What is Map AI?

Edge based mapping

Map AI is a fully automated system that processes crowdsourced street-level imagery from the Hivemapper Network into a structured global street level dynamic map - everything from lanes to road rules to dynamic events like road construction, and police activity.

It uses a combination of computer vision and spatial AI pipeline to detect, classify, and geolocate assets like traffic signs, road markings, lane configurations, gas prices, and more. This system is designed to run on edge devices like the Bee enabling cost effective, scalable, high-frequency map updates without the need for traditional manual mapping.

The Map AI pipeline integrates multiple deep learning models that perform image segmentation, object detection, and geospatial positioning alignment. These models are trained to identify street level elements in varied environmental conditions and from different vehicle-mounted camera perspectives. Once detected, each feature (speed limits, toll prices, etc.) is associated with a precise geographic allowing for accurate placement on the map and relationship to other objects.

To continuously improve performance, Map AI outputs are routed through an optional human-in-the-loop system involving AI Trainers. These contributors validate and correct model predictions through structured review tasks, feeding that data back into the training pipeline. This hybrid approach ensures high-quality feature extraction even in edge cases or geographies with sparse data, and it supports rapid model iteration as new types of features are added to the detection set.