While your two hands may not solve the rubik block, this robot can do it with just one hand

Rubik is a favorite intellectual game worldwide, not only by humans but also … robots. If humans are always trying to break previous rubik records, robots are also trained to have quick and weird ways to solve rubiks.

Most recently, San Francisco-based Open-AI has demonstrated the ability to solve robots’ rubik, notably that it can be solved with just one hand.

This robot is called Dactyl, while most of us, with both hands, struggle to solve the rubik’s cube and it can do it with only one hand. Since 2017, Open AI researchers have been training this robot in its ability to solve rubiks and notably, it has a similar design to human hands.

Of course, before Dactyl, there were robots that were able to solve rubiks very quickly, in less than a second, much faster than the 5-second record set by humans, but they did not have the same design as their hands. people.

Although there were many rapid rubik solvers, Dactyl was developed in a different direction. Researchers have built it out of nothing and put their intelligence in so that it can learn and solve complex tasks in real time. This is a first step in creating highly intelligent robots for human life, not only skilled but also thinking.

During the past 60 years, to perform the difficult tasks that humans do with their hands, it is required to design a unique robot for each mission. Dactyl is the first step in changing that. It is the forerunner of intelligent robots, better adaptable, more agile, capable of dealing with the clutter and complexity of the real world.

The robot hand itself is not really a new invention; Its many variations that have existed for the past 15 years. The new point is that researchers use the Automatic Domain Randomization (ADR) system, which is capable of creating endless random environments with increasing difficulty, used in simulation before Robots are tasked with challenging tasks in the real world.

” During the simulation, everything was gradually changed from the size and mass of the rubik’s cube to the friction of the robot fingers,” said Open AI researcher Ashley Pilipiszyn.

“ADR starts with a single fixed environment, neural networks will learn how to solve rubik from here, ” she said. “When the neural network gets better and reaches a certain capacity, the number of domains The randomization will be incremented by itself. This will make the task of the robot more difficult, because the neural network now has to learn how to get used to the random environment . “

It’s still not perfect, the researchers set a 15-fold rotation to solve the rubik and Dactyl blocks with only 60% success in the attempts. When the rubik block is disturbed to the maximum difficulty and requires 26 turns or more to solve, it only succeeds 20%.

However, Dactyl’s achievements have brought the project closer to the ultimate goal: developing learning-capable robots to perform complex and diverse tasks.

During the past 60 years, to perform the difficult tasks that humans do with their hands, it is required to design a unique robot for each mission. Dactyl is the first step in changing that. It is the forerunner of intelligent robots, better adaptable, more agile, capable of dealing with the clutter and complexity of the real world.

Reference: Motherboard