The robot hand solves the Rubik's cube in just about 4 minutes

The fact that a smart robot capable of solving a Rubik block with one hand has proven how far the robot industry has gone - but at the same time, according to experts, we still have a long way to go. discover.

The OpenAI Led system used a computer simulation to teach the robotic robot to solve the cube, operating them through the basic rules that an average person takes about 10,000 years to complete.

On average, the robotic hand took about four minutes to solve the Rubik's cube.

The robot, after being taught, can solve a slightly modified cube to help the machine know how it is being operated.

The team says the time it takes for a robot to finish varies, but it usually takes about four minutes to complete the cube.

Using machine learning and robots to solve a Rubik's block has achieved previous results. Specifically, in March 2018, a machine upgraded by the engineers at MIT completed a cube in just 0.38 seconds.

What is remarkable about OpenAI's effort is the use of a versatile robot with a human-like design rather than a machine designed specifically to handle Rubic blocks.Matthias Plappert, head of robotic research at OpenAI , telling the BBC: "The ability to solve rubic cubes with a robot hand is actually extremely difficult. You need to control your fingers very accurately and do it for a very long time and mess it up. in the middle, and there are a lot of different things that can happen during rubic solving too. '

Mr. Plappert praised the group's method of gradually adding difficulty to the process - simulation obstacles would force the robot to respond to complete the task.

The team says this technique, called automatic domain randomization (ADR) - has been used to give robots the ingenuity and consistency to handle changes in circumstances, far beyond what the machine can do. Predictable calculation and simulation.

Art of advertising

In a demonstration video posted on Tuesday, OpenAI showed how to make it difficult for robots - such as nudging the cube with a giraffe stuffed animal, or covering the cube with a black plate - not the best. block the completion of the cube (scientists say the hand can't do the same thing 100%)

The team intervened in the slight disturbances to the robot to see if it could handle the unexpected

While human instincts, dealing with interruptions or hassles, especially when handling handling and manipulating objects, are considered a major robot challenge - a problem that will need to be addressed. If advanced robots become popular in every home and business.

Peter Welinder, team leader at OpenAI said: 'We use human hands for everything, we use to solve Rubik's blocks, and also use them for cooking. This is one of the reasons we choose robotic hands, because it promises robots that can do more things. ''

'The biggest obstacle is craftsmanship like a hand'. So, we have taken the Rubik's cube as an example of how we can use our ingenuity.

But as Professor Ken Goldberg argues from UC Berkeley: "Although it is an impressive piece of performance, we should not overstate the OpenAI research." He told the BBC: "The average person is not good at solving Rubik's blocks. So when they see a robot doing that, they say, 'well, this is better than a human. But that's a bit fake. lied because the game is not real. '

He said the OpenAITHER ADR technique was a real advance, but handling objects that were more complex and unpredictable than a Rubik's cube needed more research.

Will we be able to achieve the result when a robot can pick up a deck of cards and shuffle them like a crook in Las Vegas? That could take 10-20 years.

We also have a lot of effort when replacing very complex tasks such as cooking, slicing vegetables, or even picking vegetables or washing dishes.

OpenAI was announced in October 2015, co-sponsored by Tesla CEO Elon Musk, startup expert Sam Altman and other investors (although Musk is no longer involved). Its goal is to do more in-depth research on artificial intelligence, focusing on a technique called "intensive learning" - teaching AI with repetitive tasks and "rewards" for accurate results. OpenAI is considered the main rival of Google's Deepmind.

1. See the robot arm controlling the plane as skilled as a pilot
2. Jeff Bezos "showed off" the robotic arm delicate enough to play rubik