Imagine being able to ask a computer to “show me the buildings next to the river,” and it instantly highlights them on a satellite map. Picture typing “create a neighborhood with shops, homes, and a park,” and seeing a realistic satellite-style image generated on the spot.
ARLIS, through its Intelligence and Security University Research Enterprise (INSURE) academic consortium, is funding a study with Washington University in St. Louis to make it easier for everyday users to explore and understand complex geospatial data using plain language.
Today, working with satellite imagery and geographic data often requires special tools, coding skills, or deep technical knowledge. That makes it hard for many people, including those in government, security, emergency response, and city planning, to use the powerful insights these maps can provide.
Led by Dr. Nathan Jacobs at Washington University and Triet Le, Research Scientist at ARLIS, the research team is developing two core models that will enable natural language interaction with geospatial content:
- Synthesis Model: This model creates realistic, satellite-style images based on written descriptions. It uses real map data, like roads and buildings from OpenStreetMap, to build custom scenes. This can help with things like planning new neighborhoods, training AI systems, or running what-if scenarios.
- Discriminative Model: This model helps users find things in satellite images by typing natural phrases like “show the buildings next to the river.” Instead of needing to know technical commands, users can just describe what they’re looking for, and the system finds and highlights it in the image. This makes it much easier to search wide areas quickly and accurately.
“This project is about giving people an easier way to work with maps and satellite images,” said Dr. Jacobs. “Instead of needing to learn complicated software or sort through endless data, they can just ask for what they need. This opens the door to a future where interacting with complex geospatial data is as easy as having a conversation.”
These tools build on recent breakthroughs in artificial intelligence and natural language processing, the same kind of technology behind chatbots like ChatGPT. By leveraging OpenStreetMap and transformer-based architectures, the team is designing their system to work with large datasets and complex locations that minimize training time and maximize usability for a broader range of analysts and decision-makers.
“People need tools that work the way they think and speak,” said Triet Le, Research Scientist at ARLIS. “This project directly supports our mission to make analytic systems faster to use, more adaptable to users’ needs, and more effective in high-stakes, dynamic environments.”
The project will run for a year, to create open-source software, testing the tools with real data, and sharing findings with the broader research community. In the long term, this kind of technology could help decision-makers respond faster, analyze smarter, and rely less on expert-only systems.