Wizards of ROS: Willow Garage and the Making of the Robot Operating System

How a small band of Silicon Valley engineers started a global robotics revolution

24 min read

Evan Ackerman is IEEE Spectrum’s robotics editor.

Erico Guizzo is IEEE Spectrum's Digital Innovation Director.

One of the initial PR2 prototypes, built in 2008, at Willow Garage.
One of the initial PR2 prototypes, built in 2008, at Willow Garage.
Photo: Willow Garage

Ten years ago today, an engineer at Silicon Valley robotics lab Willow Garage published a new code repository on SourceForge. The repository, made publicly available to anyone in the world who wanted to access it, hosted the codebase for a new project Willow was working on: ROS.

The ROS code repo, set up by Ken Conley, ROS platform manager at Willow, on November 7, 2007 at 4:07:42 p.m. PT, was the first time the term ROS was used as a formal, public designation for Willow’s Robot Operating System project.

Choosing an exact date for the 10th anniversary of ROS is a little bit tricky, because what we know as ROS today is both older and younger than this: The concept of a robot operating system started at Stanford University, evolved through Willow Garage, and now resides with Open Robotics. It’s a complicated story that has shaped much of the robotics industry in recent years, and as robotics research makes the difficult transition to companies and products, the influence of ROS is becoming even more pronounced.

Over the last several weeks, IEEE Spectrum has been speaking with many of the people who helped shape ROS, from its origins as part of Stanford’s Personal Robotics Program to Willow Garage and its PR2 Beta Program, and beyond. Of course, there are many other people who contributed to ROS, and we weren’t able to talk to them all. This is our initial effort to put together an oral history and tell as much of the story of ROS as we can, in the words of the people who were there, making it happen.

What follows are excerpts of our interviews with eight people:

  • Eric Berger was software lead on the Stanford Personal Robotics Program, and director of the Personal Robotics Program at Willow Garage. He is now CTO of Desmos.
  • Keenan Wyrobek was mechanical systems lead on the Stanford Personal Robotics Program, and director of the Personal Robotics Program at Willow Garage. He is now head of product and engineering at Zipline.
  • Morgan Quigley was a PhD student at the Stanford AI Lab. He is now chief architect at Open Robotics.
  • Steve Cousins was CEO of Willow Garage. He is now CEO of Savioke.
  • Brian Gerkey was director of open source development at Willow Garage. He is now CEO of Open Robotics.
  • Melonee Wise was senior engineer and manager of robot development at Willow Garage. She is now CEO of Fetch Robotics.
  • Tully Foote was software development manager at Willow Garage. He is now ROS platform manager at Open Robotics.
  • Ken Conley was ROS platform manager at Willow Garage. He is now an engineer at Aurora Innovation.

We grouped their comments into sections, trying to organize things thematically as well as chronologically, though some events may overlap or span many months or years:

1. Stanford, STAIR, and PR1

2. Scott Hassan and Willow Garage

3. Willow’s Personal Robots Program

4. ROS and PR2

5. ROS in the Open

6. Beyond Willow

1. Stanford, STAIR, and PR1

Eric Berger: Before ROS itself was a concept, Keenan Wyrobek and I were working in Ken Salisbury’s lab at Stanford, running a project called the Personal Robotics Program. We had two things that we were trying to do: Build a hardware platform, and build open-source software tools with the fundamental goal of building a robotics development platform. We were grad students, the problems we saw around us were grad student problems, and we saw grad students in robotics wasting a whole lot of time. People who are good at one part of the robotics stack are usually crippled by another part—your task planning is good, but you don’t know anything about vision; your hardware is decent, but you don’t know anything about software. So we set out to make something that didn’t suck, in all of those different dimensions. Something that was a decent place to build on top of.

“We saw grad students in robotics wasting a whole lot of time. People who are good at one part of the robotics stack are usually crippled by another part … So we set out to make something that didn’t suck”

Keenan Wyrobek: We talked to people who did impressive PhDs in robotics, and we kept hearing this same theme: “Yeah, I spent four years hacking together a robot and a bunch of code, I barely got it working, I made a video, and that was the end of that.” We heard that over and over. At some point, Eric and I were just like, what’s actually important is solving this bigger problem that people are talking about. I think both of us hate wasting our time. Maybe we’re lazy would be another way of putting it, but the notion of spending four years doing what someone else had done before… we couldn’t get excited about it. That’s where it started: Could we actually create something that would enable people to build on each other’s results rather than continuing this cycle of 90 percent duplicating what someone else has already done, with a little bit at the end of something new, if you’re lucky.

PR1 robot prototypeThe PR-1 prototype, a mobile manipulation development platform, was built by Keenan Wyrobek and Eric Berger as part of the Stanford Personal Robotics Program at Ken Salisbury’s lab.Image: Stanford

Eric Berger: We built a hardware prototype, the PR1, and we started working on software to try to get it to move. That was the original thing that had the name ROS on it, although I don’t think it actually shared any code with the thing that ended up moving forward. We spent a lot of time thinking about, what does it take to make robots work well, and writing down how ideally we’d want to make this code. We looked around at a bunch of things—we took a lot of inspiration from something that Morgan Quigley had been working on called Switchyard.

STAIR 1 robot built by Andrew Ng and his team at StanfordThe STAIR 1 robot built by Andrew Ng’s group at Stanford as part of the STanford Artificial Intelligence Robot (STAIR) project. The goal was producing a viable home and office assistant robot.Photo: STAIR Project/Stanford

Morgan Quigley: Switchyard was our first attempt at having the STAIR Project work together. STAIR was an effort that Andrew Ng started, even before I got to Stanford. It was in the tradition of these grand university AI and robotics projects that go all the way back to Shakey and Flakey. The idea of STAIR was to unify AI projects into one single robot. There were a lot of different components to the STAIR project—vision, manipulation, navigation… And everyone needed those things to talk to each other on the same robot.

One of the challenges in these big group projects is that by nature, everyone thinks that their part of the project is the most important part. So, what everyone tends to want is for their part of the project to be freely iterating, but everyone else’s stuff needs to stay still and not be a moving target for you to integrate with. The idea of Switchyard, and eventually ROS in general, was that you can have these subcomponents that are changing all the time, but you can version them to hold the rest of the system constant, and that view of the system can be different for every person using it.

Keenan Wyrobek: At the time, the STAIR program had four different robot platforms, and even among those robots they were having trouble sharing code. It was hard to build the tools and create the right abstraction layers to make it all practically work. The STAIR guys were like, “Our dream is to have a robot with all the starter software taken care of, and we can write code on it and share it—it’ll just be easy.”

PR1 wasn’t the original idea. The original idea was let’s make 10 copies of a robot and hire 10 software engineers to build the software, all at Stanford. We thought, “Hey, we’re at Stanford, we should be able to go ask rich people for money and they’ll just write us a check!” A year into asking for money, we had no checks, and that’s when we met the first people who did write us a check [Joanna Hoffman and Alain Rossmann]. They said, “Look guys, we love your enthusiasm, but you have no credibility, and you’re really bad at this.” I still remember that conversation. They wrote us a check for $50k, and said, “Get as much credibility as you can with this money.” And that’s where PR1 came from.

Once we had PR1, we had videos to share, and a story to tell. A gazillion pitches later, we met Scott Hassan. We were trying to convince him to write a check to Stanford. He was just like, “Look, has Stanford ever done anything like this before? Why don’t you come do this at my research lab instead?” I remember Eric and I were terrified—we were like, “Wow, we actually have money to do this now? We can’t screw this up!”

Stanford researchers demonstrate a teleoperated PR1 cleaning a home (and disposing of a broken Roomba) in 2007.Video: Stanford

2. Scott Hassan and Willow Garage

Steve Cousins: I had hired Scott when he was 18, as an NSF Summer Undergraduate Research Assistant at Washington University School of Medicine. He was dirt poor, and had everything he owned in the trunk of his beat up car, and over a period of not that long he went from that to becoming somebody who could afford to fund Willow. I started talking with Scott about Willow as early as 2005. He had been living up in San Francisco; in 2007 he moved down to Palo Alto and bought the building at 68 Willow Road. He brought me by and said, “I’ve got this building and I’ve got the money for 60 people, do you want to join?” And I said, “No, you haven’t told me what you’re going to do yet.” He came back a couple weeks later and said, “Okay, I’ve been thinking about this, and I want to do something around autonomous technology.” At that point, I said sure, I’ll join.

Willow Garage founder Scott HassanWillow Garage founder Scott Hassan at a 2010 event.Photo: Willow Garage

Keenan Wyrobek: There were two very powerful things about Willow. One was Scott’s own motivations and passions. The way he put it to us when we met him, which was one of the reasons we made the leap from Stanford… We were actually both very close to getting our PhDs, which we’ve since never finished… He was like, “Look, I’ve made a lot of money building web companies on an open-source stack,” and he saw creating the Linux of robots as giving back, which really resonated with us. He didn’t want robotics to start with Windows and transition to open source—he wanted open source to be there from the beginning. And of course, he put a lot of cash behind it.

Eric Berger: I definitely had some skepticism about Willow Garage, because it didn’t fit any natural mental model. It was originally pitched as Scott had decided that he would establish a robotics research lab; one of the things that he said was that he had been able to build on top of a ton of the open-source work that made internet companies possible. And in robotics, he wanted to see Linux happen before Windows, to see the open source version be the first version and the best version out of the gate, as opposed to playing a competing catch-up game. That vision was something that I believed, and that was one of the things that we used as a guidepost for where we wanted ROS to go.

“Originally when I got to Willow there were two projects happening. One of them was the autonomous car … The next thing I knew, there were these four people and a car in the cafeteria”

Brian Gerkey: It was a momentous time. I got introduced to Willow when Eric came over to SRI, where I was working at the time, to meet with me and talk about a new place he was working at. He told me about this new lab that was being built, they were going to build this cool robot, and they were going to build all of this really cool open-source robot software. That’s what I’d been doing back when I was in grad school. Eric eventually asked if I wanted to come and help out.

Steve Cousins: Originally when I got to Willow there were two projects happening. One of them was the autonomous car, and the other was the autonomous boat. Before I had completely signed on, Scott went off and hired a self-driving car team, which Melonee Wise was a part of. The next thing I knew, there were these four people and a car in the cafeteria at Willow, and they were feverishly trying to make it work for the DARPA Urban Challenge.

Willow Garage's autonomous boat project

Willow Garage's autonomous car projectWillow Garage’s first two projects: an autonomous boat for scientific data gathering, and an autonomous car, part of an effort to compete in the DARPA Urban Challenge.Photos: Willow Garage

Melonee Wise: I interviewed with Scott in a pizza shop in downtown Palo Alto. Derek King and I were in electrical and mechanical engineering at the University of Illinois at Urbana-Champaign; Team Orange was trying to put a car into the DARPA Urban Challenge, and Scott funded it. Derek and I came out to California to work on the cars.

Brian Gerkey: At first, Willow was meant to be a home for this Urban Challenge team. Along the way, it also picked up an all-electric solar recharged autonomous boat, which was meant to do long-distance voyages for oceanographic data-gathering purposes.

Tully Foote: The car team had tried to hot program the motor controller for the steering motor the night before the DARPA Urban Challenge qualifying event, and ended up with a non-steering car. The car project kept going after that, and they were thinking about going into commercial autonomous cars. But due to connections we had at Google, we knew about the Chauffeur Project, and knew that we did not want to compete with that. When the car project ended, most of the autonomous car team switched to the autonomous boat project, and the design reviews when they joined the team suggested a full restart.

Brian Gerkey: Between the time I interviewed and the time I showed up to start work, the other two projects had been cancelled, and basically the entire organization was on the shoulders of Eric and Keenan and the Personal Robots Program.

ROS and PR2 grasping taskThe combination of ROS and PR2 would emerge as a powerful development platform to demonstrate advances in sensing, navigation, and manipulation.Photo: Evan Ackerman

3. Willow’s Personal Robots Program

Keenan Wyrobek: To be clear, setting the boat and car projects aside was not Eric and my idea. Scott, Eric, and I shared an office, and one day Scott was in the office before us, which was a little bit atypical, and he was like, “Guys, I gotta talk to you about something. I’m shutting down the boat and car programs today.” By then, we had a lot of interest in the Personal Robots Program. We had a bunch of universities and companies who wanted to be a part of it. The snowball was rolling, which in hindsight is why I think Scott made the decision. It was strange for us because our world changed a little bit, but it worked out really well.

Eric Berger: Once we came into Willow Garage, we said, “Okay, let’s try to make hardware that looks like the PR1 but can actually be made in quantities of more than one without relying on a lot of free grad student labor. And let’s try to make software.”

Ken Conley: I was the first software engineer to transfer to the Personal Robots Program. I’d come to Willow Garage because I had worked with Steve before at Xerox PARC. I’d heard about Willow Garage for a while, and it sounded like something really interesting. At Stanford, Morgan Quigley had built something called Switchyard, which was what ran the STAIR robot. What happened with ROS was basically Eric sat down with me and Morgan and said, “When we build this robot, we’re going to need software for it.”

“We brainstormed a bunch of different names. Names like, PACKMOS, RMOS, PRP, ROSIX… it was almost a daily exercise of naming. I think what actually settled it on ROS was that we’d gotten the domain name somehow”

Keenan Wyrobek: It was called ROS way back in the Stanford days, but Eric and I always joked that we wanted to meet somebody who would give it a better name. I think ROS stuck because it was so generic—the name wasn’t associated with individual people or a company or a university. Everybody kind of owns it.

Ken Conley: We brainstormed a lot of different things and went through a lot of prototypes and a bunch of different names. Names like, PACKMOS, RMOS, PRP, ROSIX… it was almost a daily exercise of naming. I think what actually settled it on ROS was that we’d gotten the domain name somehow. People had definitely been talking about a robot operating system for quite some time—I guess we were the ones who took the name.

Eric Berger: We put together our first prototype, which shares way more features with modern-day ROS than I thought it would. It turned out that a bunch of those initial decisions ended up carrying through—the structure of topics, the parameter server, the process model, and the general communication message format ended up staying there. We split up the different packages among different engineers, and every Monday, I would sit down, and first thing in the morning until dinner I’d meet with one developer at a time about their current builds and where we were trying to get to. With the PR1, I’d built a really, really shitty version of everything. I hadn’t done anything right. Well, I’d done some things right. But my role at Willow was as the person who could see how things fit together.

ROS paper from ICRA 2009First page of the paper, “ROS: An Open-Source Robot Operating System,” presented at the 2009 IEEE International Conference on Robotics and Automation in Japan.Image: Stanford/Willow Garage

Morgan Quigley: There were a bunch of people working on ROS pretty early. There were one or two dozen graduate students at Stanford working on STAIR, and we tried from Day 1 to unify all that work, so the ROS prototypes that were driving PR1 around we also had driving multiple Stanford robots around. That’s one of the things that helped the design of ROS, I think: From Day 1, more than one robot was running on the code.

Now that I’m thinking about it, ROS probably would have ended up quite different if we were all in the same building all of the time. By having Willow Garage be really close to Stanford, but still a bike ride away, actually kind of helped. I never actually worked at Willow, I was a Stanford student. I’d bike over and eat their food sometimes. Sometimes Ken would bike to Stanford. It was just enough distance that it forced you to have a little more formality baked into the software to allow you to work on different components in parallel.

Gradually, because more people from Willow started to hack on ROS, the center of mass of people writing code shifted over there until it became almost exclusively where ROS core development was happening.

Willow Garage's PR2 robotWillow’s PR2 robot.Photo: Willow Garage

4. ROS and PR2

Ken Conley: The Personal Robotics Program was super exciting. A lot of what Eric and Keenan demonstrated with the PR1 was definitely very inspiring, and I think it’s a vision that we’re still inspired by in robotics. What was also inspiring was what we wanted to do with the PR2—very early on, we knew we were going to be building a bunch of these robots and giving them away, so the notion that we really wanted to move the field of robotics forward with what we were doing was there from the get-go.

Steve Cousins: Andrew Ng actually brought the idea to us. Andrew said to me, “I’ve got a project I’ve been talking about with some graduate students,” by which he meant Eric and Keenan. “We want to make 10 copies of PR1 and give them away to universities.” Scott was like, “Well, let’s make 20 of them, we’ll keep 10 here and give 10 away.”

Willow’s Milestone 1 for the PR2 consisted of autonomous navigation for pi kilometers for two straight days.Video: Willow Garage

Tully Foote: The high level vision coming into this was that we believed in software reusability. The reason that Eric and Keenan ended up building the PR1 was because they wanted to buy a robot to do research on, but they couldn’t do that, because there were no high quality research robots available. I think that Scott’s perspective was that we were going to take the Personal Robots Program, building both state-of-the-art hardware as well as open source software, with the goal of being a LAMP stack for robotics: You’d be able to take its open-source software, customize it a little bit, put your business model on top, and you’d have a startup.

“We were going to build both state-of-the-art hardware and software, with the goal of being a LAMP stack for robotics: You’d be able to take its open-source software, put your business model on top, and you’d have a startup”

Keenan Wyrobek: I think we were pretty naive in some of the early Stanford days about what it meant to make a platform. We talked to a lot of people—everybody from Woz on the original Mac to folks at Cisco—and in practice none of them made a platform, at least not at the beginning. They all had a very specific use-case and a very specific user in mind. Once we understood what that really meant, we stopped thinking as much about cool personal robotics applications and started thinking about the people who wanted this “platform” today: universities and companies doing R&D. That was a big turning point.

PR2 prototypesPR2 prototypes built in late 2008.Photo: Willow Garage

Eric Berger: PR2 was a hardware platform with capability that way outstripped the current software. That gave the software a lot of room to run. One of the goals with the PR2 when we built it was to support the next five years of software development. From computers that wouldn’t feel outdated in three years, to sensors that were beyond the capabilities of what people really knew how to process at the time, to mechanical performance and robustness… The PR2 was really that catalyst that helped get ROS going. Keenan’s vision for how to build a mechanical system that was well-rounded was one of the things that was really effective there.

From the beginning, we tried hard to make sure we were building something for multiple robots—we were very strict about keeping PR2-specific stuff labeled and isolated. We had a couple of test platforms, to make sure that we were still controlling other robots effectively. Morgan had his mad scientist vibe going on, and he had like 40 different things that ran ROS. He was always building cool things and hooking them up.

We used PR2, and the fact that it had really unique capabilities, to get researchers all around the world bought into ROS. Some of that was getting academics to take software development more seriously, and invest the hours of grad student time into building and releasing better software. Some of it was to convince companies. Bosch had never released open-source software as a company before, I believe, but releasing their code open source was one of the requirements for participation, so they figured out as a company how to release their code open source. That was a big thing that we did—we said, “Here’s the tool, but in order to participate, you need to open source your work, and you need to work together with us to release ROS packages.” That, together with the federated model of ROS, I think was a lot of what helped give it a lot of momentum. And of course the fact that Willow Garage had the resources to put dozens of engineers on building versions of all the basic pieces that didn’t suck.

Milestone 2 consisted of PR2 autonomously navigating the building, opening doors, and entering offices to plug itself into a wall outlet.Video: Willow Garage

5. ROS in the Open

Brian Gerkey: In the early days, we weren’t trying very hard to get people to use ROS, in part because it was nowhere near done yet. Across the board, the early adopters were adopting it without us evening telling them that they should—they just saw it out there, thought it was cool, and picked it up.

Steve Cousins: I went on a trip up to a university in Canada, and they were showing us their lab, and they said, “Oh yeah, we’re using your stuff.” And we were like, “What stuff?” We hadn’t released ROS yet. They were using open-source components from ROS. They had stripped off the ROS wrappers and just used the other pieces, because they needed it, and we had done it, and they’d heard about it, and they were able to use it. We hadn’t released anything. And that kind of impact beyond what we even knew about was pretty powerful. It was feedback that said, hey, we should do this, it’s working.

Melonee Wise: It was actually incredible that anyone [outside of Willow] was using ROS. I remember the first time I used it—I came from the boat project onto the ROS team, and started working on controls and testing for PR2. There was no documentation, and everyone at Willow was so much better at programming than I was. I didn’t have any formal training in computer science, so I was massively frustrated all the time… I found ROS excruciating to use, and I was constantly pushing for documentation. I was a ROS core developer, and half the time I felt like I was the stupidest person in the room.

There was a huge gap in the ROS community. There were the experts—people who had real CS backgrounds—and then there were people like me who were smart, and had a limited exposure to CS, and wanted to do cool things with robots. But reading API documentation was like running into a brick wall. A lot of people point to Milestone 3 as one of the pivotal moments of ROS: Actually creating a set of tutorials that enabled people to get through the abstract concepts and into concrete usage.

Video showcasing different robots using ROS throughout the world.Video: Willow Garage

Brian Gerkey: Milestone 3 was a bunch of user testing and documentation and tutorial writing. We’d written a ton of code, with a ton of functionality in it, but we hadn’t given any thought to who people would come in and use it. We went through a user testing phase, and it was very humbling. We brought in people who were good examples of the user groups that we were trying to hit, and then we’d sit there and watch at how massively frustrated they would get because the documentation was so incredibly insufficient. That’s what spurred us to focus on documentation and tutorials to show people how to do common tasks. It’s not the case today that everything in ROS is as well documented as it should be, but I like to think that that influence is still with us.

Ken Conley: It’s crazy that Willow Garage agreed to let us just put all our code out in the open. Not only did we do open source, we did it for real—a lot of companies will develop their stuff internally, and occasionally put out releases. But one of the things that this anniversary celebrates is that we committed to every commit that we did being in the open, on a publicly available server. I think it’s an amazing thing for a company to decide to do. We basically gave away millions worth of work product, and that’s a big deal.

PR2 robot head at Willow GaragePR2 head on a shelf at Willow.Photo: Evan Ackerman

Steve Cousins: Scott strategically was thinking a lot about, how do we, starting Willow from scratch, have success. The principles were open source focus, and let’s look for the area where we really can lead. Scott had a very radical approach to this—if we’re going to do it, let’s just do it open full time. You could ask him how that was ever going to lead to making money, and he said, “Look, you build all this expertise, you have all the know-how, its value will come.”

You could only do that if a person or group is funding the effort. And that was something that Scott uniquely brought to the table—his belief that open source was amazing. He took it to the extreme… Even the meeting notes. When we finished a meeting, they would hit save on the wiki, and it would be up on the internet where anyone could look at it. In the early days, it didn’t matter that much, because not many people knew about us, but we got a lot of positive feedback early on.

Tully Foote: The big push for adoption was the PR2 Beta Program. That was our drive to get people to pick up ROS and use it heavily. We had this vision for a universal open-source robot operating system, and we wanted to prove to people that it was useful, that you could do good research with it. PR2 was the generic robot that demonstrated how you could do state-of-the-art research in robotics with this common platform, and collaborate with other researchers around the world using open-source software.

Eric Berger, Scott Hassan, Steve Cousins, and Keenan Wyrobeck from Willow GarageFrom left: Eric Berger, Scott Hassan, Steve Cousins, and Keenan Wyrobek celebrate the delivery of Willow’s PR2 Beta Program robots.Photo: Evan Ackerman

6. Beyond Willow

Ken Conley: Originally, we thought that the vehicle for ROS was going to be the PR2, but what happened, and this is really a credit to Melonee and what she did with the Willow Garage internship program, was that because we had so many interns come through from labs that had robots, it was the case that ROS was already running on lots of different lab robots at many different universities before we ever officially released the PR2. And that’s really a credit to how Melonee brought in a huge number of interns every year and really made that program work, and we really got a lot of great roboticists that came through Willow Garage as a result of that program.

Willow’s internship program played a key role in disseminating and popularizing ROS among researchers and roboticists.Video: Willow Garage

Melonee Wise: When we started really working on the Personal Robots Program and on ROS, I went to Steve and said, “The way that you make companies grow, and the way that you get people excited about what you’re doing, is to have a really big internship program. If you want to make sure that people start using ROS, the best way to do it is to bring people here and make them use ROS, and then send them all back to their labs.” I knew this because I’d done a lot of internships, and it was obvious to me how they were successful. Internships were about selling the Kool-Aid.

The internship program made a huge impact on robotics. It wasn’t just about learning stuff; I made a real effort to make their entire time at Willow awesome. We had parties, we went rafting, wine tasting, laser tag … And that’s important, if you want to create this connection. We had jackets. I know it’s stupid, but we had to make sure that when they went back, everyone saw it. People called it the Willow Mafia.

“The internship program made a huge impact on robotics. I made a real effort to make their entire time awesome. We had parties … We had jackets. I know it’s stupid, but we had to make sure that when they went back, everyone saw it. People called it the Willow Mafia”

Brian Gerkey: That first summer, when the interns arrived, I believe there were more interns than full-time employees. Some of them were brought in thinking that they were going to work on a car or a boat, and then there was no car or boat. And then some of them were brought in thinking that they were going to work on a PR2, because in the early days there were some optimistic estimates made about when early versions of PR2 would be ready to work with. When these interns showed up, there weren’t a lot of robots around the place, by which I mean, there were none. Out of necessity, we focused a lot on simulation, because we didn’t have robots yet, but we had designs for robots. I remember having discussions with John [Hsu] and Sachin [Chitta] to pick a simulator, and we picked Gazebo. That was the beginning of Gazebo being closely associated with ROS.

2009 Willow Garage internsWillow’s second class of interns, in 2009.Photo: Willow Garage

Ken Conley: When we first started doing ROS, one of the things that was consistently said was that, “ROS is only for research robots, no one is going to use this in commercial robots.” We always knew at the time that it was more [than that], that we’d have to be patient for that. We knew that no one who had a current robotics company was going to throw away all of their software and switch to ROS. That would be an unrealistic expectation. But, what we were really trying to do was set a foundation for future growth in the robotics industry. I think we did a pretty good job of predicting that, and being ready for it.

Steve Cousins: Willow was this center of excellence. Everybody wanted to come there. We had this incredibly unfair recruiting advantage, and the way that you get from a cold start to the best professors in robotics around the world sending you their best graduate students, was that you’re generous. It was very much that we were giving stuff away to create a fertile ground. You could think of it as another approach to market development, on a big scale.

Willow celebrated the delivery of its PR2 Beta Program robots with a “graduation party” in 2010.Video: Willow Garage

Eric Berger: It was really exciting to see that people were picking up ROS and running with it. The big thing I remember was the PR2 Beta Program workshop. Each of the 11 organizations that was getting a robot came, and by the end of that workshop, seeing the things that people had hooked up to the system, that’s really when I remember that explosion of ROS becoming common in other labs happening.

Morgan Quigley: Willow had a standard physical platform that shipped with ROS already on it, and because that standard physical platform was super awesome and they shipped them all over the world, it got a critical mass going. Without that critical mass, I don’t know what ROS would have become. ROS was a huge group effort. There were just so many people writing code, especially in the beginning when people were just doing stuff, and it was super awesome. It went from nothing to all of this incredible stuff really fast because so many people were contributing code.

Willow Garage PR2 Beta ProgramRecipients of the 11 PR2 Beta Program robots.Photo: Willow Garage

Eric Berger: Brian was the one who brought a lot of maturity about how to make an open source project healthy: What the roles and responsibilities of the contributors and maintainers are; how to build and think about that carefully. He was one of the people who played a huge role in making ROS not just a software project, but a community open-source project.

Brian Gerkey: I don’t think that any of us would have honestly predicted that ROS would get to the scale that it’s at now. And I think it largely happened organically: The only real authority that Willow had to try to dictate the use of ROS was through the PR2, because that’s what it ran, but pretty quickly, the PR2s represented just a small fraction of all the robots that were running ROS. So, everybody else that was using it was using it because for some reason or another they thought it was a good idea. I’m still surprised. I’m less surprised about the use in academia, because that was exactly what we were aiming for: In the early days, when we were trying to make design decisions, we’d ask ourselves, “What would the best grad student in the best robotics lab in the world want?” What I’m still surprised by now is the level of use of ROS in products, and these days, look at its increasing use in automotive applications. That’s certainly not something that we had planned in from the beginning: We were building tools to allow researchers to do science experiments in a lab. That was the original purpose. Without our necessarily realizing it was going to happen, the scope of ROS has gotten much, much grander.

“What I’m still surprised by now is the level of use of ROS in products … Without our necessarily realizing it was going to happen, the scope of ROS has gotten much, much grander”

Ken Conley: The thing that really surprised me was the number of different areas that ROS impacted—the ROS community is much broader than I thought it would be. Originally we thought that it would only impact things that looked like the PR2, but it turns out that ROS has been able to affect the entire robotics industry, which is pretty cool. We were always ambitious to have some impact with it, but I didn’t expect self-driving car companies to be using ROS, I didn’t expect ROS to go onto the International Space Station. I think that was my fanboy moment when I had code running in space.

PR2 writing on whiteboardPR2 writing on a whiteboard.Photo: Evan Ackerman

Melonee Wise: When I joined Willow, I had no idea that it would have such a big impact. I was just excited that I was going to get to work on robots. I never thought it would become as cool as it did. I wish it had become something bigger, and make robots for the real world—I wish it had made the transition between research and product.

Steve Cousins: A proud moment for me was seeing the creation of the Open Source Robotics Foundation. There was a lot of work behind that, but once we had it created, we knew that ROS would live on its own and have a life beyond Willow. We knew that ROS was the thing that was valuable and going to last.

Morgan Quigley: It’s not my project, it’s not Brian’s project, or Ken, or Eric, or Andrew… It’s really a communal effort. It’s a big huge thing that tons of people have done lots of awesome stuff for. ROS isn’t just Open Robotics—it’s this huge effort that’s going on all over the place.

As Morgan says, ROS is much bigger than the people we were able to speak to for this article. If you were involved in ROS over the last decade, please get in touch or leave a comment and let us know what you remember.

The Conversation (0)