Graph Extraction and Planning for Long-Horizon Semantic Mapping and Information Gathering from Aerial Imagery

Extracting actionable graphs from aerial imagery to guide long-horizon semantic mapping and informative UAV data collection.

From Imagery to Actionable Maps

Aerial imagery offers broad coverage of large environments, but turning pixels into useful decisions remains challenging. This project develops methods to extract graph-based representations from satellite and aerial imagery. The resulting graphs compactly capture traversable structure, meaningful landmarks, and relationships between regions so that an autonomous system can reason about a scene at mission scale.

Long-Horizon Semantic Mapping

We use these structured representations to plan where a vehicle should observe next as it builds or refines a semantic map. Rather than treating every possible location independently, the planner can reason over connected regions, likely information value, mission constraints, and the effects of decisions made far into the future.

Informative Aerial Data Collection

Our goal is to enable aerial systems to gather the most useful information with limited flight time and sensing resources. We are investigating graph extraction, semantic mapping, and informative planning together so that the representation learned from imagery directly supports practical data-collection missions in complex, changing environments.