MapAgent: An Industrial-Grade Agentic Framework for City-scale Lane-level Map Generation

MapAgent is an industrial-grade framework designed for city-scale lane-level map generation, enhancing existing vectorized mapping methods by integrating explicit specification verification and constraint-aware reasoning. It employs a Judge-Planner-Worker loop to diagnose errors and generate corrective edits, effectively addressing challenges in complex scenes where visual evidence is insufficient. The framework has demonstrated significant performance improvements over existing baselines, achieving over 95% automation in lane-level map production for Baidu Maps across more than 360 cities, making it a practical solution for autonomous navigation infrastructure development.