Autonomous drone for factory planning

A drone that can fly autonomously in unknown indoor spaces has been developed by researchers at the Institute for Integrated Production Hannover (IPH) in the "Autodrone in Production" project.

According to the project's website, in the future the "car drone" will perform data acquisition in factory planning processes much faster and with less manpower. The drone flies autonomously through an indoor area and automatically and autonomously creates a virtual model that can serve as the basis for factory planning projects.

GPS not possible indoors

Since a GPS signal can hardly penetrate buildings, GPS only works well outdoors. Indoors, a so-called Unmanned Aircraft System (UAS) would get out of control and crash. So if a drone is to fly indoors, a completely new type of navigation is required. Since October 2020, IPH has developed indoor navigation and built the prototype of an autonomous flying indoor drone in the project "Car Drone in Production".

Computer mouse principle

The indoor navigation works according to the principle of a computer mouse. A camera that is directed at the ground and an optical flow module determine the position of the "car drone". In addition, the inertial measurement unit (IMU) ensures flight stability by measuring acceleration and orientation during flight, among other things.

Navigation in unknown spaces

The drone is equipped with a LiDAR sensor for automated collision avoidance. This laser scanner detects obstacles and thus prevents the drone from flying into walls, machines or shelves.

Immediately after takeoff, the on-board computer is only aware of the immediate surroundings. Only during the flight is the space explored piece by piece and a map is created in a 3D grid and continuously expanded.

Two algorithms for exploration

Two algorithms help to ensure that the exploration of a space proceeds systematically: The A* algorithm for planning path distances, and a point cloud filter developed in-house for identifying boundary areas and distinguishing between fixed boundaries and open edges. The point cloud filter sets a point on the open edge as the target position and the A* algorithm plans the route there. Once the target position is reached, a new point is defined - until there are only fixed boundaries such as walls, shelves or machines in the edge area of the map. The 3D model created can then serve as the basis for factory planning projects.

Scientific work honored

Project leader Andreas Seel received the "Certificate of Best Paper" award at the Iccar Conference 2022 for his research paper "Deep Reinforcement Learning based UAV for Indoor Navigation and Exploration in Unknown Environments", which describes the principle of indoor navigation and the Deep Reinforcement Learning approach that enables the drone to make automated decisions. The publication can be found here: https://ieeexplore.ieee.org/document/9782602

Safe drone flight

While autonomous drone flight works indoors in the research environment, some hurdles must be overcome for industrial use.

Hazards posed by the drone can be readily reduced. For example, the risk of a battery fire can be minimized through proper maintenance and storage. Cutting injuries from the rotors can be prevented by using a propeller guard.

The biggest hurdle the IPH team has encountered is the electromagnetic compatibility (EMC) of some sensors. In industrial environments, such electromagnetic interference cannot be prevented. Electric motors from machines or forklifts, current-carrying conductors, larger accumulations of metal - all of these can limit the drone's ability to navigate and, in the worst case, cause it to crash. Further research and development is therefore necessary before industrially suitable, autonomously flying and safe indoor drones are ready for the market.