A robotic hand has been created that imitates real movements

Openai’s artificial intelligence laboratory presented a Dactyl robotic hand, which simulates the mechanics of real movements and operates autonomously.

If you give a dactyl alphabetical cube and ask to show concrete letters, then II will independently determine the desired edge, and then start moving with hand and fingers, turning the object with a suitable side. For us, this is a fairly simple task, since Millias preceded millions of years of evolution, but for the car it is a real achievement. Previously, individual devices could successfully perform simple tasks, for example, capture, sort, push.

In just two days, artificial intelligence itself has learned to manage a robotic hand, simulating events that in real time would take 100 years of trial and errors. Such a move implies random events, so the algorithm takes into account the force of friction and gravity. After examining this data, it simulates actions in reality and copes with uncertainty.

While the AI with a robotic hand can only rotate the object, but the researchers are studying how the system can be applied to solve more complex tasks. Such models can be used in production, drones or unmanned cars.