A team of engineers created a robot for making omelets: from taking ingredients to preparing the final dish. And even improved his culinary skills to make it tasty. Researchers at the University of Cambridge, in collaboration with Beko, a home appliance manufacturer, have used machine learning to train their robot to handle highly subjective taste issues, according to IEEE Xplore.
For decades, a robot that knows how to cook has been the goal of futuristic scientists and scientists. With the development of artificial intelligence methods, commercial companies have created prototypes of chef robots, although at present none of them are commercially available, and in terms of skill level they are significantly behind their fellow humans.
Learning how to cook and prepare a robot is a difficult task because it must solve complex problems associated with manipulating the robot, computer vision, sensing, and human-robot interactions, and it must also include a consistent end product.
In addition, the taste differs from person to person – cooking is a quality task, while robots tend to excel in quantitative tasks. Since the taste is not universal, universal solutions do not exist. Unlike other optimization tasks, special cooking tools must be developed for cooking robots.
Other research groups trained robots to cook cookies, pancakes, and even pizza, but these robotic cooks were not optimized for many subjective variables related to cooking.
Omelet is one of those dishes that are easy to prepare, but difficult to cook well. Researchers thought it would be the perfect test to enhance the robot chef’s capabilities and optimize taste, texture, smell, and appearance.
In partnership with Beko, scientists taught their robot chef how to cook an omelet, from breaking eggs to putting a finished dish on a plate. The work was done in the Cambridge engineering department using a test kitchen supplied by Beko plc and the Symphony Group.
The machine learning technique developed by the team uses a statistical tool called Bayesian inference to squeeze out as much information as possible from a limited number of data samples, which was necessary to avoid omelets overflowing with people-tasters.
The problem that researchers have encountered is the subjectivity of human taste. People do not give absolute measures very well and usually give relative measures when it comes to taste. Therefore, it was necessary to set up a machine learning algorithm, the so-called batch algorithm, so that people-tasters could provide information based on comparative assessments, rather than sequential ones.
The results show that machine learning can be used to produce quantitative improvements in food optimization. In addition, this approach can be easily extended to several robotic cooks. Further research should be conducted to explore other optimization methods and their viability.