It’s San Francisco, so naturally there’s some controversy. Burgers are the number one segment of restaurant food, ripe for disruption by robotics, and the VCs are pouncing. Robots might take away fast-food jobs, or create better ones, so the labor activists are activating.
But today I just want to understand how they made it. How two young engineers could invent a complicated culinary robot that makes cheeseburgers from scratch. Really good ones.
I’m on my second visit to Creator, the world’s only robotic hamburger restaurant. Since opening doors in September 2018, it has built a rep for quality and innovation, thanks in part to its fun-to-watch robot and its tourist-trod location next to San Francisco’s Moscone Convention Center.
But the real secret sauce is the small team of makers who overcame technical setbacks and ridicule to create a single, integrated robot that makes the freshest burger ever served. And for just 6 bucks! Which even a tourist knows is unheard of in San Francisco.
Everything in a Creator burger is prepared to order in a way that’s never been done — beef freshly ground 20 seconds before it is precision grilled by artificial intelligence; whole buns automatically split, buttered, and toasted; veggies precision sliced and cheese shredded the instant before they meet the robo-toasted bun. And yes, it’s delicious, thanks to quality ingredients and gourmet techniques and seasonings.
Today the burger bot team includes alums from Tesla and Google, Berkeley, Stanford, and Caltech, Chez Panisse and Fat Duck. But it started with two California kids barely out of college: a self-taught roboticist with a vision for the future of fast food, and a trained mechanical engineer who came to trust in that vision. I want to hear how they created the culinary equivalent of landing a moon rocket on a robot barge.
Bootstrapping on I-5
Alex Vardakostas grew up flipping burgers in his family’s restaurant in the surfy Southern California town of Dana Point, and earned his physics degree from UC Santa Barbara. His dad, a Greek immigrant, saved enough to buy his own burger joints in the 1970s. His mom, also an immigrant, lost her engineering career to the Iranian Revolution and became a burger chef.
During college Alex became obsessed with the idea of completely automating the cheeseburger. He believed a robot-made burger could be faster and more sanitary, but above all, fresher and tastier, because a robot could perform modernist gourmet techniques while also freeing up money for better ingredients.
After graduation Alex began building his robot, teaching himself mechanical engineering as he went. Without prototyping tools of his own, he became a regular at TechShop, the late lamented chain of open-access DIY fabrication workshops — but this required brutal 800-mile round trips on Interstate 5.
“I built the first robot in my parents’ garage in south Orange County,” Alex says. “There was no TechShop in L.A. then, so I would drive all the way from Dana Point to the Menlo Park TechShop to do parts fabrication, crash on a friend’s couch for weeks, then drive it back to Orange County and see if it fits! It was pretty lonely driving the 5, listening to Stanford ETL lectures on audiobooks. I grabbed a bunch of textbooks — Shigley’s Mechanical Engineering Design, Practical Electronics for Inventors, Machinery’s Handbook — and Make: Magazine was inspiring.”
People just didn’t get it, Alex says. “I spent two years in the garage working alone, trying to push it along, to solve these problems of how robots need to meet food, which is squishy and variable. Everyone’s telling me, ‘You’re crazy. Robots can’t do this, there’s too much dexterity required.’ I found it more socially acceptable to say I was unemployed, rather than working on a cheeseburger robot in my garage.”
Learning robotics from scratch was a steep climb for Alex. “I had a physics background but that’s not engineering. I didn’t know stress from strain, or what a microcontroller was. The first day I learned what an endmill was was the first day I used the CNC — that’s where I met Steve, at TechShop, cranking on the mill!”
An Engineer Appears
Steven Frehn was raised in the no-nonsense city of Palmdale, California, a military and aerospace center in the West Mojave Desert. In high school he interned for NASA at Edwards Air Force Base and placed third in a national science contest to win a scholarship to Stanford, where he earned a degree in mechanical engineering. In 2010, Steve was making parts for his day job designing rooftop solar panels, while volunteering at Menlo Park TechShop in exchange for membership. A cheeseburger robot was the last thing he expected to work on.
But Alex outlined the burger bot dream with his usual enthusiasm and culinary expertise, and soon Steve was milling parts at TechShop and FedExing them down to SoCal, where Alex tested them and then sent requests back for modifications. They started hanging out to do design reviews and share technical challenges.
Steve brought the hands-on fabrication chops and maker DNA the project had been waiting for.
“Two years later I had a GUI on my computer, plugged a USB cable into it, and a burger would appear in my garage!” recalls Alex. “My best friend came over and it was weirdly quiet, like we knew that this had actually happened — it was possible.”
By 2012, Creator had a little incubator money. Alex moved the startup to San Francisco as CEO, and Steve took the leap to join full time as COO. Joined by two more engineering grads out of Berkeley and Utah, they prototyped their second robot at TechShop’s three-story fabrication palace in San Francisco. Seed and venture money from Root Ventures, Khosla, and Google followed, and in 2014 the duo assembled a team of engineers from Tesla, Google, Boston Dynamics, and Walt Disney Imagineering.
Today the fourth-gen robot team is led by systems engineer Noe Esparza, a Stanford PhD roboticist (and long-ago co-intern at NASA with Frehn); hardware engineer Michael Balsamo; software engineer Jeffrey Jensen; and R&D engineer James Brinkman, ex-Navy and proud to be Creator’s “first Craigslist hire!” I quizzed the team about the robot’s development, how it all works, and memorable roadblocks or failures.
Overclocked Dorm Fridge
“I like the culture here, failure is seen as an opportunity,” Michael says. “I came here from Tesla and it was the same way there. Figuring out refrigeration on this machine was hard, with the huge transparent assembly area that’s got these openings on the ends. We built a prototype, but how do you measure airflow? We thought we’d do it like the automakers do, so we got a smoke candle firework and tested it on the roof … ”
“And it makes this huge cloud of smoke and San Francisco Fire Department shows up with a ladder truck!” Jeff laughs.
Steve looks anxious just remembering it. “We explained, ‘We’re a food technology company, we’re testing a prototype, we apologize …’”
“They were pissed,” recalls Jeff with a grin.
“Then we were testing the grinder performance, and they don’t perform so well when they’re not refrigerated,” Michael continues. “So we had to go to Lowes —”
“Dude, you took ours!” interrupts Alex. “We had a dorm-room little fridge and Michael took that …”
“Oh yeah …”
“And gutted it.”
“We gutted it and put it on the grinder, added a shelf fan for active cooling …”
“To overclock the dorm fridge.”
“And installed aluminized fins on the evaporators to really sink the heat,” Michael continues. “We forgot to turn it off once, came back and it was a giant block of ice!”
I Love Lucy
In 2015 team had 24 hours to move their latest prototype robot and set it up for an investor demonstration at Cavallo Point, a swanky resort across the Golden Gate Bridge from San Francisco.
“This was when we still had a regular conveyor belt,” says Noe, “before the butterfly belt. It was straight out of I Love Lucy — suddenly our conveyor turns on full speed — and backwards! We’re running around trying to turn it off, and burgers are like, thunk, thunk, thunk against the back wall!”
“Everyone looked at me!” yelps Jeff. “They thought it was a software issue, but Noe found out it was a bad wire crimp that got shorted when we moved the machine.”
Alex chuckles. “The bacon aioli was really good on that one.”
Massively Embedded Computing
Presumably this giant burger beast is stuffed with microcontrollers? Not anymore.
“Initially we used a lot of Microchip PIC microcontrollers, but that was making it more difficult to be modular, to iterate, to debug,” says Alex.
“Now it’s got 20 computers in it,” Jeff says. “We’re using mostly BeagleBone Blacks, and it’s all Linux, off-the-shelf Ubuntu. Unlike most embedded design for mass manufacture, we didn’t have the same cost and space constraints, so it’s faster to use Linux throughout, and much easier to work with the mechanical engineers; we don’t have to translate down to 8-bit microcontrollers.”
Creator relies heavily on open source software for robot communications (Apache Thrift and libserial, curlpp, and Boost.Signals2 libraries) and for development (Docker, Eclipse CDT, CMake, gtest, Jenkins, React Native, and Atom). And like a lot of makers on a budget, they still design circuits in EagleCAD (and these days, Altium) and order their purple prototype boards from OSH Park.
Grilled by AI
There are a lot of proprietary bits the team couldn’t show me. In particular, the burger grilling module is still a black box, but the results are impressive. (I’m a flame griller at home but Creator’s tasty burger makes a strong case for griddling instead.)
“It’s loaded with temperature sensors and really sophisticated algorithms to get the right doneness — no two burgers are cooked exactly the same way,” Alex says. “There’s some machine learning involved.” Moisture sensors too? Probably but they would neither confirm or deny.
They did share the saga of inventing it. The first prototypes were built with gas burners. “To get enough gas ribbons and starters, we bought three gas grills at Lowes, and stuffed all the burners inside our grill unit,” says Steve. “It got so hot it would singe your hair just looking in it, and wires from 3 feet away would start to melt.”
Electric griddling was much easier to control, so they started over from scratch. “The good news is, griddling gives us a much better burger,” says Alex. “You get better heat contact and a better crust, that Maillard reaction above 310°F. It’s way juicier now. And no flare-ups.”
It’s also far more efficient. “We do electrical heating of griddle surfaces only when there’s a burger on it. A regular restaurant’s griddle is on all the time, the whole surface is hot, there’s a lot of waste. We may be running the world’s only all-electric burger joint,” Alex says proudly. Inside are 11 thermosensors watching ambient and griddle temperatures, constantly adjusted based on machine learning AI algorithms.
The air filtration is impressive, too. The air is weirdly, impossibly fresh inside this burger joint. Yes, there’s the expected grease baffle above the grill — it’s required by building code. “But we built two extra air filters into the robot,” Alex says, “one for smoke, and one for smell. The only thing coming out of the robot is warm air.”
Buns Vs. Steel
Like many parts of the robot, the bun system is on its fourth major revision. “The first robot had the buns stacked vertically in two hoppers, heels and crowns, and an arm would whack the bottom one into the toaster. In general there were a lot of issues with squishing,” says Alex.
“So we rejected that and tried a single horizontal feed for the bun slicer,” says Noe. “We found that a Gatorade bottle was the exact diameter of a bun, so we filled it with different amounts of water to test the weight.” The horizontal feed is fun to watch. More importantly the buns aren’t smashing each other and they stay fresh until sliced to order.
Noe explains the solution. “It’s pneumatic: air is pushing the wood block piston, with lasers to read the position of the bun at the end, providing continuous feedback to the air system.”
Alex elaborates, “The normal way is a conveyor belt, but that has lots of moving parts, and we wanted something easier to reload. We tried sucking the buns with a vacuum, we tried blowing them with air, but that would dry them out. We were using a Shop-Vac and manifolds we made from MDF at TechShop.”
“Then we came up with the idea for a pneumatic pusher piston,” says Noe. “That worked as far as the mechanics, but then there’s a lot of algorithmic stuff you’ve got to do, the sensors and reactions.”
“There are industrial slicers that slice a lot of buns all at once,” Alex says, “but one-at-a-time-slicing, to order? It’s tricky. Nobody had done it before. There was no place to start.”
The bun slicing unit uses infrared and laser sensors to detect when each bun has arrived, then actuates the blades as servos move another wood block piston to push the bun through the blades.
The two-sided vertical toaster conveys the bun heel and crown through a buttering station — where secret technology applies liquid butter evenly — and three griddling stations. Alex chimes in with the food nerd science: “A little layer of butter really helps thermal conductivity, caramelization, and crispiness.”
The buns are moved along and pressed onto the griddle by a series of 16 paddles, all actuated and timed by a camshaft that wouldn’t look out of place in your car transmission except for the beautiful wood and aluminum design. Noe roughed it out in CAD and then iterated by hand, cutting linear cam profiles and feeding them through the paddles to observe the resulting motion and model a final sequence that toasts the bun optimally and also looks good to watch.
Like most of the robot, the toasting system is the product of trial and error. “First we tried an infrared toaster off the shelf, then we built our own infrared toaster,” Noe says. “Then finally we built our own contact toaster.”
Four “handoff paddles” at the bottom of the toaster drop the crown and heel, butter side up, into their respective halves of the open clamshell box, which Creator calls a “case.” On our first visit this drop sequence was highly accurate, but not without the occasional inverted bun. The team is continually fine-tuning the paddle motions using motor position feedback data from the servos.
Move It Stop It
Sauce It Slice it
The saucer today is a far cry from the three upside-down condiment bottles in the early prototypes. It dispenses 15 different sauces including of course homemade catsup, mustard, and Pacific Sauce (a California twist on your usual Thousand Island-y burger sauce) but also charred onion jam, sunflower seed tahini, shiitake mushroom sauce, and oyster aioli. Positive displacement pumps, monitoring flow to the milliliter, are a hybrid of off-the-shelf systems and Creator’s own technology; challenges included dealing with different viscosities and particulates. Sauces are distributed in a 2D pattern on the bun by an X-Y “sauce head” that moves just like a plotter or 3D printer head.
At the next station, washed, trimmed whole veggies in three hoppers (onion, tomato, and pickles) are gravity fed to blades that sense their own position, then sliced to precise thickness. But veggie slicing was a thorny problem for years.
“[Root Ventures founder] Avidan Ross said the proper thickness of a sliced onion for a burger is so thin you should be able to read through it,” Alex recalls. “Once we got it dialed, we sent him a picture of a slice over the top of a message: ‘Is this thin enough?’”
Slicing a tomato may seem simple, Alex says, ”but we’ve working on the slicers longer than probably anything else. A lot of IP goes into getting that last slice! I can’t even do it right at home.”
So how much IP is in this robot? “We’re probably at two dozen patents at this point, and a lot of them are really broad ‘omnibus’ patents,” Alex says with a gleam in his eye. “No one’s ever done this.”