The rise of the Robot Operating System

Having your work covered in a documentary like How to Start a Robot Revolution — a five-part documentary in the Open Source Stories series from Red Hat — is bound to make you feel old. You look back and think, "wow, I've been doing this for a long time." Indeed some of us have been working on Robot Operating System (ROS) now for well over a decade, far exceeding the traditional Silicon Valley two-year cycle of jumping to the next thing.

Personally, the story in the film is just the latest chapter in an even longer journey. As a computer engineering undergrad at Tulane University in the mid-1990s, I met a new professor in the department named Jim Jennings. Like the other faculty, Jennings had a lab, but unlike them, his lab had robots. There were three RWI B14 robots, and students were welcome to program them. I was immediately hooked by the experience of writing code that made things move in the world.

An early project in that lab, as in pretty much every lab of that era, was to write software that we could use as an interface to the robots. We needed something on the robot that would sit above Linux and provide access to the sensors and actuators in a way that would allow us to write our robot control code quickly and easily. Because it was Jennings's prerogative, and because he had studied in a particular lab at Cornell, we did our robot programming in Scheme. So we had to somehow bridge from the C libraries of the OS and device drivers to Scheme. 

We could have directly linked the underlying C libraries into Scheme, but then we'd be limited to having that single program controlling the robot. A better idea was to write a server (in C) that would manage access to all the robot's peripherals and offer a socket that allowed clients (in Scheme, or pretty much any language) to control the robot, either locally or across the network. The result, named TRIP (for Tulane Robot Interface Process, I think), was an example of what today we call robot middleware. Two aspects of TRIP stuck with me: the importance and influence of the infrastructure that you use when programming robots, and the pleasure of collaboratively developing such tools in a team.

A few years later I was in Maja Matarić's lab as a grad student at USC. We quickly ran into the same problem: we wanted to do experiments with the robots (by this time a fleet of ActivMedia Pioneers), but first we needed something that would manage access to a robot's sensors and actuators and allow us to write our control code against a comfortable API. Strongly inspired by TRIP, we built a multi-language client-server system that we called Player. But we also took a significant step beyond TRIP: we made Player an open source project. Because we were idealistic academics and because it was the early 2000s, we attached the GNU GPL and pushed the code to SourceForge.

That step of first releasing the code is always a bit like screaming into the void, and it was even more so back then. This was before today’s ease of sharing and searching for such things online. You don't know whether anyone will notice, or care, and of course you're feeling self-conscious about all the things that you wish you had done more efficiently or elegantly (or maybe just correctly) before letting other people see it. 

But we released Player anyway, and to our great surprise, we got feedback quickly. Other labs with Pioneer robots started using Player, and before long we were receiving patches from people who had modified the code to support other types of robots in their labs. From that point onward, I was hacking on Player pretty much full-time, devoting just enough effort along the way to my classes and research to eventually graduate.

The Birth of ROS

Fast-forward another few years, when I was a research scientist at SRI, and Eric Berger asked whether I was interested in joining this new company called Willow Garage. The pitch was that we would build a new robot, plus a complete software infrastructure for it, and make all the code open source. Naturally I leapt at the opportunity, and managed to attach myself to what became the ROS project. The rest, as they say, is history, and is explained from multiple perspectives in How to Start a Robot Revolution.

We are now beginning year thirteen of ROS, but as with many open source initiatives, when did it really begin? We know that the first commit was made to the ROS project on SourceForge on November 7, 2007. We also know the first paper on ROS was presented at ICRA in May 2009, and that ROS 1.0 was released in January 2010. But the seeds of ROS were planted much earlier. Mine is just one story. Many others have found their own path to ROS and to related projects.

I'd like to say that I had a plan all this time, that I was always intending to make a career out of working on open source robot software. But looking back I have to conclude that I benefited from a series of happy accidents in the form of colleagues, collaborators, and bosses who were willing to indulge my passion for this once esoteric topic. In return I can only hope that I'm able to help support like-minded open source tool-builders.

For more information on ROS, please visit Open Robotics or the Robot Operating System site. To watch the full five-part documentary “How to Start a Robot Revolution,” visit Open Source Stories, and join the conversation on social media using the hashtag #opensourcestories.