Air Force Pilot Studies Flying Moths to Inspire Next Generation Technology

11/7/24 Pratt School of Engineering

ECE graduate student Matthew LaRosa is using machine learning and generative AI to understand strategies for making rapid decisions in complex environments.

Matthew LaRosa stands in front of Wilkinson Building.
Air Force Pilot Studies Flying Moths to Inspire Next Generation Technology

When Matthew LaRosa, an electrical and computer engineering (ECE) master’s student and graduate of the Air Force Academy, came to Duke in 2022, his advisor thought he’d enjoy a project that dealt with flight – albeit on a much smaller scale than LaRosa was used to as a soaring instructor pilot.

“I thought he would be interested in insects because there was a flying object involved,” joked Vahid Tarokh, the Rhodes Family Distinguished Professor of Electrical and Computer Engineering and LaRosa’s advisor.

Two years later, LaRosa has become an integral part of that ongoing project, and his contributions could help inspire the next generation of advances in systems integration. From more robust computers to improved autonomous robots and drones, the applications are broad.

LaRosa and Duke Engineering are part of a large, multi-university research initiative funded by the Air Force Office of Scientific Research that aims to develop new, brain-inspired strategies for making rapid decisions in complex environments.

But as fascinating as the future possibilities are, it begins with an equally fascinating mental image: lab experiments akin to a moth playing a video game.

A sensor attached to the moth acts like a joystick as the insect seeks food. As it flies left or right, the nectar moves with it. Data on the moth’s force, torque and muscle voltage is then sent to Tarokh’s lab, and LaRosa works on decoding large datasets using novel machine learning techniques.

“Overall, we want to figure out how moths think and process information, and then how that info makes its way from being received to making a decision on it,” LaRosa said. “Humans can engineer many things as good or better than insects, but we fail to connect the systems together like they do.”

Matthew LaRosa

Humans can engineer many things as good or better than insects, but we fail to connect the systems together like they do.

Matthew LaRosa ECE graduate student

Hawk Moths in Flight

The Sphingidae family of moths includes more than 1,400 species found around the world. Sometimes called hawk moths, they might be mistaken for hummingbirds because their proboscis, a long appendage common to insects, resembles a beak when they feed on nectar. Hawk moths are relatively large insects renowned for their flying ability and have been studied extensively in the past.

“Hawk moths have a relatively small brain for the large number of actions they do,” LaRosa said. “They can encode their entire environment, fly, avoid predators, all in a very energy-efficient way.”

A hawk moth in the middle of an experiment as it faces a 3D printed flower and is connected to sensors.
Collaborators at Georgia Tech measure voltage spikes from hawk moths and send that data to the Duke team. Photo courtesy of Georgia Tech.

To understand how the hawk moth – specifically the Manduca Sexta species – processes its environment, neuroscientists at Georgia Tech gather neural measurements in simulated environments, including that joystick-like scenario. A series of voltage spikes visualized on a graph is called a spike train. The initial readings are noisy due to experimental factors and because each moth is a unique creature that exerts different amounts of energy.

“We’re trying to understand how the insects make decisions in complex environments using machine learning tools and then use that understanding to inspire new tools,” said Simon Sponberg, associate professor of physics and biological sciences at Georgia Tech. “The Duke team helps us better understand the data we collect and also think about how the structures and patterns we see in that data help inspire new algorithms.”

Four spike train graphs
LaRosa uses spike trains, graphs of voltage spike data, to train generative AI.

Although LaRosa doesn’t have the pleasure of working directly with the hawk moths, their data fascinates him. After simplifying the datasets, he trains generative artificial intelligence with them and works to find the underlying distribution of moth movements with the objective of creating “virtual” spike trains.  

“Our goal is to put everything together and understand the output of force and torque with any given muscle voltage spike,” LaRosa said. “This helps us understand how natural intelligence and neural networks operate.”

Returning to the Air Force

In December, LaRosa will graduate and go from studying moth flight at Duke to learning how to fly the T-38 Talon supersonic jet at Sheppard Air Force Base, Texas. He will undergo pilot training before beginning a 10-year service commitment with the Air Force.

Other students in the Tarokh lab will continue the hawk moth research, but Tarokh hopes to bring LaRosa back for his PhD in the future, Air Force obligations permitting.

“Matt is as good as everyone at the Air Force told me he was,” Tarokh said. “He’s nice, hardworking and honest. I’m sad he has to leave, but I hope he will return.”

Much like hawk moths, LaRosa loves flying and is looking forward to getting back in the air.

“I’m looking forward to pilot training,” LaRosa said. “The Air Force community is always striving for improvement, and I am excited to return to that community.

“At the same time, I’ve had a great experience at Duke. The ECE department has been extremely generous. Working with Dr. Tarokh has been nothing short amazing, and the campus is absolutely beautiful.”

Learn More About Duke Engineering

12/3 Duke Research Blog

Farmers, Crops…and Computers?

Duke’s Forever Learning Institute featured a session with Maiken Mikkelsen who explained how hyperspectral cameras could improve global agricultural practices