The robot named Minitaur learns to walk. (Photo: thepaper.cn)
An artificial intelligence (AI) system developed by scientists in the United States has taught a robot how to walk in just two hours, reports thepaper.cn.
The AI algorithm was developed by researchers from the University of California, Berkeley and the Google Brain, an AI research team at Google. Their paper "Learning to walk via deep reinforcement learning" has been published at the online preprint library arXiv.org.
In a video released by the researchers, a four-legged robot called Minitaur attempts to traverse an even slope. At first, Minitaur walks like a toddler, staggering and sometimes only able to march on the spot. But 18 minutes in, Minitaur is capable of walking forwards, although it's a bit unsteady on its feet. But within two hours, it had mastered walking up the slope using steady and rapid steps.
Deep reinforcement learning was the AI training technique used to train the robot. It learned how to move through the unfamiliar terrain by accumulating feedback provided by trial and error.
"To our knowledge, this experiment is the first example of a deep reinforcement learning algorithm learning underactuated quadrupedal locomotion directly in the real world without any simulation or pretraining," the researchers wrote. Or, put simply – it had to learn to walk before it could run.