At Unbounded Robotics, Smart Compromises Bring Down Costs
When is one arm better than two? When you’re trying to build a mobile-manipulator robot that somebody can actually afford to buy.
One of the problems that kept PR2, a humanoid robot developed by Menlo Park, CA-based Willow Garage, from succeeding commercially was its $400,000 price tag. And one of the things that made PR2 so expensive was that it had two arms, each replete with a complex series of joints, motors, and sensors.
But as it turned out, most of the university labs and other institutions that bought PR2s never came up with applications that required both arms.
“I think there are 43 PR2s out there, and if you look at the cross section of all those universities, only three or four actually use two arms,” says Melonee Wise, who was the second employee hired at Willow Garage in 2006. It turns out, Wise says, that “you can do a lot of tasks with just one arm.”
And that, in short, is why UBR-1, the first robot from Wise’s new company Unbounded Robotics, is single-limbed. It’s one of the many engineering decisions and technology improvements that will allow the startup to sell UBR-1 for just $35,000.
Wise says one roboticist who programs robots to fold laundry—normally a job for two arms—told the company, “Your robot is so inexpensive that if I needed to have a second arm, I’d just buy a second robot.”
Today, Willow Garage is defunct, its employees having scattered to spinoffs such as Suitable Technologies, the Open Source Robotics Foundation, and Unbounded itself. And there’s a new recognition, Wise says, that robots won’t find their way into new niches until they can solve real problems at much lower cost than the expensive, research-oriented, all-purpose robots of the past.
UBR-1 is a compact, wheeled robot designed to work alongside humans in places like factories, warehouses, supermarkets, and elder-care facilities. Along with its fancy arm—which can pan, pitch, roll, and grip in ways that make a human arm seem clumsy—it’s got a full suite of imaging sensors and communications tools (check out Wise’s full demo in the video below). It’s versatile, but most of the jobs it can do don’t even exist yet, because its capabilities will be defined largely by its programming rather than its hardware.
“With robots, feature creep is so much more present than in some other fields,” Wise says. “There is always this desire to make a Swiss Army knife. But you have to make compromises, and those compromises directly impact the capabilities as well as the cost of the robot. We wanted to strike a good middle ground between building robots and writing software.”
The speed with which Unbounded Robotics brought UBR-1 to market is a sign of the four-person team’s focus on practical compromise, guided by careful market research (what Wise calls “need finding”) and thoughtful design. The company’s spinoff from Willow Garage was completed in January 2013. By October, Wise was showing off the first production version of UBR-1 at the RoboBusiness conference in Santa Clara, CA.
In a way, Wise and her co-founders—chief technology officer Michael Ferguson, lead mechanical engineer Eric Diehr, and lead systems engineer Derek King—have formed their new identity in direct reaction to their experiences at Willow Garage. Founded by ex-Googler Scott Hassan, Willow long aimed to build flexible platforms like PR2 and the open-source Robotic Operating System (ROS) that would help other robot builders accelerate their own efforts. Only in the last couple of years before it shut down did the company begin to focus on getting something to market.
In fact, in early 2012 Willow Garage CEO Steve Cousins had asked Wise, Ferguson, Diehr, and King to build a mobile robot that would be smaller, cheaper, and more affordable than PR2, and they’d built a prototype that met a number of important milestones. But by late 2012, Wise says, it still wasn’t clear how or when the new “PR2 mini” robot (my moniker, not Willow’s) would be marketed. “There was real excitement, and everyone felt like we were going in a very positive, exciting direction,” she says. “But from our team’s perspective, it was like, ‘When are we going to start showing this to people and doing things with it in the real world?’”
So the four Unbounded co-founders decided to start over, outside Willow, with a blank slate. (Well, not completely blank—more on that below.) In a mid-December interview at Unbounded’s workshop in Santa Clara, Wise shared a few of the strategies that helped the company get UBR-1 to market in just 10 months. They’re the kinds of methods that more robotics companies may need to emulate if they want to build robots that customers will find affordable.
1. Pick Your Markets Carefully
PR2, Willow Garage’s versatile, twin-armed robot, wasn’t just expensive—it was also heavy (550 lbs) and so big it couldn’t fit through some doors. “When PR2 was originally envisioned, we thought we were only ever going to make five of them,” Wise says. “It was meant to be this grand prototype.” That meant there wasn’t much of an attempt to make the robot easy to manufacture, or to figure out exactly which problems it could cost-effectively solve.
Unbounded Robotics went at the problem from the opposite direction. Based on their need-finding work for the “PR2 mini” project, the team figured there were three market niches where a mobile manipulator robot might be welcomed. “We thought about logistics—pick-and-place, packing boxes, and things like that,” Wise recounts. “We thought of elder care. And then the long-term vision is supermarkets. For all those things, mobility is a major requirement. It needs to rove around and do the same things that people can do. Most likely, it will be doing things on tabletops or picking things up off the floor. And it will probably be lifting objects that are less than 1.5 kilograms.”
Those basic requirements drove everything else, from the height of the robot to the placement of sensors to the type of motors that would be needed inside the robot’s shoulder and arm joints. At the most basic level, UBR-1 is designed to help humans with precise but repetitive tasks that involving moving things around. In that sense, Unbounded could be seen as a competitor for Rethink Robotics. The Boston-based startup, founded by famed MIT roboticist Rod Brooks, makes a stationary, two-armed manipulator robot called Baxter that excels at tasks like picking objects off a conveyor belt and putting them into boxes. The big difference, Wise says, is that UBR-1 can move around.
“We want to be more like people and have the robot collaborate with people and trade tasks,” Wise says. “People are not always in one place doing one thing.” In an automated warehouse, for example, UBR-1 robots could pick objects off shelves, deliver them to other locations, and pack them into boxes (a scenario that might put Unbounded in competition with yet another Boston-area robotics company, Amazon subsidiary Kiva Systems).
To find UBR-1 its first real job, Unbounded is working with industry partners (Wise can’t yet name names) who are testing the robot in real-world situations. “We decided that how we were going to go at the market was to find a partner, start developing basic, MVP [minimum viable product] apps for that partner, and concentrate on apps that we thought could translate well between several types of companies,” she says.
2. Build It Right the First Time
In the world of software startups, the new guiding ethic in product development is “rapid iteration.” That means building something, however incomplete; testing it on real users; gathering feedback; and starting over. “A lot of people have really bought into this idea of ‘build it, break it, build it, break it,” Wise observes. “But with every ‘break it’ you are wasting a month—and in robotics, you are wasting $70,000.”
The team at Unbounded built exactly one physical prototype for UBR-1. Wise hauled it out of a back room during our interview: it’s a science-fair-style contraption made of particle board, a few cheap motors and batteries, and a Microsoft Kinect. The team called it Clunky. “This is how we feel about iteration,” Wise says. “If it can’t be done in a couple of hours, it’s not worth doing.”
The alternative to iteration is simple, Wise says: “Careful, thoughtful design, using the tools that are available, like FEA [finite element analysis] and simulation.” Unbounded’s first purchase as a company—even before it had office space—was a copy of SolidWorks, the 3D computer-aided design program from Dassault Systèmes. From February to May, the team slaved over the designs for UBR-1’s mechanical and electrical systems. In June, the company sent the design specs out to vendors to begin fabricating the required steel, aluminum, and plastic parts. In July and August, “every day was Christmas,” Wise says, as packages containing the new parts started to come in. The last of the parts arrived in September, and as soon as Wise and her team assembled the robots and turned them on, “they worked,” Wise says matter-of-factly. “If that hadn’t happened, we would have missed RoboBusiness.”
Chalk it up to thorough planning. “Having a clear indication and specification of where the design is going; setting a schedule and following it—I know it sounds absurd, but those are the things that have enabled us to be so lean,” Wise says.
3. Build on Reusable Components
There’s another element to the fast progress at Unbounded: it didn’t have to invent software for the robot. For that, there’s ROS, which Wise helped to develop at Willow Garage. More than just an operating system, ROS is also a simulation environment and planning tool; before Unbounded ever turned on UBR-1, it had a working model of the robot in ROS, allowing it to rehearse different motions and iron out problems before sending the specs for the parts to vendors.
Now overseen by the Open Source Robotics Foundation and used by hundreds of robotics teams around the world, ROS is probably the single most important product to come out of Willow Garage. “It’s a ginormous success story,” Wise says. “Without ROS it would have been very hard for us to turn the robot on in September and end up at RoboBusiness doing a demo in October.”
Unbounded also built on the co-founders’ experience with PR2, especially the design of the arm. “The underlying mechanism is significantly different, but the kinematic configuration is the same,” Wise says. That’s a reference to the specific motions that each segment of the arm can perform: panning, pitching, and rolling at the shoulder, pitching and rolling at the elbow, and pitching and rolling at the wrist. (Those are the “seven degrees of freedom” that Wise refers to in the video above).
The big preoccupation at Unbounded now is testing UBR-1 in actual workplaces—which kinds, Wise can’t yet say, because of non-disclosure agreements with partners. The hope is that these partners will help the company discover which tasks the robot is uniquely suited for—or how it can be adapted for jobs that no one would have thought to give to a robot before.
“I think the general problem in robotics is that there aren’t enough robots for people to get experience with,’” Wise says. “We will make headway in a niche space as more and more people start trying our platform and saying, ‘This is an 80 percent solution.’ Once you start seeing those 80 percent solutions, you start seeing other needs. And then people see how the platform can be modified to make solutions for those needs.”