Bridgeman Wins Office of Naval Research Young Investigator Award

2/2 Pratt School of Engineering

Award will help unlock the potential of historically underappreciated control theory to solve modern challenges of controlling fleets of autonomous vehicles

A graphic drawing of many types of drones traveling together, such as air drones, cars and semi-trucks
Bridgeman Wins Office of Naval Research Young Investigator Award

Leila Bridgeman, assistant professor of mechanical engineering and materials science at Duke University, has been awarded a prestigious Office of Naval Research (ONR) Young Investigator Award. The program identifies and supports researchers within the first seven years of their careers who show exceptional promise for doing creative research. For the next three years, the $450,000 grant will support Bridgeman in her quest to develop new control theories that will allow large groups of unmanned vehicles in the air, on the water and undersea coordinate their movement efficiently and effectively.

“This isn’t just a problem for the Navy. Delivery drones flying in cities or fleets of autonomous vehicles on the ground or in the air face similar challenges,” said Bridgeman, who joined Duke’s faculty in 2018. “Individual vehicles within a fleet can have completely different kinds of behavior and performance specifications, but they still need to coordinate in some way. Even competitors want to avoid hitting one another.”

“This isn’t just a problem for the Navy. Delivery drones flying in cities or fleets of autonomous vehicles on the ground or in the air face similar challenges.”

leila bridgeman

Getting dozens or even hundreds of autonomous vehicles to behave in a coordinated fashion is an enormous challenge. Besides the sheer scale of the problem, communication lines between separate entities are never 100% reliable. They also often don’t want to share their plans and status, and even if they do, their own sensors have limited precision.

“Basically, we’re always working off information that is a bit wrong and a bit outdated,” explained Bridgeman. “If you’re not careful, when you optimize performance while designing highly agile maneuvers, you can easily make a vehicle that over-reacts to this wrong information. Even worse, other vehicles can begin over-reacting to that over-reaction, propagating and amplifying the mistake. Then things really fall apart.”

Four people stand in the sunshine in front of a red brick buildingTo fight the reign of chaos, control theorists develop algorithms and theorems that ensure stability, which precludes these over-reactions. The techniques they use to impose stability, however, usually couple all individuals together. The math behind them then necessarily becomes exponentially more complicated as the group gets large. Worse still, you need a unified framework to describe the systems, which is inappropriate in the case of an entire fleet, where different drones and boats might be trying to fulfill different obligations.

While the usual techniques don’t quite work, control theory has a history of more powerful results. “I feel like I’m delving back into the history of control theory and going to this treasure trove of long underused theorems,” Bridgeman said. From this `treasure trove,’ Bridgeman found inspiration for a framework that can break a large network into smaller groups of objects or vehicles coupled together. Rather than worrying about controlling 10 submarines, for example, it might tackle the more manageable task of controlling five groups of two submarines.

“In my lab alone, I see connections to control of power systems incorporating renewable resources, reducing the energy footprint of server farms, and even autonomous ultrasound design for breast cancer screening.”

leila bridgeman

“The cool thing about this approach is it allows us to break down this huge question into smaller, much more tractable questions,” Bridgeman said. “This also allows us to give different performance criteria to different assets and use more varied methods for assessing how they interact. One might use a lot of machine learning and data while another might build a model from first physics principles. Either way, it allows us to take advantage of whatever it is that we actually know.”

This line of research holds a special connection to Bridgeman, and not just because it’s her field of passion. This specific project, she says, combines all of the research projects her graduate students have been pursuing during her time at Duke. Most of the ideas for this work were inspired by applications outside of drone control, and its impacts will also go beyond this subject.

“The fundamental results we’re deriving here will contribute to so many other areas,” Bridgeman said. “In my lab alone, I see connections to control of power systems incorporating renewable resources, reducing the energy footprint of server farms, and even autonomous ultrasound design for breast cancer screening.”