It should be able to program a self-driving car with a) a database of known intersections and their conditions as a fallback and b) to evaluate the conditions of an upcoming intersection for safety even without the requisite signage.
Maybe it'd run it if there's no one else around, but a human probably would too in that case.
Another imperfect analogy is power outages and stop lights. For the most part, it doesn't result in full-speed collisions; instead, folks treat it as a stop sign.
Self-driving cars are going to need a "huh, that doesn't look right..." system. Detecting missing/defaced/misleading signs should be a part of it.
That's what you're supposed to do. In some cases, the traffic lights fail to a mode where one direction is blinking red, which is treated like a stop sign, and the other is blinking yellow, which is treated sort of like a yield sign (i.e., proceed, but with caution).
I believe the law (and common sense) states that if the traffic light it out, you treat it like a 4-way stop sign intersection.
If I see rocks I can e.g. drive around them, judge if they're really too big for my car, drive on the sidewalk, etc.. => all these options are really complicated to program.
Current cars don't stop, at least not in Europe - they might break/go slower but it's going to be mixed with steering to avoid whoever is there causing all that fuzz. And if I think that whoever is in front of me is a threat that I might just run over him/her (or "it" for "animal" - a different kind of complexity - e.g. do you stop for 1 frog? 10 frogs? 100 frogs?) , at least partially.
As opposed to placing some rocks on the street like they do now?
> why should s/he wait to cross the street if s/he knows that the car that is nearing has an anti-collision/auto-break system?
Why do them wait now? Current cars stop too.