Researchers from the University of Cambridge and appliance maker Beko have trained a robot to assess the salt content of a prepared dish – scrambled eggs with tomatoes. Because the taste of food changes when it is chewed in the mouth, the robot tasting team set up a kind of artificial chewing process and measured the salt content at different stages of this process. Grzegorz Sochacki and his colleagues now describe the technical details in the Frontiers in Robotics & AI journal.
In this research work, people are still cooking. With their project, however, the British researchers are tackling a very central problem with food processors: the lack of feedback. Because the machines that fry burgers in more and more kitchens around the world, prepare pizza or prepare pasta dishes, such as the robot cooker from the Leipzig start-up DaVinci Kitchen, all work strictly according to a formula : cooking works for them like a linear algorithm – the machine knows what it needs, when, how it should be prepared and works step by step.
The developers worked around the difficulty that food is always different, mushrooms are bigger or smaller, broccoli has a little more or less water, and Hokkaido pumpkin can be ripe at different times by thoroughly testing recipes. “For every change, the overall recipe still has to work,” says Ibrahim Elfamarawy, CTO of DaVinci Kitchen. “So we tried every possible combination. There were hundreds of them.” Only recipes that prove to be sufficiently robust in this regard are used in practice.
Aitme, a Berlin start-up that has also developed a food processor primarily used in canteens, is taking a similar approach. “We tried to work with sensors,” says Julian Stoss, co-founder of Aitme. “The robot arm can be equipped with a camera, for example, to determine the degree of browning of food, for example. But that was taking too much time and ultimately not even necessary, because we know exactly how much energy we are using . from the performance data of the induction hob, put the food”.
Measure the salt content of food
However, Sochacki and his colleagues have in mind a machine that can learn to prepare recipes in such a way that customers rate them in the most positive way possible. For this to work, however, the machine needs a measure of the food’s seasoning. Since so-called electronic tongues can only determine taste after laborious sample preparation, but the measurement must take place during cooking, the researchers opted for sensors that measure the salt content of food. Such conductivity meters are now widely used in the food industry.
In order to improve the accuracy of their measurements, the researchers also decided to simulate the chewing process. For example, when we bite into a fresh tomato in midsummer, juices are released from the tomato and at the same time enzymes are released into the saliva. This changes our perception of the taste of tomatoes.
“Flavour map” for each sample
To mimic the human chewing and tasting process in their food processor, the researchers attached a conductivity probe, which acts as a salt sensor, to a robotic arm. They made scrambled eggs and tomatoes, varying the number of tomatoes and the amount of salt in each dish. To mimic the change in texture caused by chewing, the team placed the egg mixture in a blender and had the robot test the dish again. The different readings at different “chew” points resulted in a “taste map” for each sample.
Using this method, the machine was able to determine the amount of salt and the number of tomatoes in a sample with 95% accuracy.
In the future, the researchers want to improve the food processor so that it can taste different types of food and improve the sensors so that it can taste sweet or fatty foods, for example. They want to reproduce the effect of human saliva with artificial enzymes.