Online Navigation Refinement: Achieving Lane-Level Guidance by Associating Standard-Definition and Online Perception Maps

TL;DR

This paper tackles lane-level navigation by bridging static SD maps with dynamic OP perception through map-to-map association. It introduces the Online Map Association Dataset (OMA) and a transformer-based MAT model featuring Path-Aware and Spatial Attention to align heterogeneous SD and OP maps in real time. The proposed Navigation Refinement P-R (NR-P-R) metric evaluates both geometric and semantic alignment between maps, enabling robust benchmarking. Empirical results show state-of-the-art NR-F1 performance with low latency and strong generalization across diverse OP generators, highlighting practical potential for low-cost, up-to-date lane-level navigation.

Abstract

Lane-level navigation is critical for geographic information systems and navigation-based tasks, offering finer-grained guidance than road-level navigation by standard definition (SD) maps. However, it currently relies on expansive global HD maps that cannot adapt to dynamic road conditions. Recently, online perception (OP) maps have become research hotspots, providing real-time geometry as an alternative, but lack the global topology needed for navigation. To address these issues, Online Navigation Refinement (ONR), a new mission is introduced that refines SD-map-based road-level routes into accurate lane-level navigation by associating SD maps with OP maps. The map-to-map association to handle many-to-one lane-to-road mappings under two key challenges: (1) no public dataset provides lane-to-road correspondences; (2) severe misalignment from spatial fluctuations, semantic disparities, and OP map noise invalidates traditional map matching. For these challenges, We contribute: (1) Online map association dataset (OMA), the first ONR benchmark with 30K scenarios and 2.6M annotated lane vectors; (2) MAT, a transformer with path-aware attention to aligns topology despite spatial fluctuations and semantic disparities and spatial attention for integrates noisy OP features via global context; and (3) NR P-R, a metric evaluating geometric and semantic alignment. Experiments show that MAT outperforms existing methods at 34 ms latency, enabling low-cost and up-to-date lane-level navigation.

Figures (17)

• Figure 1: Motivation for online navigation refinement and map association. The roads and lanes with the same color indicate that they are interconnected: (a) Standard Definition (SD) maps offer only road-level navigation without lane details. (b) Online Perception (OP) maps offer lane-specific details, yet they are not connected to SD maps and cannot identify the correct lane. (c) Associating SD and OP maps enables lane selection, achieving online navigation refinement and lane-level navigation.
• Figure 2: (a) The schema of SD map input: Road $\mathcal{R}$. (b) The schema of OP map input: Centerline $\mathcal{L}$ and boundary $B$. (c) The objective in our task: Mapping function $f$.
• Figure 3: (a) The visualization of SD map and GT OP map with association annotations of Boston in nuScenes. The same color implies an associative pair. (b) The visualization of OMA Train/Val set. (c) The visualization of OMA Test set.
• Figure 4: Example of TP, FP and FN for evaluate Navigation Refinement Precision-Recall.
• Figure 5: Overview of Map Association Transformer (MAT). The framework processes vectorized roads in SD map and centerlines/boundaries in OP map through $N$ stacked layers containing Spatial Attention (for global context via curve-based serialization) and Path-Aware Attention (for topological alignment via path indexing). The Association Head then aggregates road features and calculates association probabilities with centerline tokens to generate the final navigation refinement result.