Introduced artificial intelligence that learned human hand movements by itself

OpenAI, a not-for-profit company, has developed an artificial intelligence that simulates complex hand movements without human behavioral data, the company said on Wednesday.

The human hand is an extremely sophisticated tool. The simple operation of holding the cup on the table requires extremely complicated calculations to reproduce the machine. Even now, robot arms are moving at high speeds in many factories, but they only perform sophisticated programming to capture standardized objects. In other words, only bolt side bolts and nut side nuts can be caught. It is still very difficult to create a robot that can catch anything like a human being.

Open AI, sponsored by Tesla founder Ilron Musk, has developed a new artificial intelligence that adapts like a human hand to move a robot hand under these changing conditions. This program is called 'Dactyl'.

The remarkable thing about Dazzle is that he did not learn any human behavior data at all. Until now, most artificial intelligence has learned to learn human data. As I learned the Baduk score of the Alpha Goh people who fought with Lee Sedol. Instead of human data, the researchers have chosen a way to artificial intelligence by imagining various attempts on its own and applying it to real world. And to remind themselves of the repetition of 'try-fix'.

Dactyl was given only the image data of his three-directional camera to see his robot arm and target object. Based on this, Dactyl was designed to do it for himself for a few seconds before moving the robot arm in a virtual three-dimensional space in digital space. And I try to implement the method I think is best. After that, based on the data of whether it was actually successful or not, it is necessary to repeat the same process before the next attempt and finally find the optimal movement.

There is still considerable computer resource available for such operations. The researchers found that 6144 central processing units (CPUs) and eight graphics processing units (GPUs) were needed in the paper. The researchers described it as \"computing ability capable of learning about 100 years in 50 hours.\"

This shows that Dactyl can do complicated movements that conventional machines can not imitate. For example, as we rotate our apples with one hand to see if we have any hurt, we learned ourselves of the most difficult movements without turning the training cubes down. Smruti Amarjyoti, a researcher at the Carnegie Mellon University Robotics Institute in the United States, said in an online media interview that robotic movements are \"level\" and \"graceful\". On the other hand, Antonio Bicchi, professor of the Italian Institute of Technology, said it is difficult to see that artificial intelligence \"has learned to move beyond a certain set of conditions.\"