It is reported that the experiment of ``robot chef'' that can reproduce the dish just by watching the cooking video was successful
A research team at the University of Cambridge has developed a ``robot chef'' that reproduces and cooks a recipe by showing a cooking video about one of the recipes to a robot that has learned multiple recipes. In addition, it is reported that this robot chef not only reproduces the recipe but also devised the original recipe.
Recognition of Human Chef's Intentions for Incremental Learning of Cookbook by Robotic Salad Chef | IEEE Journals & Magazine | IEEE Xplore
Robot 'chef' mimics recipes from food videos
Unlike humans who can learn new recipes by observing others cooking or watching cooking videos, it is very difficult to program a robot to cook while learning multiple recipes. It is said to be costly and time consuming.
Therefore, a research team led by Grzegorz Sochacki at the University of Cambridge devised eight simple salad recipes and created a cooking video. We then trained a robot chef using an open-source neural network that can individually recognize a variety of objects, including vegetables and fruits such as broccoli, carrots, apples, bananas, and oranges.
The robot chef applies
``By correctly identifying the ingredients and actions held by humans, the robot chef was able to determine which of the eight recipes was being cooked,'' Sochacki reports. .
In a robot chef experiment, when a robot observes a cooking video in which a human holds a knife in one hand and a carrot in the other, the robot guesses that the carrot held in the hand will be chopped with the knife. , move the device that cuts the material. By interpreting and understanding human behavior, robots have successfully identified specific recipes cooked in cooking videos.
Below is a video of the robot chef in action.
Robot 'chef' learns to recreate recipes from watching food videos-YouTube
Mr. Sochacki shows the apple to the robot.
In addition, it shows how Mr. Sochacki cuts apples finely.
The robot chef who understood 'cutting the apple' put the apple in the device and succeeded in cutting it together with the carrot shown separately.
According to the research team, of the 16 cooking videos watched by the robot chef, the robot chef identified the correct recipe with a high accuracy of about 93%, even though it detected only about 83% of human actions. bottom. In addition, it has been reported that the robot chef did not regard each recipe as a new recipe, even if it was the same recipe or a human made a mistake during the video, and judged that it was all the same recipe.
Furthermore, when the research team made the robot chef learn a cooking video of a new ninth salad recipe that is not in the learned recipe, the robot succeeded in reproducing this recipe correctly.
``It's amazing how we could recognize a recipe with two chopped apples and two chopped carrots as the same recipe with three chopped apples and three chopped carrots,'' Sochacki said. ' said.
The cooking videos used by Sochacki's research team are different from general cooking videos that include cut editing and visual effects. is used. ``It is difficult for this robot chef to recognize cooking videos that hit social media such as YouTube,'' Sochacki said. On the other hand, 'If robot chefs can identify ingredients in videos more clearly and faster, they may be able to learn different recipes using sites such as YouTube in the future.' There is.'
ScienceBlog , a scientific information site, said, ``This report by a research team at the University of Cambridge shows that the content of cooking videos, which are abundant in the world, will be a valuable data source for the automation of cooking, and will be used by robot chefs in the future with higher accuracy. It may be possible to deploy cheaply,' he said.
Related Posts:
