What are technical reasons for criticisms of ROS's Reliability/Robustness/Safety?

edit

To expand on the previous answer a bit: the industrial, automotive and aerospace communities have standards for development of safety-critical software. (IEC 61508, ISO-26262, DO-178, MISRA and others).

When writing software for a safety-critical application in one of these fields all software and libraries that are used should be certified to the standard that is appropriate for that field. The library authors for safety certified libraries have substantial documentation which shows that the library was certified to the appropriate standard. (Creating this documentation is expensive, which is why library authors usually charge high licensing fees)

ROS 1 was not developed to any of these standards, and does not have any of the documentation that could show that it is suitable for use in a safety critical system. Further, ROS 1 depends on many libraries which were not developed to these standards either. This would make getting a safety certification for ROS 1 rather difficult.

ROS 2 is based on DDS, and there are some companies which sell safety certified versions of DDS. A safety-certified version of ROS 2 is therefore possible. I believe Apex.AI is working on a version of ROS 2 that is or will be certified to ISO 26262 for automotive applications.

Those who make claims are responsible to provide evidence. If you claim any safety attributes for ROS please provide evidence. The lack of any systemic safety evaluation, lack of a functional hazard assessment, lack of evidence of any functional safety integrity level (SIL) as in IEC 61508 family, makes the ROS unfit for use in safety critical systems.