving systems of several leading vehicle companies.

Figures (4)

• Figure 1: LDMapNet-U presents an end-to-end automated industrial-grade approach for lane-level map updating. Our proposed method uses BEV images and historical maps as inputs, achieving end-to-end prediction of vectorized results and change labels through the innovative design of the Prior-Map Encoding (PME) and Instance Change Prediction (ICP) modules. With these advancements, LDMapNet-U significantly enhances update efficiency and quality.
• Figure 2: (a) Overall architecture of LDMapNet-U. (b) Our proposed Prior-Map Encoding (PME) module. (c) Our proposed Instance Change Prediction (ICP) module. Please refer to Section \ref{['section:dumapnet']} for detailed illustrations.
• Figure 3: Different fusion methods for historical map embeddings. (a) w/o Fusion indicates that the historical map embeddings are not fused with the network before map association prediction. (b) Decoder Query CA refers to the fusion of historical map embeddings with the decoder queries by leveraging multi-head cross-attention. (c) BEV Feature CA refers to the fusion of historical map embeddings with the BEV features by leveraging multi-head cross-attention.
• Figure 4: Qualitative comparisons of our model with several state-of-the-art models. Instances addition, instance deletion, style change, and no change are highlighted in red, yellow, green, and light purple, respectively. Best viewed in color.