On Friday morning, Matthew Johnson-Roberson sat at his office window in Ann Arbor, Michigan, looking out at his favorite restaurant gone dark. As cities and states rush to stem the spread of the coronavirus, restaurants across the country are closing doors to eat-in diners. “The economic impact is huge,” said Johnson-Roberson, an associate professor of engineering at the University of Michigan. “I’m really worried that these restaurants aren’t going to come back.”
Many, however, are open for takeout and delivery, and that is where Johnson-Roberson’s robots come in. Johnson-Roberson is co-founder of Refraction AI, one of a growing number of startups that aim to deliver restaurant or supermarket food to your home in the tamper-proof compartment of a robot on wheels. Some big players have entered the field, from Amazon to Postmates. But the industry has been slow to grow. “There have been a lot of small and limited test deployments,” says Wendy Ju, an assistant professor of information science at the Jacobs Technion-Cornell Institute at Cornell Tech. “It’s not a business that makes money yet.”
Now, in a time of social distancing and Instacart delays, the food delivery robot could finally have a moment. According to Johnson-Roberson, his restaurant partners in Ann Arbor have tripled, and online orders max out daily. But despite the increased demand, this pivotal point may have come too soon for the delivery robot to rise to its full potential.
In some communities, people have become accustomed to food delivery robots. Residents of Milton Keynes, a town northwest of London, can send Starship Technologies robots to fetch groceries. Students at George Mason University in Virginia can send these same robots to pick up meals from vendors across campus. Hungry folks in Berkeley, California, and Denver can order food through the Kiwibot. And residents of Snohomish County, Washington, receive some Amazon orders from the Amazon Scout.
I encountered my first delivery robot in D.C. A Starship robot was paused at a crosswalk by its human handler like an obedient dog off leash. The robot, its handler told me, was learning the streets so that it could one day cruise solo. That was three years ago. Since then, I’ve ordered more dinners from Shake Shack than I care to admit. Not one came in the compartment of an autonomous cooler on wheels.
“This is definitely a growing industry. But the growth hasn’t been as fast as people expected for many reasons,” says Tinglong Dai, an associate professor at the Johns Hopkins Carey Business School who specializes in operations management and business analytics.
Dai points to the suburbs as one barrier. Distances between homes and restaurants are too vast for robots rolling along sidewalks at the pace of a strolling human to traverse in any reasonable amount of time, though some companies like Nuro are experimenting with autonomous delivery cars.
Then there’s the challenge of building the infrastructure for what is still a relatively new business type. Kiwibot’s head of business, David Rodriguez, says that his company has spent the past two years developing expertise in this area. “The logistics, the know-how, the app, all of the support,” he says. “The robot is only a small element of the entire system.”
The industry has made great strides in developing the A.I. technology that enables robots to learn and navigate terrain. “Almost anyone,” says Ju, “can make a robot that can traverse on its own and not get lost.” Human interaction and unexpected activity in the environment remain the real problems, she says.
Johnson-Roberson’s team designed its REV-1 robots to operate in bike and car lanes so that the robots don’t need to understand the nuanced rules of sidewalk right of way. But the company is still working on teaching the robots to recognize when a pedestrian is about to jump into the bike lane. “As drivers, we get really good at that,” he says. “It’s a really hard thing to teach robots.”
Though many delivery robots in development now roam without a human handler at their side, none are truly autonomous. Employees sit in a command center, ready to assume control whenever a robot comes to a tricky intersection or unexpected road closure. “Our goal,” says Johnson-Roberson, “is that the ratio of drivers to robots is something that makes this economically sustainable.”
Dai and Ju both expect that the delivery robot industry could advance during this uneasy time when a comfortable distance between friends is 6 feet or more. “People trying to minimize human interaction is a scenario in which the robots are perfect for,” says Dai.
“[Delivery robots] may in fact help with reducing some amount of risk,” says Barun Mathema, an infectious disease epidemiologist at the Columbia University Mailman School of Public Health. He suggests that any measure to limit contact between an infected person and a susceptible person is worthwhile. But he also notes a recent study in the New England Journal of Medicine that suggests that the virus can live on cardboard for up to 24 hours and plastic for up to two to three days. Even when taking human interaction at the door out of the equation, a person is still involved with the packaging.
Whether or not replacing humans makes deliveries safer, having robots shoulder some of the delivery load could make deliveries more timely, since current food orders far outpace the human capacity to fill them.
“I think 10 years from now we will be in a situation where this kind of event is something that we could just pull out all of the robots for,” says Ju. Today’s fleet of food delivery robots is too small to be widely deployed. But within the small footprints where they currently operate, food delivery robot companies are attempting to meet the increased demand.
Refraction AI is building new REV-1 robots—at a pace of about one every other day—to serve the 400 customers in a pilot group. In China, the delivery app Meituan Dianping released autonomous delivery robots onto public roads for the first time. Starship expanded its delivery area in Milton Keynes to offer service to an additional 180,000 people. At George Mason University, the robots kept running after most students had left campus. “I honestly appreciate it,” said student Clinton Carlson Jr. last Friday. “It’s delivery without having to have that physical interaction with other people, so I feel like it’s less risk.” But all restaurants on George Mason’s campus closed Sunday, so the robots are no longer running. (Update, March 26, 2020: Though the robots were down for two days, they have now returned to work to deliver food from two restaurants, possibly with more to follow.)
Kiwibot has added face masks, hand sanitizer, and gloves to its regular menu of takeout food options. “Our goal is to mitigate what’s going on,” says Rodriguez. “But moving forward, we want to be agents that advance recovery because this is going to hurt many businesses.”
But expanding during a pandemic is difficult, even for a service that’s in demand. These businesses are working under the same constraints that the rest of us are under, says Dai. “They require investments and people to work together, not just from home.”
Assembling new robots at Refraction AI is usually a team project. But respecting recommendations for social distancing, Johnson-Roberson now only allows a single employee to work on a robot at any one time. When that person leaves the office, another tags in. His remote operators continue to watch over the robots, though now from the safe distance of their own homes.
Even under these conditions, Johnson-Roberson expects that his operation could expand within the next few months to handle most of the deliveries in this college town. Reaching beyond Ann Arbor, however, seems unlikely. “There’s no way we can take over the [national] delivery supply chain,” says Johnson-Roberson. “But I do think that this is a moment to show people that there are ways of doing this differently.”