MapTCL: Temporal Consistency Learning via Bidirectional Alignment for Vectorized HD Map Construction

2026-08-05

AI Summary

This work addresses the challenges of online high-definition map construction in dynamic urban environments, where moving objects and occlusions introduce geometric noise and temporal jitter. To enhance temporal consistency, the authors propose a plug-and-play optimization framework that introduces, for the first time, Bidirectional Vector Consistency Learning (BVCL) and Raster Map Consistency Learning (RCL). By modeling geometric-semantic alignment between vectorized maps across current and historical frames and enforcing consistency in bird’s-eye-view (BEV) features, the method jointly improves the temporal stability of online vectorized maps. Evaluated on nuScenes and Argoverse 2, the approach achieves performance gains of +3.7 mAP / +2.8 C-mAP and +3.1 mAP / +2.5 C-mAP, respectively, without incurring additional inference overhead.

Abstract

Constructing reliable online HD maps remains challenging in dynamic urban environments due to moving objects and occlusions. While recent works employ feature-level temporal fusion to address this, they rely solely on per-frame ground truth supervision. Consequently, they lack an explicit objective to directly penalize the geometric noise and temporal jitter between consecutive online HD maps. To address this, we propose MapTCL, an auxiliary training strategy that formulates temporal consistency loss between current and past frames via bidirectional alignment. Specifically, Bidirectional Vector Consistency Learning (BVCL) models the geometric and semantic discrepancies between associated past and current vector instances as an auxiliary loss. We also employ Raster map Consistency Learning (RCL) as an additional loss to stabilize dense BEV features. By jointly training with these dual losses, MapTCL improves the temporal stability of generated HD maps. Extensive experiments on two standard benchmarks demonstrate the effectiveness of our approach. As a versatile plug-and-play module, MapTCL consistently enhances existing baseline models, achieving gains of +3.7 mAP & +2.8 C-mAP on nuScenes and +3.1 mAP & +2.5 C-mAP on Argoverse 2 without additional inference overhead.

Problem

• temporal consistency
• HD map construction
• geometric noise
• temporal jitter
• dynamic urban environments

Innovation

• Temporal Consistency Learning
• Bidirectional Alignment
• Vectorized HD Map
• Auxiliary Training Strategy
• BEV Feature Stabilization