can be your best friend!
Many robotic researchers have focused on enabling robots to solve the Rubik’s cube quickly on computers. A little more than a year ago, OpenAI managed to use artificial intelligence to help a neural network learn how to solve the Rubik’s cube, and demonstrated it through a robotic appendage resembling the human hand. While the results are humble, and the robotic technology cannot solve the cube quickly, it is still a landmark achievement in the area of robotic sciences.
In this article, we take a brief look at the technology used, and what it means for the cubing community.
Brief history about OpenAI and its Rubik cube robot solver
OpenAI is a company founded by Elon Musk, Sam Altman, and others. After Musk resigned from the organization, Microsoft funded the company and its projects. The OpenAI has been working on robotic models that learn new tasks for some time now. The company’s objective has been to train a human-like robotic hand to learn how to solve the Rubik’s cube on its own.
The researchers at OpenAI tasted initial success when they solved the cube in simulation in July 2017. Back in 2018, they explained how Dactyl, a system for object manipulation, could learn to manipulate a single block. Dactyl is the robotic system that enables objects such as Shadow Dexterous Hand to move and manipulate objects. The robot has since learned how to solve a Rubik’s cube single-handedly.
The robot’s ability to learn how to solve the cube with a single hand is a milestone in the field of robotic appendages and artificial intelligence. The system’s achievements translate to something bigger: robots can now learn to perform tasks based on either incentives or trial and error, much like how humans learn to perform tasks.
The technology behind the robot solver
The robot’s learning strategy mimics human learning and is called reinforcement learning. Reinforcement learning is a type of deep learning that is based on incentives. Scientists also used a technique called Automatic Domain Randomization (ADR) to help the robot learn randomly during the simulated learning process.
Here is some specific trivia about the process:
- The robotic hand was designed to mimic that of a human and is operated by a pair of neural networks.
- These networks are trained exclusively during simulation in a random way, which makes the robot apply what it learns while solving the cube.
- The robotic hand was exposed to a variety of stimuli that it had never seen before, and it actively adapted to that stimulus.
- Researchers prodded the system with a stuffed giraffe in an effort to test its reaction. It could actively change the way it was solving the cube in response to that.
Why this Rubik’s cube robot solver by OpenAI matters
The technology is still nascent, and robotic systems like the one OpenAI has been working on cannot solve the Rubik’s cube at speeds that computers can. They also lack the finesse and ease with which human hands can manipulate the cubes. In fact, news articles and blogs have reported that OpenAI’s system took a few minutes to solve the cube in demonstrations.
However, what makes this scientific development remarkable more than the robot’s speed is its ability to learn how to solve a cube on its own. In other words, developments such as these hold promise, especially in the field of health sciences.
The implications of this technology are huge. Robotic science is an emerging field in the health industry, and experts believe AI-enabled robots can one day help disabled people to live more adaptive lives. In addition, robotic appendages such as the one that solved the Rubik’s cube can someday help treat health conditions and assist in surgeries. Scientists at OpenAI also believe that this lays the groundwork for building general-purpose robots that can be used in a number of situations.
What the OpenAI Rubik’s cube solver means for the cubing community
OpenAI’s Dactyl is certainly not the best Rubik cube robot solver. However, it heralds a new era of robots that are built with artificial intelligence and use deep learning to perform new tasks independently. Speed cubers will probably not benefit from this particular robot solver directly. However, it corroborates our theory that solving the Rubik’s cube may develop an interest in other STEM (science, mathematics, engineering, and technology) related fields.