It's a big step to go from interstate autopilots to having no human in the loop end-to-end. If autopilot systems improve safety and/or are autonomous enough that drivers can sleep and thereby make fewer stops, they'll certainly make sense. However, it may also make economic sense to keep a human on board to handle unanticipated issues or last mile tasks that may be harder to automate.
If the driver can sleep, than the car needs to handle everything that can happen on the interstate, including deer or people on the road. So I would guess that a interstate autopilot has the same difficult challenges, but 'simpler' things like navigation may be easier.
Basically my argument boils down to, it is easier to drive on an interstate on average, but for autonomous vehicles we are not interested in average and the hard challenges are similar to hard challenges for a fully autonomous car.
>If the driver can sleep, than the car needs to handle everything that can happen on the interstate, including deer or people on the road.
Interstates are less complicated overall. You don't get a lot of jaywalkers or bicycles doing unexpected things. Nor negotiation of left-hand turns, etc. What you do have is events happening at higher speeds but handling the car in front jamming on its brakes is probably easier in general for a machine that's always paying perfect attention. As a result, I do expect to see autonomous systems on limited access highways before I see them in a busy city.
That said, I fully agree that once you have full automation you can't depend on human backup that can take over quickly. Whether or not people are supposed to, they will doze off, read, text, etc.--even more than they do today.
