Quite the qualification for self-driving: "as long as there are clearly painted lane lines."
The endless list of "exception" cases is why I'm continuous skeptical about any sort of full-self-driving claims.
Sure, you can cover everything they can think of. But there are so many cases you can't predict, or which don't have an obvious solution, and it often comes down to a human judgement call that doesn't always have a programmatically-clear right answer.
I notice that tesla seems to choose (highlight in blue) the left and/or right lane lines or the car in front of you. But it still seems to work without lines.
That was specifically for their existing Gen2 highway assist expansion. Not the Gen3 custom silicon full autonomy that they were discussing for the rest of the presentation.
+3.5 million miles out of ~4.x million miles of existing roads -- not too shabby.
Seems like a good start to me, and I'd rather they approach as cautiously as possible.
Fully autonomous driving in all conditions in all locations is clearly a hard problem that's still not solved.
I'm very happy with any company that clearly spells out the situations where their tech works, and the situations where it doesn't yet work.