Artificial intelligence seems to be getting smarter each day, often accomplishing things that humans can’t. One such accomplishment comes from a robotic arm that has taught itself how to solve a rubik’s cube.
Artificial intelligence research group OpenAI, which was founded by tech mogul Elon Musk, has been working on enhancing the skills of their neural networks in different ways. Their latest project, a one-handed robot showcased unbelievable intelligence and dexterity when it taught itself how to solve a rubik’s cube using only one hand.
The pair of neural networks that control the arm were instructed to make all the sides of the cube the same colour, with no further instructions. The AI then used computer simulation to learn how the different hand movements affected the cube, along with how gravity affected the cube as the arm manipulates it. Dactyle, (the name of the robot arm) was then able to transfer those simulations into real life, with amazing accuracy.
An advantage to using this form of machine learning is that multiple simulations can be run at the same time. Meaning this method of training has allowed the AI to gain thousands of years worth of experience in a very short period. As the robots skills improved with each simulation, the researchers ramped up the difficulty.
The research team increased the difficulty by placing a rubber glove over the hand to hinder its movement, physically moving around the rubik’s cube as the arm tried to solve it, and even went so far as to wrap rubber bands around the robot’s fingers. This technique of incrementally upping the difficulty is called ADR (Automatic Domain Randomization) and aims to teach the AI network how to deal with an ever changing environment, like rubber bands tired around its fingers for example.
The significant part about OpenAI’s effort is the fact they made use of a multi-purpose robot hand that anyone can purchase, instead of a machine designed specifically designed to solve a rubik’s cube.
OpenAI say this type of research is crucial if scientists want to improve on multi-purpose robots that have dexterity on par with humans, although handling items that are more complex and unpredictable than a rubik’s cube will require far more research.