Become a fan of Slashdot on Facebook


Forgot your password?

Video Willow Garage Makes Open Source Robots for Researchers (Video) 22

We're not talking cheap here; Willow Garage PR2 robots list for $280,000 with the academic discount, $400,000 without. Still, spokesman Ryan points out that it can take a PhD candidate two or more years to build a robot chassis and create new software equivalent to Willow Garage's open source robotware. The thought, too, is that if a university buys the robot a lot of students can share it. Sounds good, doesn't it? But much though we might want a robot, it's probably a good thing Slashdot doesn't have one because we'd probably spend all day fighting over who got to use it next.

Tim: So Ryan, we are standing next to a device that is not running Android, and it is not running Chrome, what are we looking at here?

Ryan: Right. So this is the PR2 robot. Basically, this is a robot development platform that we sell. Normally, we sell these into university research labs for people doing research on the premise that if you are doing your PhD, before this guy you’d have to spend two or three years making the robot before you actually got to the application part of it. So we try to solve that problem for you. So right out of the box we give you a really feature-rich robot. And we also have our robot operating system ROS that is open source and free. It doesn’t just run on this robot it runs on the PR2 it runs on 50 to 100 other robots. So you can download, mess around with it, play with it. Yeah, so right out of the box you can kind of put up whatever application you have, whatever you can imagine basically you can program it to do.

Tim: Now Willow Garage obviously does a lot of work to make the robot. How much of it is off-the-shelf parts, how much of it are things that are made custom only for Willow Garage robots?

Ryan: Right. So most of the machine, sheet metal and mechanical parts are custom. So I think it is about 60 percent of the total robot is custom. Stuff off-the-shelf, a lot of the sensors are off-the-shelf, a lot of the computing components are off-the-shelf, the batteries are off-the-shelf, while mostly the mechanical stuff is all custom and sourced locally by machine shops _____1:30 shops.

Tim: Can you talk a little bit about the grip from the end of this arm here?

Ryan: Sure. So this is the gripper. A lot of people have different preferences in grippers. We went for a simple and more robust design. So it is just a two finger gripper. Right now we have dummy fingertips on here, but normally we will have pressure sensitive tips on here, so you can we have a demo where it will with this gripper it will crush the egg, but with the sensorized tips on, it will be able to hold the egg without crushing it. So this gripper also has an accelerometer in it, so it will do cool demos where it can do fist pumps, high fives, and the laser the Star Wars demo you saw before. It also has an LED in here for tracking with the cameras.

Tim: Now this arm, how long is it? It is about 3 feet long?

Ryan: Oh, I don’t know the actually dimensions, probably it looks like maybe 1 meter and a quarter or so, yeah, it is a 7 degree of freedom arm, so you can just pretty much it is really flexible for programing manipulating getting stuff. The gripper is actually 8 degree of freedom. So we have a lot of people who are doing tabletop manipulations so they bring the robot in front, and have the robot look at the scene, recognize the objects and then attempt to grasp them, pick them up, move them around, that kind of thing.

Tim: What kind of hardware does the robot use in order to recognize things?

Ryan: So it has a multitude of sensors, starting at the top, you’ve got a 3D this is basically a connect sensor so it gives RGB, picture plus depth data. And this guy is nice. When we designed the robot, this wasn’t actually out, and we’ve added it to the robot since. Because it offloads some of the computing power from the actual CPU. So this is a really kind of game changer. It is really cheap, really nicely featured sensor, so we use that quite a bit now. We also have two sets of stereo cameras, one with a narrow, one with a wide field of view, and with stereo we can do depth perception also.

We also have a 5 megapixel global shutter camera. With global shutter it basically takes a snap of all the pixels at the same time, so you don’t get motion blur if you are moving. And the other thing is actually not a camera, this is actually an LED with pseudo random pattern on it so that we can project onto objects. Like say you have a white sphere in front of a white wall, you may not see it with a camera, but if you project on it, you can see the distortion in the pattern.

Tim: Can you talk about the actual computing guts of it?

Ryan: So this guy pretty much in terms of computing it is really nice for software developers. So in the base there are two servers, they are both identical. Both of them have two quad cores Xeon processors, so it is a total of 16 cores for the whole robot, each box has 24 gigs of RAM, and I think 1.5 terabytes of hard drive each. There is a lot of computing power. In fact, the biggest drain on the batteries is from the computers.

Tim: How many of these are in use around the world?

Ryan: So we have about 40 of them out in the world right now.

Tim: And if I wanted to put it on my Amazon credit card, how much would it hurt?

Ryan: So the list price for this guy is $400,000. Now if you’re shooting through the open source community, _____4:55 to ROS, you may qualify for a discount, and then it drops the price down to $280,000.

Tim: That’s quite a discount.

Ryan: Yeah, we are trying to encourage people to use ROS and unify the software process, so everybody can leverage off each other and leverage from what everybody’s using.

This discussion has been archived. No new comments can be posted.

Willow Garage Makes Open Source Robots for Researchers (Video)

Comments Filter:
  • by Anonymous Coward on Wednesday May 29, 2013 @04:38PM (#43854673)

    ... costs a lot less than $280k. It barely costs 1/5th of that, and schools tend to treat PhD students as if they have all the time in the world. This company needs a better pitch line than telling us that it saves a grad student two years of work.

    Sorry, that's just plain wrong. I don't know where you're pulling your numbers from. Perhaps a local excrement port.

    On the low end--in the US--an engineering student or physical science student should be making ~21,000 USD/year. (Been there, done that, recently. This is complete and utter crap in comparison to industry) The advisor/project head is expected to cover ~45% overhead for each salary in tuition, health care, and other costs--so roughly 30,000 USD per year. At dirt cheapest, you're looking at 60,000 (more than 1/5 the cost, barely)

    On the high end--here in Switzerland-- a PHD will make close to 58,000 USD base (Still awful compared to local industry rates). After accounting for overhead (mandatory pensions, taxes, accident coverage...) that shoots up to about 72,500 a year. A net cost of about 145000 or about 52% of the list price to an academic.

    In either case, you get a homebuilt solution that has a useful expectancy of exactly one generation of students. (The thing will be so poorly functional that no one but the builder will be able to use it) For your two years of startup costs, you get at best two years of productive use out of it but have to pay for three (dissertations take about 1 year of funded time to write, regardless). So now you're footing another 90,000 to 220,000 in costs.

    On the low end, how does 105,000 (21000*5, cheap US) to 360,000 (72,000*5, expensive, CH) USD every 5 years make sense to anyone against 280,000 USD every 10 years (typical capital depreciation timeframe). Especially when the new instrument can be split amongst multiple grad students and projects?

    None of this includes the additional costs for the Professor/project head, who is paid between 3x and 4x as much per year as the grad student, or support staff costs who generally total to about 0.25x grad student costs/project?

    Damn_registrars needs to get an actual clue about how much research actually costs before criticizing the the company's cost/benefit performance. As far as commercial equipment goes, this is remarkably cheap/undersold...

  • by HizookRobotics ( 1722346 ) on Wednesday May 29, 2013 @05:23PM (#43855003) Homepage

    Actually, the fully-loaded cost for a PhD student is $75k-$80k / year (ie the amount charged to a faculty member's grant). You have to remember, PhD students' tuition is usually incurred as part of the cost since they're working in exchange for (1) a minimally-viable living stipend and (2) fully-paid tuition.

    We had two PR2's in our lab (Georgia Tech's Healthcare Robotics Lab). There were ~2 people working on each at any given time.... so the $$ makes sense. And the PR2 was a great platform!

    Source: my work on the PR2 []

"Luke, I'm yer father, eh. Come over to the dark side, you hoser." -- Dave Thomas, "Strange Brew"