It’s said, “Cooking is an art, but all art requires knowing something about the techniques and materials.” And why learn such tedious techniques when we can train a machine to do so! Scientists from University of Maryland, NICTA, and Australia’s Information and Communications Technology Research Center of Excellence are finally on the verge of launching robot chefs. But the more astonishing fact is that there cookery training isn’t any piece of complicated program rather these Robotic chefs learn cooking like humans do by looking at YouTube videos. Who thought YouTube can teach robots too?
As YouTube videos has diverse backgrounds ,colors , lights and other such effects , only few clear and unambiguous videos can be used to train a robot. 88 such selected videos, which when viewed by robots are first broken down and converted to understandable commands and then learned.
This remarkable invention is accomplished by using the Deep learning technique of Artificial Intelligence. These scientists used Convolutional Neural Networks (CNN) to identify, observe and imitate activities like holding a frying pan and differentiating between a potato and a beetroot.
In CNN a mathematical function, called an artificial neuron acts like a living neuron. It uses two visual recognition modules. One focuses on hand gestures of the chef in the video to learn about the grip like a tight vacuum grip or a gentle grip. Other focuses on the object like some vegetable, the cook is holding or the cream cook is blending. All these information will be consolidated within the robot chefs.
The next step is Identification of common cooking scenario. Common Scenarios like cutting, rotating, chopping etc. once identified can be copied easily by breaking these scenarios into series of simple steps. This series is called a “sentence”. This great invention will be put forth on 29th annual conference of the Association for the Advancement of Artificial Intelligence. Till then, happy cooking!
Source: The Gizmodo