Atkeson’s house is littered with pictures and sculptures of fish and robots. So many robots. He and his wife Jessica, also a robotics professor at Carnegie Mellon, create sculptures of robots in their spare time out of found items—trashcans, cogs, antennae; anything they can get their hands on that might make something look remotely robot-shaped—as if they did not spend enough time around robots at work.
Atkeson was born in 1959, which means he was 18 when the first Star Wars film was released. “There were no computers when I was in high school,” he reminded me. “Nobody really had heard of robots, and frankly, computers.”
Atkeson’s passion as a child was something far simpler: He was fascinated with tensegrities, objects that were first conceived of by the American engineer, architect and futurist, Buckminster Fuller (although Atkeson will tell you that it was actually a student of Fuller’s, Kenneth Snelson, who first came up with the idea). They are simple structures that essentially combine rigid objects with flexible ones to create a sturdy construct that relies on tension alone to support itself. He used to build simple tensegrities out of wooden sticks and string, and amongst the fish and robot sculptures, you’ll find tensegrities dotted around his home, including some that Atkeson built during his high-school days.
Growing up in Washington, DC, Atkeson was part of a large extended family. He had six brothers, revered for their prowess in sports, including wrestling and cross-country. He was the second-eldest. “As a child I was very clumsy and chubby,” he said. “Moving and being coordinated was a big challenge, and thus was fascinating to me.”
While they were winning medals, Atkeson spent his spare time building bigger and more complicated tensegrities. He was enthralled by the way he could use materials to create structure and balance in artistic sculptures. But his brothers’ physical prowess also stuck with him: “At some level I spent the rest of my life trying to automate all the things that my brothers were better at.”
“There are people making robots essentially just like this where the strings can be muscles and the thing can walk and move,” Atkeson said. “There is, at some level, a very tight connection between this and robots, and yeah, it doesn’t look soft, because it has sticks, but yes, this is a precursor of sometimes thinking about soft robotics.”
Atkeson was also the son with the driving license who was willing to help out, which meant he often was helping out his family. “My grandfather had ALS, and that basically meant that his brain was fine, but it couldn’t tell his muscles what to do,” he said. “He would slide off his chair and fall on the ground.” His grandmother wasn’t strong enough to pick him up, so she would call Atkeson and ask him to come over and help his grandfather back into his chair, or help with whatever other menial physical tasks his grandparents couldn’t do on their own. “My grandmother told me what to do, so she was the brains of the operation,” he said.
During his college years, he started to see what robotics could do for the world. Looking back on the help he provided his family, Atkeson said, “I could have easily been replaced by a robot, and we could build that robot now.” As was the case with his grandfather, there are millions of people who are mentally sound, but physically weaker than they’d like to be. “Many couples, older couples living alone, trying to stay in their own home have this kind of problem,” Atkeson said. If you could provide them with something that they could direct to do the tasks that they can’t do on their own it could change their lives. “I think everybody’s better off both economically and in terms of happiness, not going to some kind of institution or assisted living facility, but staying in their own home as long as possible,” Atkeson added.
Soft start
Atkeson went to Harvard to study biochemistry and mathematics, graduating in the top 5% of his class, and then on to MIT to work on a doctorate in cognitive science. “When I was in college, I took an artificial intelligence course that covered robots and I thought robotics were really stupid, because the robots I saw were incredibly slow and clumsy, and I had no interest in that,” Atkeson said. “I also took a course on how the brain worked, neuroscience, and the idea of having a computer in your head processing information was fascinating to me,” he said. He started taking courses in computer science and programming. “I got really excited about being able to create things using a computer. Eventually, that led to me putting together an interest in brains and computers, and I naturally gravitated towards making stuff move.”
“My math teacher in high school said I would go far in life because I was so incredibly lazy,” Atkeson said. Being lazy—and smart—can be a powerful combination, as it means a person will search for the simplest answer to solve a problem, as Bill Gates famously argued. “We need lazy robots,” Atkeson added.
After spending time as an assistant and professor at MIT, and at the Georgia Institute of Technology, Atkeson became a professor at Carnegie Mellon, splitting his time between the university’s Human-Computer Interaction and Robotics institutes, both of which are within the school of computer science.
Carnegie Mellon has arguably the top computer-science program at a US university—in turn, one of the top in the world—and has had alumni and faculty win armfuls of Nobel Prizes, and produced countless world-changing inventions, from wifi, to self-driving cars, to Kevlar, to… Juicy Couture.
Atkeson was headhunted by CMU, according to its current dean of the computer science school, Andrew Moore. He previously studied under Atkeson, and went on to work for nearly a decade at Google, where AI has reshaped the company’s entire business. “The Robotics Institute, wisely, realized that they had to have someone who could think about both the control theory [of robots], which is one of Chris’s specialties and the biologically-inspired control of limbs,” Moore said. “And so CMU made a short list of who could do this, and that short list contained one name.”
Atkeson has worked on a range of projects over the years, including Carnegie Mellon’s entrant into the DARPA Robotics Challenge in 2015, a Pentagon-funded competition intended to inspire roboticists to build machines that could help in disaster relief situations. The team, which was a joint venture with the Worcester Polytechnic Institute in Massachusetts, used Alphabet’s multi-million-dollar Atlas robot to try to overcome an obstacle course concocted by DARPA. The team finished seventh out of 23 entrants.
“We’re very good at hard robots now,” Atkeson said, referencing the Atlas robot used in the competition, and a Sarcos robot, created by a subsidiary of the defense contractor Raytheon, that he has in one of his labs.
“The problem is they are overweight,” he added. “They might weigh 300 lbs., and a robot like that, if it falls down in your house, it’s going to put a dent in the floor. If your dog trips it up it might kill your dog or your child.”
For the first 25 years or so of Atkeson’s career, he, like most in the field of robotics, focused solely on metal robots. “I would say at the time it was him and Marc Raibert [the founder of Alphabet’s Boston Dynamics, the maker of the Atlas robots]. They were the énfants terribles of robotics because they were really pushing on dynamics,” Moore said.
As a PhD student at the University of Cambridge in the late 1980s, Moore submitted what he thought was a very smart thesis on teaching robots how to learn—only to find a paper by Atkeson in Cambridge’s library on that exact topic. At first he was horrified at the prospect of having to rewrite his thesis, but then thought to write to Atkeson to ask if he could join him at MIT. “It was like winning the Willy Wonka Chocolate Factory golden ticket when he wrote back and said ‘yep,’” Moore said.
Atkeson helped Moore, as he did all his graduate and undergraduate students, to better understand how robots work in theory and in practice. “He was quite an intimidating taskmaster,” Moore said, “because when I was used to doing things mathematically and doing proofs, he was saying ‘Look, let’s just build a robot—that’s the only way we’re going to know if this stuff really works.’”
“Chris was very influential in the first set of real systems on real robots where there were real dynamics,” Moore said, making Atkeson one of the first researchers to explore how robots might actually operate in the real world, rather than just in theory. “In that sense he really pulled a lot of people, including myself, out of theory-land into something which worked with real dynamic systems.”
But when it eventually occurred to Atkeson that robots need to be as malleable as we are, he shifted his attention. “In order to safely get close to people and touch them, we’ve got to be soft, we’ve got to be lightweight,” Atkeson said. Since the 2000s, this is what a large chunk of his research has been geared toward solving. Combining the logic of tensegrities—that rigid structures can, in part, be made out of flexible materials—with his desire to help the aging, he has been building robots that can one day be used safely in a real-world setting.
He applied the same approach he’d taken with Moore—just building robots to see if they worked—to the technology that inspired Disney to feature his work on the silver screen.
Mickey finds out
Baymax came a few years after Disney bought Marvel Comics. The entertainment giant had decided that it wanted to make a more kid-friendly film adaptation of one of its comics, Big Hero 6, about a team of Japanese superheroes. The setting was changed from Japan to a near-future Japanese-inspired San Francisco, and the protagonist, a 13-year-old boy named Hiro, was given a new sidekick. In the comic, he creates Baymax, a giant lizard-shaped robot, but Disney wanted something a little more Disneyesque for the film. The studio organized tours of the country’s top robotics labs to provide the writers and animators with inspiration for what a futuristic robot might look like. “One of the places they came to was Carnegie Mellon and one of the people they talked to was me,” Atkeson said.
At the time, he and his team of researchers and graduate students were working on an inflatable robotic arm as a proof-of-concept that robots could be made out of materials that wouldn’t hurt a human to interact with. It was a pretty simple idea, but ended up being foundational to Disney’s concept of a soft robot, and to the work that Atkeson has inspired others to do.
The arm isn’t much to look at. It’s essentially a bent rubber tube hooked up to a large actuator—a mechanical device that controls motion—that can flap it up and down. But the tube is filled with air, and connected to a thin metal wire that controls how it moves and bends. As Atkeson pulls on it, you quickly realize that it’s mimicking the way a human arm moves, with the wire acting as a tendon, and the rubber tube a combination of the muscle and bone. The only rigid, weighted part of the machine is at the base. The team from Disney had found what they were looking for, Atkeson said: “Although they didn’t indicate it or show it at the time, basically the minute they saw this inflatable robot they said, ‘Okay. Now we have a robot that’s completely different.’”
Atkeson told the Disney crew that he was exploring soft robots as a way of caring for humans, and from that, an inflatable, health-conscious robot called Baymax found its way onto the silver screen. “It was just a short one day visit; we had dinner with them and talked about robots, and art, and science fiction,” Atkeson said. “They took that idea and ran with it.”
Big Hero 6 went on to gross over $650 million, according to Box Office Mojo. Atkeson signed a contract that allowed Disney to use his idea in the film, although he was not actually paid for his contribution. The company has sent him a few toys and Baymax costumes, though.
Atkeson doesn’t seem to mind about the money. In fact, he’s generally thrilled about his contribution to the movie. There’s a giant poster from the movie outside his office, and given only the slightest provocation from me and my colleague, Atkeson donned one of the Baymax costumes he had, inflating it using a small compressor, like the one you’d use to inflate an air mattress, and walked around the robotics department pretending to be Baymax. We followed him as he ambled along the department’s corridors.
How soft robots look now
Out of his Baymax outfit (which has also doubled as his Halloween costume in years past), Atkeson took us on a whistlestop tour of pretty much the entire Robotics Institute. We ducked into one lab where graduate students were trying to build radio-controlled trucks to make their way through a cardboard maze. Atkeson watched and laughed as one student successfully navigated the maze in one try. He showed us HERB—the “Home Exploring Robot Butler”—one of the university’s better-known (but still very metal) robotics projects, explaining how it’s being used to research how robots can help us in our homes, completing simple gopher tasks, much like his own research. “But of course this will all be made obsolete by soft robotics,” Atkeson joked, walking out.
We continued on, past ancient remnants of robots long decommissioned.
“We could make blockbuster movies here but we choose not to,” Atkeson said, as we walked into the university’s motion-capture lab. As the name suggests, this lab is used to digitally capture the way humans move their bodies. But unlike similar setups at film production facilities that study actors’ movements to help design how a computer-generated character should move, Carnegie Mellon’s lab helps its researchers build more natural robots.
Next up on our tour is a room full of shelving, housing a broad-shouldered robot, a Baxter model made by Rethink Robotics.
In place of hands, it had what looked like green salad tongs, wrapped in cling film. Atkeson introduced us to one of his postdoctoral fellows, Akihinko Yamaguchi, who was using the robot to test one of Atkeson’s more recent soft-robotics ideas. The plastic wrapping on the robot’s tongs was printed with a matrix of dots, and underneath the plastic were cameras pointed at the dots. These cameras can essentially “see” the sense of touch: they look for the slightest change in position of the dots when the tongs grip something, and send a signal to the tongs to adjust the pressure they’re gripping with accordingly. Yamaguchi showed us the robot picking up an empty Coke can without making a single dent in it. It’s something that you or I might be able to do with ease, but a machine made of hard metal and plastic would usually struggle to do this easily. Yamaguchi then placed something even more delicate into the robot’s grip—a small origami paper crane. The robot held it perfectly between its tongs.
The robot was wearing the chef’s hat because Atkeson and Yamaguchi envision their research being used to develop robots that could prepare dinner for someone in their home, with all the dexterity and deftness of a proficient human chef. It draws on some of Atkeson’s recent work using a soft robot’s “skin” in the way that we use our own, inserting sensors below thin plastic surfaces that could potentially mimic the way we have nerve endings under our own skin.
Thanks to Atkeson’s research and other advances in robotics in recent years, soft robots may well be moving from science fiction to reality in the near future. Atkeson himself thinks we are “very close” to having human-scale soft robots that can help us around the house, although he didn’t provide a specific timeframe for when he thinks we’ll be able to purchase soft robot helpers. There are, however, already soft robots developed by Vause’s team that are being used in factories around the US, such as robots that can pack produce at farms without damaging it, or sort tomatoes for shipping. Vause told Quartz that his company is also working on developing applications for factories that produce goods that aren’t evenly shaped (where squishy grippers may make it easier to grab items than the claws of the average factory robot). They might not be the robots that Atkeson is dreaming of, but they are already at work in our factories.
The evangelical roboticist
The next day, Atkeson looks up from a podium, the large lecture hall dotted with students. They’re half-awake and idly tapping on computers as they decide whether the talk is worth paying attention to. Atkeson has been invited to talk about using AI to train robots. He explains the work he’s done in the past, and starts out by asking the audience a question: How do you open a stuck jar? He surveyed the class for ways that they would go about solving this problem. Some said they would use a towel, some would run it under hot water; others would tap the edges of the lid. With every answer a student gave—Atkeson encourages students to speak up and interrupt him with questions as he talks—he put up an image for that answer on a slide behind him.
Every answer they’d thought of was somewhere on his slide. He was trying to prove a point, that essentially that they were all correct, asking: Does it really matter how the jar was opened, if it wasn’t broken in the process? There’s no one correct way to open a jar, Atkeson said, and he wanted the AI students in front of him to think about their research that way. No one taught us how to open a jar—perhaps we read something online or our parents had their advice, but in reality, we found the solution that works for us, and we use it. What if robots thought that way?
Atkeson has spent the majority of his career trying to create robots that can learn how to operate safely in our world, but no one really teaches us how to do that—we generally learn by doing, by making mistakes. And that’s how our robots should act.
Atkeson’s lecture was like many of the YouTube videos he’s uploaded over the last few years, and probably what drew Disney to him: honest, endearing, funny, and really smart. And he seems to like dressing up—a lot. After an hour and half, the students filed out. I overheard one say to another, “I’m so happy I didn’t miss this.”
“He’s exactly the kind of mentor the United States really needs right now,” Moore told me, thinking back to his own time under Atkeson’s tutelage, and what Atkeson did for him. “He turned someone who is good and theoretical into someone who could really execute on a big scale.”
Atkeson’s optimism for turning his dreams into a reality, even after so many decades in the field, has clearly rubbed off on those who have worked with him. “Chris’s intellectual children and grandchildren have gone out there to do robot control systems which are more biological or natural,” Moore said.
“He’s very popular,” Moore added, “both with the grad students who like to know they’re getting strong training from him just like I did—but the undergraduates get to see someone who sort of shows them the joy of understanding intelligence by looking at real biological systems.”
“Undergraduates, when you look at the comments coming back from students, they’re often in the form of, ‘Wow, I never knew I was going to be interested in this, but now this is what I want to do with my career,’” Moore said.
One of those is Atkeson’s former student, Joohyung Kim, who worked with him on the DARPA Robotics Challenge, and now works at Disney Research. While it might be surprising that the company best known for animated movies and theme parks also has a string of high-tech research facilities (it has research labs in Los Angeles, Zurich, and on Carnegie Mellon’s campus), someone needs to build those creepy animatronic robots on the It’s a Small World rides. The lab in Pittsburgh also carries out fundamental research in robotics, artificial intelligence, mobile technology, materials, and a range of other areas that may or may not directly benefit the moviemaking or theme park arms of the company.
We visited Kim’s office overlooking the edge of Carnegie Mellon’s campus. He showed us the areas of research that he’s focused on now, such as modular robots that can keep walking even if they get limbs knocked off, and a Michelin-man shaped robot that looks like it might have been inspired by Baymax. The adorable gelatinous robot can hold objects deftly in its nubby arms, and is designed as a proof of concept of what a robot that can safely interact with a human might look like. Recent patents filed by Kim and his team suggest they may be trying to build a full-sized Baymax—something Atkeson and his students have also played around with—although he couldn’t comment on the purpose of the research. What Kim did show us, however, was a package containing a plastic Halloween costume. Much like the one that Atkeson had paraded around in earlier, it was a Baymax costume that Atkeson had given to Kim when he left to work for Disney.
On the back of the package, Atkeson had scrawled: “MAKE ME WALK.”