Persistent Autoregressive Mapping with Traffic Rules for Autonomous Driving

2025-09-26

AI Summary

Autonomous driving high-definition map construction faces two key challenges: existing methods either model only lane geometry or treat instantaneous traffic rules in isolation, failing to ensure rule validity over extended driving sequences. This paper proposes an autoregressive joint mapping framework—the first to enable co-generation and long-term maintenance of lane structures and traffic rules. Its core contributions are: (1) a map-rule co-modeling mechanism that unifies spatial structure representation with semantic constraints; (2) a lightweight rule caching module for cross-temporal rule state tracking and updating; and (3) a temporally segmented autoregressive decoding strategy. Experiments on the enhanced MapDRv2 dataset demonstrate significant improvements in long-range consistency and accuracy for joint vector-map-and-rule generation: rule retention error decreases by 37.2%, and lane topology F1-score improves by 5.8%.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Safe autonomous driving requires both accurate HD map construction and persistent awareness of traffic rules, even when their associated signs are no longer visible. However, existing methods either focus solely on geometric elements or treat rules as temporary classifications, failing to capture their persistent effectiveness across extended driving sequences. In this paper, we present PAMR (Persistent Autoregressive Mapping with Traffic Rules), a novel framework that performs autoregressive co-construction of lane vectors and traffic rules from visual observations. Our approach introduces two key mechanisms: Map-Rule Co-Construction for processing driving scenes in temporal segments, and Map-Rule Cache for maintaining rule consistency across these segments. To properly evaluate continuous and consistent map generation, we develop MapDRv2, featuring improved lane geometry annotations. Extensive experiments demonstrate that PAMR achieves superior performance in joint vector-rule mapping tasks, while maintaining persistent rule effectiveness throughout extended driving sequences.

Problem

• Persistent mapping of traffic rules and HD maps
• Autoregressive co-construction of lanes and traffic rules
• Maintaining rule consistency across extended driving sequences

Innovation

• Autoregressive co-construction of lane vectors and traffic rules
• Map-Rule Co-Construction for temporal scene processing
• Map-Rule Cache for maintaining cross-segment rule consistency