AI food tech startup Beyond Honeycomb is fully launching its “full-stack robotics” strategy, which involves manufacturing actuators in-house. The strategy aims to reduce robot production costs by developing everything in-house—from actuators to robot hardware, control software, and AI—thereby expanding deployment in the field and leveraging data collected from real cooking environments as a core asset for next-generation physical AI.
Jeong Hyun-ki, CEO of Beyond Honeycomb, said, “In the era of Physical AI, data accumulated in real-world physical environments determines competitiveness,” adding, “We will build one of the world’s largest real-world cooking data platforms to further advance our Robotics Foundation Model (RFM).”
“You Can’t Collect Data with Expensive Robots”… In-House Actuator DevelopmentBeyond Honeycomb is an AI food tech company that received initial investment from Naver D2SF. The company developed “GRILL X,” a cooking robot that learns chefs’ cooking methods using AI and molecular sensors, and has supplied approximately 300 units to hotels, restaurants, and corporate cafeterias over the past two years. Notable examples include the Andaz Seoul Gangnam Hotel, Hanam Dwaejijip, and the cafeterias at Naver and KTCorporation.
As the company expanded its business, it realized that the robot’s price was the biggest obstacle to commercialization. The company determined that using external collaborative robots and actuators would drive the product price up to tens of millions of won or even 100 million won, making widespread adoption in actual stores difficult and ultimately preventing the collection of real-world data.
Consequently, Beyond Honeycomb decided to develop its own actuators, building on its existing designs for gearboxes and manipulators. Rather than simply in-sourcing components, the company also designed the production processes and assembly methods to minimize mass-production costs.
CEO Jeong explained, “By designing the entire structure from scratch, we can significantly reduce production and assembly costs while maintaining the required performance.”
Mass Production of 15,000 Units Next Year… External Sales Also PlannedBeyond Honeycomb plans to launch its actuators in the first half of
next yearand produce 15,000 units annually in South Korea. It will first introduce two 3-kg-class models suitable for commercial and service robots.
Most of the production volume will be used in the dual-arm cooking robot the company is currently developing. A single dual-arm robot requires at least 12 actuators. The company is also considering supplying a portion of its output to other robotics firms. It explains that purchase inquiries from domestic and international companies have been pouring in since it unveiled a prototype at “NextRise 2026” last June.
To differentiate its products from lower-priced alternatives, the company has applied impedance control and physics-model-based force control technologies. CEO Jeong stated, “Traditional industrial actuators and those required for Physical AI have different performance requirements,” adding, “We redesigned the structure to be optimized for Physical AI so that it can understand and respond to physical changes while coming into contact with people or objects.”
“Cooking Is a Treasure Trove of Human Behavior Data”
What Beyond Honeycomb ultimately aims to secure is human behavioral data.
Currently, GRILL X acts as the “chef’s brain,” using sensors to assess the cooking status and determine the optimal cooking method. Going forward, the goal is to implement a “body” that learns even the movements of human hands and arms—such as picking up, moving, and flipping ingredients.
CEO Jeong emphasized that the cooking environment is one of the most suitable fields for physical AI learning. This is because each ingredient differs in size, shape, weight, elasticity, and friction, and requires the use of various tools—such as tongs, forks, and spatulas—allowing the system to learn both precise tasks and rapid movements.
Beyond Honeycomb plans to unveil a new cooking robot capable of using both hands like a human by the end of this year. While the existing Grill X focused on grilling meat, the new robot will perform pre- and post-processing tasks, such as moving ingredients, sprinkling seasonings, and spreading butter.
The company plans to finalize the development of the manipulator this September, then integrate it with control software and AI to unveil the finished product by the end of the year. Subsequently, it will deploy the robots in commercial kitchens to accumulate real-world cooking behavior data and further refine the development of RFM.
CEO Jeong stated, “While we have so far created the ‘brain’ that implements a chef’s judgment, we are now moving on to the stage of creating the ‘body’ that learns human behavior,” adding, “We will grow into a Physical AI company that secures top-tier real-world data through large-scale field deployment.”
