Xiao Lu Launches AI System for Autonomous Cooking
The Beijing company introduced a multimodal cooking model and two robots that adjust recipes by watching food change in real time.
From fixed recipes to adaptive cooking
Beijing-based Xiao Lu Robot unveiled a multimodal cooking foundation model and two machines built around it at the 2026 World Robot Conference. The launch moves commercial cooking automation beyond replaying fixed instructions toward responding to conditions inside the wok.
The company says CookingMuse combines kitchen imagery, process data and decision-making software. Its first hardware implementation, a 3K-vision cooking robot, uses three 40-megapixel industrial cameras to monitor ingredients at 30 frames per second. The system can alter heat, cooking time, stirring speed and seasoning when it detects differences in moisture, temperature, portion placement or thawing.
In demonstrations, the robot reportedly compensated after extra water was added to mapo tofu and recalculated the cooking process for partially frozen meat. These tests were designed to show adaptation rather than simple recipe execution, although the results were presented by the company and have not been independently benchmarked.
A mobile automated kitchen
Xiao Lu also introduced Fresh-Cook Ark, a movable micro-kitchen that combines refrigerated ingredient storage, two parallel cooking stations, serving, fume treatment and self-cleaning. The company is targeting offices, campuses, festivals and other locations where constructing a conventional kitchen would be expensive or impractical.
Xiao Lu says its existing cooking robots operate in more than 13,000 outlets across nearly 350 Chinese cities and over 20 countries, while its recipe system contains close to one million digitized dishes. That installed base could provide unusually broad operational data, but the company did not disclose pricing, availability or independent measures of food quality and reliability for the new products.
The launch matters because cooking is a difficult embodied-AI environment: ingredients vary continuously, safety requirements are strict and small errors directly affect the finished product. If Xiao Lu can reproduce its demonstrations at commercial scale, it would offer a concrete example of specialized multimodal models creating value through physical deployment rather than general-purpose conversation.