In a recent interview, Yang Jiancheng, founder and chairman of Xianglu Robotics, highlighted a fundamental shift in the culinary automation industry. He pointed out that traditional cooking robots relied heavily on pre-programmed standard operating procedures, essentially memorizing recipes without any real grasp of the cooking process. However, with the launch of CookingMuse, a multimodal AI foundation for cooking, these robots can now visually perceive real-time changes within the wok and transition into true comprehension-based cooking.

According to Yang, CookingMuse integrates advanced visual perception, cooking process data, and intelligent decision-making capabilities. It continuously monitors the state of ingredients in the wok, dynamically adjusting heating power, stir-fry duration, mixing speed, and seasoning quantities based on live feedback. This system genuinely understands the current state of a dish, the reasons behind that state, and the optimal next steps for preparation.

Using "shredded potatoes" as an example, Yang illustrated the limitations of conventional robots, which require preset parameters for meat quantities, potato variety, cutting specifications, and even target texture preferences. All these factors are traditionally executed through a rigid, time-based sequence of mixing and cooking. He argued that this approach is fraught with problems, as variations like meat fat ratios or potato moisture levels often deviate from pre-set protocols, leading to inconsistent dish quality and unstable output.

With the integration of CookingMuse's AI vision and culinary neural engine, the new Xianglu 3K visual AI cooking robot shifts from executing fixed steps to understanding the desired outcome. It can identify when chilies develop their characteristic blistering appearance, rather than waiting for a predetermined cooking time to elapse. This feedback-driven, real-time control system replaces the outdated single-step logic of traditional SOPs, focusing on user experience and final dish quality standards rather than rigid procedural compliance.

At the WRC exhibition and launch event, the Xianglu 3K successfully passed two challenging "major tests" that traditional AI cooking robots struggled to overcome. The first test, called "the disappearing cup of water," involved two 3K units cooking the same Mapo Tofu dish, with one wok receiving an extra 100 grams of water mid-process. The system detected the change in wok condition, automatically reassessed the sauce reduction process, and paused the next step until the dish returned to its target state.

The second test, titled "the wok energy returns," featured one unit using partially frozen meat while the other used fresh ingredients to cook pepper-fried pork. The 3K dynamically recalculated the cooking path in real time, compensating for the temperature differences and ensuring both woks produced identical dishes with consistent quality and flavor.

"Cooking has no single correct answer—children need proper nutrition, the elderly require lower salt intake, and those away from home long for their mother's taste," Yang explained. "That's why we bring AI into real kitchens, to see how ingredients change every second, to capture the decades of experience from master chefs, and ultimately to understand every individual's true needs. Our hope is that someday, no matter where someone lives, how old they are, or where they come from, everyone can easily enjoy a truly suitable meal. Because what robots truly need to see is never just a dish—it's every person."