What about machines whose behaviour comes from a learned model?

What is checked

What DatumProof checks is what is written in the design: the states, the sequences and the conditions under which interlocks engage. The questions are of the form “is this state reachable?” — for example, whether any sequence exists in which two axes enter a hazardous zone at the same time.

What is not checked

A learned model is not a design document of that kind. DatumProof does not check a learned model or predict what it will decide.

Where the two meet

A machine driven by a learned model is still a machine that someone designed: it has states, sequences and interlocks that were decided when it was built. Those are the part that can be written down and asked about.

Whether that is useful for a particular machine depends on the machine. It is decided one machine at a time, when the scope of an early-adoption program is agreed in writing.

Back to the question list.