Chef Robotics—a San Francisco-based startup deploying AI-enabled robotic systems for meal assembly in food manufacturing plants—has raised a $43.1 million Series A round led by Avataar Ventures.
The round was also backed by Construct Capital, Bloomberg Beta, Promus Ventures, MFV Partners, Interwoven, HCVC, MaC Venture Capital, Red and Blue Ventures, Tau Partners, Siddhi Capital, and BOLD Capital Partners.
It includes $20.6 million in equity and $22.5 million in equipment financing debt, which will be used to cover the financing of Chef’s robotic systems for Robotics-as-a-Service (RaaS) so Chef’s customers don’t have to front CapEx for their robots, said Chef Robotics founder and CEO Rajat Bhageria.
“Robotics is really having a moment right now,” said Bhageria, who has raised $65.6 million since founding the business in 2019.
“Innovations in AI have unlocked the potential of Embodied AI [AI integrated into a physical form enabling it to perceive, act, and learn in the real world through interaction] for robotics. We’re in the pole position to scale given all the real-world production training data we already have.”
“AI in the physical world is happening right now with robotics. Food is one of the largest markets in the world. Industrial AI is already winning, and food packaging automation is quietly transforming how we get our meals. Chef has quickly cemented its standing as the industry leader in AI-enabled robotics for meal assembly at over 40M servings produced and counting.” Mohan Kumar, founder and managing partner, Avataar Ventures
Embodied AI and on the job learning
Chef, which currently serves customers in the US and Canada, and plans to expand to the UK in 2025, has deployed its ‘bots at food manufacturers including Amy’s Kitchen, Sunbasket, Chef Bombay, and Cafe Spice to serve more than 40 million meals.
Historically, observes Bhageria, robots have been designed to automate specific tasks in ‘low mix’ production lines optimized for mass production of a single product. But they do not suit ‘high mix’ environments where companies are handling hundreds of SKUs or producing customized meals.
In these more complex environments, or in situations where products are too delicate to be handled by traditional dispensers, he says, “traditional automation simply doesn’t work” and companies still rely heavily on manual labor.
Chef Robotics has found its sweet spot with these companies, which have a certain level of complexity, but a meaningful level of throughput, where it can deploy robotic arms armed with proprietary utensils trained to dispense accurate portions of food into trays through rapid ‘on-the-job’ learning, facilitated by AI.
Sensing includes a few depth cameras, as well as a weight-sensing platform for the food container to ensure consistent amounts of food are picked, says Bhageria, a master’s graduate of Penn’s Robotics and Machine Learning Lab who started his first company in high school and launched an early-stage venture capital company called Prototype Capital in 2018 that has invested in several robotics startups.
Training data
Chef Robotics’ systems are based on advances in diffusion models and deep learning such that its robotic arms are adaptable enough to pick and plate almost any ingredient, notes Bhageria.
But to have highly capable AI, you need training data, and in food assembly, there are no off-the-shelf training data or physics-based simulation engines because “food can be deformable, sticky, wet and inconsistent,” he observes.
To generate useful training data, therefore, you need to deploy robots in actual production environments.
So how do the robots learn?
For ingredients that are similar to those Chef Robotics has handled before, it can essentially take an existing policy, establish some key parameters such as weight, and then fine tune on the job, using an approach known a ‘KNN‘ (k-nearest neighbors) he says.
For new ingredients that Chef’s ‘bots haven’t handled before, more sophisticated imitation learning comes into play, he explains. Here, Chef’s staff will demonstrate, say, a scooping motion using one robot, and the second robot will mimic it, says Bhageria. “You can do 30 to 40 demonstrations and use that to train a diffusion policy and get this very articulate, very dexterous scooping motion.”
He adds: “This approach is really useful to do new tasks. Let’s say you want to spread mayo on bread, a task we couldn’t do before, where there’s no code written and if you did write code for it, it would be thousands of lines of code and might take weeks or months to perfect. Now, we just do a series of demonstrations and it might take 15 minutes.
“The system is spreading the mayo completely autonomously, but there wasn’t any code written, and this is the key differentiation, compared to what I would call the old way of doing robotics, which is all about rules. If you see a red light, stop your autonomous vehicle. If the queso is stuck, shake the utensil three times.”
The problem with food, he says, is that “there’s an infinite number of ingredients, so to accommodate that, your number of rules balloons and you end up with thousands of parameters, and it’s really unwieldy. Now, we just do a bunch of demonstrations, and the robot learns how to spread or scoop, and if things change, it can also adapt.”
Chef also has a lot of different sensors that allow it to understand things like how much pressure and force to apply to a given material as its robotic arms pick it, he explains. “Each robot has a different interface where you can attach different utensils and cameras pointing at the ingredient being picked up from a tub/container and the conveyor containing trays in which food is deposited.”
Return on investment
When it comes to ROI, he says, the biggest benefit to customers is making more revenue. “If you have 10 lines but you can only run seven of them because you don’t have enough people, or someone didn’t show up for work, you’re missing out on revenue.”
In addition, average throughput goes up with robots, which don’t get tired, don’t need breaks, and always show up for work, he notes. They also reduce giveaway and increase yield, as humans tend to “over-deposit.”
And finally, he says, there’s the labor saving itself, and the fact that you can free up humans to do other tasks.
“The ROI differs per customer. For some, they literally cannot hire people like, so if you have our robots, it means you can now run line one at full speed, and the benefits of that are almost instantaneous. In other cases the issue is high staff turnover and training costs, whereas for others they’re just wasting far less food.”
Asked about the equipment financing segment of the latest round, he said, this enables Chef Robotics to grow without spending equity dollars on capex.
“Customers pay a deployment fee upfront, but that’s relatively small. On an ongoing basis, they pay us monthly, quarterly, or whatever has been agreed. What equipment financing essentially allows us to do is say, we have a signed contract with a customer, so we have predictable cash flows. We can then go to the bank and say we signed this contract and they’ll front us X amount of dollars and then we’ll pay back the bank on a monthly or quarterly basis.”
What’s next?
For now, Chef Robotics is focusing on high-throughput food manufacturing environments doing fresh and frozen meal assembly. But moving forward, he says, it will be able to operate in increasingly complex, lower-throughput environments from stadiums to cruises to prisons and ghost kitchens, with the ultimate goal being fast casual and restaurants.
“The more robots we deploy, the more training data we collect, the better the autonomy gets.”
Chef Robotics is steadily picking up new customers including “a very large fresh food player, one of the largest in the world” but is also scaling up with current partners, which is a strong validation of its technology, says Bhageria.
“We’re also actively deploying into the UK this year, where they love prepared meals and have some very sizable players, and we’re very excited about airline catering and fresh fruit preparation.”
