Researchers have devised a deep-learning technique for drones to deal with unpredictable wind conditions.
Engineers at Caltech have created a deep-learning technique that enables drones to adapt in real-time to wind conditions that are novel and unknown. Research describing Neural-Fly was published on May 4 in Science Robotics. It was evaluated using its Real Weather Wind Tunnel, a unique 10-foot-by-10-foot array of over 1,200 small computer-controlled fans. The device, dubbed Neural-Fly, employs deep neural networks to learn how to adapt to aerodynamic forces like gusts and turbulence. Following a flight route, the error rate of the drones is around 2.5 to 4 times less than that of drones equipped with matching adaptive control algorithms.
Researchers at the CAST Real Weather Wind Tunnel in Cambridge, Massachusetts, have created Neural-Fly, a drone capable of detecting and responding to changing wind speeds. The team has shown that a drone’s flight data can be transmitted to another drone, creating a knowledge base for autonomous vehicles. Test drones were subjected to winds of up to 12.1 meters per second, or around 27 miles per hour, or a Beaufort six.
Source: https://www.sciencedaily.com/releases/2022/05/220505085644.htm
