AI facilitates autonomous drone flights

Drones with four rotors, known as quadcopters, are able to fly through forested areas autonomously and collision-free at a speed of 40 kilometers per hour. This totally new technology developed by researchers from the Robotics and Perception Group at the University of Zurich (UZH) functions exclusively on the basis of on-board sensors and computations. Computer software and experienced pilots are no longer required for these aspects. A specially trained AI can acquire such abilities practically overnight, according to a press release.

The robotics experts have achieved this by letting the AI of this drone learn from the algorithm of another drone. This algorithm had flown a computerized drone through a learned, simulated environment. Interestingly, the simulators used here do not have to be an exact replica of the real world, says PhD student and co-author of the study, Elia Kaufmann: “If using the right approach, even simplistic simulators are sufficient”, she explains.

In a recent study that has been published in the trade journal “Science Robotics”, Professor Davide Scaramuzza and his team trained an autonomous quadcopter to fly through previously unknown environments such as forests, buildings, ruins and trains without bumping into any obstacles. This could not only help to save lives, but also improve the performance of autonomous cars. Moreover, this technology could even pave the way towards training AI systems for operations in areas where collecting data is challenging or even impossible.

The aim will now be to improve the system and sensors to allow the neural network of an autonomous drone to generate even more information about an environment in as short a time as possible. The hope is that this would then allow the drones to safely fly at speeds in excess of 40 kilometers per hour.