In a giant warehouse in Reading, Massachusetts, I meet a pair of robots that look like goofy green footstools from the future. Their round eyes and satisfied grins are rendered with light emitting diodes. They sport small lidar sensors like tiny hats that scan nearby objects and people in 3D. Suddenly, one of them plays a chipper little tune, its mouth starts flashing, and its eyes morph into heart shapes. This means, I am told, that the robot is happy.
Proteus, as Amazon calls this machine, is not like other industrial robots, which are generally as expressive and aware of their surroundings as actual footstools. “Wait, why would a robot be happy?” I ask. Sophie Li, a software engineer at Amazon, explains that being able to express happiness can help Proteus work more effectively around people.
Proteus carries suitcase-sized plastic bins filled with packages over to trucks in a loading bay that is also staffed by humans. The robot is smart enough to distinguish people from inanimate objects and make its own decisions about how to navigate around a box or person in its path. But sometimes it needs to tell someone to move out of the way—or that it is stuck, which it does by showing different colors with its mouth. Li recently added the heart eyes to let Proteus also signal when it has completed a task as planned.
“Proteus will hopefully make people happy,” Li says, referring to the workers who will toil alongside the robot, transferring packages from bins into trucks. “And if not, well, at least it should do what they expect it to.”
I find myself wondering if some people might, in reality, find the robot’s cheeriness a bit annoying. But perhaps putting a friendly face on the new wave of automation about to sweep through Amazon’s fulfillment centers isn’t a bad idea.
Proteus is part of an army of smarter robots currently rolling into Amazon’s already heavily automated fulfillment centers. Some of these machines, such as Proteus, will work among humans. And many of them take on tasks previously done by people. A robot called Sparrow, introduced in November 2022, can pick individual products from storage cubbies and place them into larger plastic bins—a step towards human-like dexterity, a holy grail of robotics and a bottleneck in the automation of a lot of manual work. Amazon also last year invested in a startup that makes humanoid robots capable of carrying boxes around.
Amazon’s latest robots could bring about a company-wide—and industry-wide—shift in the balance between automation and people. When Amazon first rolled out large numbers of robots, after acquiring startup Kiva Systems and its shelf-carrying robots in 2012, the company redesigned its fulfillment centers and distribution network, speeding up deliveries and capturing even more business. The ecommerce firm may now be on the cusp of a similar shift, with the new robots already starting to reshape fulfillment centers and how its employees work. Certain jobs will be eliminated while new ones will emerge—just as long as its business continues growing. And competitors, as always, will be forced to adapt or perish.
Proteus isn't the only robot being put through its paces at the Reading facility, which houses Amazon Robotics, a laboratory and foundry for the company's warehouse robots. Nearby, a small platoon of blue mobile robots, each about the size of a push lawn mower, are going through some algorithmic choreography. I watch as they drive, one by one, into large machines that test the performance of their wheels and other features. Those declared fit for service then trundle under a walkway and into packing crates destined for Amazon fulfillment hubs.
The visit provides a rare glimpse of how Amazon’s develops its industrial robots. I am accompanied by Xavier Van Chau from Amazon public relations, who arrived on a red-eye from the company’s Seattle headquarters and is highly enthusiastic and impressively caffeinated. While Amazon Robotics engineers show off machines that will significantly shift the line between what humans and machines can do, my chaperone supplies a stream of anecdotes about workers who love their robot coworkers or their new robot-related roles.
Some workers in Amazon’s fulfillment centers have of course shared their own anecdotes about the company pushing them hard in the name of efficiency, although the company maintains staff welfare is a top concern. In January the company was called out by US regulators for poor workplace safety and it has faced industrial action and walkouts in several US states and the UK. Leaked documents obtained by Vox suggest that Amazon expects it to become more challenging to find enough people to hire in the US as warehouse workers, due in part to high staff turnover. Accelerated adoption of robotics may help the company soften some of the challenges posed by its human workforce.
But to replace human labor, these robots need to be built. And much of that work is done by humans. At a nearby production line, Amazon workers are busily putting robots together, hefting large pieces of steel around with the help of mechanical arms and installing electronics, sensors, and motors.
Jobs in robot manufacturing and maintenance have multiplied at Amazon since it began ramping up its use of robots. The company also opened a new manufacturing facility dedicated to making robots in Westborough, Massachusetts, in 2021. But the addition of manufacturing workers and engineers means that other jobs at Amazon are changing—or disappearing altogether.
Amazon’s first robots, from the acquisition of Kiva, were low-slung orange brutes—Cro-Magnon ancestors to Proteus—that blindly followed preprogrammed routes inside large caged-off areas. The robots rolled beneath shelves of cubbies stuffed with different products, and carried them over to human pickers on the edge of the automation zone. The humans would grab products to assemble customer orders, placing them into bins that were sent for packaging and shipping.
That automated retrieval system let Amazon store more goods in the same space, and move them to customers more quickly, helping the company ascend to the pinnacle of ecommerce in the eyes of customers, investors, and competitors. Between 2010 and 2020, sales on Amazon rose 10-fold from $34 billion to $386 billion, and its robot workforce soared too. Between 2013 and 2023, the cumulative number of robots made by Amazon grew from 10,000 to 750,000.
Today, three quarters of all Amazon’s products—every conceivable item you could need and plenty you probably don’t—are handled at some point by one of the company's robots. The 750,000 mobile robots at more than 300 Amazon fulfillment centers worldwide can trace their lineage back to the first Kiva machines. Amazon also employs more than 1.3 million workers at these locations. Van Chau of Amazon declines to say how it expects the number of robots it uses to grow in the years ahead but says it will “continue to grow very rapidly.”
The most common robots at Amazon today are the blue machines that I saw rolling into shipping crates, called Hercules. They are members of a pantheon of cage-bound machines created at the company with names borrowed from Greek mythology.
Hercules, a heavy lifter of course, lugs shelves over to human pickers and has largely replaced the older Kiva bots. Pegasus, a wheeled robot with a tilting conveyor belt on top, drops packages into chutes that lead down to loading bays. And Xanthus, a slimmed-down version of Pegasus, serves as a general dogsbody, taking on tasks like ferrying stacks of emptied crates back to wherever they’re needed.
The newer and more expressive Proteus, which went into service at a fulfillment center in Nashville, Tennessee, earlier this year, is Amazon’s first mobile robot to venture outside of safety cages. It is designed to be more general purpose than the company's previous mobile robots, with software upgrades adding new capabilities over time. It is part of a new generation of robots now arriving at Amazon thanks to recent leaps in AI.
Across the room from where Proteus showed me its happy dance is a large robotic arm called Cardinal, which is being tested in tandem with the mobile robot. Cardinal looks like the kind of robot you’d find on any automotive production line. It is fixed in place, but instead of the usual metal claw wields a gripper covered with vacuum suckers reminiscent of a giant squid. Cardinal uses AI vision to identify and determine how to grasp heavy packages from a jumble of them rolling along a conveyor. It hoists them with its suckers, and deposits them into a tote carried by Proteus, which will ferry the load over to waiting trucks.
Xinye Liu, senior technical product manager for Cardinal, watches the robot as it picks packages. She recently returned from the Nashville fulfillment center where Proteus and Cardinal are being trialed. Liu tells me the center was so short-staffed during a spike in demand that she decided to help load boxes herself. She says that the scale of Amazon’s operations—the absurd variety of products it stores, and the vast number it processes each day—make it a uniquely compelling place for roboticists to work.
Robin, another of Amazon’s newer robots, also incorporates a large robot arm. It transfers packages from conveyors onto the back of waiting Pegasus bots. The company has deployed more than 1,000 Robin robots, and during an April 2023 earnings call revealed that Robin has now handled over a billion packages.
Amazon's rapid robot rollout belies just how big the gap between humans and machines remains. There are still many tasks done by people that the company can't automate away. Take the fiddly business of retrieving an item from a storage shelf at the beginning of its journey from warehouse to customer. If a customer has ordered a tiny pair of eyebrow tweezers you first have to spot a tiny item amongst a pile of others, then know how to pick it up, rotate it, and read the label, all without dropping it.
For a human, all of that feels simple, despite the complex feats of perception and control required. Programming those capabilities into robots has proven extremely challenging. But over the past decade, progress in computer vision, robotic grasping, and robot hardware have removed the need for human hands in some situations.
From 2015 to 2017, Amazon ran a contest with cash prizes that invited researchers to build robots capable of picking a wide range of objects from its shelves. A lot more robotic grasping research was evidently going on behind the scenes. Last November, the company revealed Sparrow, a smaller robot arm that can reach into totes and reliably grasp 65 percent of the more than 100 million items in Amazon’s inventory. That’s a large enough percentage for the company to reconfigure some operations around the robot.
Tye Brady, chief technologist at Amazon Robotics, explains that the sheer number of products that Amazon handles gives it a competitive edge. “Machine-learning techniques allow robots to teach themselves what to pick and how to pick,” Brady says. “And because they're connected through the cloud, all those learnings can be shared instantaneously and propagated with all the arms.”
In other words, Amazon’s robots are going to get faster and more reliable over time as more data is shoveled into the AI models they depend on. And this continual learning is almost certainly paying off already.
During my tour of the robot foundry in in Reading, I got a glimpse of how Amazon is working to expand robot picking. Wandering through a network of caged-off robots, we came across a small team of engineers working on a new robot arm that was picking items from a large plastic bin. Unlike Sparrow, the setup used what I recognized as a so-called collaborative robot arm, designed to work in close proximity to humans. In the future, perhaps humans and robots will share picking work side-by-side inside Amazon.
“Err … that’s nothing,” Van Chau said as he hurried us along. “It’s just an experiment.”
Amazon's long-term vision for a more roboticized future is quite different from the current very messy, human-dominated reality.
A couple of hours drive from Reading, in Windsor, Connecticut, the company has built a new, 3.8 million-square-foot facility known as BDL4. There is no sign of robots taking over when we arrive. The car park is almost full, and workers are filing in for a new shift, many toting their belongings in transparent plastic backpacks. It’s not a fashion trend—clear backpacks make it faster to get people through security checks. After donning a safety vest and toe protectors, we venture inside.
The workers I get to meet at BDL4 are unfailingly cheerful and helpful. There is Allison Kim, a senior operations manager, who gives a tour with the aid of a golden toy microphone that boosts her voice above the constant whir of machinery. And Alex Sabia, an Energizer bunny of a man who keeps mentioning his meat-rich diet. His job is to prevent workers from injuring themselves by encouraging them to take regular breaks, ensuring good ergonomics, and giving them physical therapy exercises to do. Talking to him makes me feel quite tired.
The ground floor of the building is dominated by conveyor belts ferrying packages in one direction or another. On the second floor up through the fifth, humans are busy picking items from shelves that Hercules robots ferry over from an enormous, fenced-off storage area where only robots roam. It has few lights because the machines do not need to see to navigate. The scale is dizzying. The line of human pickers retrieving items from shelves along the boundary between the human and robot zones, the light and the dark, shrinks into the distance.
Among the pickers, I notice one worker wearing a utility belt and shoulder strap mounted with a flashlight, carrying a tablet and what looks like a short hockey stick. He unlocks a door that leads to the robot area, walks in, and closes it behind him. This worker’s job is to assist when a Hercules has dropped something, which often means retrieving an item from between several robots with the hockey stick. I’m told that these workers are known as “amnesty specialists,” because in Amazon corporate lingo, items dropped by robots are termed “in amnesty.” As I watch the human emissary set off into the robot zone, their light gradually disappears into the darkness.
On a floor lower down, where packages filled with items are routed to trucks below, the ratio between humans and robots is tilted firmly towards machines. The center of the room is a hive of motion, as some 12,000 Pegasus robots, similar in size to the lawn-mower-scale Hercules models I had seen earlier, zip around, dropping packages down holes in the floor that lead to the loading docks. Each time a robot sheds its load, it returns to a parking spot somewhere on the floor. At the edge of this area the robots are loaded up again not by human pickers, but by Robin robot arms grabbing parcels from conveyors.
As this floor demonstrates, automation can take over certain tasks previously only achievable by human workers. In some cases, certain jobs can even disappear. For an individual, replacement by machine might be devastating, but the picture across the job market is more complicated. One US study found that each robot adopted in manufacturing replaced about three workers. However, other research shows that companies that deploy more robots sometimes add more jobs overall.
“I'm not really worried about us running out of jobs for humans anytime soon,” says Erik Brynjolfsson, director of the Stanford Digital Economy Lab and an authority on the impact of AI and automation on jobs. “If unemployment increases in the next couple of years, I don't think it's going to be because of automation."
The key questions are how many new jobs will emerge, and what kinds? Amazon’s robotic push may create better-paid manufacturing jobs for some—but could also lead to greater demand for delivery drivers who tend to work for outside contractors.
Amazon’s new robots may also affect its workforce in other ways. Greater use of robotics at BDL4 makes it possible for Amazon to squeeze more kinds of work into the facility, in the same way the Kiva acquisition allowed more items to be stocked in the same space. In Amazon’s largest markets, after orders have been packaged they are generally sent to regional sorting centers that organize packages for delivery by geographical location to make distribution more efficient. But the new BDL4 incorporates sorting, eliminating the need for another facility and speeding up the whole process.
In recent years, labor organizers protesting punishing working conditions inside Amazon’s logistics operation have pressured the company through walkouts or other actions at sorting centers, which can be choke points in its distribution network. When sorting work done is on the same site as picking, fulfillment centers like BDL4 could perhaps lessen Amazon’s vulnerability to industrial action. On the other hand, says Rand Wilson, a labor organizer who has worked with Amazon employees, the workers who keep their jobs after the adoption of more automation may in fact have greater leverage because their work is more specialized and they are more difficult for Amazon to replace.
Whatever you think of Amazon, it’s hard not to marvel at the company’s ruthless efficiency. The company's bold push to adopt more robots will no doubt delight many of its customers by speeding up delivery times. But it will also have ripple effects for millions of workers and thousands of other businesses who compete with Amazon, which commands more than half of all US online purchases.
A couple of weeks after the visit, I received an email from Amazon’s Van Chau. He tells me that he has injured his arms in a traffic accident and has to temporarily write his emails using voice-to-text software. “Robots are helping me write this email,” he jokes, then hints at Amazon’s many future automation advances to come. “The latest end-effector tools are crazy cool,” he said. “But nothing I can share on the record.”
