There are some great examples of how artificial intelligence can accomplish things that people cannot do, but there is nothing colder than seeing AI married to robotics. OpenAI, the artificial intelligence research group that counts Elon Musk among its founders, is working on respecting the skills of its nervous networks in a variety of ways, but this one-off robot exhibition is perhaps the most significant thing yet.
The Rubik Cube bot was entrusted to solve with one hand of human style. All this might not show great, but it becomes much more interesting when you realize that it never told your bot creators how to achieve their goal.
The pair of acral networks working together were instructed to observe the cube and control on the arm each side of the game making a cube. The AI then trained himself in computer simulation, learning how the movements of the hand affect the cube and the effect of gravity on the cube as it is being manipulated.
Multiple simulations can be done simultaneously and OpenAI says that this training method has allowed the AI to experience thousands of years in a very short time. Today, the robot can solve the cube quickly under one condition.
As the robot became more trained to complete the task, the researchers put the difficulty up by adding external factors. The team tried out a number of items, such as putting rubber gloves on hand to block his movement, physically putting his hand to the hand while trying to work, or even rubber bands around the fingers of a pot. packing. All this helped the robot to be more reliable to solve the puzzle.
OpenAI says that this type of work is crucial in the advancement of multi-purpose robots that will require uniformity to a person. This will certainly be useful when the robots get up and push them.