I agree though - if your cities are already gridlocked, this won't make them somehow less gridlocked. You still need mass transit, cycling and walking no matter who's driving the cars.
I disagree. Human drivers use roads very inefficiently: they overreact, have high latency and drive overly selfish.
Computers can drive cars closer together, can maintain higher speed with the same safety margins and without causing traffic jams by overbreaking, and they have no problems obeying traffic laws (and thus won't cause a literal gridlock).
The only reason why self-driving cars won't notably reduce road utilisation is because of induced demand
So either the traffic patterns allow pickup to be optimized to do much better than doubling the car-miles, or traffic patterns are so that it doesn't matter.
(of course that's overly simplistic, but I don't see how it's a less valid simplification than yours)
The important case of commuting toward the city is interesting. Car-sharing breaks completely there - everybody starts work at about the same time (commute traffic confirms this). So we'll need as many shared cars as we already have. No savings at all.
Any savings comes down to off-commute traffic. Its a second-order effect. Maybe even too insignificant to bother with the complexity of sharing.
If people are going from everywhere to everywhere, there will be almost no extra cars.
Yeah, when all directions are congested it will add a lot of very short trips.
Something interesting to think about is logistics optimization. I have nothing against the big brown trucks but it should somehow be possible to haul my amazon packages from distribution center in the city by the railroad tracks to the burbs. Or distro center to distro center package movement.
Author here. You are completely right, I didn't include self-driving time in the second table. To correct my omission, let's add a column for 3X utilization where self-driving cars spend half of the time driving alone.
Shared self-driving car (3x)
-----------------------------
Used: 3 x 5% x 2 = 30%
Parked: 70%
Number of cars in the city: N/3
Parking places needed: (N/3) x 70% = N x 23.3%
Parking reduction: (N x 95%) / (N x 23.3%) = 4.07x
-----------------------------
So, in that theoretical case, we would need 4x less parking. However, this is just a rough calculation, for better precision you would use some traffic simulation application.
Thanks for feedback!
Of course with proper scheduling, this worst-case could be chipped away to get better numbers. The best case is nearly zero deadheading (with likelihood of nearly zero). Some kind of curve could be constructed? I'd love to see that too!
In many European cities, parking area could be converted into additional lanes by removing a few marker lines. Reduced demand for parking area can increase throughput too, in these cases.
Also, any savings in parking needed is going to be eroded by road congestion - in a perfect scenariao cars are all going to be moving around most of the time. That seems like a much worse problem at present - roads are often at peak usage, while reducing parking is only a cost-saving feature.