http://nyti.ms/1L69N0m Cover Photo The National Robotics Engineering Center, part of the Robotics Institute at Carnegie Mellon. Credit Christopher Payne for The New York Times Uber Would Like to Buy Your Robotics Department When the company wanted a team of roboticists, it raided a university lab to get them. Can high-tech academia survive today’s Silicon Valley talent binge? By CLIVE THOMPSON Continue reading the main story I n the center of the lab, CHIMP stretched out one huge arm, then gracefully unfurled its three metal fingers, as if about to beckon someone. On its head, two rapidly rotating laser scanners enabled it to monitor its surroundings, but the engineers standing nearby were wary nonetheless. If the robot — a five-foot-tall, crimson-colored humanoid machine designed to emulate the complex movements of the human body — moved suddenly and hit one of them accidentally in the chest, it could easily break some ribs. The contraption weighs 407 pounds and is powerful enough to bench-press 150 pounds. SEPT. 10, 2015 ‘‘It’s a hazard for humans to be around — we wear steel-toed boots,’’ Michael Vande Weghe told me. A tall, lanky man who was indeed wearing steel-toed footwear, Vande Weghe has spent the last two years as one of the engineers building CHIMP for the National Robotics Engineering Center at Carnegie Mellon University. (CHIMP stands for ‘‘C.M.U. Highly Intelligent Mobile Platform.’’) The robot is one the best-known creations to emerge from the center: Earlier this year it placed third in a competition sponsored by the federal government to create an automaton able to perform several humanlike tasks, like cutting a precise hole in a wall with a hand-held electric saw, climbing a small set of stairs and opening a door. Though it finished behind teams from South Korea and Florida, CHIMP made robot history by falling and then returning to its feet . Other robots fell over without getting back up. ‘‘I aged 10 years when that robot fell down, and I gained it back when he got up again,’’ says Herman Herman, the director of the center, who has worked there since its founding 20 years ago. The lab has come by its storied reputation by producing robot technology for clients like the American military and heavy-machinery manufacturers like John Deere and Caterpillar. Today the place is crammed with projects like ‘‘Crusher,’’ an armor-plated military robot that can find its own way across craggy, wooded terrain, and a smaller wheeled robot designed to explore mines for the company Anglo American. Plaques hanging along a wall commemorate some of the hundreds of patents received over the years by the center’s 100-person staff. Continue reading the main story The New York Times Magazine Newsletter Sign up to get the best of the Magazine delivered to your inbox every week, including exclusive feature stories, photography, columns and more. There was something missing from the lab today, though. In February, a group of some 40 employees — including several longtime senior lab members — resigned. They had been lured away, en masse, by a new employer: Uber. The San Francisco firm, famous for its popular car-dispatch services, is getting into the robotics business itself. In February, barely a mile away, it opened the Advanced Technologies Center in Pittsburgh, where the former university researchers are now developing technologies to help Uber extend its reach over the roads. These include producing better maps and safer guidance systems — and most lucrative, even if they’re still years away, Uber’s very own fleet of self-driving cars. It was a startling raid of talent. By offering private-sector salaries substantially higher than university equivalents (as well as a chance to earn equity in a fast-rising tech firm), Uber was able to acquire a hefty chunk of the center’s brain trust, including some top experts in autonomous vehicles. In doing so, the company has placed a bet that self-driving robots are no longer the stuff of scholarly visions but valuable intellectual property.
Advertisement Continue reading the main story Carnegie Mellon’s experience is a familiar one in the world of high-tech research. As a field matures, universities can wake up one day to find money flooding the premises; suddenly they’re in a talent war with deep-pocketed firms from Silicon Valley. The impacts are also intellectual. When researchers leave for industry, their expertise winks off the map; they usually can’t publish what they discover — or even talk about it over drinks with former colleagues. In the long run, raids can generate symbiotic relationships; researchers who return to academia years later bring their real-world experience into the classroom and can draw on their network of wealthy industry contacts to fund university research. But as Carnegie Mellon’s roboticists are finding, reaching that end point can make for a bumpy ride. There’s a useful high-tech concept called the Technology Readiness Level that helps explain why Uber pounced when it did. NASA came up with this scale to gauge the maturity of a given field of applied science. At Level 1, an area of scientific inquiry is so new that nobody understands its basic principles. At Level 9, the related technology is so mature it’s ready to be used in commercial products. ‘‘Basically, 1 is like Newton figuring out the laws of gravity, and 9 is you’ve been launching rockets into space, constantly and reliably,’’ says Jeff Legault, the director of strategic business development at the National Robotics Engineering Center. Today’s early-stage inquiry — so-called basic research, the Level 1 work, where scientists are still puzzling over fundamental questions — is financed almost exclusively by the federal government. It’s too far out, too speculative, to attract much investment; it isn’t clear if anyone will make any money on it. This wasn’t always the case. Decades ago, corporations were more willing to engage in Level 1, moonshot research. Bell Labs supported the work that led to the transistor when it was far from clear that there would be a market for it; Xerox supported research into the ‘‘windows’’ style of computing years before the market existed for such an interface. But in the last few decades, the vista of corporate R.& D. has shrunk as markets and executives have focused more on short-term profit, says Marc Kastner, an M.I.T. physicist. The far-off research questions have been left to university labs, though they struggle, too: The percentage of the federal budget devoted to basic research is about half of what it was in 1968. These days, private industry gets involved mostly when a field of research has matured to the midpoint of the NASA scale. In the ’90s and early ’00s this happened to ‘‘machine learning,’’ the science of getting machines to recognize patterns. It had long been an academic concern. But once online firms like Google began grappling with ‘‘big data’’ — search-engine requests, social-network behavior, email — the field became suddenly lucrative, and Silicon Valley started frantically hiring experts away from Stanford.
Advertisement Continue reading the main story Photo Carnegie Mellon’s CHIMP robot, assisted by the roboticist Prathamesh Kini, demonstrating its fine motor skills. Credit Christopher Payne for The New York Times
Advertisement Continue reading the main story Researchers who leave for industry are paid better, certainly, and often get sizable research budgets. But the intellectual register of their work changes. No more exploring hard, ‘‘basic’’ problems out of deep curiosity; they need to solve problems that will make their employers money. When the computer scientist Andrew Moore left Carnegie Mellon to become a vice president of Google in 2006, his new job, he says, was to be ‘‘as useful as possible to someone trying to buy something’’ — which is to say, to help Google’s algorithms learn that the search query ‘‘my basement is smelly’’ should point to a selection of dehumidifiers. At Carnegie Mellon, robotics scholars spent decades slowly approaching this moment. In 1979, the university founded its Robotics Institute to tackle the basic problems in the field, like how to interpret sensor data so a robot could ‘‘see.’’ But by the ’80s, government agencies and private firms struggling to create industrial and military robots were asking Carnegie Mellon’s roboticists for help. To capitalize on this demand, the school established its National Robotics Engineering Center in 1995 and staffed it with a few faculty members and a large complement of full-time engineers, often young robotics graduates. In effect, Carnegie Mellon used the NASA scale to carve up its robotics research. The Robotics Institute would handle research from Levels 1 to 3 or 4, while the center would take technology from there and move it to 7. If John Deere approached the center for help with a self-driving tractor, for example, the center would produce a prototype that could be mass-produced while publishing its research publicly. ‘‘Typically,’’ Legault says, ‘‘people who call us, they’ve been looking around, and there’s nobody else who can solve their problem.’’ By the late 2000s, the field of “wheeled robots” was taking off. Google had started its effort to make self-driving cars, and in 2007 roboticists at Carnegie Mellon won a military competition to produce a car that could navigate city streets; automakers were developing systems to help drivers stay in their lanes. A sort of Silicon Valley of robotics began to emerge in Pittsburgh as investors funneled money to firms like 4Moms (which makes robotized strollers) and Aethon (robots that deliver supplies in hospitals), heavily staffed by former Carnegie Mellon robotics experts. Today, nearly a third of the university’s faculty in the field are involved in a start-up on the side. ‘‘If you’re well versed in the area of robotics right now and you’re not working on self-driving cars, you’re either an idiot or you have more of a passion for something else,’’ says Jerry Pratt, head of a robotics team in Pensacola that worked on a humanoid robot that beat Carnegie Mellon’s CHIMP in this year’s contest. ‘‘It’s a multibillion- if not trillion-dollar industry.’’ This precise epiphany was dawning on many at the National Robotics Engineering Center. One was John Bares, who ran the lab from 1997 to 2010. He had always loved the intellectual challenge of the work but felt frustrated that his projects produced only prototypes for clients, who might do nothing with them. ‘‘We’d do a fantastic job engineering, do an advanced prototype, so-called throw it over the fence — and it would sit,’’ he told me. Bares wanted to invent something and then sell it himself. ‘‘I just had the itch to do product,’’ he says. In 2010, he left the center to found a start-up, Carnegie Robotics. Within a year, his team built a robot that could detect land mines, which the Army bought. Photo Andy, a Carnegie Mellon robot intended to manipulate objects with little supervision. Credit Christopher Payne for The New York Times
Advertisement Continue reading the main story In 2014, Bares got an email from Uber that led to a meeting with its chief executive, Travis Kalanick. Kalanick said he planned to open a lab in Pittsburgh and told Bares that Uber’s mission was to increase the efficient use of cars, which would have environmental benefits. Bares was persuaded, and in January of this year he joined Uber. He knew that many of his former colleagues at Carnegie Mellon shared his entrepreneurial impulses. Back at the lab, in fact, a group was already preparing to break away. One of Uber’s recent hires from the National Robotics Engineering Center told me that he and his 20-person team had long found their lab work rewarding, but that they had been stunned by the rush of money into robotics in recent years: Google bought Nest, the Internet-connected thermostat firm, for $3.2 billion; Makerbot, a start-up that creates inexpensive 3-D printers, was acquired by another company for $604 million. ‘‘We were just flabbergasted,’’ he says, adding that his team could have invented those products ‘‘in two weeks.’’ ‘‘I have this big, well-tuned machine of a team, standing there looking at me saying, ‘Hey, man, when the hell are we going to do something big?’ ’’ he says. His response was: ‘‘I’m an entrepreneur at heart. Let’s do it.’’ He visited venture capitalists in Silicon Valley and lined up tens of millions of dollars to fund a robotics start-up. By summer 2014, he informed the National Robotics Engineering Center that he and his team meant to exit on Jan. 1 this year. Then Bares made his own offer: Join Uber’s new lab, where, he said, ‘‘their efforts would impact the world.’’ The team would get bigger salaries, and Uber could produce their inventions quickly. The team leader agonized over the choice; he’d spent the last year dreaming of his own start-up. But he opted for Uber. ‘‘They have the road map, and they have the fuel,’’ he told me. ‘‘They just need some tech. Here they are with an actual market for what we do.’’ The scale of Uber’s recruitment surprised nearly everyone I spoke to. ‘‘I know of no place that was raided like that,’’ says Julio Ottino, dean of the McCormick School of Engineering and Applied Science at Northwestern University. People at Carnegie Mellon did not want to talk about it; they declined even to confirm how many staff members had left. ‘‘In an academic institution, people do come and go,’’ Herman says. ‘‘The only unique thing here was the number is slightly larger.’’ In a typical year, he says, between five and 10 researchers leave for industry. ‘‘I tell people, ‘After you work here for a few years, you can get a job anywhere.’ ’’ The departures left Herman and Moore, who last year returned from Google to Carnegie Mellon as dean of the computer science school, scrambling to find replacements. They’ve already hired a dozen new researchers, from institutions like Columbia University. In the long run, though, Moore says, they have to develop their own young scientists, encouraging robotics grad students to apply to the center: ‘‘We’re responsible for building all these duplicates of all these luminaries in robotics.’’
Advertisement Continue reading the main story Many familiar with Carnegie Mellon say the raid had a silver living, because it signals that the university is a hot place to work. One of those is Richard Florida, the urban-studies scholar who argues for the economic role of the ‘‘creative class’’ in a city’s economy. Florida himself was a social scientist at Carnegie Mellon for years, and his 2002 book, ‘‘The Rise of the Creative Class,’’ was inspired in part by his noticing, in the ’90s, that engineers there with a good high-tech idea had to leave for the East or West Coast to find investors, and they rarely came back. ‘‘Pittsburgh has always been this caldron, but they couldn’t get scale,’’ Florida says. ‘‘What Uber provides is immediate scale.’’ And funding: This month, the company announced a $5.5 million gift to Carnegie Mellon to support a new robotics chair and three fellowships. Continue reading the main story A booming local industry gives — but also, inevitably, takes. Stanford gains luster from Silicon Valley, along with money for research. But the valley lures away promising scholars; in recent years, several have decamped to oversee online-course start-ups, for example, like Sebastian Thrun (who founded Udacity) and Daphne Koller and Andrew Ng (who created Coursera). ‘‘Nobody is happy about that, but those are decisions that the faculty make, and they make them for rational reasons,’’ says Jennifer Widom, who led Stanford’s computer-science department from 2009 to 2014. Short-term faculty stints in industry can be a good thing, she points out, because those who return can share with students their hard-won practical knowledge. The deeper impact of Uber’s hiring may be how it affects what ideas roboticists will pursue. After all, 20 years ago, if you wanted to push the frontiers of robotics, you had primarily abstract academic options to consider. But now that the field is booming, a faculty member or grad student with an ambitious idea has to ask questions: Where’s the best place to pursue my research — at a university, or in the corporate R.& D. lab at a place like Uber? What type of engineering work is so far out, so hard and so unsolved that it can only be done at a university? Moore says there is still blue-sky work to be done in robotics. One example is the problem of ‘‘grasping.’’ Sure, wheeled robots can steer themselves through an Amazon shipping center pretty accurately; a self-driving car can trundle around San Francisco without running people over. But no robot can yet match the dexterity of a human hand. It can’t fluidly and confidently manipulate objects on a table — like picking up a coffee mug (assuming it can first identify it). This is, Moore notes, a huge research challenge in the humanitarian and health fields, because such robots could transform the lives of people with mobility issues, including both the elderly and people with spinal-cord injuries. ‘‘There’s maybe about one million people in the United States who, if they dropped their TV remote on the floor, just have to wait for a caregiver to come along and pick it up for them,’’ Moore says. Grasping is a classic early-stage technology, one that the Ubers of the world won’t spend any time developing because the payoffs are too far away. Moore says he also wants this research done in a university setting so the results will be published openly and benefit the world at large.