CleanMAP is a novel framework that uses Multimodal Large Language Models to filter and refine crowdsourced HD map data, addressing a critical challenge in autonomous vehicle navigation.

• Implements a confidence-driven distillation approach to handle inconsistencies in crowdsourced map data
• Effectively manages common data quality issues including motion blur, lighting variations, and lane marking degradation
• Creates more reliable, real-time HD map updates for intelligent connected vehicles
• Bridges the gap between raw crowdsourced data and high-quality map information needed for autonomous systems

This research is vital for engineering reliable autonomous navigation systems that can operate safely in diverse, real-world conditions with continuously updated spatial awareness.

CleanMAP: Distilling Multimodal LLMs for Confidence-Driven Crowdsourced HD Map Updates