Researchers at the University of Cambridge (UK) have created a robot "chef" that learns recipes by watching videos of cooking various dishes. The respective article was published in the IEEE Access scientific journal.
The programming of this robot is based on the simulation machine learning method. The robot looks at what the person is doing in the video and then tries to reproduce what it saw. Researchers have explained that, as a rule, people improve their cooking skills in the same way.
As part of the research, eight simple salad recipes were downloaded into the robot. Then they showed it the cooking videos of those dishes. With hand movements and ingredients, the robot determines the recipe and starts repeating the stages of preparation.
Out of 16 videos viewed, the robot correctly identified the recipe 93 percent of the time. Moreover, it was able to recognize the specific actions of the human chef only in 83 percent of cases.
The researchers noted that so far this robot is unable to replicate the dishes of famous chefs, whose videos often become popular on social media. The problem is that this robot is unable to identify fast human movements. However, researchers believe that as technology improves over time, this robot will be able to learn to cook on YouTube.