Why Is It So Hard to Make a Robot Chef?

"The robot in the kitchen was upset.

Beep. Beep. Beep. Beep. Beep.

Steve Ells ignored the noise as he expounded on the future of the restaurant industry (hint: robots!) and how his food chain, Kernel, will play a leading role.

Mr. Ells is best known for starting Chipotle, the burrito giant, in the mid-1990s. Now, with Kernel, which has two locations in New York City and a third on the way, “we are reinventing the kitchen job,” he said.

On a warm day in late October, he nibbled on a chicken-salad sandwich that had been partly prepared by a robot that looked as if it might have relatives welding cars at Toyota. The machine was quickly swiveling, extending its arm to remove chicken thighs and carrots from an oven heated to 515 degrees and deposit them on a steel counter where human employees assembled sandwiches and salads before bagging them for customers, also presumably human.

But as the robot continued to beep its displeasure, Mr. Ells stopped speaking, cocked his head and listened carefully as he chewed. “Well, that’s a new sound,” he said, sighing.

The robot revolution still has some kinks to work out, at least when it comes to cooking.

In recent years, restaurants have dabbled with ways to automate their operations.

Labor typically eats up more than a quarter of restaurant chain revenues, so McDonald’s and Dunkin’ have installed kiosks where customers punch in their orders, reducing the number of cashiers.

Others are looking to artificial intelligence programs for drive-through orders or inventory management.

In the post-Covid environment, where restaurant chains continue to struggle with labor shortages and soaring costs as well as orders that come out wrong, many are now seeking ways to automate at least some functions in the kitchen.

Chipotle, where labor costs could reach $2.7 billion this year, or a quarter of its revenue, has been testing the Autocado, a machine that can cut, pit and scoop out an avocado in 26 seconds. White Castle has installed a system called Flippy to make fries and other items in 18 locations in the United States.

And Sweetgreen, the fast-casual salad chain, will have an automated salad-making system, the Infinite Kitchen, in about a dozen of its 240 restaurants by the end of this year.

But despite high interest in robots in the kitchen and years of efforts, they have yet to take off.

The beeping that interrupted Mr. Ells’s train of thought as he ate lunch turned out to be a reminder to check if supplies needed to be replenished. No big deal, especially compared with the early days, when the brioche buns kept getting stuck in the toaster, shutting the entire cooking system down. The automated system, he explained, was still a work in progress, requiring tweaks here and there. Little did he know that just a few weeks later, a much bigger adjustment would be needed.

Bespoke Barriers to Entry

Unlike car factories or Amazon warehouses, which rely on robots to perform repeatable actions, restaurant kitchens run on multitasking. Flipping pancakes requires a different system from one that dispenses coffee or makes spaghetti. Moreover, producing potentially sticky, gooey or cheesy foods means robots need to be easy to clean to maintain food-safety standards.

“These systems are largely bespoke,” said Sharon Zackfia, group head of consumer research at William Blair, rattling off reasons that robots won’t replace fry cooks and prep chefs anytime soon.

Moreover, while automated systems may reduce labor costs over time, many have a steep upfront price tag — typically six figures.

“The restaurant has to have deep pockets,” Ms. Zackfia continued. But, she cautioned, “it has to be largely company-owned and not franchised.” At a chain like McDonald’s or Burger King, franchisees, who own most outlets, would have to bear the costs.

McDonald’s opened a nearly fully automated test system at a company-owned restaurant outside Fort Worth in 2022, but executives say it’s unlikely that a slew will follow.

“We’ve spent a lot of time, money, effort looking at this, and there is not going to be a silver bullet that goes and addresses this for the industry,” Chris Kempczinski, the chief executive of McDonald’s, told analysts and investors on an earnings call in 2022. “The idea of robots and all of those things, while it may be great for garnering headlines, it’s not practical for the vast majority of restaurants.”

Nor is it easy to squeeze an automated system into already tight kitchens or retrofit an entire kitchen.

Sweetgreen took seven weeks to overhaul its location near Madison Square Garden in New York with an automated system. It is now focusing more on new outlets, where it can build for automation from the start. Of the 40 restaurants the company plans to open in 2025, about half are expected to be automated, at a cost of as much as $550,000 each.

In 2021, Sweetgreen acquired Spyce, a Boston-based robot restaurant founded by students at the Massachusetts Institute of Technology. The purchase sped up Sweetgreen’s efforts to automate its salad-making.

With the automated Infinite Kitchen, employees still cook chicken and brussels sprouts and slice vegetables. They then feed those ingredients into large containers that dispense the food into bowls on a conveyor belt beneath them. But the system had to be modified so that the conveyor spins the bowls as they move along the line, allowing the individual ingredients to be placed in different parts of the bowl and avoiding a lava-like overflow. Furthermore, the system needed to slice hard-boiled eggs and adjust proportions of the different types of locally grown kale.

Even with all these adjustments, some delicate foods, such as avocados and cooked salmon, are placed in bowls in the final stages by human hands.

“If all it was doing was dispensing almonds into a salad, that would be easy,” said Jonathan Neman, a co-founder and the chief executive of Sweetgreen. “But we needed a machine that could be easily set up, cleaned and scaled at a cost that made sense to us, a cost that could be used at all of our locations.”

The system is worth it, Sweetgreen executives say. It can churn out 500 bowls in an hour, compared with a top human speed of about 300, and locations using the Infinite Kitchen are considerably more profitable than the average. The automated Sweetgreen in Naperville, Ill., has a restaurant profit margin of more than 31 percent, well above the 20.7 percent average for the chain. (The chain itself has yet to turn a profit, although losses are narrowing.)

A move toward automation in restaurants could save the quick-service industry as much as $12 billion a year in wages, the consulting firm Aaron Allen & Associates estimated. Those wages are often meager (the Bureau of Labor Statistics put the median wage for food service workers at $14.29 in 2023), but they offer a lifeline for millions of people without a college education or other skills.

Nor can automated systems replace human problem solving or the experience of talking to another person, noted Elyanna Calle, president of the Restaurant Workers United union. “It’s just different having a conversation with a human rather than punching buttons on a screen,” she said.

‘A Wickedly Complicated Endeavor’

When Mr. Ells started Kernel, he and his team looked around for the perfect robot system. When they couldn’t find one off the shelf, he brought in Tom Cortese, a co-founder of Peloton, the connected-fitness company, to help create and design the assembly-style system.

Mr. Ells had experience building a brand from scratch. At Chipotle, he expanded the chain to more than 2,350 stores before a number of serious food-safety issues led to his replacement as chief executive in 2017. He stepped down as executive chairman of the board of directors in 2020.

Mr. Ells sees Kernel as a quality fast food chain that, like Chipotle, can grow. But he also views it as an opportunity for the company to license a robotic kitchen system to other restaurants.

“This is the platform of the future for the restaurant industry,” he said in October.

The industry’s robotic future, however, is going to be a bit delayed. Two months after the lunch, Mr. Ells decided that Kernel and its automated system needed a reboot. Rather than continue to make incremental changes, he said, he is closing the restaurants in a few days “to go to version 2.0.” That process is not likely to be completed before March.

“This is a wickedly complicated endeavor, automating restaurant work,” Mr. Ells said on a phone call in mid-December. “After 10 months or so with this technology, we have all of this data, and now it’s time to build the new system.”

The essence of Kernel and its automated system will remain the same, he insisted. The food will still be chopped, seasoned and par-cooked (by humans, who now earn between $22 and $25 an hour) at a central kitchen in Manhattan and delivered to Kernel’s restaurants, where employees load it into the bins, racks and pans that make up the automated system.

The robot will still do much of the heavy lifting, which, at Kernel, involves pulling pans of partly cooked meats and vegetables from racks and placing them in an oven for a few minutes to roast. But Mr. Ells wants the new system to be able to heat larger batches of food, rather than the current one-plate-at-a-time system.

Most of the workers now at Kernel will be out of a job when it shuts down for the reboot. But Version 2.0 of Kernel’s kitchen will not eliminate the need for human labor. The robot cannot yet dice, chop or sauté. It can’t taste for correct seasoning. It can’t make sure the side dishes are in the right bag.

The kitchen, for now, is still mostly the human domain.” [1]

1.  Why Is It So Hard to Make a Robot Chef? Creswell, Julie; Karlsson, Brian.  New York Times (Online) New York Times Company. Dec 22, 2024.