Soaring over the large bodies of water and land is no simple feat for large birds. While utilizing minimal effort, they rely on thermals to glide through the sky and use wind currents. And while scientists understood the physics behind this capability of birds, questions remained on the navigation strategies responsible for it.
Two years ago, researchers from the University of California, San Diego (USD), and Salk Institute unveiled their efforts in creating mathematical models to demonstrate and recreate how birds soar. The team focused on reinforcement learning—a branch of artificial intelligence (AI) that allows machines and software to determine a behavior within a specific context for maximizing performance—to develop algorithms for a navigational strategy. Using a glider to simulate a bird’s behavior, the team trained it to navigate different environments based on soaring performance and environmental cues.
The researchers have continued their work and put their simulations to the test. Soaring up to heights of nearly 2,300ft, the autonomous gliders can navigate atmospheric thermals, taking into account vertical wind accelerations and roll torques used by birds.
“This paper is an important step toward artificial intelligence (AI)—how to autonomously soar in constantly shifting thermals like a bird. I was surprised that relatively little learning was needed to achieve expert performance,” said Professor Terrence Sejnowski, head of Salk’s Computational Neurobiology Laboratory and one of the paper’s authors.
For testing, gliders with 2m wingspans were equipped with a flight controller that provided the on-board autonomous flight policies for bank angle and pitch control. The gliders’ navigation strategy was based on collective experiences gained during several days of field use employing exploratory behavioral strategies.
“We establish the validity of our learned flight policy through field experiments, numerical simulations, and estimates of the noise in measurements that is unavoidably present due to atmospheric turbulence,” said USD Physics Professor Massimo Vergassola. “This is a novel instance of learning a navigational task in the field, where learning is severely challenged by a multitude of physical effects and the unpredictability of the natural environment.”
The algorithms allowed the gliders to estimate vertical wind acceleration and vertical wind velocity gradients across their wings, as well as estimate the noise in gradient sensing from atmospheric turbulence.
“These results are significant because we were able to successfully apply our previous simulation work to a real-world glider,” Sejnowski said.
The team’s work has the potential to take unmanned aerial vehicles (UAVs) to new heights using less energy and better navigating the unexpected.
Interested in more AI innovations? Check out A Healthy Future for Artificial Intelligence in Healthcare and Mystic - AI Controlled Drone.