Sept. 20 (UPI) -- Scientists at the University of California, San Diego have succeeded in training autonomous gliders to recognize and ride warm air currents, or atmospheric thermals.
Eagles, falcons and hawks use air currents to soar to great heights, but until now, scientists weren't sure how birds locate thermals. The latest research provided clues.
To teach gliders to soar, scientists used a type of artificial intelligence training called reinforcement learning. Inspired by behavioral psychology, the machine learning style replicates the process of learning through experience.
"This paper is an important step toward artificial intelligence -- how to autonomously soar in constantly shifting thermals like a bird," Terry Sejnowski, researcher at UCSD and the Salk Institute for Biological Studies, said in a news release. "I was surprised that relatively little learning was needed to achieve expert performance."
Before gliders could learn through trial and error, scientists had to provide a jumping off point. Researchers developed exploratory flight behaviors for the gliders to use in the air.
The gliders used their baseline instructions to experimentally adjust bank angle and pitch in response to environmental cues.
Because turbulence produces noise, it's difficult to predict how exactly a glider will receive and interpret environmental cues -- and recognize constantly shifting thermals -- in real time. Learning through experience helped scientists meet the challenge.
The gliders learned to recognize and interpret thermals by measuring wind acceleration and the roll-wise torques the currents exert on the craft. Through trial and error, the gliders learned how to adjust their flight accordingly, using the thermals to soar to heights of nearly 2,300 feet.
The research -- published this week in the journal Nature -- revealed the different combinations of environmental cues birds use to ride thermals.
"Our results highlight the role of vertical wind accelerations and roll-wise torques as viable biological mechanosensory cues for soaring birds, and provide a navigational strategy that is directly applicable to the development of autonomous soaring vehicles," said UCSD physics professor Massimo Vergassola.