Brooks McKinney

Jan 18th 2019

Biomimicry Gives a Lift to AI in Aviation


Who among us hasn’t stared up at a hawk or a vulture circling lazily in the sky and wondered how they stay aloft so long? Or wondered how sky-darkening flocks of migrating birds can travel thousands of miles so quickly and so effortlessly?

Researchers are now able to tackle these questions more systematically using biomimicry, the process of imitating nature’s systems to solve complex human problems. Their goal is to develop an artificial intelligence algorithm that will allow gliders — either piloted or autonomous sailplanes — to mimic the flight behavior and endurance of birds. Soon, AI in aviation will transform the types of missions that gliders perform.

Gliding Like Birds

Popular Mechanics has documented the fact that birds use updrafts (zones of rising air created by topographical features such as mountains and cliffs) and thermals (rising columns of warm air) to stay aloft for extended periods of time. How they identify and navigate these zones of rising air — which form, disintegrate and shift constantly in three-dimensional space — is not well understood. Experienced glider pilots typically use visual cues from lofty cumulus clouds (often fueled by thermals) and warmer, darker areas of terrain to discover the presence of thermals.

Field Training

Gautam Reddy, a graduate student in physics, and a team of his colleagues at the University of California, San Diego, have studied these questions and reported their findings in a recent issue of Nature magazine.

They equipped a two-meter-wingspan, radio-controlled glider with a flight controller and instruments that could measure vertical wind accelerations along the length of the glider and across its wings. Their working theory was that birds use similar sensory information to identify and navigate thermals.

Between 2015 and 2017, they conducted more than 500 flights near Poway, Calif., each one lasting 20 to 30 minutes. The goal was to train the glider in a concept called “reinforcement” to navigate thermals autonomously. They used their measurements to build a simple algorithm that would steer the glider toward zones of higher lift.

Taming Unpredictability

Unfortunately, explained Reddy, atmospheric conditions vary so much hour to hour and day to day, even in the same location, that building a robust AI algorithm for glider aviation using field measurements alone is quite challenging.

“AI algorithms require a lot of data,” he said. “One way to add fidelity to the algorithm would be to use numerical simulation and a process called model-based learning.”

This approach, which his team did not use, would involve “flying” a simulated glider repeatedly through a simulated atmospheric environment containing realistic thermals, convective currents and turbulence. Atmospheric data and the glider’s response to those conditions could be collected on each simulated flight, then used to systematically update and refine the AI algorithm.

Reddy is not sure, however, if biomimicry or trying to learn from birds is the best approach for building an effective AI algorithm for gliders.

“We see birds navigating thermals, but we’re really not sure what they are sensing or feeling as they fly,” he said.

Data From Gliders

A better strategy, Reddy proposes, would be to extract visual and other sensor information from gliders themselves.

“State-of-the-art gliders actually have better aerodynamic properties than birds,” he said. “We can also use instruments to gather higher accuracy measurements of temperature, wind velocity, humidity, barometric pressure and other relevant data.”

AI in the Cockpit

Before turning glider controls over to AI algorithms, however, Reddy suggests that we first try out AI as a navigational aid for pilots, the same way we use Google Maps today while driving.

“A pilot would still fly his sailplane as before,” he said, “but he would use the AI algorithm to suggest the fastest, most efficient route to his destination, taking advantage of the fastest-flowing streets (thermals) along the route.”

New Rules of Engagement

AI in aviation offers some tantalizing prospects.

“If we can come up with a glider that’s as efficient as a bird, it could travel across continents,” said Reddy. “That capability could be extremely useful for surveillance or remote sensing missions.” The one limitation, he admitted, is that gliders rely on thermals, which are normally present during daylight hours only.

Nonetheless, AI-controlled gliders could be a perfect complement to high-altitude surveillance systems such as the U.S. Air Force’s Global Hawk unmanned aircraft.

“Gliders are very small, unpowered and virtually silent,” said Reddy. “So they would not need to fly at high altitude. I think they could provide our military commanders with some very interesting options.”

Northrop Grumman is a leader in the development of autonomous aerial reconnaissance systems. If you’d like to explore opportunities to help your career soar, check out our Careers page.