I highly doubt that selecting traffic signs in pictures does anything to help Google's self-driving efforts. The models used in autonomous vehicles are extremely state of the art and object detection is largely a solved problem (many people make object detection apps as Hello World projects).

I'm no expert in this area, but as far as I'm aware object detection at the level required for self-driving vehicles is far from a "solved problem" (and requires significantly lower error rates than a "Hello World" project…).

So, if it is largely a solved problem, the captcha is near useless, no?

Yes, people have used Google Cloud's own public APIs to bypass reCAPTCHA v2.

It helped digitize books... for Google. AFAIK they didn't release it to anyone else?

Also, for a while they used it to determine street numbers for Google Maps.