AI Models Enter the Kitchen: wan AIChef Launches Kitchen Intelligence System

As large AI models become standard in consumer electronics like smartphones and computers, the home appliance industry is also beginning to see their application.

At the AGIC Conference held at the Shenzhen World Exhibition & Convention Center, wan AIChef unveiled its first Kitchen Intelligence System powered by the self-developed Zhurong Cooking Model. Unlike traditional microwaves, this system is equipped with an NVIDIA GPU and a 13-megapixel high-definition camera to identify ingredients. It automatically controls temperature and heating time based on advanced model algorithms.

"Since 2000, innovation in the home appliance industry has hit a bottleneck. We wanted to identify pain points in the kitchen scenario and solve them through technological innovation," said Wang Ning, Operations Director of wan AIChef.

Wang Ning explained that traditional kitchen appliances operate on fixed programs, requiring user input and lacking the ability to personalize or learn user habits. Even high-end appliances are limited to recipe selection and executing fixed programs, still relying on user input. Based on this logic, wan AIChef aimed to develop a product that could fully replace human cooking, acting like a human chef to achieve true machine autonomy.

The emergence of ChatGPT and large AI models in early 2023 made this vision possible. The technical logic of ChatGPT demonstrated a feasible path of 'achieving breakthroughs through scale.' By continuously increasing computing power and data, OpenAI has shown that AI model performance can keep improving, with no visible 'ceiling' in sight.

"We decided to apply this capability to innovate microwave cooking. Ordinary microwaves can't be used for cooking because heating is uneven. The first problem we solved was how to make microwave heating uniform, which involved adding numerous sensors and stirrers to control the microwave's behavior and logic," said Luo Xingchen, Manager of Applied Algorithms at wan AIChef, in an interview with Titanium Media APP.

For the self-developed cooking model, the biggest challenge was data training, as there was almost no readily available cooking big data on the market. To obtain high-quality cooking data, wan AIChef established a team of hundreds of cooking experimenters. Using tools like temperature probe arrays, RGB cameras, power devices, and odor sensors, they collected data on temperature, color changes, and odors during the cooking process. With these sensors, data could be captured every second throughout the cooking process. Each dish has its own temperature requirements—for example, eggs need to reach 72°C and maintain that temperature for three minutes to solidify the protein and kill E. coli. After data annotation and cleaning, the model was pre-trained, and parameters and algorithms were fine-tuned.

Ultimately, wan AIChef developed the end-to-end Zhurong Cooking Model, which integrates ingredient perception, cooking decision-making, and device control into a single system, enabling fully automated cooking.

wan AIChef stated that its Intelligence System, developed over four years, was trained on 1 million iterations of 1,000 dishes created by over 100 renowned chefs. It can recognize 1,500 home-style dishes and perform automated cooking. Additionally, through the combination of sensors and stirrers, the system achieves "different temperatures in the same cavity," allowing the machine to control temperatures in different zones based on the ingredients.

To precisely control cooking, wan AIChef equipped its system with the Egg Nest 4D Ultra-Sensitive Temperature Control System. Using high-resolution thermal imaging sensors and intelligent stirrers, it captures real-time temperature changes on the surface of ingredients and automatically adjusts cooking parameters to ensure optimal results for different ingredients in the same pot.

It is reported that wan AIChef will also launch an unmanned kitchen commercial solution, capable of serving 120 people simultaneously and operating 24/7. By 2026, the company plans to cover 10,000 core locations, collaborate with 300 food companies, develop 2,000 dishes, and drive 15 billion yuan in food sales.

Source: TMTPost(钛媒体), September 5, 2024.