Google’s DeepMind Teaches Itself Parkour

Jul 12, 2017 at 4:36 PM ET

Now, Google is teaching its robots to scale city walls.

To create robots that can navigate the real world, Google’s DeepMind artificial intelligence is learning “parkour,” the visually stunning sport of navigating real-world urban environments using acrobatic jumps and gymnastic balance.

In what looks like the weirdest video game ever, researchers gave the AI a sense of balance and the desire to move forward. Then, they let the programming loose in a “Super Mario”-like environment.

Through reinforcement learning, DeepMind programmed every move for the AI, such as grabbing ledges with its knee and knowing how far to jump to clear gaps.

Projects like this could teach our future robot co-workers how to walk up stairs, navigate an office, and ultimately take over our jobs and lives.

But for now, no worries when they look like this.