A humanoid robot hand developed by artificial intelligence research lab OpenAI has learned to solve a Rubik’s cube without assistance.
The 24-joint robot – named Dactyl – achieved the feat in around four minutes using a new training technique developed by OpenAI that could lend itself to building a new generation of more dexterous and adaptable autonomous machines.
The training method – called automatic domain randomisation (ADR) – allowed Dactyl to learn new tasks using virtual simulations that gradually injected simulated hindrances and complications into the process to force the robot to adapt in order to complete the cube.
According to OpenAI, ADR was used to give the robot consistent, intricate dexterity that can handle changes in the environment that go beyond what a computer could predict and simulate.
“One of the parameters we randomise is the size of the Rubik’s cube,” said Peter Welinder, a research scientist and robotics lead at OpenAI. “ADR begins with a fixed size of the Rubik’s cube and gradually increases the randomisation range as training progresses. We apply the same technique to all other parameters, such as the mass of the cube, the friction of the robot fingers, and the visual surface materials of the hand.”
Using machine-learning and robotics to solve a Rubik’s cube has been achieved before. In March 2018, a machine developed by engineers at MIT solved a cube in just 0.38 seconds. Later that same year, Japanese YouTube channel Human Controller developed its own self-solving Rubik’s cube using a 3D-printed core attached to programmable servo motors.
However, what’s significant with OpenAI’s effort is the use of a multi-purpose humanoid robot design, rather than a machine specifically developed to handle a Rubik’s cube and nothing else.
When the model was tested in a real-world environment, the robotic hand it piloted managed to solve the Rubik’s cube while wearing a rubber glove, with a fingers bound together and even when researchers tried to knock the cube to the ground using various objects.
“Plenty of robots can solve Rubik’s cubes very fast. The important difference between what they did there and what we’re doing here is that those robots are very purpose-built,” said Welinder.“Obviously there’s no way you can use the same robot or same approach to perform another task.
“The robotics team at OpenAI have very different ambitions. We’re trying to build a general purpose robot. Similar to how humans and how our human hands can do a lot of things, not just a specific task, we’re trying to build something that is much more general in its scope.”
OpenAI believes the training technique could potentially enable industrial robots, drones and other autonomous machines to dynamically modify their behaviour when encountering an unforeseen obstacle or hindrance – an important step towards realising self-learning AI.
The Rubik’s cube solver is one of the first major projects OpenAI has detailed since signing a US$1bn partnership with Microsoft in July. Under the deal, Microsoft will provide OpenAI with capital and cloud infrastructure to support research in exchange for intellectual property access.
