An AI allows a drone to fly alone in a forest

researchers at the Institute d ‘artificial intelligence from the University of science and arts of Southern Switzerland (SUPSI) and Robotics group and perception of the University of Zurich have just put online a video illustrating their research in a specific application of the machine learning. They have imagined operate artificial intelligence to allow a drone to find his way in a forest. Able to follow a trail of himself, the drone can be used in order to search for a missed person.

Slowly but surely, the drone finds its way through its neural network

L'image captée par la caméra du drone est analysée par les 10 couches du réseau neuronal. En fin d'analyse, 3 bits seulement : aller à gauche, à droite ou avancer.

The image captured by the camera of the drone is analysed by 10 layers of the neural network. At the end of analysis, only 3 neurons: go left, right or forward.

Nothing to do with racers race organized in the forest. The drone zigzags on the trail, seems reluctant to every second, and yet there are trail and argues tirelessly. No not the remote control driver. Researchers in artificial intelligence it were equipped with an AI that allows it to analyze its environment from the pictures from his camera and find his way. To achieve this result, the drone is based on a neural network composed of 500,000 neurons arranged in 10 layers, representing 57 million connections. Researchers train the brain simply by setting three cameras on the head of a Walker. From 20,000 photos taken during his hike, Deep Learning algorithm has literally learned the road to follow. The result is there, not only the drone can be launched on the trail but on an image analysis, AI proves to be more efficient than a human.

Translation : Bing Translator

Sources:

“Drones recognise and follow forest trails in search of lost people”, RoboHub, 10 février 2016

“A Machine Learning Approach to Visual Perception of Forest Trails for Mobile Robots”, IEEE Xplore, 17 décembre 2015

Commentaires Facebook
Twitter Facebook Plusone Pinterest Linkedin
This entry was posted in Aéronautique, Développement and tagged , . Bookmark the permalink.