In the clip linked, just after the time-point linked to, Lex asks him about autonomous vehicles and LIDAR vs Cameras only. He goes on to say his robots have a harder environment to navigate than cars. Take that as you will.
Taking this with a lot of salt, at least cars are expected in decently defined pathways. Interpreting the activity around the roadways is definitely complicated, but the number of possible paths to take is fairly well defined. It doesn't have to find the walls that restrict it's path. I can see a bit of validity to his thought process. The obvious bit of overlooking the fact the robot doesn't have the ability to kill people does make it look silly
His argument is that a t-shirt can be on the floor and there are stairs etc. But a truck could drop anything on the road, any sort of animal could cross the roads, sink holes could form in roads etc. I feel he is conflating what is often a problem for his robots with what is often a problem for autonomous vehicles. Rather than what each of them must be able to handle irrespecive of how often they happen. Then there is weather etc etc.
