According to a study accepted last month by the AIAA Journal of Aerospace Information Systems, researchers at the University of Maryland (UMD) have demonstrated a feasible method for autonomous, probe-and-drogue refueling of unmanned fighter aircraft. The work, reported by UMD's Maryland Today, addresses a critical challenge for the U.S. Navy's plan to integrate unmanned aircraft into carrier air wings within the next decade.
The research was led by Donald "Bucket" Costello, a UMD assistant professor of aerospace engineering and a former naval aviator with 25 years of experience. Costello, who is also the director of test and evaluation of autonomous systems at UMD's MATRIX Lab, noted that midair refueling is a high-stakes maneuver even for human pilots, requiring precise control with a margin for error of just four inches. "If you don't practice, you can crash," Costello told Maryland Today.
The core challenge for autonomous refueling is gathering enough data to train machine learning algorithms. Costello explained that collecting data in the air costs the government $45,000 per flight hour, and training a reliable algorithm could require tens of thousands of images under varying conditions like different light, cloud cover, and weather. He described the required flight hours as "unobtanium."
To bypass this cost, the UMD team developed a workaround using synthetic data. Instead of conducting expensive flight tests, the researchers used UMD's HoloCamera—a setup with 300 cameras capable of producing 3D renderings—to create a detailed model of an F/A-18 refueling drogue. They then superimposed images of this virtual drogue onto video footage of actual midair refueling missions, adding in various environmental conditions to build a highly realistic training dataset. Their testing indicated that an autonomous refueling algorithm trained on this synthesized data could likely perform in real-world conditions.
The potential savings are significant. Costello estimated that while a single flight test hour costs $45,000, the synthetic approach could reduce that to the cost of electricity. "Instead of spending hundreds of thousands of dollars on flight tests, I can spend 65 cents on electricity to have computer models train with augmented reality videos," he said.
The study was funded by a grant from Lockheed Martin Corp. and was a collaboration between the MATRIX Lab and UMD's Institute for Health Computing (IHC), which manages the HoloCamera. The first author of the study is Ryan Lowe, who studied robotics under Costello as an undergraduate at the U.S. Naval Academy and is now training to be a naval aviator. Lowe said the technology could support future autonomous jets, known as collaborative combat aircraft, which are expected to fly alongside piloted planes on missions including reconnaissance, strikes, and patrols. He noted that passive sensing, enabled by machine learning, would be more stealthy than active communications, especially in sensitive operations where GPS or radio communications are unavailable.
For university students, particularly those studying aerospace engineering, robotics, or artificial intelligence, this research highlights a growing intersection between academia and military technology. It also demonstrates how synthetic data and virtual environments can solve real-world engineering problems, a skill increasingly relevant in fields from autonomous vehicles to healthcare. The work underscores the role of university labs in developing the algorithms that will underpin next-generation defense systems.