Silicon Valley Can’t Quit Its Pizza Robot Obsession
At Stellar Pizza, a group of engineers from SpaceX and other rocket makers have turned their attention to a daunting experiment in food automation.
(Bloomberg Businessweek) -- Early in the pandemic, Jaya Iyer began seriously considering a career change. By that point, she had two degrees in mechanical engineering and eight years of experience working in aerospace. Iyer, the lead spacecraft structures engineer at a subsidiary of Boeing Co., consulted with someone she had worked with at SpaceX, Elon Musk’s rocket company. Her friend suggested she get in touch with their former colleague Benson Tsai, who’d been tapping some of the bright minds from the company and other aerospace firms around Los Angeles to solve a novel technical challenge.
When Iyer was at Space Exploration Technologies Corp., Tsai was developing battery technology for the company’s satellites and spacecraft. Now he makes pizza. Perplexed and a little intrigued, Iyer agreed to a meeting. “It was outlandish to the point where, ‘Please, tell me more,’ ” she says. Tsai regaled her with his vision of a robot on wheels capable of making a pizza from scratch. Someone would drive the truck to an office park, shopping center or residential area; customers would order through an app; and the robot would do the rest. Tsai had already persuaded venture capitalists—a group that would soon include a firm run by Jay-Z—to invest millions of dollars. He’d assembled a team of 40 or so rocket scientists and other technologists to, as Tsai describes it, “solve pizza.”
But first they needed to solve mushrooms. By 2020, when Iyer first visited the headquarters of Stellar Pizza Inc. in Hawthorne, California, the team had already figured out how to knead dough, spread tomato sauce and scatter green peppers. Mushrooms were proving extremely annoying. The machine’s design called for them to be pre-sliced and loaded into a chute. But mushrooms, more than diced mozzarella or pepperoni, are fleshy and moist. When cut and packed into the machine, the pieces clumped together and clung to the sides. Tsai compares them to wet paper towels sticking to a wall.
Mushrooms notwithstanding, Iyer was impressed by Stellar’s progress on its robot. She’d spent her entire career on spacecraft, but the prospect of working on something more whimsical captivated her. “It seemed like a really cool place to come and just pick up new things from an engineering perspective—um, and have pizza,” she says. Iyer joined as a mechanical engineer overseeing configurable toppings, meaning anything that goes on a pizza that’s not sauce or cheese.
To the untrained eye, the so-called robot doesn’t look much like a robot at all. It’s more of a food truck with a miniaturized assembly line inside. For the purposes of recruiting talent, investors and customers, the idea of building a robot sounds a lot cooler than replicating something Henry Ford did but for pizza. With all the intricacies of the machine and the possible messes that can result from a misfire, Tsai says it’s almost like trying to get into orbit. “If you look at it hard enough,” he says, “it’s not easier than rockets.”
At SpaceX the ultimate mission was to colonize Mars. Tsai’s conquest at Stellar is a little more local: “Our competition is Domino’s.” Stellar’s machine doesn’t produce the world’s best-tasting pizza. That’s by design. “We’re not trying to be the Italian, fresh-out-of-the-oven, Neapolitan pizza,” he says, between bites of a sausage slice and cheesy bread ball, both made by his robot. Domino’s Pizza Inc. generates more than $4 billion a year in revenue. Stellar can outdo Domino’s, Tsai says, by trading the costs of physical real estate for mobile pizzerias and by employing fewer humans.
For now, most of those humans are in the lab. Iyer spent her first year at Stellar building prototype chutes, testing fans and blades and experimenting with different cuts of vegetables. All the while, the mushroom issue was lingering. Mushrooms aren’t a problem in a frozen pizza factory, because frozen mushroom slices don’t stick together the way fresh ones do. Stellar is determined to use unfrozen ingredients. Mushrooms, Iyer learned, are a must. They outrank onions, green peppers and black olives as the top-requested veggie topping, according to the research firm NPD Group. “It is unfortunate that mushrooms are most popular,” she says in a moment of wry frustration.
On numerous occasions, Iyer consulted with a colleague who overcame similar problems with cheese gunking up the machinery. But the solutions for dairy didn’t translate to mushrooms. Reversing the direction of a paddle in the chute did nothing. At one point, they considered adding a light greasing to the sides of the chute, but she abandoned that after trying various products, including canola and olive oils, because they left an undesirable residue on the fungi.
The problem was vexing her when a reporter visited the lab in January and again in March. The task turned into a punchline when friends asked about her job. “I’m spending my morning cutting mushrooms,” Iyer recalls saying. “Putting that degree to good work.”
The solution finally came in May. Iyer redesigned the stirrer—the details of which, Tsai says, are proprietary. More crucial was the switch from freshly cut button mushrooms to ones that were stored in the fridge for a few days after being sliced. This process produced a drier and easier-to-manipulate topping. When staff sampled the finished pies, no one could taste the difference.
Much has been written about the space race between Musk’s SpaceX and Jeff Bezos’s Blue Origin. Not quite as much has been said about the exciting breakthroughs being made in the emerging field of pizza robotics. Why they’re happening is anybody’s guess; pizza doesn’t really seem high on the list of pressing issues. But engineers love pizza and solving their own problems. Who knows, they may have even taken inspiration from the Neal Stephenson sci-fi novel , which opens with a high-speed pizza delivery scene in a dystopian version of Los Angeles. Whatever the reason, the idea of a pizza robot has for years held an almost mythic sway over a certain category of technophile.
The most ambitious effort in the past decade was at a startup called Zume Pizza. Alex Garden, a former president of the video game company that made , started Zume in 2015 with the goal of building a pizza-making robot on wheels. One of its biggest champions was Masayoshi Son, the founder of the Japanese technology conglomerate SoftBank Group Corp. Son is evangelical about the potential for robots. He’s said that in 25 years robots will have an IQ of 10,000 and will, at some indeterminate time, replace “the entire working population.” SoftBank committed $375 million to Zume in 2018.
Zume acquired robot arms and programmed them to assemble pizzas at its kitchen in Silicon Valley. One crucial flaw in the plan called for the pies to be baked on the road to minimize idling time. The result was that the cheese ran everywhere and made a mess. Zume switched to stationary cooking and added cars and mopeds for last-mile delivery. But the dream of robot-made pizza was ahead of its time. In 2020, Zume shifted to designing sustainable food packaging and fired more than half its staff. The robots lost their jobs, too. Zume stopped making pizza and removed the word from the company name.
Tsai had followed Zume’s bumpy ride. For five years, he’d been designing the architecture for SpaceX’s batteries, but by 2019 he was ready to pursue an interest in food and robots. He considered boba, the trendy Taiwanese tea with tapioca balls, but the drink wasn’t consumed by nearly enough Americans to make business sense. Burgers occurred to him, but companies such as Creator and Miso Robotics Inc., which makes a burger bot called Flippy, each had a sizable head start. Pizza was the obvious candidate. Incredibly, about 13% of Americans eat it on any given day, according to the US Department of Agriculture. And much of the country isn’t content with the big four. Domino’s, Pizza Hut, Papa Johns and Little Caesars all lost points in the 2022 American Customer Satisfaction Index from the prior year’s poll.
To start the business, Tsai, 38, leaned heavily on his network from SpaceX. He took inspiration from Musk’s verticalized approach to product development: Stellar employees custom-designed and built most of the parts for their prototypes in house, he says, just as Tesla did for its cars and SpaceX for its rockets. For the first version of their pizza bot, Generation Zero, Tsai and his employees took frequent trips to Home Depot and Lowe’s Home Improvement.
One problem with Generation Zero showed up early on. A robotic arm holding a large, metal spatula called a peel, which is used to slide the pizza in and out of the oven, was malfunctioning. It was entering the oven a few millimeters higher than it should have, cleaving the tops of the pizza crusts. Gooey cheese ran everywhere and incoming raw dough backed up on the conveyor belt. That’s when the team called in Josh Villbrandt, the head of software. As staffers tossed the decapitated pies, scraped out the ovens and wiped down everything, Villbrandt reprogrammed the arm to come in a little lower. The memory seems almost painful to recount. “We have to exit all of the pizzas, then pause the machine, then restart it all,” he says. “That’s a pretty big efficiency drop.”
Avidan Ross was one of the first investors in the company, known early on as Serve Automation. A founding partner at the VC firm Root Ventures, Ross says he was inspired by the company’s grand ambition. “There’s a set of engineers who are not only unintimidated by challenge but love it when someone says, ‘That’s crazy,’ ” he says.
Telling people you want to build a pizza robot does sound pretty crazy. Ross rationalizes his $6 million Stellar investment in economic terms. The food industry is experiencing the twin effects of labor shortages and increasing prices. Robots can solve that, he says. “The greatest engineers are going into food automation because of its universal impact.”
Numerous other companies are making their own pizza robots. Seattle-based Picnic Works Inc. sells an automated assembly line that it says can make 100 pizzas in an hour. In Fremont, California, formerly home to Tesla, Middleby Corp. makes the PizzaBot, designed to assemble a pie in under a minute. A third, Pazzi Robotics, sought to build autonomous pizza restaurants in Paris but was forced to liquidate its business in October. Philippe Goldman, the chief executive officer, blamed Pazzi’s demise on VCs’ weak stomach for hardware investments and the public’s general distrust of robots.
Ross dismisses the competition as gimmicky. “Pizza and a show,” he says. Stellar employees refer jokingly to an internal metric called “minimum viable pizza,” defined as a pie that’s sufficiently tasty, and judge success partly by the number of pizzas successfully “launched” into an oven. Entertaining patrons with robotic performances isn’t part of the company’s mission. “You literally just place the order for the pizza and it comes out the other side, and you don’t see any of the internal mechanics,” Villbrandt says. “Proof of concept of the kitchen working is not seeing the kitchen.”
The restaurant industry sees opportunity in automation, but Tsai remains an outsider. “Way too much credit is being given to the founders of Stellar Pizza for being former SpaceX engineers,” says Brittain Ladd, a retail and strategy consultant who previously advised a Philadelphia business called Muncho that has a Picnic pizza machine. “Pizza isn’t rocket science.” A robot could succeed, he says, if it can consistently deliver a better experience at a price comparable to human-made alternatives.
Consistency has been one of the greatest challenges for Stellar. Arik Jenkins, a former automation engineer at SpaceX, joined Stellar in 2020 and quickly became exposed to the darkest fear from his rocket-making days: Things kept blowing up. Under the pressure of the cooking process, the dough overinflated and then burst. This happened again and again. Hardened fragments, shrinking and then blackening, dotted the walls and ceiling of the 900F oven.
The explosions stemmed from air pockets that formed unpredictably and grew so large that the sauce and cheese slid off them. Desirable in some styles of pizza, puffy crusts didn’t work for Stellar. The resulting bare sections looked ugly, and the displaced ingredients piled up in other parts of the pie. Engineers had to lurk outside the oven’s windows, armed with a dinner knife tied to a stick, ready to throw open the door and pierce any emerging bubbles.
Making things worse, nobody knew which batches would burst and which would come out fine. The engineers approached the problem as they would any they encountered during their time at SpaceX. They started with an assumption—that the layers of dough weren’t bonding together firmly enough after flattening under the press—and tested their hypothesis. They adjusted the dough-rolling process, which they refer to as docking. They changed the nubs on the rolling pin to be shorter and then longer, sharper and then blunter, but none of those made a difference. More data was needed.
To enhance structure and flavor, Stellar’s pizza dough sits for about two days, mostly in the refrigerator, before going into the oven. This leaves room for variables that could make some dough more susceptible to bursting than others. The engineers bought thermocouples to measure the temperature inside each ball of rising dough. “Like, every second of that dough ball, as it went up from counter temperature and then went down to the fridge temperature,” says Jenkins, Stellar’s head of test and integration.
The research revealed the precise yeast content and series of temperatures required at every stage of the dough-rising process to enable each pie to bake the same way. Around the start of 2022, they finally arrived at a consistently bubble-free crust.
Jenkins’s first response was disbelief. The team had grown accustomed to dealing with a bubble debacle every few pizzas. After a while, they finally felt comfortable retiring the dinner knife militia that routinely patrolled the ovens. “We could really step back and be like, ‘Huh, we just cooked 200 pizzas without touching anything,’ ” Jenkins says. “It was such a relief.”
The current version of the pizza robot is Generation A, the first one capable of cooking fully autonomously and reliably on a truck. The robot is essentially a set of interlocking tools, sized to slide in and out of an Isuzu NRR box truck in one piece. (Stellar picked the Isuzu because the team liked the aesthetic.)
Inside the vehicle sits a refrigerated chamber with trays of dough. The fridge has capacity for exactly 420 dough balls, a number that would make Musk proud but Tsai insists is a coincidence. Each tray slides down, one by one, beneath a metal claw bearing a sticker of the three-eyed alien from . When it’s time to bake, the claw grabs a ball and moves it along the assembly line.
Next, a press flattens the dough. A conveyor belt advances the circular slab toward the front of the truck. Within seconds, sauce dribbles over the top in a dotted pattern that, once in the oven, will melt into one uniform coating. Then the conveyor belt pauses under a chute from which a cloud of diced mozzarella falls.
Depending on the customer’s order, the next step could involve a band saw cutting in one motion through 19 separate logs of pepperoni dangling overhead. Additional chutes dispense veggies or sausage. Last, the pie moves near the cab of the truck, where an oven column sits. A mini elevator takes the pie up to one of the four ovens, and a robotic arm holding an automated pizza peel slides it inside.
Along the way, sensors check for roundness and placement of the toppings. If any one element doesn’t measure up, the pie drops off the line into the trash. This happens occasionally when the team is experimenting with a new topping or other modification. The failure rate in the field is essentially zero, says Tsai. The whole process is videotaped, and each pizza is assigned an identification number and QR code that can be referenced if a customer complains later.
A reporter for was pleasantly surprised after sampling a veggie slice. The crust was slightly charred, neither too crispy nor too chewy. The cheese was the right amount of gooey and played well off the tang of the black olives. It was no gourmet pie, but it tasted a lot better than Domino’s. That’s pretty much what Tsai is going for. To ensure quality, Stellar paid for the advice of Noel Brohner, the famed chef who’s helped Los Angeles restaurateurs and celebrities such as Tom Hanks improve their pizza games.
Stellar began stress-testing operations in September by sending its truck to the neighborhoods around the University of Southern California campus, where students are more than willing to download an app in exchange for free pizza. On one of the first visits, 180 customers placed orders within five minutes, resulting in two-hour delays. Stellar started charging a base price of $8 for a 12-inch pie in late October, which reduced wait times. One topping you can’t get from the Stellar truck at USC is mushrooms. That’s because college students usually don’t order them, Tsai says. He maintains that mushrooms are “largely resolved from a technical perspective.”
Two employees sit in each truck, mostly to answer questions from customers and put finished pies into boxes. Often there’s a third employee present: Tsai. He likes to observe how people interact with the app and the staff. He says a next-generation robot, expected in 2023, will box the pies itself. Even then, Tsai wants the head count to remain at two humans per truck. That way, one can take a break without leaving the assembly line unattended. The workers aren’t pizza chefs, though. If any part of the machine breaks down, they’ll have to turn the truck around and return to home base, no matter how many unsold pies are aboard. So far, that hasn’t happened, Tsai says.
He isn’t concerned with making the truck drive itself, which would require an entirely different set of expertise and put his startup on a collision course with Alphabet Inc. and Tesla Inc. “Stellar is focused on self-cooking,” he says.
The idea is to have a system in place that can be replicated easily and cheaply. Currently, a new garden-variety food truck costs about $100,000. Tsai declines to say what it cost to build the Stellar prototype. Factoring in salaries, rent, supplies and more than two years of work, it was at least $6 million, says a person familiar with the expenses who wasn’t authorized to disclose the number and requested anonymity.
Over the past couple of years, the Stellar team consumed a great deal of pizza. Too much, maybe? “That’s like asking me if I would ever get sick of rice,” says Tsai, the son of Taiwanese immigrants. “Like, no.” About an hour later, an interview with Villbrandt is interrupted by a courier toting a bag from the salad chain Sweetgreen. Villbrandt’s face shows a tinge of pizza remorse. “You can’t eat it every day,” he says.
Tsai says he’s pleased to have overcome the engineering challenges associated with miniaturizing the equipment. He’s won over the discerning palates of the student body at USC and secured the backing of Jay-Z, whose venture capital firm led a $16.5 million investment in October. But Stellar has many more challenges ahead. It needs to reproduce the robot on multiple trucks, prove the economics of the business and expand beyond Los Angeles. The startup is determined to succeed, Tsai says, for humanity: “Solving problems here on Earth is really important to me.”
More stories like this are available on bloomberg.com
©2023 Bloomberg L.P.