I think there's definitely something to be said for constraining the scope of the problem in order to achieve usefulness quicker. But constraining scope also limits opportunity, and what the big names in the self-driving space (Waymo, Uber, Tesla) really want is to be the first to claim their slice of the pie, because whoever gets there first gets the largest slice.
But full self-driving is such a complex problem that it seems to suffer from Zeno's paradox: the effort required to solve it scales non-linearly as we get closer to a solution, so the closer we get the slower we go, and the solution never quite seems to be within reach.
I'm skeptical that non-constrained full self-driving can be achieved within the next ten or even twenty years. But I'd love to be wrong!