If the future of restaurants will involve robots, then it’s being cooked up in Boston.
Two separate teams of engineers are chasing the holy grail of complete order accuracy, while also trying to surpass the speed of a human worker. One team works for a startup, Dexai Robotics, and the other for a publicly traded company, Sweetgreen, which in 2021 bought a local startup, Spyce, in a deal that could be worth over $50 million. The teams are both based in Charlestown.
Dexai recently landed an order for 10 robots from the Defense Logistics Agency, for installation at military bases around the United States. Sweetgreen plans to roll out the technology it is developing in two locations later this year, according to cofounder Nicolas Jammet. Bostonians got a glimpse of an earlier generation of the Sweetgreen meal assembly tech at restaurants that Spyce operated in Downtown Crossing and Harvard Square.
On a recent Wednesday, I dropped by Dexai’s office at the invitation of Dave M.S. Johnson, the company’s chief executive. Johnson had told me earlier that a well-trained human in a kitchen can assemble a salad in about 90 seconds, and that Dexai had managed to break the 2-minute barrier, putting together a salad with five ingredients in 1 minute and 45 seconds, on average. (I’d seen one of the robots at work last April, at a conference, and it moved somewhat ponderously.)
Dexai had four robots running, each assembling the same salad in a beige compostable bowl: a bed of couscous, topped with lettuce, onions, black olives, and feta cheese. Supervising the bots was Brian Krieger, the company’s cleverly titled chef de technologie. He previously worked at a Cambridge restaurant and at the Charles Hotel. The Dexai system is built around a robot arm made by Franka Emika, a German company.
First, a small lever with a suction cup at the end of it picks up the paper bowl. It makes three attempts to try to separate the top bowl from the others in the stack, but even then, it sometimes fails and picks up two. (Krieger says each bowl costs 56 cents, so every mistake adds a bit to the cost of the salad.) Then, the white robot arm identifies a series of specially designed kitchen implements, using a camera that can spot black-and-white QR codes printed on the end of them. It can connect to salad tongs or scoops, and place food in the bowl. The part of the robot that holds the bowl can instantly weigh it, to determine whether enough lettuce or olives have been added before the arm moves on to the next ingredient.
Krieger monitors a central station that alerts him when a robot runs out of an ingredient. He is also responsible for cleaning up spills at each salad station — the robots don’t seem much messier than a human would be — and for putting a plastic lid on top of the bowl. Occasionally, there are other issues, like feta that gets mashed into a corner of its plastic bin and needs to be loosened up so the robot can get to it with its scoop. Krieger also picks up the finished salads and stacks them on a cart. For this lunchtime production run, each robot is making 15 salads.
I time a few salads, to see how fast the robots can work. The first one takes about 2 minutes and 22 seconds; Johnson says that the slower times can happen when the robot needs to make extra grabs or scoops of ingredients to get the right amount into the bowl. But the next two salads take 2 minutes and 7 seconds, from bowl in “hand” to finished salad. Johnson says the team has ideas about making the process faster, like by having the robot take pictures to assess its environment while it’s moving, rather than stopping to do that.
The salad tastes great, dressed with balsamic vinaigrette. I probably would’ve added chicken, chickpeas, or croutons, but the robots weren’t set up to take custom orders.
Dexai has raised $12 million since it spun out from Draper Laboratory, a nonprofit research and development firm in Cambridge, in 2018. The company has 15 employees. Johnson says it’s doing ongoing pilots and demonstrations with potential customers, and has one robot installed at Travis Air Force Base in California, as part of its deal with the Defense Logistics Agency. He says many organizations that run big cafeterias — colleges, hospitals, companies, military bases — seem to like the idea of automating some of the more routine and high-volume parts of their daily work. While Dexai has been focused on salads, Johnson says the technology can also be used to assemble various kinds of meals.
Dexai’s rivals in the robotic lunch race got started a few years earlier, in 2015, as a startup called Spyce. After opening two restaurants in Boston and Cambridge, Spyce was acquired by Sweetgreen, a publicly traded California company that runs 180 healthy food restaurants around the United States. Both of Spyce’s local restaurants were closed a few months after the acquisition, and Sweetgreen grew quiet about its plans for the technology.
After messages to Spyce’s founders went unanswered, Jammet, the Sweetgreen cofounder, said he would speak with me. The Spyce team has largely remained intact, he says, with roughly 20 employees. He says it’s been working on a third generation of the technology, advancing what Spyce had developed for its Harvard Square location. That system could work with hot and cold ingredients — even adding a sear to precooked chicken or broccoli.
A key difference between the Dexai and Spyce approaches is that Dexai’s robot can be installed in an existing kitchen, and perform a specific task. Spyce, now part of Sweetgreen, is more of a complete assembly-line system that requires a restaurant be designed around it.
“If you walk into any typical Sweetgreen, you can see how much of our energy is around assembly,” Jammet says, noting that there are often separate production lines for orders placed digitally, versus in-store. The new technology will merge those two lines into what Jammet calls “the Sweetgreen engine.”
“It allows us to rethink the customer experience, and redeploy people to hospitality moments” where they can interact with customers rather than toss items into a bowl, he says. “The team member gets to be friendly, and the engine gets to be fast and accurate.”
Jammet notes that at busy times, even with well-trained staff, mistakes inevitably happen. He also says the average turnover in the quick service restaurant industry is about 150 percent, meaning a restaurant like Sweetgreen will have to replace the average cashier or prep cook 1.5 times in a year. “No job will be eliminated with this technology,” he says. Instead, “these machines require teaching skills around maintaining them — how to be a technician, for example. That creates new career paths.”
The company looked at both Dexai and Spyce as potential acquisitions, but was more impressed that Spyce had shown how its technology could work in a real-world environment. “They had customers and revenue,” Jammet says.
At some point in 2023, Sweetgreen plans to introduce the new assembly technology in two stores, Jammet says, though he doesn’t want to disclose where they will be. He says the team in Charlestown has been running tests with customers in its office.
As Sweetgreen adds new locations, he says, senior executives are “hyper-sensitive to our product getting worse as we scale,” and they regard technology as a countervailing force. “We really believe that there is this future, evolved state,” he says, “and we think automation is going to play an important role.”
Lauren Abda, a partner at Branch Venture Group, agrees. “Finding people willing to perform all back-of-house and front-of-house tasks” in quick service restaurants has been a major challenge “well before COVID,” she says. “The need for some level of automation to quick service restaurant workflows is almost becoming a necessity.” (Branch is a Boston-based network of individual investors who put money into food startups.)
“The honest truth is that many of the line jobs in food service are extremely repetitive and often back-breaking — and too often lead to more of the same for years and even decades,” says John Pepper, a cofounder of the restaurant chains Boloco and B.Good.
Pepper and several other local investors, like Steve Fredette, founder of Toast, the Boston restaurant technology company, have put money into Hyphen, a San Jose, Calif., startup that is taking a similar approach to Sweetgreen, designing an automated assembly line system for food.
As robots and assembly lines begin to show up in restaurants, the biggest question is how they will be regarded by workers. As time-saving devices, like a good dishwasher, or as job eliminators? In recent months, chains like Starbucks and Chipotle have been the target of successful unionization efforts.
Jammet says as Sweetgreen has been developing its new production technology, which it calls the “infinite kitchen,” “every step of the way, we brought the operations team and our team members along with us.”
Pepper says automation is “inevitable” in the restaurant industry, which is perennially searching for ways to keep a lid on food, energy, and labor costs. Unionization efforts, he says, “will obviously accelerate the move to automation.” He envisions a future where restaurants offer “fewer, but more fulfilling job opportunities for those who love hospitality.”
For now, though, the robots are still figuring how to match a human’s speed at making lunch.
Scott Kirsner can be reached at firstname.lastname@example.org. Follow him on Twitter @ScottKirsner.