Google taught AI parkour, so it looks like our future robot masters will be pretty lame
Google's DeepMind wanted to push its AI to do more than just win all the time -- they wanted to teach it "sophisticated motor control" like jumping, crouching, and flipping.
As described in academic papers submitted last week, the company taught its AI body to climb over a wall, jump between planks, and get back up after falling. These are considered complex skills that allow the computer to navigate through a "rich environment."
In a blog post about the AI's new skills, the company said it trained the figures to handle diverse terrains (aka digital obstacle courses). It was able to develop skills to handle those obstacles "without receiving specific instructions, an approach that can be applied to train our systems for multiple, distinct simulated bodies."
The figure doesn't just walk forward, but can jump over obstacles, get over gaps and spaces, and basically become a parkour master. In the post, the AI team said their goal is to "produce flexible and natural behaviours that can be reused and adapted to solve tasks."
This Tweet is currently unavailable. It might be loading or has been removed.
Parkour!
Topics Artificial Intelligence
Sasha is a news writer at Mashable's San Francisco office. She's an SF native who went to UC Davis and later received her master's from the UC Berkeley Graduate School of Journalism. She's been reporting out of her hometown over the years at Bay City News (news wire), SFGate (the San Francisco Chronicle website), and even made it out of California to write for the Chicago Tribune. She's been described as a bookworm and a gym rat.
