The new workers zipped around the office completing mundane tasks like fetching coffee, delivering meals and handing off packages. They did not get in anyone’s way or violate personal space. They waited unobtrusively for elevators with unfailing politeness. And, perhaps most enticingly, they did not complain.
That’s because they were robots.
Naver — a soup-to-nuts internet conglomerate in South Korea — has been experimenting with integrating robots into office life for several months. Inside a futuristic, starkly industrial, 36-story high-rise on the outskirts of Seoul, a fleet of about 100 robots cruise around on their own, moving from floor to floor on robot-only elevators and sometimes next to humans, rolling through security gates and entering meeting rooms.
Naver’s network of web services, including a search engine, maps, email and news aggregation, is dominant in South Korea, but its reach abroad is limited, lacking the global renown of a company like Google. The company has been on the hunt for new avenues for growth. In October, it agreed to acquire Poshmark, an online secondhand retailer, for $1.2 billion. Now, Naver sees the software that powers robots in corporate office spaces as a product that other companies may eventually want.
Robots have found a home in other workplaces, such as factories and in retail and hospitality, but they are largely absent from the white-collar world of cubicles and conference rooms. There are thorny privacy questions: A machine teeming with cameras and sensors roaming company hallways could be a dystopian tool of corporate surveillance if abused, experts say. Designing a space where machines can move freely without disturbing employees also presents a complicated challenge.
But Naver has done extensive research to make sure that its robots — which resemble a rolling garbage can — look, move and behave in a way that makes employees comfortable. And as it develops its own robot privacy rules, it hopes to write the blueprint for the office robots of the future.
“Our effort now is to minimize the discomfort they cause to humans,” said Kang Sang-chul, an executive at Naver Labs, a subsidiary developing the robots.
Yeo Jiwon, who works in the company’s social impact team, recently ordered coffee on Naver’s workplace app. Minutes later, the “Rookie” exited the elevator on the 23rd floor, zoomed through the security gates and approached her desk. Once nearby, the robot opened its storage compartment with a cup of iced coffee that had been prepared at a Starbucks on the second floor.
The robots are not always perfect, Ms. Yeo said, sometimes moving slower than expected or occasionally stopping too far from where she sits.
“They do feel like a beta release sometimes,” she said, using the tech parlance for software that is still under development. The deliveries save her time, though, she said, and help her focus on her work, eliminating the distraction of walking to a coffee shop.
Technology firms often encourage employees to test out their own products, but with its robots, Naver has turned its entire office into a research and development lab, deploying its employees as test subjects for future workplace technologies.
When Naver employees drive to the office, which finished construction this year, the company automatically sends them reminders of where they parked on the workplace app. Employees walk through security gates that use facial recognition, even while masked to prevent the spread of the coronavirus. At Naver’s in-house health clinic, artificial intelligence software suggests areas of focus for employees’ annual health exam.
And then there are the robots.
Naver designed the office from the ground up with the robots in mind, starting construction in 2016. Every door is programmed to open when a robot approaches. There are no tight hallways or obstructions on the floor. The ceilings are marked with numbers and QR codes to help the robots orient themselves. The cafeteria has lanes dedicated for robots to deliver meals.
As part of its research, Naver has also published studies in the field of human-robot interaction. After a series of experiments, for example, Naver concluded that the optimal spot for a robot in a crowded elevator with humans was the corner next to the entrance on the side opposite of the elevator buttons. Putting the robot at the back of the elevator made humans uncomfortable, Naver’s researchers found.
The company’s engineers also designed animated eyes that gaze in the direction that the robot is headed. They found that employees were better able to anticipate the robot’s movement if they could see its gaze.
None of the machines look human. Mr. Kang said the company did not want to give people the false impression that robots would behave like human beings. (Some robotics experts believe that humanoid robots make humans more, not less, uncomfortable.)
Naver, of course, isn’t the only tech company trying to advance robot technology. Rice Robotics has deployed hundreds of cartoonish, boxy robots that deliver packages, groceries and more in office buildings, shopping malls and convenience stores around Asia. Robots like Optimus, a prototype that Tesla unveiled in September, are designed to be more like humans, and carry boxes, water plants and more, but they are a long way from being deployed.
Victor Lee, the chief executive at Rice Robotics, said he was impressed when he saw videos of the machines and Naver’s robot-friendly building. While Rice’s delivery robots function differently, Naver’s approaches “made sense,” he said. “Naver obviously has way more development budget on these moonshot projects.”
Naver said one distinctive feature of its robots was that they are intentionally “brainless,” meaning they are not rolling computers that process information inside the machine. Instead, the robots communicate in real time over a high-speed, private 5G network with a centralized “cloud” computing system. The robots’ movements are processed using data from cameras and sensors.
Each robot has several cameras that record images of its surroundings. There was some disagreement within Naver about what exactly the robots needed to know, and how the data being collected would be used. When prototypes were being developed, engineers initially wanted the robots to record a wider field of view to assess their location more quickly and more accurately, according to Lee Jin-kyu, Naver’s chief data protection officer.
Mr. Lee worried that this would result in data that could be used to track employees without their knowledge, creating legal problems for the company in South Korea, which has strict labor and privacy laws. Mr. Lee and the engineers agreed to capture only one photo per second from a front-facing camera, and use the other cameras only when more than one image was necessary.
The cameras can see only below people’s waists, and the images are deleted as soon as the robot has oriented itself. An emergency mode kicks in if a robot is knocked over or camera angles change suddenly. In such cases, the robot announces that it may record people’s faces.
Despite Naver’s precautions, privacy experts worry that potential customers might modify the robots or create their own policies on how they collect data. Kim Borami, a privacy lawyer in Seoul, said that many South Korean companies were opaque about their data policies, and that she had encountered examples of companies breaking privacy laws.
She also noted that it was impossible to know for certain whether Naver was following its own privacy policies without looking more closely at its software — something Naver does not share publicly.
“You don’t typically find out about privacy violations in a company until there is a whistle-blower or a leak,” Ms. Kim said.
Naver said it was in compliance with South Korean laws on employee data privacy and surveillance. But part of the challenge with new workplace technology is creating rules on the fly.
“There is no benchmark for the kinds of privacy policies we need,” Mr. Lee, the Naver engineer, said. “We had to start from scratch. That was the most difficult part.”
