OakBot releases cooking multimodal model and autonomous kitchen, using real-world data flywheel

新智元 · wechat · 2026-08-23

At the World Robot Conference, OakBot launched the "Ark of Live Stir-Frying" autonomous kitchen system and the CookingMuse multimodal AI base. The "Ark" features dual-wok parallel cooking, three-stage self-cleaning, and built-in smoke purification, achieving a fully unmanned workflow from storage to serving. Its capability stems from CookingMuse, a model trained on real kitchen process data from 13,000 stores and 600 million servings. Using 3K vision AI, it monitors the wok in real-time to dynamically adjust heat and ingredients, evolving from "executing recipes" to "understanding cooking." The company believes industry competition will shift to data flywheels, iterating models through real-world data to turn culinary experience into scalable digital assets.

Original post →

More from Embodied

Embodied channel →