Researchers at the University of Zurich, Switzerland, have developed a new autonomous navigation system that gives drones the ability to fly in complex, unknown environments at high speeds, without the need for have a human operator in control.
- China used autonomous submarine drone to track and bomb targets
- Autonomous drone may have attacked human beings of its own “will” for the 1st time
- Drone guided by algorithms leaves human pilots eating dust in duel
With this approach, quads can find the correct path without using a map previously loaded into the system. An artificial intelligence (AI) algorithm enables collision-free navigation even in places with many hostile obstacles such as buildings and trees.
“To master agile autonomous flight, you need to understand the environment in a split second to guide the drone without collisions. This is very difficult for both humans and machines. Human pilots can reach this level after years of perseverance and training, but machines still can’t,” explains robotics professor Davide Scaramuzza, lead author of the study.
Want to catch up on the best tech news of the day? Access and subscribe to our new channel on youtube, Canaltech News. Every day a summary of the main news of the tech world for you!
The engineers trained an autonomous four-engine to fly through environments never seen before, such as forests, buildings and ruins. They managed to maintain a constant speed of km/h without hitting trees, walls or other obstacles that appeared from surprise along the way.
The objective was achieved using only on-board cameras and a computer system installed in the drone. The neural network learned to fly by observing a simulation model, consisting of an algorithm that guided a computer-generated drone through a virtual environment with several complex obstacles.
“At all times, the algorithm had complete information about the state of the four-engine with the readings of its sensors, being able to count on enough time and computational capacity to always find the best trajectory without colliding with obstacles”, adds Scaramuzza.
In the real world
As they don’t need an exact replica of the real world, these artificial intelligence systems can take advantage of high-performance simulators to learn complex navigation skills that trained human pilots would take years to develop, much faster.