OpenAI taught its robot to solve a rubik's cube one-handed
The AI startup OpenAI has trained a robotic hand to solve a rubik's cube. The company described the accomplishment as a step toward adaptable dexterity in machines.
"We believe that human-level dexterity is on the path towards building general-purpose robots," they wrote in a blog post.
To train the robotic hand, engineers used a new kind of difficulty-setting scale they call Automatic Domain Randomization (ADR). ADR defines the size of the rubik's cube and any interferences the hand has to deal with during training. Interferences included tying two fingers together, forcing the hand to wear a glove, or pushing an object into the hand in the middle of the process.
Such variances in environment mean that the AI is able to perform dexterous tasks under stress, rather than only in controlled spaces. According to OpenAI, this "prepares the network to transfer from simulation to the real world since it has to learn to quickly identify and adjust to whatever physical world it is confronted with."
During training, the team also relied on the same reinforcement learning algorithm that it used to train AI agents to play and win against the world's best players in Dota 2, the extremely popular video game.
If you want to see where human-machine competition is fiercest, look to games. Complex, rules-based games present the ideal environment to train AI systems, where the industry's top startups compete against the world's best players.