Automating food production, one serving at a time

The path to success in food robotics is littered with interesting concepts that failed to deliver. Chef Robotics is a young startup that has built a healthy portfolio of clients and iterated quickly over the last two years to mature the capabilities of its autonomous meal assembly solution.

The company is crossing the chasm from early adopters to gain its initial foothold in what promises to be a huge market for food manufacturing automation. 

From its San Francisco headquarters, the company has been flying below the radar while doing all the hard work to build, fail, and iterate on the path to a viable roadmap.

Inspiration

Chef Robotics co-founder and CEO Rajat Bhageria was an entrepreneur from an early age, and he brings a unique set of experiences to the job as a startup founder. While he has a bachelor’s degree from The Wharton School and a Master’s Degree in Robotics and Machine Learning, he is also an articulate communicator. He launched his first startup in high school and wrote a book about his high school experiences before heading to college.

Before launching Chef Robotics, Bhageria approached food automation from a market analyst perspective. He completed a market research study, dissecting the market to better understand where the bottlenecks are in food production and to identify the opportunities.

As he likes to tell it, “Fine dining is extremely complex for AI and robotics. In a typical commercial kitchen, operations are divided into prep, cook, and plating. Interestingly, we learned that assembly or plating the food (what you visually see at Chipotle) is 60-70% of the labor in a non-fine-dining commercial kitchen. The reason is that the labor to do food prep scales sub-linearly with volume since there’s a lot of traditional food processing equipment (like Robo Coupe and similar industrial equipment). And cooking you can do in batches so a few people can cook for hundreds. The best place for robotics to start helping food companies overcome the labor shortage and increasing production volume is food assembly.”

He said the biggest takeaway from the market research was that early food automation companies got it wrong: they focused on cooking with robots. But cooking isn’t the labor bottleneck in food production, the most labor-intensive is plating, assembly, and portioning.

Big market

The global robot kitchen market size is projected to grow from $2.73 billion in 2023 to $3.35 billion in 2024 at a compound annual growth rate (CAGR) of 22.7%. The market is anticipated to reach $5.94 billion by 2028, driven by advancements in artificial intelligence, integration with smart homes, and the rise of cloud-connected kitchens.

According to the U.S. Bureau of Labor Statistics, the food industry has the highest labor shortage of any U.S. industry with 1,137,000 unfilled jobs, a sign that people don’t want to work in food preparation or food service. This presents a huge opportunity for automation.

There have been several failures of companies over the years in this market including Zume Pizza, Karakuri, Spyce, and Chowbotics. All of these companies attempted some form of commercial kitchen automation.

In evaluating the market opportunity, Bhageria realized there was a real opportunity in the food manufacturing and plating segment because of terrible worker conditions: long hours inside a refrigerated room, scooping individual servings of peas, potatoes, or rice for hours on end. This fulfilled one of the key criteria for a successful automation business: eliminating dull and dangerous jobs.

Also, in a restaurant, low-volume, high-mix meal preparation requires generalized skills. While automating high-volume, high-mix applications can be a target market, it’s a difficult problem to solve, and it’s the hardest problem to train an AI for. Chef Robotics decided to start with high-mix, high-volume applications, where labor availability is a concern, and from which large amounts of data can be gathered to continuously improve the operation and capabilities of the solution.

“To be able to automate a restaurant, you need an extremely intelligent AI system,” said Bhageria. “But if you can start in manufacturing, over time you can develop that and get to something more complex like a restaurant. This is similar to what Tesla did – start with the Roadster and over time get to the Model 3.”

Register now.

The COVID-19 pandemic also created an opportunity as many consumers became accustomed to preparing and eating meal kits and ready-to-cook meals. This opportunity brought dozens of new vendors to the market, all providing meal services to their customers.

In addition to meal kits and ready-to-cook meals (i.e. frozen “TV dinners”), any time you eat on a plane, or get a fresh salad from Trader Joe’s, or are served a meal as a patient at the hospital, your meals have been prepared in an industrial kitchen.

For the medium term, Bhageria thinks the biggest place to help is food assembly. It’s 70% of labor in any commercial kitchen (except fine dining). What Bhageria would like to do is slowly go from high-volume operations (food factories) to medium-volume operations (like ghost kitchens) to small-volume operations (fast casuals, prisons, hotels, stadiums, K-12, universities, cruises, etc.). In each of these cases, the learnings of how to manipulate food from the previous sector helps the following sector.

Chef has raised $22.5 million in equity and debt since its founding in 2019. Investors include Kleiner Perkins, Promus Ventures, Construct, Bloomberg Beta, BOLD Capital Partners, Red and Blue Ventures, Gaingels, Shox VC, Stewart Alsop, Tau Ventures, and others.

Enter Chef Robotics

To build a profitable roadmap, the company is laser-focused on high-volume, high-mix food manufacturing applications. The ideal application for a Chef Robotics pod is to repeatedly scoop starches (including rice and potatoes), vegetables, and sauces in low-temperature operating conditions. 

The pods use Universal Robots’ UR5e cobots a custom, modular gripper. Chef Robotics designed an exchangeable gripper system that can repeatedly handle the dispensing of measured portions of items as diverse as sauces, rice, mashed potatoes, and cooked vegetables like beans, peas, carrots, and more.

Bhageria said, “The hardest part of the process is developing an AI policy and software that can handle the day-to-day variation in the food item that we’re handling. Changes in how the item is sliced and cooked from batch to batch can’t be a failure mode for the solution”.

After many 3D-printed iterations of gripper designs, the company now has a palette of 6-7 classes of utensils, each with different portion sizes and designed for different classes of food. The software and AI enable Chef to deal with variances in the ingredients. During startup or shift changeover, the human operator selects the food item and meal choice, and then the robot controller informs the operator which gripper to install for that particular production run. 

The robot uses vision perception, real-time force sensing, and item scales to ensure each meal gets the proper amount of food. The first generations were built leveraging ready-built work cells from Vention, but the latest generation is custom-designed, made from stainless steel, fully IP67 sealed, and can withstand the cleaning process necessary at the end of each shift.

Data is king

Bhageria is clear-eyed on how to achieve the long-term objectives: mature the AI at the heart of the system by leveraging real-world data. Bhageria said a big issue for food production applications is that simulation isn’t good enough to provide training data to evolve the system. According to Bhageria, it’s too difficult to simulate the stickiness of the food items when handled by the robot gripper. For this, the system needs iterative real-world data, and that only comes from production.

“We can bootstrap an AI model that can do one ingredient, like five ingredients, like a small seedling of ingredients. If we can get something to market, then over time, we have this flywheel that forms. The more robots we deploy, the better the system gets,” described Bhageria. He went on to say, “And the better the system gets, the more flexible and harder applications we can do. So that’s when we stumbled upon food manufacturing and food production. And this is a great market because it has a much higher volume. You’re making tens of thousands of meals a day, and hundreds of thousands, sometimes millions of meals a week.”

Chef Robotics operates under a robots as a service (RaaS) business model. This means the company monitors all of its robot fleet in real-time, tracking each scoop of food. With this data, the company was able to track (in real-time) the milestone of serving 20 million meals in July 2024.

Automating food production, one serving at a time

Over the last year, Chef Robotics has quietly assembled a customer list that includes many well-known brands and products you’ve seen at your local supermarket or your doorstep. Two brands Bhageria can talk about are Intelligent Foods and Chef Bombay.

The company differentiates itself from a systems integrator. “We have a bunch of customers all over North America,” Bhageria said. “We deploy one hardware platform at every single customer. We have one library of utensils, and we deploy the same library of utensils to every single customer. We have one code base, and every customer has the same code base. And then we have a bunch of different AI models per ingredient. Of course, those can change for customers. But the whole point is this is all one AI brain, so that the more ingredients we manipulate, the better that system gets. So I think the main way we’re different from a systems integrator is it is a product.”

That real-time connection also means the Chef Robotics support team can monitor and capture errors, and often fix the problem before it impacts production. Also, the company can push updates out to the systems and collectively improve the performance of the system for all of its customers.

What’s next for Chef?

The solution has proven successful in the application of portioning and plating prepared meals. The next step for Chef Robotics is to scale the AI-enabled robots. Bhageria also plans to continue teaching the system to handle a wider variety of items to broaden its capabilities. Over time, Chef would like to deploy to lower-volume operations like ghost kitchens and fast casuals.

Chef is on a measured roadmap to find opportunities that enable the current AI models to support and then learn from new applications. High volume is the key to this approach, ensuring the system has the opportunity to integrate new data back into the AI models that are the intelligence of the system.

Bhageria recently appeared on The Robot Report Podcast, check it out to hear the vision firsthand.