Can Willow Garage’s “Linux for Robots” Spur Internet-Scale Growth?
Robot builders have a lot to learn from Internet entrepreneurs.
That’s one of the main arguments you’ll hear from the engineers at Willow Garage, a unique startup in Menlo Park, CA, that’s developing hardware and software for a new generation of personal robots. You can’t name a single Internet company, they say, that would have succeeded if it had been forced to recreate all the basic tools underlying the Web, from the Linux operating system to the Apache HTTP server to the MySQL database system to the Python, Perl, and PHP programming languages—the ingredients of the so-called LAMP stack. Yet most robot companies still try to reinvent the wheel every time, building robots that require putting together a tangled mess of proprietary, one-off software systems.
Which is exactly why robots are, for the most part, stuck on factory floors welding car parts rather than assisting in our homes, offices, hospitals, and other settings, argues Brian Gerkey, Willow Garage’s director of open source development.
“What we need is a LAMP stack for robotics,” Gerkey says. “Before the LAMP stack, [Web developers] had to know everything about how to write an operating system and manage processes and write a file system and manage sockets and so on. The ability to come in with software engineering skills and an idea about how to use the Internet, and be off to the races—that is what enabled the Internet boom.”
In the same way, Gerkey explains, today’s roboticists “have to come at the problem with a very deep expertise in all aspects of robotics, from state estimation to planning to perception, which automatically limits the number of people capable of building new things. But by providing a basic toolset analogous to the LAMP stack, we can get to a point where all you need to know is how to write code and what you want your robot to do.”
That point still seems quite a ways off—and if you walk into Willow Garage, the most visible product is actually not a piece of software but a humanoid robot called PR2. It’s almost six feet tall and has stereo vision, a wheeled base, and a pair of arms with hand-like grippers; imagine a cross between Wall-E and the “Lost in Space” robot, and you’ll have a good picture.
Academic roboticists love PR2’s modular, extensible design. But it was never intended as a mass-market product—as its $400,000 price tag attests. Only in the last six months, says president and CEO Steve Cousins, has the company started selling more PR2s than it gives away. The real show at Willow Garage, it turns out, is the software running under PR2’s hood.
Called the Robot Operating System, or ROS, it’s a collection of algorithms that handle standard tasks required of every mobile robot—things like making sense of a visual scene, or planning a path around obstacles. Unlike PR2, ROS is completely free, and is already being adapted by hundreds of robotics labs and companies around the world. It’s spreading so fast that Cousins says Willow Garage is considering creating a non-profit foundation, similar to the Apache Software Foundation, that could organize the developer community, collect donations, and act as an independent steward and champion for the software.
In Willow Garage’s world, PR2 is just a big, rolling demo. If ROS is like the LAMP stack, then PR2 is like a high-end blade server from IBM or HP—the reference hardware that shows what can be done with the software and inspires innovators outside the company to program and build their own new kinds of robots.
“PR2 is not going to be the thing that makes Willow Garage successful in the long term,” says Cousins, who previously led research projects in human-computer interactions at Xerox PARC and IBM’s Almaden Research Center. “It is a catalyst. It helps researchers and the academic community solve the problems that need to be solved to get this industry created. PR2 and ROS together help them solve those problems faster and more efficiently.”
[I sought to learn more about Willow Garage recently as part of my preparation for the next big Xconomy San Francisco event. Both Cousins and Gerkey will be on stage at The Future of Robotics in Silicon Valley, a special Xconomy forum planned for May 3.]
You can trace the beginnings of Willow Garage’s Internet-inspired philosophy to Scott Hassan, the dot-com tycoon who founded the company in 2006 and is still its main financial backer. As a computer-science doctoral student at Stanford, Hassan helped Larry Page and Sergey Brin build the first version of Google, which earned him a big chunk of Google shares. He also helped to build Alexa, the Web traffic monitoring service bought by Amazon in 1999 for $250 million. And on top of all that, he started FindMail, an e-mail list management service that grew into eGroups.com and fetched $413 million when Yahoo bought it in 2000.
If you’ve earned a fortune three times over thanks to the Internet, you’re automatically a big proponent of open source software—and you’re likely to believe that creating a shared platform like the LAMP stack is the best way to kickstart innovation in almost any technical field. “This is a classic chicken-and-egg problem,” Hassan said at the 2010 launch event for Willow Garage’s PR2 Beta initiative, a program aimed at giving away 11 PR2s to research teams around the world. “Without robots we can’t develop useful apps, and without useful apps to run, there are no robots,” Hassand said. “So the question is, how are we going to make this happen? What we are bringing to the table is a platform—a sturdy, robust platform,” meaning the combination of PR2 and ROS.
Gerkey was a logical choice to lead Willow Garage’s effort to develop ROS. Aside from being a veteran of Stanford’s Artificial Intelligence Lab and SRI’s Artificial Intelligence Center, he’s the founding developer of Player, a widely used open-source interface that lets researchers run robots over a network connection. He says ROS began in 2007 as a collaboration with researchers at Stanford. “The motivation was very much parallel to the motivation for building the PR2. It was natural to say that if we are going to build this hardware platform to give people a better starting point, we should match that with a software platform that gives them the nuts and bolts for everything from talking to sensors to inter-process communication, so they don’t have to worry about those things.”
ROS runs on many robot models, but there’s no question that it co-evolved with PR2 as Willow Garage readied the robot for public demos intended to show off its capabilities. “The best example of that was the tabletop manipulation demo,” says Gerkey. “The robot looks at a tabletop with no idea of what’s there, and it identifies objects, recognizes them, decides how to grasp on to them, picks up an object, moves it, and puts it down. That pick-and-place functionality is a basic capability that you would expect out of any robot like PR2—but if you don’t have a system like ROS you have to build an entire system of algorithms just to test it.”
The deep idea behind ROS, Gerkey says, is to save engineering time and let researchers leapfrog over solved problems. If your expertise is in computer vision, you want to put your effort into fine-tuning your object recognition algorithms—not wrestling with your robot over how to get the cameras to talk to the central processor, or how to move the gripper arm toward an object.
“Now you have the possibility of directly and empirically comparing competing approaches,” says Gerkey. “It’s one of the dirty little secrets of robotics that this hasn’t been done regularly. The story of robotics in the research community tends to be, you do a demo and show a video and everyone says ‘Isn’t that cool.’ There isn’t a lot of the head-to-head comparison of results that would really serve the community better.”
Researchers jumped on the ROS bandwagon from very first “0.4” release in 2009. The first stable 1.0 release came in February 2010, and every six months since then, Willow Garage has pushed a new release, each one named after a variety of turtle. (That’s a reference to another Willow Garage product, a squat little robot called TurtleBot. It’s a mashup of iRobot’s Create chassis, Microsoft’s Kinect sensor, and an Asus EeePC netbook.) Willow Garage puts no restrictions on the distribution of ROS, and Gerkey says he has no way of knowing exactly how many roboticists are using it. But he says there are 100 to 150 ROS code repositories around the world, containing at least 3,000 “packages” or user-contributed components that handle functions such as mapping, perception, or simulation.
In late May, Willow Garage and seven other companies will sponsor a two-day ROS developers’ conference called ROScon. It’ll be in St. Paul, MN, right after the IEEE’s International Conference on Robotics and Automation (ICRA), the world’s most important academic robotics conference. “That looks like it is going to be a big hit,” says Cousins. “We have sponsorship from a number of robotics companies, which is really exciting,” as it’s a display of industry solidarity around an emerging standard. The sponsors include Bosch, Clearpath Robotics, Heartland Robotics, Yaskawa, Coroware, Schunk, and Yujin Robot.
To make more room for collaboration, Willow Garage plans to cede stewardship of ROS to the community itself, in the form of a non-profit foundation. “The role of a ROS Foundation would be to champion the ROS project, including getting more developers, getting more users, running a blog, and highlighting interesting use cases or companies or research labs,” Gerkey says. “It would also act as a place that can take in funding from stakeholders, whether they are the research arm of a big company, or a smaller company that wants to see some targeted development, or government funding agencies.”
Willow Garage’s commercial status makes it tricky to handle such contributions or contracts right now, Cousins says. “We have always wanted ROS to be a community thing,” he says. “We are going to continue to support it, but we would like to open it up to a broader base of support, and there are a lot of people who would like to contribute but who don’t feel right sending their money to a for-profit company.”
Meanwhile, ROS is beginning to make its way out of pure research environments and into settings where robots are performing serious work. Cousins says Willow Garage invited manufacturing systems engineer at the Southwest Research Institute in San Antonio, TX, to spent several months at Willow Garage last year. They were able to prove that ROS can be used to run robot arms used for cutting, welding, painting, packaging, assembly, and many other tasks. That was a bit of a surprise, since such arms, built by companies like Yaskawa’s Motoman Robotics in Miamisburgh OH and Adept Technologies in Pleasanton, CA, typically come with their own specialized control software. But it turns out that it’s not so hard to convert them to run on open-source code.
“This code, which is fundamental to what we are trying to do at the cutting edge of research, is all of a sudden available to all of these different industrial arms, and that is exciting to people in the community,” says Cousins. “It’s exciting to see this bridge forming between the academic community and the industry community.”
Of course, if you visit Willow Garage, what you’ll see is robots, not the software running inside them. The PR2 is still the company’s flagship product, and on any given day, a handful of them are rolling around the building on various research assignments (or perhaps just searching for power outlets where they can recharge). These days, you’ll also see quite a few Texai robots—“remote telepresence” bots built to help remote workers to connect with their colleagues by enabling them to send a camera-equipped robot roaming around the office. (This New York Times video explains the concept.) So many people have expressed interest in Texai that Hassan has started a spinoff company, called Suitable Technologies, to commercialize them.
And creating even more spinoffs like Suitable is part of the plan at Willow Garage. But unlike most of other Silicon Valley startups, the company is not in a rush to build money-making products. Thanks to Hassan’s deep pockets, it can take its time building a solid foundation, in the hope that a future Willow Garage—or its spinoffs, or even its competitors—will be able to build the upper floors.
“Software is a huge lever,” says Cousins. “In the Internet boom, companies with one or two people could put together a piece of software that made a big difference in people’s lives. The business argument I make is that if we invest a few million now, we can cause the future to happen five years sooner than it would have.”