Chef Robotics is developing a new bi-manual physical AI system for meal assembly at prep tables. The system uses two robotic arms and is designed for environments where meals are assembled in smaller volumes and with greater variation than on industrial conveyor lines. Target applications include ghost kitchens, fast-casual restaurants, airline catering, schools, hospitals, military facilities, prisons, stadiums, corporate dining and hotels.
The announcement extends Chef Robotics’ current focus. The company’s existing robots are mainly used for high-volume meal assembly on food manufacturing conveyor lines. In those environments, tasks can be divided across separate workstations, for example by ingredient or process step. Prep table assembly is different: one worker, or in this case one robotic system, has to assemble an entire meal.
According to Chef Robotics, this makes automation more complex. The robot must not only pick and place ingredients, but also deal with changing shapes, textures and conditions. Food items can be wet, sticky, soft, irregular or otherwise variable. The system must also be able to work with different utensils and end effectors.
For that reason, the new system will use two robotic arms to perform tasks that require coordinated movement. Chef Robotics gives back-of-house burger and burrito assembly as examples. The company says the robot is intended to perform dexterous manipulation comparable to human arms and hands, but within a foodservice environment.
A central part of the system is Chef’s Food Foundation Model, or FFM. According to the company, this AI model has been developed specifically for food manipulation. Chef Robotics argues that off-the-shelf vision-language-action models and general physical AI models are not sufficient for food handling, because they are mostly trained on rigid objects. Food requires models that can handle deformable materials and a wide range of physical states.
The FFM is intended to combine multiple functions in a single model. Instead of using separate models for picking and placing food, detecting trays and compartments, or handling scoopable and discrete ingredients, Chef Robotics says these capabilities can be supported through one underlying AI model. The model learns through imitation learning, meaning that it is trained from demonstrations rather than programmed step by step.
Chef Robotics also says the model is designed to transfer across different robotic hardware platforms. In practice, this means the same AI layer could potentially be used on systems with different robotic arms, end effectors or configurations. The company describes this as building a physical AI layer for food.
Some of the announced capabilities are forward-looking. Chef Robotics expects the model to eventually support zero-shot or few-shot ingredient onboarding, allowing it to adapt to new ingredients with little additional training. The company also says the system may improve itself over time to increase yield and consistency.
The new robotic system is being built with proprietary hardware for the food industry. Chef Robotics says it will be food-safe, wash-down capable and able to operate in different temperature and humidity conditions. It is also intended to work safely alongside employees and to be easy to use through language prompting.
“We started Chef by focusing on high-throughput food manufacturing, but a large part of the industry still relies on manual prep table assembly,” said Rajat Bhageria, founder and CEO of Chef Robotics. “These environments are more complex and less structured, which makes them harder to automate. With this new physical AI system and our Food Foundation Model, we will extend physical AI to handle those real-world conditions and unlock a much broader set of applications in the food industry.”
