When large models became standard equipment for consumer electronics such as mobile phones and computers, application cases for large models began to appear in the home appliance industry.
At the agiC conference held at the Shenzhen International Convention and Exhibition Center, Wonder Kitchen released the first kitchen intelligence equipped with Zhurong's self-developed large model of cooking. Different from traditional microwave ovens, this kitchen intelligence is equipped with NVIDIA's GPU, which can identify ingredients through a 13-megapixel high-definition camera, and automatically control temperature and heating time based on model algorithms.
"Since 2000, you will find that innovation in the entire home appliance industry has hit a bottleneck. We hope to find pain points in the kitchen scene and solve them through technological innovation."
Wang Ning, operations director of Wonder Kitchen, said that around the different logics of user initiative and machine initiative, traditional kitchen utensils are mostly operated by fixed procedures and lack the ability to personalize and learn user habits. All are user initiatives. Some high-end kitchen appliances are also limited to the selection of recipes. Cooking dishes based on fixed procedures is also the initiative of the user. Based on this logic, Wonder Kitchen is trying to develop a product that can completely replace human cooking, cooking like a human chef, and realizing true machine initiative.
In early 2023, the emergence of ChatGPT and large models made this possible. It can also be seen from the technical logic of ChatGPT that the large model has found a feasible path to "make miracles happen with great effort". By continuously increasing computing power and data, OpenAI can continuously improve the model's AI performance, and currently the industry has not seen a "ceiling" for this capability improvement.
"So, we decided to put this capability into the innovative upgrade of microwave cooking. Ordinary microwaves cannot be used for cooking because the heating cannot be uniform. The first problem we solved was how to make microwave heating uniform, here It involves adding a large number of sensors and stirring blades to control the methods and logic of microwaves," Luo Xingchen, manager of the Application Algorithm Department of Wonder Kitchen, told Titanium Media APP that the most difficult thing for self-developed large-scale cooking models is data training. There is currently almost no directly available cooking big data on the market.
In order to have high-quality cooking data available, Wonder Kitchen has established a cooking experiment team of several hundred people, using tools such as temperature probe 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. Through a large number of sensors, data during the entire cooking process can be collected every second. Each dish has its own temperature. For example, eggs reach 72 degrees, which is maintained for three minutes to allow the protein to solidify and E. coli to be killed. After data annotation and cleaning, model pre-training is performed, and finally the model parameters and algorithm are adjusted.
In the end, Wonder Kitchen developed an end-to-end Zhurong cooking model, which can integrate the perception of ingredients, cooking decisions and equipment control into one system to achieve automated cooking.
Wonder Kitchen said that through four years of research and development, more than 100 famous chefs developed recipes, and 1 million trainings on 1,000 dishes, the Wonder Kitchen intelligent agent can recognize 1,500 kinds of home-cooked dishes and complete automatic cooking. At the same time, through the superposition of sensors and stirring blades, the intelligent body can also achieve "different temperatures in the same cavity" - the machine controls the temperature of different areas according to different ingredients.
In order to control the cooking heat, Wonder Kitchen is equipped with the Egg Nest 4D super-sensing temperature control system, which uses high-resolution thermal imaging temperature sensors and intelligent stirring blades to capture the temperature changes on the surface of the ingredients in real time, and automatically adjusts the cooking parameters to ensure that different ingredients are cooked properly. The best cooking results can be achieved in the same pot.
It is reported that Wonder Kitchen will also launch an unmanned kitchen commercial solution, which can provide 120 people with dining at the same time and 24-hour supply at any time. It is expected to cover 10,000 core points by 2026. At the same time, Wonder Kitchen will unite 300 food companies to develop 2,000 dishes and drive 15 billion food sales.