A new algorithm, developed by specialists of the laboratory OpenAI, founded by the head of SpaceX and Tesla, greatly facilitates the learning process of robots. If earlier in order to program the robot, it was necessary for him to write special algorithms, now everything looks much easier — the robot just need to show what needs to be done, and the robot after some time, learn to repeat the necessary sequence of actions and be able to do without human help.
Simulation training — a new milestone in the work of the laboratory. Now experts are drawing up development on the robot-manipulator, which quickly learned to put the pyramid out of blocks, looking through the visual neural networks as it does one of the developers.
The development builds on the previous lab programs, but unlike previous programs, the creators of the algorithm used are not real photographs and objects, deciding this time to use a virtual reality headset and a program that recreated the table, lighting, and cubes with different textures. The lighting, the colors of the cubes and desks, as well as elements of the environment were constantly changing, but the robot copes with a task, over and over again repeating the action of the operator.
The creators want the algorithm to finish, so that AI could solve more different problems in completely different conditions. New method learning like a game with a child, that too, at first is not always good, but then improves skills and begins to perform certain tasks better and better.
In OpenAI Elon musk taught robots to repeat for people
Vyacheslav Larionov