Clicking over to the homepage of WikiHow this morning, I found entries with instructions on melting marshmallows, becoming a Bollywood actress and determining the sex of a turtle—if there is a question out there, someone has almost certainly submitted an answer to Wikihow. And as part of an ongoing project, researchers in Germany are now using that vast database to teach robots to understand human language, specifically human language about breakfast preparation.
The goal of the RoboHow project is to develop robots that can follow simple human instructions. The thought being that if robots are ever to become commonplace in our lives, they won’t do us much good if we all need a programmer on-call who can rewrite a robot’s code every time we need them to perform a new task.
So the researchers had their robot, PR2, study many simple Wikihow articles on cooking and “watch” live demonstrations from people performing tasks like pancake flipping. The pancake flippers wore gloves that tracked all their movements, movements they then loaded into the robot’s database.
The results are either impressive or terrifying, depending on your view of robots and the possibility that they will one day rule us all. Through a process called semantic parsing, PR2 was able to take the words in a simple set of instructions and translate them into actions.
So far, successful experiments have been limited to rudimentary activities—finding and squeezing premade pancake batter onto a griddle and cooking it. But we may be able to call it now: The rise of the machines began over breakfast.
Check out a video of PR2 in action below.