I’d like to ask about how well ROS 2 scales when managing a large fleet of mobile robots—specifically fleets with over 1,000 robots. I’ve used ROS 2 in many projects, but typically only with one or two robots.
When working with 1,000+ robots, each with around 10 hours of daily uptime, the chances of hitting edge cases or rare bugs naturally increase (as was the case with ROS 1).
I’m really interested in hearing about your experiences. What’s the size of your fleet? Are you using community-supported packages like ros2_control or others not maintained by the core OSRF team?
Are you using community-supported packages like ros2_control or others not maintained by the core OSRF team?
In my experience, medium to large robotics companies (say 50+ deployed robots) will typically only use community ROS packages that they feel equipped to fork and maintain themselves – with few exceptions, such as nav2. This is actually a risk-mitigation strategy supported by this paper a few years back which pointed out how poorly maintained many community packages are: It Takes a Village to Build a Robot: An Empirical Study of The ROS Ecosystem
I believe the essence of the problem is the lack of a business model for open-source packages. Many community ROS packages are the result of a PhD student’s thesis work, and that’s great, but once they graduate they often don’t have the time to maintain it. So unless there is a commercial entity behind the package, i.e., developing and using that package for their own purposes, you are taking a risk when you build your robotics company on it, and that risk needs to be mitigated.
Steve at Open Navigation and me at Transitive Robotics are pursuing two different, possible pathways for solving this problem, i.e., doing open-source development in a (financially) sustainable way. But there are also other commercially successful companies that contribute back to ROS, and I believe Kiwi is one of them (please keep me honest, @charlielito).