A robot chef able to watch and learn from cooking videos and recreate the dish itself. To develop a new machine able to understand the recipe after seeing it performed by a human being and make it in turn, were researchers at the University of Cambridge. Their results, published in the journal IEEE Access, demonstrate how video content can be a valuable source of data for the robotic chef that also allows him to invent a new recipe on his own.
Robot chefs have been studied for some time, but the models developed so far lag behind in terms of skills. Chefs in the flesh, in fact, can learn new recipes through observation, whether it is watching another person or a video on YouTube. "We wanted to see if we could train a robot chef to learn in the same way as humans, incrementally, by identifying the ingredients and how they combine in the dish," explains Grzegorz Sochacki, from Cambridge's engineering department, first author of the paper. The researchers devised eight salad recipes and filmed themselves preparing them.
To train their robot chef they then used a neural network that had already been programmed to identify different objects, including broccoli, carrots, apples, bananas and oranges. Using computer vision techniques, the robot analyzed each frame of the video and was able to identify different objects, such as a knife, ingredients and hands. In the 16 videos it watched, the robot recognized the correct recipe 93% of the time and also created a new one, adding it to its recipe book and preparing it. He is not yet able to understand the videos about food that go viral on social media, because because, for example, he would have difficulty identifying a carrot if it is covered by the hand of the user. "As these robot chefs become better and quicker at identifying ingredients, they may be able to use platforms like YouTube to learn a whole range of recipes," Sochacki said.