I suspect the real problem will turn out to be extremely deep and difficult. That human drivers drive not only with great pattern recognition around shapes, but

I suspect the real problem will turn out to be extremely deep and difficult. That human drivers drive not only with great pattern recognition around shapes, but also with an actual understanding of the idea of a truck and the idea of a tunnel and the idea of a sign. There is a substantial chance that will turn out full self-driving is a fully general, human level AI problem. That said, even the existing partial self-driving is extremely useful. I would be hard-pressed to consider another brand of vehicle until the others catch up with this.