It is now accepted that robots can learn to perform basic tasks. Whether by viewing videos on Youtube or by trial and error, Deep Learning algorithms today give them this ability reserved little beings living. However this requires the robot a profusion of efforts, time and energy totally disproportionate. Nevertheless, artificial intelligence world is changing quickly, and researchers now propose a much more rapid to enable robots to learn approach. This technique, called “Deep Reinforcement Learning” to multiply by 100 the speed of learning of the robots of the future.
Osaro is adapting the approach to industrial robotics
Sean Parker, Peter Thiel, and Jerry Yang wagered $ 3.3 million in a totally unknown startup: Osaro. The vocation of this startup is to develop algorithms of artificial intelligence for machines. Its researchers intend to develop algorithms “Deep Reinforcement Learning” or “deep neural network in reinforcement learning” in french for robots. DeepMind, an another startup today bought by Google DeepMind had demonstrated the interest of this technique in December 2013 with… of the 1980s arcade games.
This technique allowed the computer to acquire a level equivalent or even superior to humans always fans of Atari 2600. Asked by MIT Tech Review, Derik Pridmore Osaro president claimed that the industry will be the first market of the algorithms designed by his team. A rather risky choice for the startup. Currently algorithms are developed in simulation. The first implemented among industrialists are announced for 2017.
Translation : Bing Translator
“A Supercharged System to Teach Robots New Tricks in Little Time”, MIT Technology Review, December 2, 2015
“Robotics: Ethics of artificial intelligence”, Nature, May 27, 2015