AI Improves Drone Stability in Air Ducts

When the wind starts picking up, maintaining control over a drone becomes a very challenging proposition. Even highly experienced operators have a tough time staying on course when that wind comes in irregular gusts, as it so often does. Fortunately, for many flights these gusts may not actually matter all that much. With plenty of open space around the vehicle, bouncing around a little from time to time is not going to cause too many serious problems. But in confined spaces, those gusts can easily cause a crash.

Of course confined spaces generally also have a blocking effect on wind, rendering its effects negligible. But these enclosed areas give rise to another related problem — the drone’s own rotors stir up the air, causing unpredictable gusts that destabilize it during flight. This is especially problematic in air ducts, where small drones are frequently deployed to inspect areas that humans cannot otherwise reach. These ducts may be little more than a foot across, so there is precious little room for error.

To give drone operators a leg up in these situations, a group led by engineers at the Université de Lorraine in France has developed a novel control algorithm. With the help of a clever test rig, they were able to collect a large dataset describing the turbulence drones experience when flying in air ducts. They leveraged this data to train a machine learning model that helps drones to remain stable in these environments.

To get started, the team mounted a quadrotor on a robotic arm equipped with a force/torque sensor. By moving the drone through hundreds of positions inside an air duct, they measured the forces acting on it at each point, effectively creating a force map of the environment.

This force map revealed that the center of the duct is not actually the safest location for the drone. Depending on its position, the rotor wash can recirculate and push the drone toward the walls, significantly increasing the risk of collision. The researchers identified a sweet spot where these recirculation forces largely cancel out, allowing for a more stable hover.

But identifying the safe zone was only half the battle. Air ducts are dark, featureless environments, making it difficult for a drone to determine its position relative to the walls. To address this, the team equipped their 18-centimeter, 130-gram quadrotor (based on the Bitcraze Bolt platform) with a suite of small time-of-flight sensors. These sensors provided basic distance measurements to nearby surfaces. A neural network, trained on motion capture data, interpreted these measurements to estimate the drone’s position in real time.

By combining the aerodynamic force map with this lightweight positioning system, the drone could autonomously maintain itself in the safest possible location inside the duct. In tests, the prototype successfully hovered in ducts as small as 35 centimeters in diameter and navigated through them without crashing.

If everything goes according to plan, future prototypes will integrate mission-specific payloads, such as cameras, thermal imagers, or gas sensors, to make them practical tools for building maintenance and emergency response. And that would be a meaningful step toward enabling small, fully autonomous drones to perform tasks in small spaces too dangerous or inaccessible for humans.