Drones have provided humanity with many simple solutions and conveniences. These flying devices have helped prevent the spread of diseases, regenerate agricultural land and forests, track wildlife for conservation purposes, aided search and rescue missions, and much more.

Drones commonly fly in formations, feeding each other information to increase the efficiency and safety of their mission. A problem these drones face is the fact they cannot fly in all weather conditions, with a spurt of wind sending the formation off balance.

To solve this problem, a team from Caltech has developed a deep-learning method called Neural-Fly. This system can help drones adapt to new and unexpected wind conditions in real time, simply by updating a few key parameters in their software.

Commonly when drone software is developed, it is tested in stable conditions. To combat this, Neural-Fly was developed in Caltech’s Real Weather Wind Tunnel, a custom 10-foot-by-10-foot array of 1,200 tiny computer-controlled fans. The simulator can replicate wind conditions between a gale to a light gust.

“The issue is that the direct and specific effect of various wind conditions on aircraft dynamics, performance, and stability cannot be accurately characterized as a simple mathematical model,” corresponding author Soon-Jo Chung explains.

They continued: “Rather than try to qualify and quantify each and every effect of turbulent and unpredictable wind conditions we often experience in air travel, we instead employ a combined approach of deep learning and adaptive control that allows the aircraft to learn from previous experiences and adapt to new conditions on the fly with stability and robustness guarantees.”

Using a “separation strategy,” the team was able to bypass previous challenges on this front and allow the drone to learn how to stabilize and communicate their position with each in real time. In turn, this allows the whole swarm to be able to adapt to the changing environment. Implementing this strategy allowed for the drones to withstand their formation in winds up to a “moderate gale,” where the wind would cause whole trees to sway.

The team developed the software to not only perform well but also be accessible. The Neural-Fly system can be uploaded to a standard off-the-shelf $20 flight control computer, commonly used by the drone research and hobbyist communities.

Source study: Science Robotics – Neural-Fly enables rapid learning for agile flight in strong winds