Google’s AI just learned to walk. And it’s quickly moved past babysteps. It is particularly good at jumping.
Learning how to walk is a natural thing. Most people can do it, but they don’t remember learning how. We see babies go through the process, when they go from crawling to walking (and very, very quickly to running). All along they’re bumping into things, falling over, crying laughing. It looks goofy but their brains and bodies are forming new connections, a complex choreography of shifting neurons connecting balance, muscles, motor skills and vision.
Scientists have been trying for decades to program robots to move with such ease, but they just can’t seem to get it right.
Those are robots destined for the US Department of Defense, for either search and rescue operations in As that video shows, it’s not as easy as just programming in “left foot forward, now right foot.” And you can’t just wait for it to teach itself. Until now.
Meet Google’s self-taught walking, running, and jumping artificial intelligence program. Or, as we’re calling it, the Walkinator.
This video from TechInsider shows the Walkinator in action. It’s hilarious, looking a bit too excited, and sometimes flabbergasted, trying to make it to the end of the obstacle course. Even though it looks goofy, keep to things in mind: (1) It taught itself to walk. This has never been done before. (2) Next step is to put the program into physical robots. Fast robots. Robots that look like humans. And, maybe, ones that move like spiders.
The program has learned very quickly, a result that the researchers attribute to giving it more complex tasks. They conclude that the more complex the environment, the better the program is at learning how to conquer it. Since the real world might be the most complex and unpredictable environment of them all, how much faster will the program conquer us? One small step for Google’s AI, one giant leap into the abyss for mankind.