Mathematically, on demand vehicles are probably harder to schedule because of random variation in queuing. Consider a population of workers each of which sometimes works from home. Some days an unusually high, but not statistically unlikely, number of riders all stay home: the idling vehicles need to sit somewhere. Other days, a statistically likely but atypical number of riders don't work from home and there are not enough vehicles to go around.

Public transit addresses fluctuation with a low nominal cost of providing surplus capacity to normal operation in the form of standing room. Public transit can be sized more closely to peak demand. On the other hand, from 20:00 to 06:00 the peak demand capacity of an autonomous car fleet needs to be parked somewhere. Between 06:00 and 08:00 and again between 18:00 and 20:00 the fleet needs to be moved to storage.

To put it another way, a premise that autonomous cars will work in places like Southern California is that an autonomous car will be at your doorstep whenever you need to leave. The ideal fleet to achieve this is one car dedicated to only serving your schedule. Anything else has all the drawbacks of party line telephone service.