Big models are taking over the kitchen! How do kitchen appliance companies teach AI to cook?
Author：Eve Cole
Update Time：2025-02-09 09:36:01
While people are still discussing how AI large models can change family life, a large number of kitchen appliance companies have launched AI large models one after another, aiming to capture the most personalized cooking scenes, with great ambitions. In the past, the intelligence of kitchen appliances usually revolved around intelligent interaction and standardized cooking. However, due to the inability to improve cooking efficiency and the inability of dishes to meet personalized tastes, consumers called it "tasteless". However, Yangcheng Evening News’ all-media reporters discovered that this year’s wave of AI kitchen appliances has made the products soar in terms of practicality, directly realizing the functions of self-selected cooking methods and dynamic temperature control, and has the potential to take over the kitchen. New brands enter the market to develop kitchen intelligence. On November 9, a cooking experience event was held in Liede, Guangzhou. There were no chefs on hand, just a lot of cooking equipment that looked like microwaves. Neighbors put the cut dishes into the device, and the device can quickly identify the dishes and match the cooking method. In a few minutes, the hot dishes are freshly baked.
These cooking equipment are AIChef kitchen intelligence developed by the Wonder Kitchen brand of Shenzhen Wonder Kitchen Technology Co., Ltd. (hereinafter referred to as " Wonder Kitchen"). In August this year, Wonder Kitchen launched this kitchen intelligent body at the 2024 agiC Shenzhen (International) General Artificial Intelligence Conference, becoming another case of kitchen appliance companies developing large-scale AI models. Luo Xingchen, head of the Algorithm Department of Wonder Kitchen, told the Yangcheng Evening News all-media reporter: “Although cooking scenes are very common, there is no ready-made cooking big data in the industry. Therefore, to develop a large cooking model, we must first combine cooking knowledge with data. The form settles." In order to overcome this problem, Wonder Kitchen established an experimental team of several hundred people, using tools such as temperature detector arrays, RGB cameras, power devices, and odor sensors to collect the temperature and color of each dish during the cooking process. Change, smell and other dimensions of data. After data annotation and cleaning, model pre-training is performed, and finally the model parameters and algorithm are adjusted. In the end, they developed the Zhurong cooking model, which integrates the perception of ingredients, cooking decisions and equipment control into one system, which not only realizes automated cooking, but also adjusts it according to the user's taste preferences. Taking the automatic sensing of ingredients and recipes as an example, Luo Xingchen told reporters: "The device will first identify it based on the local database and identify it within 1.5 seconds; if the dish cannot be identified locally, the data will be pushed to the cloud. Let the large model in the cloud perform secondary recognition, which is equivalent to combining the cloud and edge to identify the dish." Cooking data is the key to developing "cooking skills" in large models. Since this year, faced with pressure on the industry, kitchen appliance companies have invariably extended their tentacles to large AI models. At the end of October, FOTILE released the world's first AI healthy cooking system Healthy CookingGPT and the Ruiyin series of full-scenario kitchen appliances equipped with this "intelligent brain"; and in June this year, Robam Electric launched the world's first AI cooking system The model, the large model of the God of Cookery, will be open for internal testing in October this year and will be officially launched in December.
Just like cooking requires ingredients, the development of large models also requires a large amount of data training. Before developing large-scale cooking models, kitchen appliance companies without exception have to overcome the large-scale but very fragmented collection of cooking data. Zhou Haixin, vice president of Robam Electric, believes that the most important things in building a large model of the cooking industry are data, algorithms and scenarios. "Our knowledge base integrates cooking culture, recipe nutrition, health management and electrical product knowledge, and also covers laws, regulations, policies and standards in related fields, ensuring the comprehensiveness and practicality of the data." Fotile also built Kitchen appliance knowledge map including dietary health knowledge map, healthy meal combination plan, ingredient storage guidance, cooking techniques, cooking curve database, etc. An all-media reporter from the Yangcheng Evening News found that several cooking models that have been released so far are more down-to-earth in function; automatically identifying ingredients and adjusting cooking methods in real time have become the main focus of these cooking models. FOTILE's Healthy CookingGPT is equipped with an intelligent cooking system and AI smart eyes, which can not only directly identify ingredients, but also continuously monitor the cooking status of food, intelligently adjust the cooking time, and ensure that the ingredients reach the ideal maturity and best taste. Boss’s God of Cookery model can quickly generate personalized recipes based on the user’s health status, nutritional needs and dietary preferences, and supports one-click cooking functions. "Looking to the future, we will further develop technology to recognize dishes through pictures, restore cooking steps, and realize the reproduction of dishes." Zhou Haixin said. The imagination space is even more "beyond the dining table". The large-scale cooking model must fully take over the kitchen. It must not only rely on massive databases to improve "cooking skills", but also use cooking as a link to find scenes and channels to connect consumers. FOTILE is targeting the public's demand for healthy diet. Facing the dietary needs of thousands of people, Healthy Cooking GPT can use AI algorithm support to provide users with personalized healthy diet based on the basic personal information, sports and health data entered by the user. suggestion. For example, Fotile launched the first healthy meal application specifically designed for pregnant women.
Wonder Kitchen started from the big model and conceived a new business model. In addition to reaching consumers from the C-side, it also plans to extend the B-side business, through intelligent transformation, to help traditional hardware and food manufacturers achieve product upgrades and service innovations, and jointly explore the new blue ocean of digital intelligent cooking. At the same time, it is also planned to launch an unmanned kitchen commercial solution, which can satisfy 120 people dining at the same time and provide 24-hour supply at any time. The relevant person in charge of Wonder Kitchen told the Yangcheng Evening News all-media reporter that with the further popularization of its smart devices in the future, they also hope to collaborate with food manufacturers to penetrate safe and traceable meat and other products into thousands of households through this channel. household.
