Dealing with the wind is part of flying in the air. Gusty winds can be a challenge for pilots to overcome as they land their planes, or on a smaller scale, the gust can push a drone around a small area of its airspace.
To provide better maneuverability for drones in wind, a team of CalTech engineers have developed a deep neural network – an artificial intelligence tool – to allow the drone to be agile in the presence of blowing air. IN videoresearchers show a drone with a quadcopter, which thanks to this software can perform maneuvers with the number eight and fly through a small gate, all in the presence of wind speeds of 27 miles per hour in a wind tunnel.
The scientists first had to gather data so they could train a neural network to get the flying machine to perform the stunts. It didn’t take long: only 12 minutes of flight time. This is very little data, says Michael O’Connell, a graduate student in the aerospace department at CalTech and one of the authors of new study describing the work published Wednesday in the journal Scientific work. This work, run by AI, is called Neural-Flyand follows another similar work called Neural Lander and Neural-Swarm.
During training for this latest Neural-Fly experiment, the drone flew in a wind tunnel in the presence of six different wind speeds, the fastest being 13.4 mph. “Basically, we teach a drone, ‘That’s what it looks like when you get hit by wind at 5 miles per hour, wind at 10 miles per hour,'” O’Connell said. “The drone is able to learn what the wind looks like, and then when we fly on our test trajectory at number eight, it uses that experience and says, ‘I’ve seen this before.’
From this data, the team created the deep neural network, which then allowed their aircraft to be skilled in maneuvering in the same wind tunnel, such as approaching through an eight-gauge port or passing through two elliptical gates. The speed the drone tested was faster than the speed it had encountered in training: about 27 miles per hour. This is the maximum wind speed that this wind tunnel can produce, said Guanya Shi, another author and CalTech student. In addition to requiring only a small amount of data, the software only works with the Raspberry Pi, a low-cost computing device.
Soon-Joe Chung, a professor of aerospace and control and dynamics systems at CalTech and co-author of the same paper, says the error rate they see with the new system is between 2.5 and 4 times better when compared to the existing “state of art ”technology for precision drone flying. The deep neural network that controls the drone also has “adaptive control,” Chung said, calling it a “breakthrough method.” This means that AI can react adaptively to what happens in real time with the wind.
Chung sees applications for this machine learning system when it comes to a future in which our skies can be filled with more drones. Companies like FedEx are looking to use large drones to help move packages from one place to another, and Alphabet-owned Wing delivers consumer goods through small drones in Texas. Meanwhile, other companies are working on electric flying machines that can transport people – these are air taxis that can take off and land vertically. Plans for them range from ships designed to fly alone with a passenger on board to those currently planned around pilots.
Drones, which must have an “ultimate safety guarantee”, can take advantage of software like this, Chung said. “The final example is obviously flying cars, because they have to carry human passengers.
“We hope to make this the future in which we can have safe drones that can survive all kinds of winds – tornadoes, hurricanes and severe storms,” Chung added. “I can’t say that we can achieve this immediately using our Neural-Fly, but we are taking a great step forward towards this goal.
After all, whether the drone is transporting packages or people, it must land safely on its site, even if the wind blows in an unpredictable way. Without the promise of a safe landing, the mission may need to be cleared before it lands, or the flying machine may be diverted to another location if it is already buzzing in the air.
Watch a short video on the technology below: