Great news, humans! We got the Terminator T-800‘s hand.
There have been robots programmed to solve the Rubik’s Cube but none taught itself how to do it from scratch — and much less with a human-like hand. Until now. That’s what OpenAI has done with Dactyl. This a big step for putting humanoid robots on the street acting just like us.
The AI research company is looking to develop robots that can learn on their own from scratch, analyzing what they see and feel to take decisions and behave like real human beings.
They do it by running simulations in software, presenting a neural network with many randomized variations of the same problem. In this case, the problem is how to solve the Rubik’s Cube with one hand.
According to the researchers, this is an ideal test to demonstrate how they approach to build generic robots that can do anything in the real world works: “Solving a Rubik’s Cube one-handed is a challenging task even for humans, and it takes children several years to gain the dexterity required to master it.”
And it worked. Dactyl — their new artificial intelligence that controls a “robotic hand that has been around for the last 15 years” — has successfully passed the test, as the above video shows.
Like OpenAI admits, it’s not perfect. First, it’s slower than humans for sure. And the AI-powered hand hasn’t perfected the technique, as it only achieve a solution in 60% of the times and “20% of the time for a maximally difficult scramble.” Still, that’s still 100% more times than me and most people.
And of course, this is only the beginning. Like Boston Dynamics’ Big Dog evolved in just a few years to become the scarily dexterous and agile Spot, this robotic hand technology will advance at an equally fast — or even more so — pace. Buckle up. The next decade is going to be a very wild ride.