Hanging on the wall of Postmates’ stealth R&D laboratory, there’s a framed photo of an iconic scene from Star Wars, Luke Skywalker bent down beside R2D2. Except someone has used Photoshop to replace Luke’s face with Ali Kashani, Postmates’ VP of Robotics. Nevermind that Kashani has never seen Star Wars (he considers this a point of pride). Kashani recognizes the symbolism of his face in a world where robots roll around next to people, where bots act almost like friends.
Kashani joined Postmates a year and a half ago, with a special mission to bring robots to the company. In the seven years since its founding, Postmates has been on the forefront of the on-demand revolution, averaging 4 million deliveries each month in over 550 cities. Now, it’s turning its sights to something bigger: learning how to design and build its own delivery rovers inside Postmates X, its new in-house laboratory.
You can see this vision whirling around in the company’s production garage. Postmates’ rover, called Serve, stands about a meter high. It’s shaped like a child-sized shopping cart, with bright yellow paint. On top, a touch screen tells you it’s on the clock—“ON DELIVERY,” it says—and a flashing strip of LEDs around its body function as turn signals. Most noticeably, it has two big saucers for eyes, which blink, lending an uncanny resemblance to the animated robot WALL-E.
If you ask Kashani, Serve won’t replace Postmates’ fleet of human workers so much as it will economize their routes. Ninety percent of Postmates’ deliveries happen in cars, but more than half of them are within walking distance. When someone orders a burrito from a nearby taco shop, a Postmate usually hops in a car, sits in traffic, circles the block looking for parking—then has to repeat the process when they drop the delivery at the customer’s house.
“Somehow, as a society, we are OK with the fact that we are moving a two-pound burrito with a two-ton car,” says Kashani.
Serve is designed to fix that. Imagine: a fleet of them rolling into restaurants and convenience stores, picking up food, then hopping into a car with a human, who drives a few miles more before sending Serve to bring the delivery the last mile to reach a customer, who unlocks the cargo by using the touch screen or their mobile phone.
‘There’s this perception issue, this dystopian view. We have five seconds to change people’s minds.’
Ali Kashani, Postmates VP of Robotics
Postmates will soon begin rolling Serve into its key markets, starting in Los Angeles, where mayor Eric Garcetti has been exceptionally welcoming. But Kashani likes to think about how, if they get it right, the rovers could become so much more than just take-out couriers. It could deliver medication. It could fight crime. It could pick up the day-old bagels from a bakery and ferry them to a food bank, helping to eliminate food waste. It could roam around developing accurate maps of streets and sidewalks and cities. “It could patrol the neighborhood,” says Bastian Lehmann, Postmates’ co-founder and CEO. “Or you could use it for evil things, like it could write parking tickets.”
It’s a heroic view of what the team here is building, and one that makes Postmates X feel less like a robotics R&D lab, more like a bid to save the world. But first, Kashani likes to remind himself, they have to figure out how to deliver burritos.
In the delivery world, robots have come to represent both a holy grail and an inevitability. Replace some or all of the human labor and you get faster drop-offs, more pick-ups per hour, and cheaper goods for everyone. Big companies, like Uber, Amazon, and Alibaba, have each invested heavily in developing such technologies, from delivery drones to autonomous bots; venture capitalist have poured millions into delivery robot start-ups like Marble, Boxbot, and Dispatch, all of which want their piece of the robo-delivery future. By some estimations, sidewalk bots like Serve will make up 85 percent of last-mile deliveries by 2025. Whichever company gains a foothold now will have a huge advantage in the future on-demand economy.
But building an autonomous sherpa is no easy task. Take the challenges of self-driving cars, then add a few sidewalk cracks, curbs, pigeons, piles of trash, popsicle stands, and—worst of all—people, who are historically hard for robots to understand. A delivery robot has to fit into the elaborate choreography of public space, picking up on the subtle cues pedestrians give when they want to turn or stop or walk a little faster. And then, it also has to signal to cars when it’s entering a crosswalk, navigating streets more like a self-driving car.
“Within unstructured environments, like the sidewalks of San Francisco, all of these challenges are really exciting. It also brings up a lot of interesting questions that we have to deal with,” says Hadas Kress-Gazit, a professor at Cornell who researches autonomous robots. A sidewalk bot has to make quick judgements: “Is this a person who’s trying to get out of the way, or someone who’s not noticing you? Is it a dog? Is it a child? The main thing here is safety. You have to make sure these robots, which could be fairly big, can be safe around people.”
Safety concerns have caused some cities to severely limit where sidewalk robots can operate. San Francisco—where Postmates is headquartered—now requires permits for start-ups to run their robots on public sidewalks, and even then, they’re limited to areas with low foot traffic. (Postmates, for its part, was the first to apply for a commercial permit in San Francisco.) It’s a valid concern: Robots are hard to predict; people can trip over them. For the elderly and disabled, sidewalk bots represent an even greater annoyance than scooters.
When Kashani joined Postmates, he wanted to see these challenges out in the real world as soon as possible, so he sent his team to start building. (Postmates had previously partnered with Starship Technologies, one of the premiere startups in the space, to pilot test delivery robots. During the pilot, Lehmann realized Postmates’ key advantage—it has a ton of data about how actual deliveries work—and decided to build the robots in-house. “Why would we help make someone else the winner in this space that we believe we can build ourselves better?”)
Postmates X’s first rover, low to the ground with four rugged wheels, looked like a Roomba with a GoPro stuck on top. The team named it ET (“because it was the first of its kind out in the world”) and put it on the streets of San Francisco. They saw almost immediately that it was too small. If you were, say, staring at your phone while walking down the sidewalk, you could easily trip over the thing.
The next version, called Curie, stood a meter high, tall enough to appear in a pedestrian’s periphery. To make it even more noticeable, they added speakers, which played music like the Super Mario theme. Curie made the first real delivery—a batch of iced tea for the engineering team—which felt like proof that if they just kept iterating, they could actually make this work.
Then one day, one of the early rovers was cruising down the street—this one called Roberta, for Roberta Bondar, Canada’s first female astronaut—when it rolled into the view of Ryan Hoover, the founder of Product Hunt. Hoover tweeted a photo to his followers, inviting them to “name that robot.” It took three minutes before someone compared Roberta to a bomb disposal robot from The Hurt Locker.
To Kashani, that represented a real problem. Here they were, trying to build a rover that would bring people joy, pick up their iced tea, deliver their late-night burritos, pipe tunes throughout the sidewalk—but what did that matter if people still associated it with doom, or wanted to kick it, or tried to dismember it, or push it into the street?
“There’s this perception issue, this dystopian view,” Kashani says. “We have five seconds to change people’s minds.”
‘If there’s ever going to be an emoji for a rover on the sidewalk, it better look like Serve.’
Bastian Lehmann, Postmates co-founder and CEO
So Kashani developed a new objective: The rover had to work, yes, and successfully navigate its way into restaurants without running anyone over. But it also needed to make people happy. The next prototype, Valentina—named for Valentina Tereshkova, the first woman to fly to space—looked like a Fisher-Price play truck. The team added googley eyes, and lips. Painted it in bright colors, found more cute music to play. Then someone sketched up Serve, more like a child-sized shopping cart, and they knew they had it right.
Creating an anthropomorphic robot is a design gamble. Give the bot a face and people expect it to behave like a person, which can annoy people when it doesn’t. Plus, Serve has a sort of ice cream truck effect. Lehmann says kids always want to run up and play with it.
But Postmates says the humanist elements help to telegraph intent (you can tell where Serve is headed, for example, by following its gaze) and improve public perception. “There are two ways to go,” says Kashani. “Either you can say, ‘Get out of my way,’ or you can say, ‘We’re going to embrace this.’ This is part of the charm.”
Serve uses a combination of lidar, sonar, computer vision, time-of-flight cameras, and GPS to navigate the world. It can carry a payload of 50 pounds, and goes 30 miles on a single charge. When you see it, though, none of that will matter. Because instead, you’ll see it as a singing, blinking, brightly colored machine tootling down the sidewalk.
“We wanted to create something that you’d see it, and it’s like the iMac in this space,” says Lehmann. The ultimate goal: “If there’s ever going to be an emoji for a rover on the sidewalk, it better look like Serve.”
Postmates has dispatched versions of its rovers on thousands of deliveries in cities like Los Angeles and Toronto. For the most part, customers wouldn’t know: the rovers pick up items from restaurants, drive them to Postmates cars, and leave the humans to do the final hand-off. The rovers are remotely supervised by a Postmates employee, who can intervene with a game controller when necessary. But these days, they mostly just sit back and watch, taking in the view of the sidewalk from the rover’s little camera.
Serve still has to prove itself out in the world. But Kashani fully believes in getting the thing out there. He likes to tell his team about a design contest called the Marshmallow Challenge. Teams of four get dry pasta, tape, and strings, with the goal of creating a structure that holds the marshmallow highest. It reveals a lot about the nature of teams: Engineers tend to do fairly well, MBAs don’t. The group that does best, every time, without fail? Kindergarteners.
“Adults design and start building the foundation and then put the marshmallow on top and the whole thing comes crashing down, because they put it too late,” says Kashani. “Kids, at the very beginning, pick up a bunch of pasta and stick it into the side of the marshmallow. And then they make it stand. So now they have 20 minutes to get it higher.”
He thinks of his work at Postmates X like this. The company won’t know for sure that they’ve built the right thing, or found the right use-case, until it’s rolling out in the world. All they can do now is get Serve as stable as it can be, and then, like the marshmallow, then take the concept higher, and higher, and higher.