Chef Robotics: Harnessing computer vision and AI to bridge labor gaps and boost F&B production
12 Feb 2025 | Chef Robotics
California-based Chef Robotics has developed an AI-driven robotic assembly system to help food companies tackle labor shortages and to increase production. CEO Rajat Bhageria walks us through how the company’s technology uses robots and computer vision to handle food ingredients while maneuvering various technical and business challenges. He also highlights its consistent output, waste reduction, and cost-saving benefits.
Hi, this is Inha Noreen from CNS Media, a journalist at Food Ingredients First.
California-based Chef Robotics has developed an AI-powered robotic assembly system that can help companies address labor shortages and boost production volume.
To provide us with more insights into this technology's impact on the food industry, Rajat Bhaeria, the founder and CEO of Chef Robotics, is joining us today.
Welcome, Rajat.
Thank you, Esha.
I'm very grateful to be here.
Great to have you too.
So Raj, can we start off by discussing how you think, the trends that are evolving in the food industry and specifically AI is impacting food manufacturing?
Yeah, I think there's two macro trends that are really impacting the industry.
One is simply labor.
I don't think this is a new trend.
It's, it's obviously been around forever, but I think it's getting perhaps worse, right?
So just to kind of put some color here, you know, a lot of plants that at least we work with are cold rooms, right?
They're around 34 °F, a couple of degrees Celsius.
On the other hand, they're often very hot too in the kitchen, for example.
So in either case, it's very hard to hire for these environments.
And there's ancillary problems around retaining that labor, you know, we've seen multiple like 100% turnovers, like 300, 400% turnover sometimes.
So this of course makes it very hard for these companies to meet production volume, you know, satisfy their customer demands, and, and, and of course that's kind of one big macro trend.
The other big macro trend is really AI.
And the way to kind of think about this is, you know, automation in the food industry is not a new idea, right?
So if you go to many food plants, you know, you'll see a lot of case packers and palletizers.
What you pretty quickly realize though is that these are, These are very hardware-driven systems, right?
The traditional automation, right?
They don't have much brains.
They don't have very many sophisticated sensors like cameras or LDAs.
They're essentially doing the same thing over and over again.
There's some very basic sensors.
What, and, and, and the result of that, by the way, is that they're not flexible enough to automate a lot of the human tasks today, which is why you still need the people, which is why the labor shortage is so impactful.
The advent of AI makes it so that you can take the same hardware basically, but with the AI and software now the systems are able to become as flexible as a person, which is to say they can do lots of different ingredients and portions and things of this nature.
And therefore, you know, you can kind of combine these two trends and say, OK, leveraging AI with this existing hardware, now you can create a system that can actually overcome the labor shortage.
And, and really help these companies and food companies really increase production volume, and those are the kind of two trends that I think are More macro really affecting the industry and that we're trying to actually kind of ride the coattails of as.
OK, great.
And coming to chef robotics, can you tell us more about how the robotic systems are helping food production?
Yeah.
So, what we kind of realized is that if you think about making food on a, on a high level, like let's say making a meal, there's basically three parts.
You have to cut food, cook food, and then assemble food or plate food.
What we kind of realized is that visiting a lot of plants, you know, we, we, we found that around 60 to 70% of labor is actually in the food assembly side.
This is not obvious because of course when you're cooking at home, you know, cooking is the thing that takes the most time and labor, but, but really in a, in a more scaled setting, it's, it's assembly.
So we said, OK, we're going to really focus on the assembly problem.
And you know, if you were to go to one of our customers today, what you would find is these long assembly lines.
And you know, on either side of the assembly line there's maybe 1015 people.
Each person has a big tub of food of their particular ingredient, and it's really an assembly line.
So like if you're making a salad, you know, the first person that's putting on leafy greens, and then, you know, maybe they're putting on some diced chicken and then you're putting on some some vegetables, etc.
Etc.
You get the idea.
And so what we basically do is say, OK, we're going to create a flexible automation system, chef, and these chef modules we call them, they're the same footprint as a person.
So they're physically quite small and it's a robot on wheels, basically.
So the robot kind of you can slide the robot onto the line.
And then it has a vision system, so it figures out where in the tub to pick from dynamically.
And then it'll pick from that pose using a bunch of computer vision and AI.
Then it'll, once it picks, it'll figure out how much weight it picked up so we can be really consistent.
We have scales underneath the pan so it can be really consistent.
And then same idea on the, on the, on the placement side, we'll use vision to figure out where to place.
So we'll detect the trays, track the trays, and then we can place it in the right part of the tray, let's say the right compartment of the tray.
And the ultimate goal here then is that You know, we can really help these companies relocate, reallocate their labor.
So now, you know, that person who is doing that station can now be freed up to do something else or better yet, sometimes they can't even run a line because they don't have enough people.
So now they can run the line that previously wasn't possible to run.
Thereby really helping them increase volume.
That's kind of the ultimate goal, while of course also helping them with consistency and giveaway and really food wastage as.
When you were building this system, what were some kind of challenges that you faced, and if you would like to tell us about them?
Food is a lot harder than maybe it meets the eye, harder to manipulate than meets the eye, right?
So for us, for humans, to scoop, let's say, you know, some cheese and sprinkle it onto a meal, that's trivial.
It's very simple.
For a robot it's quite difficult, as you might imagine, right?
Every cheese is a little bit different.
If you cut it a little bit differently, different materials, there's different stickinesses, even the same cheese day by day can be different depending on how it's cut and cooked.
So there's just such a variety of ingredients and For a chef, because it's a flexible piece of automation, we have to do a very large number of ingredients, and if we can't do a very large number, then we're not adding value to our customers.
Or if we can only do one or two ingredients, we're basically as good as the traditional automation I was alluding to earlier.
So that flexibility is key.
And yet how do you build something that can do essentially any ingredient at any portion size without damaging the ingredient, without cutting the ingredient.
And without spilling the ingredient while placing reliably into the right part of the tray with, you know, there's like and and and and and, right, so that's, I think, a big technical challenge that we really have to deal with.
And what's interesting is that it's a novel robotics challenge.
A lot of the robotics research that really exists today is really focused on what's called rigid body manipulation, which is like Essentially things that you can't compress.
Like imagine like a box or like like an iPhone box or like, you know, something like this.
You can't squish it.
So you can model in physics and there's a lot of simulation tools you can use, but for deformable, sticky, wet objects, there isn't a lot of prior art.
So we really had to like develop the robotics framework and, and really like learn from production by shipping robots in the field.
That's kind of really how we learned and how we trained our models, computer vision and AI models in the field.
There's another challenge that we kind of face on the business side actually, which is, You know, a lot of our customers are very used to buying Capex, right?
They, you know, it's, it's, you know, they're a Capex approval process and they're very good at leveraging debt from their lenders at low interest rates and so they really like the Capex process.
Because chef is mostly software, right, really the core innovation is, is, is, is, is these computer vision models and these AI models, it necessitates a software-based.
Pricing as.
So in other words, we do robotics as a service.
The cap price is also basically nothing.
You don't pay really anything upfront outside of some deployment fees.
And really the way we make money is a recurring robotics service fee and it's kind of an end to end.
Service with pride.
That's kind of necessitated by the software driven approach, just like you pay for Netflix, and the reason you pay for Netflix on a monthly basis, there's new content and things like that.
Same thing for chefs, there's new ingredients we're onboarding.
There's new software we're shipping to you, but a lot of customers, of course, aren't familiar with it.
So there was definitely an uphill battle in the early days where we had to really convince our early customers about why this makes sense.
And in the food industry, what specifically are the segments that can use your robots, and are there any particular feedback that you have received from them?
Yeah, so right now I think we're really focused on a few, and so one is frozen prepared meals.
So we work with customers like Amy's Kitchen, for example, and others, fresh food manufacturing.
And so for chefs, like what's important in those cases is being able to work with lots of different ingredients, very high changeovers, lots of different trays, you know, it's even more flexible than let's say a fresh or frozen food manufacturing facility.
And then direct to consumer meals.
So those are kind of the four segments we're kind of in today.
Now what we're exploring actively is things like airline catering.
And meat packing, fruit packing, actually, interestingly, and pizza production, you know, the vision of the company really is to leverage all that training data about how to manipulate food to go to smaller and smaller volume kitchens.
So you can imagine down the line we can use the same essential idea robot arm plus cameras plus AI to be able to do manipulation, for example, for fast casuals, for ghost kitchens, prisons, hotels, stadiums, etc.
Really kind of going from industrial kitchens to commercial kitchens, but today really, we're really focused on the, the industrial side because that's really where we think we can add ROI, to our customers and, and really accelerate that trading data flywheel, we are working on.
That's really interesting.
And what do you think, you know, your future plans are for expanding your chef robot, and, do you also plan to use them for preparation or packaging processes?
Yep.
So I think the way we kind of think about Expansion is in a few different dimensions, right?
So first of all, like we, we really focus on Finding, I would say like mid-market enterprise customers, fairly big customers, where they have multiple plants, right?
So, what that essentially entails then is we really try to Make them successful.
Like let's really invest in their success so that they can scale within their plants.
Within itself, right?
The second approach is just scaling to a bunch of net new customers in the existing industries, the 4 existing industries I mentioned.
So that's kind of near term and that's what we're doing today, right?
Scaling within our current customers, scaling to more customers in the current 4 industries I mentioned.
I think in the next couple of years, we're, you know, we really like to get into that, the next set of industries I mentioned like pizzas and fruits and things like that.
And then in the next like, let's say 4 to 5 years, it's really gonna be like scale into some of the low volume industries like ghost kitchens, fast casuals, everything else.
So I think, I think, I think that's kind of like the industry perspective.
The other thing, of course, we're doing is geographic expansion.
So, for example, this year, like we have plans to expand into the UK, you know, it's actually a fairly big market for prepared meals.
So how cost effective do you think this technology is and what potential does it have in our future?
Yup, it's a good question.
So, the way we kind of think about cost is, is really from an ROI basis.
So, usually our current customers, at least in this segment, and of course this is going to change if we go to Fast casuals and ghost kitchens, right?
Usually in our, in our current segments, our customers run 2 shifts a day.
So let's say per shift they're paying all in including like workers' comp and health insurance.
It's like $10 let's just say arbitrarily, right?
So over two shifts they're paying 2X, right?
You know, the way we kind of price chef is that it should be substantially less than 2X.
And of course it depends on the complexity of the application and everything else, but let's say, you know, if the, if the status quo is 2X, we might be charging.
You know, 1.5x, something like this, right?
So it should be less than the cost of status quo labor, right?
And it really depends on the complexity of the application, how much support they're going to need and things like that.
Now, the other thing I'll say here is like what is the ROI?
So right off the bat, yes, we're cheaper than people, of course, that's, that's apparent, but I, I actually think that's not the main ROI.
There's a few ROIs that are more impactful.
First of all, as I alluded to, some customers just simply do not have the people to even run particular lines.
So if you have, let's say, a plant and it has 10 lines and because of the labor shortage you can only run 7 of them or 8 of them at a time.
You're just leaving a lot of revenue on the table.
So if we can help you overcome this hiring problem and help you increase volume, then now you can run line 8, you can run line 9.
That's a ton of revenue for you, and that's really driven by the labor equivalent aspects of our robots.
So really increasing volume is, is something we think a lot about from an ROI perspective and that kind of, you know, if that's the case, the cost saving is like negligible.
It's not even important relative to the revenue increasing.
On a similar vein, oftentimes we can actually help our customers increase average throughput, and I, I use the word average, and the reason is that, you know, if you think about production manufacturing, what's really important is a steady rate.
You don't want to be jumping in terms of your production output, and, and yet humans are, do that.
Humans are like, you know, if you look at the, if you look at output or let's say the shift, it's all over the place, sometimes very high and then they slow down and then they increase and Whereas a robots just consistent and, and so that consistency actually over the long run actually allows you to get more output out.
The third one we think a lot about is, is what's called giveaway in the food industry or food wastage, essentially.
You know, humans are very inconsistent.
So for example, they often scoop with their hands.
Every person's hand is a very different size.
Sometimes they'll overpick, sometimes they'll underpick, and, and, and they're getting a lot of pressure from their customers and their kind of end consumers to not Underpick, they, you know, they don't want to have a customer who gets too little avocado, for example, right?
So, you know, the production manager is like, hey, make sure you don't underpick, make sure you overpick essentially.
So what's, what's, what this results in in the industry is called giveaway.
You, you basically have an average target weight that the humans are picking over.
They're putting too much food into the tray on average.
So, Again, because our robots have all these scales and we have all these vision systems and we have all these sensors, we can really help them reduce food wastage.
And so that's kind of the 3rd ROI and the 4th 1, of course, is, is cost savings.
So that's, that, you know, whenever we meet a customer, we kind of do that ROI analysis with them.
We kind of look at their status quo, we compare that to the cost of chef, and say, look, like, you know, this is how, how you can think about an ROI and why you should kind of make the jump towards robotics.
OK, thank you for explaining that so.
I think to wind up this interesting discussion, it would be great to know your thoughts on how in future you plan to, or do you even plan to incorporate AI into your system and what are your thoughts on the ongoing discussion about AI replacing humans in the workforces.
So if you would like to comment on that.
Yeah, so chef actually already uses quite a bit of AI, like.
What we do would not be possible without AI, like, and I, and I think it kind of harkens back to the original discussion around like traditional automation for low mix production versus what we do is high mix production.
The reason those, those traditional pieces of automation, like the palletizing system, for example, and others like case packing, they cannot be flexible right off the bat is because they don't leverage AI and software as much.
Whereas for chef, we essentially have the same hardware.
If you were like, look at the hardware, it's very similar.
But, but on our system, there's a really beefed up GPU and CPU and we have all these AI models that are running that allow us to do, deal with these different ingredients and allow us to place onto different trays flexibly without having to hard code or write custom software per, per, per ingredient or tray or conveyor or things like this, right?
So, we actually already leveraged quite a bit of it.
What I would say around the jobs point is I think it's more, I don't, you know, I think I definitely understand the fear and, and the reason I think it's top of mind, of course, with movies like Terminator and otherwise, that's like very top of mind.
But I don't think it's the reality of the situation, specifically because you have to remember there is a labor shortage.
There's no jobs per se to be replaced here, right?
It's, it's usually what we find is that our plants are 60 to 70%.
You know, capacity, they're running way under capacity.
It's really driven by the labor shortage or by the way, different flavors of labor shortage, which is like maybe there's not a labor shortage, but they rely on temp workers, so their turnover rate is like 400%.
It's just a different flavor of it.
You try to solve the problem, right, and you use temp workers, but the core root cause is still, it's really hard to hire people to work in a 34 °F or 1 °C room doing motions all day long with dense material that just your arm's hurting, your back's hurting, so.
That's kind of the way we think about this, which is like if we can help these companies overcome their labor shortage.
There's no jobs being replaced by any stretch of the imagination.
