Pizza and Robots? How Robots Were Able to Make Pancakes and Pizza After Reading Wikihow Articles
When you want to learn how to change a tire or whip up some delicious strawberry buttermilk frosting, you likely turn to YouTube or a how-to wiki on the web.
Robots it turns out, can be programmed to do the same thing. A team of researchers in Germany taught a robot to turn to the web for instructional videos and step-by-step tutorials to complete simple tasks, like making pancakes and cooking a pizza. The incredible part of this is how the researchers programmed the robots to understand what they were seeing and reading. Engineers have to translate our world into language robots can understand. That requires breaking down sentences and actions in videos into mathematical equations which robots can use to tease out patterns in their “brains.”
The project is part of a European Union-wide collaboration called RoboHow, which wants to give robots general knowledge about the world. (Obviously, they’re not worried about Skynet). Once a RoboHow bot has mastered how to make pancakes or pour coffee, its know-how is uploaded to an online database dubbed Open Ease. That knowledge bank is then accessible to all robots connected to it. It’s a bit like an app store for robotic tasks. If you need your robot to dispense medications, you don’t have to put together all the instructions yourself. You can just query Open Ease.
RoboHow is part of a movement called cloud robotics, a concept popularized in 2010 by Google self-driving car expert James Kuffner. The idea actually goes back to the 90s, when Masayuki Inaba wrote a paper entitled, “Remote-brained Robots.” These machines’ brains would be hooked up to supercomputers spread across the internet.
In 2011, another Euro-based cloud robotics initiative launched called RoboEarth; it’s open source, and people interested in plugging their bots into it can access the source code online. There’s also RoboBrain, a collaboration between several U.S. universities, which is an online service similar to RoboEarth that booted up last year. It taps into more than 100,00 different data sources that feed into artificial intelligence systems tuned to help robots learn simple tasks like making salads or pouring coffee.
For now robots are reliant on human crowds. They are using our online tutorials, written over decades. RoboBrain reaches out to humans directly by using Amazon Mechanical Turk, an inexpensive virtual workforce, as well as volunteers, to write instructional data for robots on the cheap. By visiting the Robobarista webpage, people can help robots learn to use a toaster or microwave, for instance.
The hope of all these projects is our robot pals will be more adept at helping us around the house one day. But the dawn of Rosie the Robot from The Jetsons is still a long ways off. Making an espresso took this robot a whopping eight hours. The pancake-flipping bot is also super slow. Nobody wants to wait that long for a caffeine kick or breakfast.
For more information regarding Robots and Pizza, visit Fusion.net.