GPS only works in the open air. Unmanned Aircraft Systems (UAS) can navigate very reliably as long as they fly outside. In enclosed spaces, they would get out of control and crash.

Flying drones indoors therefore requires a completely different type of navigation - and a completely new safety concept. Neither of these existed when IPH began researching autonomous drone flight in the production environment in October 2020. Over the past two years, the scientists in the "Autodrone in Production" project have developed an indoor navigation system, built a prototype of an autonomously flying indoor drone and developed a safety concept. The "car drone" can autonomously explore unknown indoor spaces without being remote-controlled by a drone pilot and without first being equipped with a map of the surroundings.

Navigation without GPS

Indoor navigation works on the principle of a computer mouse. The "car drone" uses an optical flow module and a camera pointed at the ground to determine its position. When it moves, it recognizes the relative deviation from its starting position. The inertial measurement unit (IMU) also ensures flight stability: among other things, it measures acceleration and orientation during the flight. Together, the IMU and the optical flow module replace the GPS.

To enable autonomous flight in unknown indoor spaces, the drone is also equipped with a LiDAR sensor for automated collision avoidance - a laser scanner that detects obstacles and thus prevents the drone from flying into walls, shelves or machines.

But how can the "car drone" find its way around unfamiliar areas independently and without a map of its surroundings? When it takes off, it only knows its immediate surroundings. During the flight, it explores the space bit by bit and the on-board computer automatically creates a map in a 3D grid, which is continuously expanded. The researchers have implemented two algorithms so that the exploration of space can take place systematically: The A* algorithm for planning routes and a self-developed point cloud filter. The latter identifies border areas of the map and distinguishes between fixed borders and open borders. The point cloud filter defines a point on the open border as the target position, while the A* algorithm plans the route from the drone's current position to this target position. Once the target position has been reached, the point cloud filter sets a new target - until there are only fixed boundaries such as walls, shelves or machines in the edge area of the map. Then the entire room has been explored.

One possible use case for this autonomous reconnaissance flight is layout recording: the drone flies autonomously through factory halls and creates a virtual model that can then serve as the basis for factory planning projects.

Safety in drone flight

Autonomous drone flight works in a research environment - but there are still a few hurdles to overcome before it can be used in industrial practice. During the development of the safety concept, the researchers encountered risks that are currently preventing its use on the market and make further research necessary.