Algorithm navigates drones through the city

Today’s commercial drones navigate by means of GPS. This is practical for higher altitudes, but drones navigating at low altitudes between buildings or along city streets, where pedestrians or cyclists might cross their way, need to be able to react quickly to unforeseen events.

Researchers at the University Zurich (UZH) and the National Centre of Competence in Research NCCR Robotics have now developed an algorithm that can safely drive a drone through such situations. 

Called DroNet, the algorithm produces two outputs for each single input image: one for navigating while avoiding obstacles, and one for collision probability to recognize dangerous situations and promptly react to them. When such static and dynamic obstacles are recognized, the drone slows down to avoid crashing into them.

Instead of relying on sensors, the drone developed by the researchers uses a normal camera with a very powerful artificial intelligence algorithm. These two components allow the drone to interpret the scene it observes and react accordingly.

“This is a computer algorithm that learns to solve complex tasks from a set of ‘training examples’,” UZH Professor Davide Scarmuzza said in a statement. 

To gain enough training examples for their algorithm, the researchers collected data from cars and bicycles driving in urban environments. The researchers also showed that their drones learned to not only navigate through city streets, but also in closed spaces like buildings and parking lots. This means that the algorithm could use the learning in unknown environments.

“With this algorithm we have taken a step forward towards integrating autonomously navigating drones into our everyday life,” said Scaramuzza.

DroNet could be used for monitoring and surveillance, parcel delivery, or even rescue operations in disaster areas. But it must first overcome many technological issues “before the most ambitious applications can become a reality,” added PhD student Antonio Loquercio.