You could put self-driving tech into one of two categories:
1) So hard, even with incremental progress, we can't predict the problem will be solved even in a decade or two, e.g. Nuclear fusion, General AI (AGI).
2) Hard, but the progress over decades has been notable, implying convergence in a decade or two. Voice recognition and Natural language processing seem to fit this category.
I'd put self-driving in category 2, but I can see why hubris around projects like nuclear fusion and AGI might make you think this is of the same type.