AI gliders learn to fly using air currents just like birds

Birds do not always flap their wings to fly; sometimes they fly away taking advantage of rising hot air columns called thermal. With large wings, they can stay in the air for hours while spending a minimum of energy. The exact way they do it – navigate through tiny changes in unpredictable drafts – is not well known. But scientists are now using artificial intelligence to learn their tricks and, hopefully, they can teach our planes to do the same.

As described in an article published this week in the journal Nature, researchers from American and Italian universities have used machine learning to form an algorithm to control a glider to navigate the thermals. This is not the first time that artificial intelligence has been used for this task (Microsoft released similar work with gliders last year), but this is the first time that actual flight data is used for

The work suggests that future autonomous aircraft could take advantage of the thermals, rather than resort to noisy and energy-laden powered flights. He also suggests that the AI ​​might be able to help us understand exactly How the birds in flight do what they do so well. When designing their algorithm, scientists found that factors such as vertical wind acceleration and lateral torque were important for the system glider to navigate smoothly. The same thing, they suggest, could be true for birds.

Image: Jerome Wong-Ng and Gautam Reddy

To create their AI system, researchers used reinforcement learning. This is a training tool that works like trial and error. The system receives a certain number of inputs and is asked to act in such a way as to maximize a certain reward. It starts without any knowledge of the task and learns to behave properly over time. In this case, the introduction consisted of flight information, such as glider height, yaw, ground speed, and speed. The reward sought was to maximize his rate of climb (the speed at which he gained height).

The researchers formed their algorithm first in a simulator and then in real life. They flew about 240 flights into the heavens above Poway, California, which lasted about three minutes on average. They steered their kite to a fixed location using a manual controller, then the AI ​​took over, using the thermal currents (which can travel several meters per second) to climb into the sky. .

"In good situations, the glider can stay in the air for about 45 minutes," says Gautam Reddy, one of the newspaper's authors. The edge by email. "We had flights where the wind was too strong for the glider to handle and we had to retrieve it prematurely [and] we had some with eagles attacking the glider and some with eagles and glider hovering together.

Image: Gautam, Wong-Ng, Celani, Sejnowski, Vergassola

There is still a lot of work to be done in this area before being able to use AI to control gliders in full flight. Thermals are just one type of updraft that big birds enjoy. Others are created by air currents that overflow on mountain ridges or by the collision of air masses in "convergence zones" – places such as shores and boundaries. of the desert. In other words, it is not because the AI ​​is able to drive a thermal that it is ready to face the varieties of wind that the world has to offer.

However, Gautam and his colleagues are confident for the future. They say that it would not be too difficult to create autonomous gliders that use AI to navigate long distances. "We are looking forward to doing so in the future," says Gautam. These gears could be used for long-term scientific studies and for ambitious projects, such as monitoring bird migrations at the end of the wings. By learning to fly like birds, we can also learn more about their lives.