Earlier today, Unbounded Robotics announced UBR-1, a $35,000 mobile manipulator. We spoke with the Unbounded Robotics team, including CEO Melonee Wise, CTO Michael Ferguson, lead systems engineer Derek King, and lead mechanical engineer Eric Diehr, about why their robot is so cheap, why it isn't, how it's going to make the world a better place, and why it only comes in orange.
IEEE Spectrum: Let's start off with the most important question: why is UBR-1 only available in orange?
Melonee Wise, CEO: Because orange is a great color! And because this robot is the Model-T of robots. Just like the Model-T, it's only available in one set of colors: orange and white.
Derek King, lead systems engineer: It's available in other colors, as long as you buy 100.
What makes your robot better than anything and everything else?
Melonee Wise: That's a complicated question! I would say that this platform is an advanced platform. It has a lot of powerful capabilities, like the PR2 does, and it's available for a tenth of the cost of the PR2, which is a big deal. You’re now transforming a community that so far has had a hard time growing due to the sheer cost of the platform: the PR2 community has less than 40 universities. If you were to try and develop the PC 30 years ago with only 40 institutions working on it, you would have a hard time getting to the place where you are today, and by having a platform like this, you can multiply the community by 100 or 1000 users, and really start the robotics revolution. With UBR-1, we're at the very beginning.
Why only one arm?
Melonee Wise: If you look at the type of research being done today and the applications that people are using robots for, a lot of them only use one arm, and when we talk to professors that are using two arms with their robots, they came back and they said, "when I really think about it, I don’t really need that second arm, and I could just buy two if I really needed a second arm."
Mike Ferguson, CTO: With the PR2, even though you’ve got two arms, they’re stuck twelve inches apart. If you think about folding things, PR2 can do small pieces of clothing, but if you want to do something like bedsheets, you typically need a second person to help set things up. If you've got two robots, those arms can be any distance apart. Also, if you have a second arm, you'd have to have a much larger robot. The PR2 is scary large. It has a hard time getting through a lot of spaces.
Melonee Wise: And of course, having one arm does help reduce the cost and size of the robot in general. If you look back to the PR2, a single arm of the PR2 weighs 80 pounds. Our robot only weighs 160 pounds total. You're talking about an arm that weighs half the weight of our entire robot. So that wouldn't leave much room for downsizing your robot.
You've made some comparisons to the PR2, but what about other platforms that might be considered competition?
Melonee Wise: Other platforms that might be competition are the Care-O-Bot; it’s not commercially available, but it has the same capabilities and costs about $380,000, I think. You could maybe say that Baxter is competition, but it would be a hard comparison since Baxter isn't mobile. UBR-1 has much of the same functionality and it's mobile.
Mike Ferguson: And Baxter also doesn't have the same type of integrated sensors. The capabilities generally available open-source in ROS today really depend on having a Kinect-like device integrated into the robot, and that's not available out of the box with Baxter, so you end up having to sort of strap that on and run it through an external computer.
Melonee Wise: One of the key aspects of this platform is that it is a platform. It's completely integrated, it's not a mobile base with an arm attached. If you go towards a product like that, or a combination of products like that, you struggle sometimes because there’s still a lot of integration work that has to be done. We're offering a cohesive platform with software integration, and we're even shipping with ROS MoveIt! straight out of the box, which is a big request from a lot of people. There are no other robots doing that today.
How is that going to make the UBR-1 user experience better?
Melonee Wise: If you look at the development of robotics with ROS, PR2 shipped with navigation on board and running, but it didn't really ship with arm controller that would do pick and place. It had controllers for the arm, but it was pretty much left up to the researcher to develop those programs. Today, with MoveIt!, you're able to do real-time pick-and-place. And that's a big advantage for people who are just doing vision research, for example. They don’t want to know about robot arms; once they tell the robot where the object is, they just want the robot to pick it up. And that’s really valuable because it allows people to focus on what they're good at, whether it's in research or even in business applications.
So UBR-1 is going to come with some basic capabilities like grasping and navigation, and going forward, you'll add more?
Melonee Wise: From our business perspective, what we see ourselves doing is obviously delivering a state-of-the-art research platform, but also we plan on partnering with businesses to develop applications to do business automation: stocking shelves, bin picking, and inspecting parts. Those are examples of applications that we plan on developing with our business partners, but there’s always going to be a part that we're going to want to put back into the community to create additional standard base capabilities for the robot.
Derek King: One of the things that having a little bit more of a business focus gives us is that we’re committed to robust software, where it’s about more than just a good demo. And that helps researchers when they’re trying to build higher level applications on top of these more robust lower layers. I think at Willow Garage, there wasn’t as much focus on extremely robust applications as there could have been.
Does that make UBR-1 a research robot, or a commercial robot, or what?
Everyone: It's a platform!
Mike Ferguson: You’re going to see it take off more in that research space first, but, being a platform, you can use it for all sorts of different applications. And having a lot of tools available to make this robot do a lot of things is what’s going to allow it to move beyond the lab.
Melonee Wise: It will also help with technology transfer. Right now, with the PR2, since it's a $400,000 robot, it’s very hard to transfer the technology that’s being developed for it to any place else. UBR-1 is a robot that’s just as capable, that’s affordable for businesses. We worked with people while we were at Willow Garage to try and do business automation, and we were successful in doing several things, like tidying an office or doing some bin picking. But, eventually the conversation had to get real: "Do you want to buy a $400,000 robot?" And at that point, things fell flat. With this $35,000 platform, we’re able to start engaging with people, and we’ve starting talking with businesses about using UBR-1 as part of their business plan. If you look at the scope of things that this robot is capable of doing, it’s not just limited to manufacturing, it has applications in other fields like elder care or even hospitality service. We actually talked with a hotel while we were at Willow, about how robots could do room service. With a robot like UBR-1, the possibilities open up a lot more. It’s more tangible. People can see a role that it would play in their businesses.
Cost is obviously a major factor here. Why is UBR-1 so cheap, and why is it so expensive?
Melonee Wise: That's a great question. So, why is it so cheap? We’ll start there. In terms of affordability, we’ve come down in price versus capability. The PR2, which is the closest comparison that can be made in terms of overall capability, is $400,000. When we started Unbounded Robotics, we focused on making this robot production-grade, and also focused on cost reduction. If you look at the progression of the PR2, it really wasn’t designed for production. And it was five years ago: Kinect didn’t exist, so the PR2 had something like $5,000 worth of sensors just in the face. There are advances like those, and then we focused on cost reduction and really bringing this down as much as we could to make it more affordable to people.
Derek King: Initially, at Willow Garage, the PR2 was never going to be made in the quantity it ended up being built in. I think the first number that might have been made were five robots total. Then ten and eleven, and finally, fifty. But that drove a lot of decisions early, where they could have used more time to design it more cheaply, but it didn’t seem worthwhile if you’re only going to make five. And that’s part of the reason the PR2 is expensive. The other thing we’ve gained from working on the PR2 is that we know what works and what doesn’t, and some of the things that were expensive in the PR2, we were able to eliminate and go with cheaper options.
Okay, so why is it so expensive, then?
Melonee Wise: I think that since we're going for a general-purpose platform, there are some things that we added because we wanted to make it very capable. If you really wanted to cost reduce this robot even more, you would focus very specifically on the features that you need for a specific application, and then just start eliminating other things that add cost to the platform. But then you start limiting yourself, and we want to be able to have hundreds or thousands of these robots available to people to program on and to use in their businesses and then hopefully create a whole app infrastructure around. And that’s enabled by this being a generic platform with many capabilities.
UBR-1 has just two primary sensors; the PrimeSense in the head and the laser scanner in the base. Is this really enough to make applications with?
Melonee Wise: Yep. At the end of the day, most of the researchers just end up using the Kinect.
Derek King: The other thing we have on our robot are modularity points, so if you did want some higher level sensor, you could just mount it and have the USB 3.0 connection to attach it to the computer. So for the people who want that extra sensor that they can’t do without, it's easy to add it to this platform. One of the other modularity points is the gripper. We put an 80 percent gripper: it can do 80 percent of the things you might want to do, but there’s definitely a lot of people that want either compliant grippers, or more fingers, or suction cups, or electrostatic. So to handle that, we made the gripper as modular as possible for either us or outside vendors or even customers to replace our gripper with their own designs.
You'll be releasing two slightly different versions of the robot, right?
Melonee Wise: Yes, we’re going to be releasing two versions of the robot. One is the base platform which we assume will go into research, because the areas that researchers are probably going to be working in are relatively small. But if you want to go into a warehouse or a supermarket or something, you’re going to need a longer-range laser sensor, and so we’ll put a 30 meter laser in it, and then that combined with a zippier processor and a higher top speed will enable the robots to work much more effectively in a bigger environment.
Why is the world in desperate need of an affordable, mobile manipulator? What is this going to enable?
Melonee Wise: I think one of the the biggest barriers to technology in general is collaboration. And until the device that you’re using to enable that technology to take off is affordable, it’s very hard to get that order of magnitude increase in collaboration. We saw it with the PR2 and ROS: over the last five years, robotics collaboration has really taken off. We went from almost no collaboration to lots of collaboration, and there are thousands of packages and robot software available for collaborating, but it’s still handicapped in some sense by robots being "different." We’re still struggling there. PR2 is starting to make a good headway in that direction, but like I said, there’s only 40 or 50 of those robots out there in the world. With UBR-1, you could get 400 or 4,000 labs collaborating. Robotics has been inhibited by not having a common platform and a common software infrastructure. The software infrastructure has now become mature, but a robotics platform really hasn't broken through yet. UBR-1 will be one of the first of its kind to do that.
Last question: let’s talk about UBR-2. What can we expect in the next generation?
Mike Ferguson: At this point, the limiting factor is not going to be the hardware. For a while, it’s going to be the software. It iterates one after another. Eventually, we’ll push the software to the point where the hardware is a limiting factor, and we'll start to look at the UBR-2, but really, it’ll be a software push probably before we do another robot.
Melonee Wise: I think that UBR-2 is going to be heavily driven by the applications that we decide to really hone in on. It could go both ways: we could make the platform more capable in the generic sense, by adding more abilities to customize the robot and make it more modular. Or, we could go after cost-reducing the robot for a very specific set of applications, but that really depends on what the future holds, and how applications evolve on the platform.
Read more about Unbounded Robotics' UBR-1 here. Special thanks to Melonee Wise, Michael Ferguson, Derek King, and Eric Diehr for speaking with us.