Assuming empty airspace, it's obvious why drones are easier to program. Assuming a single package it's obvious why drones are more efficient. Take away either of those assumptions and things get less obvious.

Those are very wrong assumptions. If a serious drone delivery program (or multiple competing ones) starts, airspace will suddenly become congested and you'll have to do heavy air traffic coordination to avoid mid-air collisions. And when you comapre with cars, you need to compare one delivery car with a hundred drones. Suddenly you see that drones are not efficient at all, which is kind of obvious, as it takes a lot of energy to keep stuff in the air with active propulsion.

I always pictured trucks driving from distribution centers, and deploying drones that cover an area. That way the drones only have to cover a small distance, reducing the number of drones that have to fly over a given area.

Mobile drone hubs seems like a very interesting solution, which could even be more energy-efficient than just trucks or just drones. Somehow I never thought of that, thanks!

Airspace is big. A central registry where drones announce their flight plans, allowing other drones to selfishly avoid them, would allow for plenty of room for drones to make sure they stay out of each other's way.

I'm skeptical on drones, and I think all this is an effort by Amazon to get in the news. But this doesn't seem like the problem.

Sort of, but as long as all drones (in this airspace) are automated, it's still an easier problem than self-driving cars. Just detecting where a road is is a feat, and then throw in other traffic, and then say that traffic is driven by people, who tend to behave irrationally by a computer's/programmer's typical definitions of "rational". Then there are traffic controls, construction, and the like.

Amazon's idea is to have airspace dedicated to autonomous drones, who can go from a to b in a straight line and mostly just have to worry about not hitting other drones (or buildings or birds or what have you). You can do that with some decent sensors and some simple physics simulations.

That's not to say that airspace is totally unregulated, but the regulations tend to be much broader (AFAIK; I'm not a pilot).

Collision avoidance is far easier in three dimensions than in two. Those extra degrees of freedom are crucial. (Of course, traffic capacity is much higher.)

It is only the severity of an accident that increases, not the likeliness.

Except that they don't need to do the R&D on the cars.