Musk's self-driving taxi idea also isn't new. It's essentially the same thing as Personal Rapid Transit [1] which was originally envisioned back in the 1950s as a way to solve real world public transit system issues:
>Most mass transit systems move people in groups over scheduled routes. This has inherent inefficiencies. For passengers, time is wasted by waiting for the next vehicle to arrive, indirect routes to their destination, stopping for passengers with other destinations, and often confusing or inconsistent schedules. Slowing and accelerating large weights can undermine public transport's benefit to the environment while slowing other traffic.
>Personal rapid transit systems attempt to eliminate these wastes by moving small groups nonstop in automated vehicles on fixed tracks. Passengers can ideally board a pod immediately upon arriving at a station, and can – with a sufficiently extensive network of tracks – take relatively direct routes to their destination without stops
You have to have extremely short bus headways before the road space efficiency tilts back towards its favor.
(1) This is a bad metric because it means everything is equivalent until traffic jam occurs. (Why stop at 20 cars? How about 2,000 cars?)
(2) More importantly, everything cannot be equivalent because traffic jams do occur in any respectably-sized city. Having 20 cars instead of one bus absolutely does contribute to having more frequent and severe traffic jams.
We stop at 20 cars in 10 minutes because that's the equivalent of the bus. If the bus ran twice as often (i.e. 5 minute headways), we'd have 40 cars in 10 minutes, but that's only one car every 15 seconds which is far from a congested state, especially with no traffic signals (like a rail/guideway based PRT or a hypothetical future self-driving system where cars coordinate with each other).
PRT is essentially just a bus system with a large number of small buses and a large number of routes as opposed to a traditional bus system with a small number of large buses and a small number of routes.
Not really. In the case of a guideway-based system, the minimum headway is determined by the time it takes to a) slow down to a safe diverging speed for the switch, b) physically pass over the switch at that speed and clear it, c) physically reset the switch to a different direction, and d) the following car has to be have sufficient distance to be able to stop in case the switch faults during step c. And you generally want to have some padding in there so that you can recover the schedule in case things go wrong. (Or you can dispense with switches entirely, in which case it's ruled by the amount of time vehicles have to dwell at a station for passengers to embark and disembark. Which, yeah, 15s is going to be too tight for that.)
The idea of using smart technology to improve capacity of roads by doing with safety margins is horrifying to me because a) the technology is really nowhere near that, b) the design is fundamentally hostile to pedestrians and cyclists and quite frankly anyone other than robodrivers, c) you still have the problem in urban environments that there's no room for grade-separated junctions, so the possible capacity gains are sharply limited, and d) safety margins exist for a reason! Just ask a certain submersible CEO (for whom you'll need a medium, for he recently stopped being biology and started being physics).
I think you're underestimating the capacity of a bus system significantly. I live in Dublin, which is a city of about 1.5 million people, not super-dense, hardly the world's greatest metropolis. The double-decker buses used almost exclusively in the Dublin transport system hold 90 people (from time to time Dublin bus experiments with bendy buses, but they tend to get stuck on roundabouts). Let's say the average car holds 1.5 people (I think this is generous, it's probably closer to one, but let's go with it). So a bus is equiv to 60 cars.
I can think of a few streets nearby which have stops for ~20 bus routes, with other routes running through them but not stopping. Some will be every-20-minutes routes, but some will be closer to every-five-minutes at peak. At peak time, there will be multiple buses per minute on many streets. You will see buses queuing up behind each other to load and unload. There is _no way_ that the equiv capacity in cars could even _fit_ on the road, even if traffic wasn't a concern.
These, by the way, are normal bus routes, not BRT; that would be even worse.
Here's the practical equivalent of 240 cars embodied in buses (plus 272 in the tram): https://s3.eu-west-1.amazonaws.com/prod-mh-ireland/607c5cf1-... - It's hard to see how you'd fit those cars in the equiv space. At peak times, all of those buses will be full.
The OP's main point is that in dense urban areas, replacing high-capacity vehicles with small-capacity ones presents a geometric challenge, because the surface area covered by streets is fixed. Were we to replace every bus carrying an average of, say, 40 people, with 20 small self-driving vehicles, each carrying an average of two people, the street surface area occupied by the small vehicles will be significantly greater and traffic will slow down to a crawl. Imagine a gazillion small vehicles stopping everywhere to pick up and drop off passengers. As the OP correctly points out, finding ways to use space efficiently is the core problem of urban transportation.