OpenAI, an Elon Musk-backed nonprofit research lab based in San Francisco, recently created a new algorithm that teaches robots to learn the way humans do. The system is based on a one-shot imitation learning technique that guides a Fetch robot through an action with just one demonstration.
A person would first perform a manual task in virtual reality. The robot would see the movement through a neural network made entirely of simulated images and then mimic the action.
“Imitation allows humans to learn new behaviors rapidly. We would like our robots to learn this way too,” says Josh Tobin, a technical staff member at OpenAI.
Ultimately, the goal is to have an AI that can learn new human-like behaviors quickly and adjust its movements as its surroundings change.