The Crazy-Hard Technical Challenges of Robots Delivering Falafel

Self-driving cars have it easy—streets are mostly orderly, predictable places. Sidewalks, on the other hand ... .

It’s the most San Francisco scene in San Francisco. A dude with a guitar and a dude with a scooter stand at a Mission District bus stop. Next to them, someone has laid out fishing rods for sale. Behind them, a white semi-autonomous delivery robot, looking not unlike a rolling dishwasher, hesitates behind the crowd before powering through and nearly running over a dog.

Think of Marble like a self-driving car, except its domain is the sidewalk. Instead of delivering humans to places like restaurants, Marble delivers food from restaurants to humans. If you live in San Francisco’s Mission or Potrero Hill neighborhoods and you order delivery through Yelp Eat24, you may just get a text asking if you want Marble to roll the food on over to your pad. And of course you’d say yes---it’s a robot filled with food, after all.

The Marble robot doesn’t see the world like you do, unless you’re special and see with lasers. The robot uses lidar and cameras to detect its surroundings in fine detail. And boy, does it need detail. The Mission is a beautiful madhouse, packed with buskers and hipsters and street vendors cooking bacon-wrapped hotdogs that I really don’t recommend eating at the end of a night out. If Marble can make it here, it can make it anywhere.

For such a young technology in a befuddling environment, the robot makes it, though in the demonstration we saw, a human chaperone was there to take remote-control as needed. That's normal, Marble's makers say---the robot for now can't make it through the city on its own. Navigating streets as self-driving cars do, it turns out, may be a much easier problem for a computerized brain to solve than threading the unpredictability of sidewalks. Streets are relatively ordered, structured environments. They have lanes and signs and stoplights. Sidewalks are pure chaos---in San Francisco especially. People are streaming out of stores willy-nilly, with no traffic signal to help Marble predict what's coming. The Marble robot, then, is necessarily semi-autonomous.

So Marble’s parents have given it a boost into the world. “We've created high-resolution 3-D maps of the entire neighborhood in which we're operating,” says Marble co-founder Matt Delaney. That helps ground the robot in its surroundings so it can focus on the neighborhood’s many surprises.

But for the time being, human handlers will escort the robots, taking control when the machines run into problems. So good luck stealing one without getting your picture taken. Even down the, er, road, when the robots become more independent, Marble plans for humans to monitor them from afar, both to ensure no one tries to swipe them and to intervene if something goes wrong.

They See Me Rollin'

If humans are still in command, what's the point of having a robot do the work in the first place? Delaney says these robots will work as complements to traditional delivery drivers, not replacements, to handle short-hop deliveries. But robots like Marble will inevitably take human jobs, says Burt White, VP of the industrial sector for Chainalytics, a supply chain consultancy: “That's a route some high school kid had at some point, right?”

Ultimately, the vision for Marble involves organizations considerably larger than your local Mediterranean joint. Venture capitalists are pouring money into delivery robots with the aim of solving what the logistics industry calls the last-mile delivery problem, bringing ever-greater efficiency to the deceptively complex process of getting a package from a distribution hub to a home. “So yes, while it is starting with things like food, I would also see it scaling up to doing things like delivering medicine or products that people need to their homes,” says Greg Reichow, a partner at Eclipse Ventures, which just led a $4 million seed-funding investment in Marble.

But realizing that grand vision may involve overcoming more obstacles than roving canines or tipsy pub crawlers. Cities will also have to decide whether these robots are safe enough to share sidewalks with humans. "At first glace it doesn't appear as if [the robot] is a clear violation of any particular existing code or regulation," says Officer Robert Rueca of the San Francisco Police Department. Because the robot transports food and not people, he says, it seems similar to a dolly, which people use on sidewalks every day. Still, unlike a dolly, this automaton operates without a direct human escort and seems big enough to do serious damage.

Delivery robots will have to win over not just the authorities, but the people with whom they share the sidewalks. Tech money, as you've probably heard by now, has thrown San Francisco's economy into chaos. The average rent sets you back around $3,500, and the historically working-class Mission in particular has grown into a flashpoint of anti-gentrification ire. Nothing embodies the city's current tech-boom moment quite like a robot rolling through your neighborhood to bring you food at the tap of an app. Crash into a bacon-wrapped hot dog cart, and the city may never forgive you. Get a robot to deliver bacon-wrapped hot dogs, on the other hand, and the city may love you forever.