This AI-powered robotic hand can solve a Rubik’s cube


From summer camp to grade school, solving a Rubik’s cube has always been an impressive way of showing off. Today, the fastest Rubik’s cube solver can do so in just 4.22 seconds. Meanwhile, people have learned to solve the classic 3×3 cube with their feet and even while blindfolded.

Now, an AI-powered robotic hand is able to solve the Rubik’s cube about 60 percent of the time. A team from Elon Musk’s co-founded company, OpenAI, is behind the project. Simply watching the hand manipulate the cube is impressive enough. Seeing it solve the puzzle is downright amazing.

Incredible Dexterity

OpenAI’s robotic hand has been training for years to get to this point. After going through countless simulations, the fake hand can now solve the Rubik’s cube all on its own. The team claims that it has enough dexterity and problem-solving skills to complete the puzzle with a 60 percent success rate. However, that does decrease to just 20 percent when it attempts to solve a “maximally difficult scramble.”

Manage your supply chain from home with Sourcengine

For anyone that has tried to solve a Rubik’s cube one-handed, this should come as no surprise. It should also make the robotic hand’s feat even more impressive.

The video from OpenAI shows just how nimble the hand is. Aside from the occasional calculated, jerky movements, the way that the fingers move is almost identical to a human. This goes to show how far the field of humanoid robotics has come.

Recently, Boston Dymanics’ Atlas robot wowed viewers with an impressive gymnastics routine. It seems like it is only a matter of time until robots like that are merged with the talents and dexterity of this one from OpenAI.

Trial and Error

Interestingly, OpenAI actually programmed an algorithm capable of solving a digital Rubik’s cube back in 2017. This should come as no surprise considering that the firm is behind other impressive artificial intelligence feats like beating pro gamers and creating compelling ad copy. While its Rubik’s cube software worked extremely well, the real world is full of complications.

Now, after years of refinement, the company has its robotic hand functioning despite real-world distractions. In a research paper documenting the feat, OpenAI notes that the hand is able to solve the cube even while the team tries to throw it off. For example, they tormented it with a stuffed giraffe, tied its fingers together, and even covered up the cube.

A blog post from the company states, “We set this goal because we believe that successfully training such a robotic hand to do complex manipulation tasks lays the foundation for general-purpose robots.”

Indeed, while humanoid robots like Atlas are groundbreaking, it will take a mixture of large-scale movement and precise manipulation for these robots to autonomously solve problems in the real world.