A robot makes its way along a city sidewalk. It stops, shifts, crawls forward. It blinks a pair of perfectly round, LED eyeballs. A few pedestrians stop and stare. But the bot, which resembles a picnic basket on wheels, is unbothered. It pauses at a crosswalk, then trundles on. Inside its insulated belly is someone else’s lunch.
It’s like a scene out of a science-fiction novel. Except it’s not. In big cities across the country, from Los Angeles to Chicago to Miami, sidewalk-crawling food delivery robots are becoming part of the scenery. Ray Bradbury, eat your heart out.
There are roughly 2,000 of these machines roaming U.S. sidewalks and college campuses today, equipped with cameras, sensors and sophisticated software that help them get around. They’re operated by companies like Coco, Serve and Starship Technologies, with delivery platforms such as DoorDash, Uber Eats and Grubhub connecting the dots between restaurants, customers and robots.
With financing flowing into the space, the number of bots is set to more than double this year and increase exponentially in the years to come.
Starship has more than 1,200 delivery bots in the U.S. | Photo courtesy of Starship Technologies
The bots will have to multiply if they are to fulfill the considerable hope being placed on their boxy shoulders. The robotics companies say that their fleets, once at scale, could reshape not only the restaurant delivery market but also the urban American streetscape. As the cost of building and operating robots comes down, customers will be able to have food delivered for as little as $1, they say, increasing demand for the service and easing the logjam of carbon-emitting cars on the road.
Robot delivery holds distinct promises for restaurants, many of which have tolerated delivery for years despite the considerable expense and operational headaches it brings. Hundreds of restaurants are now plugged into robot delivery networks, including large chains like Shake Shack and White Castle, in hopes that they will make the process more efficient.
“If a business can find a way to bring down delivery fees, to be more environmentally conscious and bring down costs both for restaurants and the consumer, it’s a win-win-win,” said Seth Cohen, president of Los Angeles-based Sweetfin, which uses Coco’s delivery bots. “I think that’s great. There’s optimism there.”
At the same time, there are some hurdles ahead for delivery bots, which are still limited by technology, geography and demand. With low speeds and delivery radiuses of just a couple of miles, there are only so many markets that make sense for sidewalk bots today, and only a handful of those have meaningful coverage.
“We have barely scratched the surface,” said Touraj Parang, COO of Serve, who noted that the global food delivery market is expected to reach $450 billion by 2030. “We are excited by it, we are energized by it … but we have lots of ground to cover, literally.”
Food delivery robots have been around since at least 2015, when Starship Technologies began test-driving a prototype in Europe. They began to gain traction during the pandemic as demand for food delivery exploded. And over the past year or two, they have found another gear. Robotic delivery companies are closing big funding rounds. Small pilot programs are evolving into full-blown rollouts. The bots are even beginning to show up in pop culture, acting as mascots for the AI era.
“Whether it’s the headlines around the industry, or the actual physical product sinking in out in the wild, I think that gut feeling [that delivery bots are progressing] is spot on,” said Aaron Campbell, founder of Operation Autopilot, a consulting firm that works with autonomous vehicle startups.
Much of that progress has been a function of financing. Both Starship and Coco recently closed eight-figure funding rounds, while Serve raised $200 million last year after going public in April.
The influx of capital has allowed the companies to rapidly accelerate production. Starship currently has more than 1,200 robots in the U.S.; Coco has hundreds and expects to have over 1,000 this year and 10,000 by the end of next; and Serve has about 300 and is under contract with Uber to reach 2,000 by year’s end.
“What we’re seeing in the past year is going quickly from dozens of robots to now hundreds of robots,” said Megan Jensen, head of autonomous delivery operations at Uber. “We’re seeing a really big ramp up.”
It helps that delivery robots have also become easier and less expensive to manufacture. When Harrison Shih was developing sidewalk bots for San Francisco-based Marble in 2017, he was cobbling together parts from mobility scooters and giant batteries that weighed hundreds of pounds.
That was before electric scooters and e-bikes became staples of modern urban life, which helped to establish a supply chain for so-called micro-mobility vehicles. Now, the parts needed to build a good sidewalk robot, like small motors and lithium-ion batteries, are cheap and readily available.
“Today, just off the shelf, you can buy things, and there’s manufacturing already set up, and costs have come down drastically compared to what we saw a decade ago,” said Shih, who is now head of product for DoorDash Labs, which leads the company’s automated delivery efforts.
Building more bots is a key piece of the automated delivery equation, because it will create economies of scale that will bring down costs for everyone involved.
But the bots’ success will also hinge on their ability to complete deliveries quickly and safely. And that comes down largely to a question of artificial intelligence.
Today, delivery robots are not completely autonomous. All three companies interviewed for this article still rely on remote human pilots to step in when their machines get into trouble. The bots tend to need help crossing busy streets, for instance, and they can get caught up by common obstacles such as curbs, tree roots and potholes.
Still, their autonomy has made strides in recent years thanks to watershed developments in AI, particularly OpenAI’s launch of ChatGPT in late 2022.
The company’s big breakthrough had to do with the way it trained the model behind the human-like chatbot. It used massive data sets and some hands-on coaching that enabled the AI to keep getting smarter and smarter over time. That approach has since spread to other AI applications, including delivery robots, which have generated millions of miles of valuable training data already.
Previously, there was only so much an AI model could learn from that data before it began to “plateau,” said Zach Rash, CEO of co-founder of Coco, which is backed by OpenAI CEO Sam Altman. That training limitation meant that robot delivery companies had to manually create rules for their bots, like that they must stop at a stop sign.
But navigating city streets is not that cut and dry. There’s an almost infinite number of scenarios that a bot may have to respond to. It can be programmed to stop at a stop sign. But what if a car blows the stop?
“You want a system that just learns, like a human does, how to operate in those environments without creating hard rules,” Rash said. “These new AI systems are really good at doing that.”
A Coco robot crosses the street in Santa Monica, California. | Photo courtesy of Coco Robotics
The hope among robotics companies is that the bots’ autonomy will advance to the point that fewer humans will be able to monitor increasingly larger fleets, which will help bring costs down. But whether the bots will ever be completely self-sufficient is still unclear.
“That’s, like, a trillion-dollar question,” said Campbell. The answer will depend on both technological advances and human factors such as regulation and safety.
On the second point, Campbell said that sidewalk bots have proven to be relatively low-risk. They’re unlikely to hurt someone by running into them. The worst that can happen is one of the bots gets hit by a car, he said. From a regulatory perspective, that could make the path to total autonomy easier.
“For sidewalk deliveries, I could see a world where humans are mostly, if not entirely, in a mission-completion sense, out of the loop,” he said. “But do I think we’re there right now and that’s a wise move or a way to get costs to zero? Not necessarily.”
For now, the urban jungle remains full of pitfalls for delivery robots. Online, there are countless videos of bots imperiled by curbs, trees, snow and even fellow robots.
Robot operators maintain that these problems are navigable, especially with human oversight. Starship’s bots have managed to survive in Finland, where the winters are long and cold.
“We have snow and black ice and all sorts of things. We have very narrow streets and very large ones,” said Valentin Naidja, Starship’s SVP of revenue. “Pretty much anything you can think of, we’ve probably operated in.”
The question he gets the most, though, is about vandalism. Do the defenseless bots become targets for troublemakers? Naidja said robot abuse is extremely rare, affecting less than 1% of deliveries. “It just doesn’t happen,” he said.
On the contrary, he said, people love the bots.
“Just look at them. They’re just cute,” he said. “There are parts of the country where you will have a lot of snow and a robot will be on the side, a little bit forlorn and stuck, and you will have a good Samaritan helping the robot.”
The robots’ friendly appearance is actually quite calculated. Serve wanted its bot to look approachable, although not so much so that it would be treated “like a little toy on the sidewalk,” said Touraj Parang, president and COO. It took inspiration from other wheeled objects people are familiar with, like strollers and shopping carts.
“We had to hit that happy medium between what’s functional and what’s also neighborly,” Parang said. The result is like “if a Minion and Wall-E had a baby.”
Serve's robot is designed to look friendly but functional. | Photo courtesy of Serve Robotics
The bot’s big googly eyes and rounded edges have helped it make friends, Parang said, including in Hollywood, where Serve’s robots have become a frequent sight. Last year, producers from the Netflix talk show “Everybody’s Live with John Mulaney” reached out to Serve about featuring a robot on the program. The bot, dubbed Saymo, has become a recurring character on the show, with at least one critic praising the bot’s “star turn.”
“Every celebrity I’ve ever wanted to meet has met our robot,” Parang said. “So I’m very jealous of the robot in that sense.”
The anthropomorphism hasn’t worked on everyone. Some pedestrians have complained that the bots are an impediment, particularly for people with disabilities. And naturally, there are also concerns that the robots could end up taking jobs from human delivery workers. In 2022, a pair of courier unions in the U.K. sounded the alarm about the rise of automation in their industry.
Robotics and delivery companies said that currently, demand for delivery outweighs the number of couriers available. Robots, they said, can help close the gap by handling short-distance orders in dense areas, leaving longer and more lucrative trips for human drivers.
They also said that by making delivery more affordable, robots will help increase demand for the service, creating even more opportunities for delivery people.
“We really do believe that the number of human couriers and delivery robots are going to continue to grow side by side for quite a long time,” said Jensen of Uber.
For restaurants in big cities, the bots do present some advantages over human delivery people. At 16-unit Sweetfin, Coco’s delivery bots typically complete deliveries faster than drivers, said Cohen. Even though they’re physically slower than cars, they get a head start because they linger outside the restaurant, ready to go as soon as an employee loads in a meal.
“That’s just a benefit of utilizing the technology. We’re not waiting for drivers to show up,” Cohen said. This is important for Sweetfin, whose fresh poke bowls are best consumed ASAP.
Long-term, the potential cost savings created by robot delivery could have an even bigger impact for the brand, where about a quarter of sales come from delivery. Today, the restaurant pays about $3.99 or $4.99 per Coco delivery, Cohen said, which is slightly less than the cost of regular third-party deliveries. But it is still far from Coco’s stated target of $1 per delivery.
To get there, “the cost to produce the robots has to be insanely cheap, and they’d have to be insanely efficient,” completing many deliveries per hour to generate enough revenue to pay themselves off, Cohen said.
“Again, I’m a supporter,” he said. “I’m just asking questions.”
Rash said Coco is “well on our way” to hitting the $1 benchmark and believes it can get there in a matter of a few years. Supply is growing, manufacturing costs are coming down, and demand is increasing thanks to third-party delivery networks.
“So then the key enabler is, you need the level of autonomy to be really high, and I think that’s a really easy line to extrapolate with the kind of progress we’ve had in the industry lately,” Rash said. “It’s just a matter of time and data before that happens.”
But there’s another, more fundamental obstacle standing between delivery bots and mass adoption, Campbell said: Getting into new markets. Every city has its own regulatory, labor and business landscapes to navigate, not to mention the streets themselves. Support teams have to be established along with restaurant and delivery partnerships. All of that takes time and money.
Campbell said he’d like to see cities doing more to support the growth of delivery robots by changing zoning rules or offering subsidies, especially given the bots’ potential to improve quality of life by easing traffic and pollution.
“It would help drive up the ability to supply the market, and that demand there would ultimately help lower the cost,” he said. “Right now, they have to price in the fact that they can’t produce 10,000 units if they don’t have 10,000 cities or places or streets for them to go.”
Uber is currently supporting robotic delivery in 10 markets, and Jensen acknowledged that the policy backdrop has something to do with where it has chosen to go live. But she has no doubt that automated delivery will continue to rumble through roadblocks over the next few years.
“We’re so early on the scaling side, and the gates that we have to continue to move through are adoption, both on the customer side and the restaurant side,” she said. “I think it’ll become much more the norm. Less early adopter, more mass adoption and appeal.”