ON THE pavement of a busy Atlanta street Deja, Jiwon, Mu, Niska and Pelin stand at attention. Their eyes blink slowly as they look straight ahead. Inside Gusto, the salads-and-wraps restaurant they face, workers take notice. “It feels like we’re being watched,” says one. At any moment the platoon will break into action—trundling away to pick up and deliver late-night munchies to college kids studying in dorms or sushi dinners to finance bros. These are no ordinary couriers. They are machines.

Thousands of cooler-sized food delivery robots now roam American streets. At the start of last year Serve Robotics, the maker of Deja’s posse, had only 100 bots. (Getty Images.)
Thousands of cooler-sized food delivery robots now roam American streets. At the start of last year Serve Robotics, the maker of Deja’s posse, had only 100 bots. (Getty Images.)

Thousands of cooler-sized food delivery robots now roam American streets. At the start of last year Serve Robotics, the maker of Deja’s posse, had only 100 bots. It has since deployed 2,000 to 20 cities. Coco has a fleet of 1,000 and Starship Technologies has 2,000. The machines map their surroundings using the same cameras and sensors as self-driving cars. Artificial intelligence then helps them decide how to cross the road, dodge runners and scale snowbanks. Thanks to contracts with Uber Eats, DoorDash and Grubhub, they are among the most visible examples of AI taking human jobs.

The bots promise efficiency. “Moving a two-pound burrito in a two-tonne car doesn’t make a lot of sense,” says Ali Kashani, the boss of Serve. Roughly a quarter of American car trips, he notes, are “last-mile” errands and shopping runs. These journeys not only clog streets but are pricey: why should moving a good from Chinatown cost $10 when shipping it from China might cost $2? Small robots ferrying more takeaways could also lift local economies. Thunder Said Energy, a consultancy, found that a Starship bot is 100-times more energy-efficient than a motorcycle.

The only snag? People seem to hate them. Social-media videos of pedestrians assaulting them have gone viral. Some seem to be shaking them down for pad thai; others are releasing rage by tipping the poor things over. In one clip a Miami man tries to chuck a bot off a bridge. The anger is translating into activism. In Chicago 3,300 residents have signed a petition asking their city to ban the bots. The editorial board at the University of Notre Dame called on students to boycott them: “Why should we tolerate mechanised Grubhub robots terrorising our walkways?”

This disdain for robots is not new—especially in America. In 2014-15 HitchBOT, a full-body robot, successfully hitchhiked across Canada, Germany and the Netherlands. Two weeks into its American trek, however, it was found stripped, dismembered and decapitated in Philadelphia. Years later a security robot patrolling San Francisco’s Mission District was discovered “besmirched by bar-b-que sauce” and “befouled by faeces”, according to a news report. A recent survey by Pew Research Centre found that Americans are far more concerned about AI intruding on daily life than people in other rich countries.

To mitigate the backlash, Serve has given its machines names and puppy-dog eyes. The company trains the bots to be “courteous” and “delightful”—to slow down near pedestrians and swivel their wheels to signal where they intend to go before taking off. Mr Kashani says the online angst is overblown: 99.8% of his robots successfully complete their trips. He hopes to widen their remit to picking up pharmaceuticals and doing shopping returns. To do so the robots will have to continue to cope with the real world, he says with a laugh. “And the real world is a tough place.”