The drone uses sensors and learns to fly in air ducts. It avoids crashing by adjusting to airflow and knowing its position.
Engineers at Université de Lorraine in France have developed a control algorithm that helps drones fly through spaces like air ducts. By using sensor data with a machine learning model trained on turbulence, the system lets drones hover and move in ducts as narrow as 35 centimeters without crashing.
The team started by building a force map of an air duct. They mounted a drone on a robotic arm with a force and torque sensor and moved it through many positions inside the duct. At each point, they measured forces from airflow. The map showed that the center of the duct was not always the best spot. In some areas, rotor wash pushed the drone toward the walls. But the team found a zone where these forces balanced out, letting the drone hover safely.
Finding this zone was only part of the work. Air ducts are dark and have few features, so it is hard for drones to know their position. To solve this, the team gave their 18-centimeter, 130-gram drone—based on the Bitcraze Bolt platform—time-of-flight sensors. These sensors measured distance to nearby surfaces. A neural network, trained with motion capture data, used these readings to estimate the drone’s position.
By combining the position system with the force map, the drone stayed in the safest part of the duct. In tests, it hovered and moved through narrow spaces without crashing.
While ducts block outside wind, they bring a new problem: the drone’s own rotors create turbulent airflow. These gusts are hard to handle in tight spaces, where small shifts can lead to crashes. The system learns from these gusts and adjusts flight in response.
In open areas, wind is easier to handle. Skilled pilots can deal with gusts, since there is more room to adjust. But in small spaces, even trained operators face problems when airflow changes.
Next, the team plans to add tools like cameras, thermal sensors, or gas detectors. This will help the drone do tasks such as inspecting buildings or entering places people cannot go.
