Is ROS 2 stable and ready for production use?

edit

It depends on what parts of ROS2 you need for your system. You'll only discover these things by trying to build a system and seeing where it falls apart. Here are a couple of pain points for _me_ right now:

ROS2 Foxy (LTS) depends on Ubuntu 20.04. Nvidia only supports Cuda 11 on Ubuntu 20.04. Pytorch doesn't yet support Cuda 11. So if you are working on AI-stuff and need pytorch, you cannot use the latest version of ROS2 yet.

If you want to use a micro-service model with a swarm of docker containers across multiple hosts, 'macvlan' network is your only option which will expose every container directly on the network.

I find that the ros2 node discovery is not super reliable, sometimes not all topics show up in ros2 topic list or fail to echo and require a couple attempts to get it working. (but it does usually work with enough attempts)

My current frustration is that the bag recording utility fails silently when recording larger sensor data topics, such as lidar and raw camera data. I mean that it will start recording but after a few minutes it will stop recording to disk but not display any error messages. I have to watch the filesize of the bag file to know when it stops recording and then restart recording.

I've put my eggs in the ROS2 basket and am trying to fix issues and send pull requests to fix ROS2 for everyone. If that sounds like your kind of fun, it's ready!

I'm not really trying to self-advertise a keynote talk, but this is a good reference. I go over the current status of ROS2 and some of the options you have. I think this talk would largely answer your question or give you enough context to make your own decision.

https://www.robobusiness.com/the-life...

tl;dr: Yes-ish. Yes it can be used, but yes, it will take some effort of your end.