Uni-PrevPredMap: Extending PrevPredMap to a Unified Framework of Prior-Informed Modeling for Online Vectorized HD Map Construction

TL;DR

This work tackles the challenge of online vectorized HD map construction under safety-critical constraints by unifying two complementary priors: temporal perception buffers and simulated outdated HD maps. It introduces Uni-PrevPredMap, featuring a tile-indexed 3D vectorized global map processor for efficient 3D prior updates and a tri-mode optimization paradigm that maintains robustness across non-prior, temporal-prior, and temporal-map-fusion-prior scenarios, while integrating simulated outdated maps to reduce reliance on precise priors. Empirical results on nuScenes and Argoverse2 show state-of-the-art performance in map-free settings and strong gains when simulated outdated maps are provided, demonstrating effective, error-resilient prior fusion and the complementary nature of the two priors. The approach promises improved robustness and safety for autonomous driving by enabling reliable, 3D-aware, online HD map construction in diverse environments and map-fidelity conditions.

Abstract

Safety constitutes a foundational imperative for autonomous driving systems, necessitating the maximal incorporation of accessible external prior information. This study establishes that temporal perception buffers and cost-efficient maps inherently form complementary prior sources for online vectorized high-definition (HD) map construction. We present Uni-PrevPredMap, a unified prior-informed framework that systematically integrates two synergistic information sources: previous predictions and simulated outdated HD maps. The framework introduces two core innovations: a tile-indexed 3D vectorized global map processor enabling efficient refreshment, storage, and retrieval of 3D vectorized priors; a tri-mode operational optimization paradigm ensuring consistency across non-prior, temporal-prior, and temporal-map-fusion-prior scenarios while mitigating reliance on idealized map fidelity assumptions. Uni-PrevPredMap achieves state-of-the-art performance in map-absent scenarios across established online vectorized HD map construction benchmarks. When provided with simulated outdated HD maps, the framework exhibits robust capabilities in error-resilient prior fusion, empirically confirming the synergistic complementarity between previous predictions and simulated outdated HD maps. Code will be available at https://github.com/pnnnnnnn/Uni-PrevPredMap.

Figures (5)

• Figure 1: Black arrows delineate the non-prior-based baseline pipeline for online vectorized HD map construction, while green and orange pathways respectively denote temporal perception buffers integration and alternative cost-efficient maps incorporation. The proposed Uni-PrevPredMap systematically unifies these complementary prior sources.
• Figure 2: The overall architecture of the proposed Uni-PrevPredMap. The upper dashed box depicts the tile-indexed 3D vectorized global map processor, implementing efficient 3D prior data refreshment, storage, and retrieval. The lower dashed box corresponds to the tri-mode operational optimization paradigm, where dash-dotted arrows indicate independent stochastic selections of temporal and map priors.
• Figure 3: Visualization of adjacency selection during retrieval. The central grid indicates the target tile with indices $(i_t, j_t)$. Orange and green star markers denote distinct vehicle UTM coordinate positions within the target tile, with corresponding shaded grids indicating respective adjacent tiles.
• Figure 4: Comparison between simulated outdated HD maps (orange dashed lines) and ground truth annotations (green solid lines). Columns share identical random seeds to ensure reproducibility and fair comparison.
• Figure 5: Prediction comparison of Uni-PrevPredMap in three modes: Uni-PrevPredMap$^1$, Uni-PrevPredMap$^2$, and Uni-PrevPredMap$^3$ denote non-prior, temporal-prior and temporal-map-fusion-prior modes, respectively. Corresponding priors are illustrated to demonstrate their influence. Green, orange and blue lines represent road boundaries, lane dividers and pedestrian crossings, respectively.