I can imagine this being at least two orders of magnitude higher than most human drivers (one in ten thousand misses versus one in one hundred). How many orders of magnitude better does it need to be before it's good enough?
Missing one in a hundred intersections? That's insane. Most human drivers are much better than that and wouldn't keep their license (or their life) for long if they weren't.
My 99.99% was just a number for the sake of argument. The real accuracy of classification networks is usually in the range of 90-99%. It's not bad for a lot of tasks, but for self-driving I'd say it's far from sufficient.
