Science & Technology, UK (Commonwealth Union) – Scientists from the University of Cambridge have programmed a robotic chef with a ‘cookbook’ that has 8 simple salad recipes. After viewing a video of an individual displaying one of the recipes, the robot was able to know which recipe was in preparation and carry it out.
On top of that, the videos provided assistance to the robot to incrementally provide more details to its cookbook. When the experiment was completed, the robot had brought in a 9th recipe taking its own initiative. The findings that appeared in the journal IEEE Access, revealed the way video content is a valuable component of data for automated food production, and can make it possible for the simpler and more economical deployment of robot chefs.
Researchers indicated that the featuring of robotic chefs continuously in science fiction for decades has made it look simple, however, in reality, cooking is a difficult task for a robot. Many commercial firms have put together prototype robot chefs, even though none of them are commercially available right now and they are far behind their human counterparts in regards to the skills they display.
With the availability of online videos on cooks and many reviews online to further enhance their skills to perfect cooking skills, researchers indicated that programming a robot to make a range of dishes will take up a lot of time and is not so economical.
“We wanted to see whether we could train a robot chef to learn in the same incremental way that humans can – by identifying the ingredients and how they go together in the dish,” explained Grzegorz Sochacki from the University of Cambridge, Department of Engineering, who is the papers, 1st author.
Sochacki is a Ph.D. candidate at Professor Fumiya Iida’s Bio-Inspired Robotics Laboratory, where his colleagues formed 8 simple salad recipes and videoed themselves producing them. This was followed by the utilization of a publicly available neural network for the training of the robot chef. The neural network was already programmed to take note of a range of various objects, this consisted of the fruits and vegetables utilized in the 8 salad recipes that had broccoli, carrot, apple, banana as well as orange.
With the application of computer vision methods, the robot evaluates each frame of video and could mark various objects and features, like a knife and the ingredients, along with the human demonstrator’s arms, hands, and face. The recipes and the videos were changed to vectors and the robot carried out mathematical operations on the vectors to find out the similarity between a demonstration as well as a vector.
The accurate identification of the ingredients and the actions of the human chef, provided the robot the capability to seek out which of the recipes was being made. The robot had the ability to conclude that when the human demonstrator had a knife in their hand and a carrot in the other hand, The robot had concluded that the carrot would be sliced.
From the 16 videos viewed, the robot knew the accurate recipe 93 percent of the time, even though it just noted 83 percent of the moves of the human chef. The robot was further capable of identifying those slight variations in a recipe, like making a double portion or normal human error, which were variations with no new recipe. The robot further accurately recognized the demonstration of a new, 9th salad, which was an addition to its cookbook, and made it.
Sochacki indicated that it was amazing the amount of nuance the robot had the ability to identify. “These recipes are not complicated they are simply chopped fruits and vegetables, but there was effectiveness at recognizing, for example, that 2 chopped apples and 2 chopped carrots were the same recipe as 3 chopped apples as well as 3 chopped carrots.
