Researchers have turned pilot-to-pilot radio calls around airports with no control tower into data that drone aircraft can use to steer clear of other planes.
Interest is growing in using small, regional airports as bases for autonomous flying to expand the reach of drone package delivery and other services. They make up nine out of 10 airfields in the U.S. and around the world, but they operate with no active traffic control towers. To fly safely, pilots coordinate with each other over the radio.
Now Georgia Tech researchers have turned those radio calls into data that autonomous aircraft also can use to avoid collisions.
The team found they could “listen” to the pilots’ transmissions, infer their intent, and use that data to predict flight paths with greater accuracy than the current state-of-the-art approach. They reduced the average error in trajectory predictions by more than half — from nearly a kilometer to about 400 meters.
“This is how humans have operated safely for decades around these airports. So, if we start flying robots here, they should operate in the same way,” said Sundhar Vinodh Sangeetha, a robotics Ph.D. student and first author of the study. “We humans shouldn't have to change the way we act because there's a robot operating around us. That was a big motivation.”
Sangeetha presented the work this month at the IEEE International Conference on Robotics and Automation, the world’s largest gathering of robotics researchers.
While the goal was to allow robot vehicles to understand human pilots’ flight paths, the Georgia Tech approach also could be developed into a backup safety system for non-towered airports.
“The number of accidents is increasing at these airports. If you have a solution like this where you have a computer that's monitoring radio traffic and what aircraft are doing, you can potentially warn pilots before accidents happen,” Sangeetha said.
Georgia Tech researchers used information in pilot-to-pilot radio transmissions to reduce the average error in predicting a pilot's intended flight path, or goal, by more than half — from nearly a kilometer to about 400 meters. In this illustration, dots represent predictions using current state-of-the-art methods; diamonds show predictions that incorporate radio calls. (Illustration Courtesy: Sundhar Vinodh Sangeetha)
To validate their approach, Sangeetha and his colleagues used flight data and radio calls from a non-towered airport in Pennsylvania. Their method used a speech-to-text model to transcribe the calls and then a modified large-language model to interpret the pilots’ intentions. They paired that data with existing prediction systems that use information about where the human-piloted craft have been to estimate where they’re going — or what the researchers called their goal.
“We're thinking about this in terms of mixed human-autonomy settings, so one day — and I hope this is soon — we introduce autonomous aircraft in the same space as human-piloted aircraft,” said Sarah Li, assistant professor in the Daniel Guggenheim School of Aerospace Engineering and a co-author of the study.
Li’s research focuses in part on aircraft coordination and trajectory planning, and she sees this study as the first step to also helping autonomous aircraft communicate with the humans flying around them.
“We've shown we can turn language into position,” she said. “Can we go backwards and generate that language so the autonomous aircraft can announce its intention and coordinate with humans on the same channels?”
Sangeetha said future efforts will include taking the more accurate flight path predictions from their model and using them to help the drone aircraft plan its own movements. The researchers are also thinking about how their model could be generalized to other small airports.
Along with Li and Sangeetha, the research team included electrical and computer engineering Research Engineer Chih-Yuan Chiu and Shreyas Kousik, assistant professor in the George W. Woodruff School of Mechanical Engineering.
Related Content
Georgia Tech at ICRA 2026
With more than 90 faculty members and students co-authoring 48 papers, Tech is a leading contributor in multiple areas. New research represents foundational advances in machine learning and computer vision being deployed to real-world systems in areas such as manufacturing, healthcare, logistics, and more.