The core problem with these ideas, IMO, is that you are simply replacing a car with a different car. My cost example, as low as it was, was my most expensive per-mile car. We have a BEV that we use for the majority of our driving and the incremental cost of keeping it around is trivial. Even when Uber was subsidized by investor cash and by drivers not understanding how much their (required) newer cars were depreciating and wearing, they were much more expensive than just driving our own. And for me and a lot of others, cost isn't the primary factor--as I said, you could triple my costs and it wouldn't affect my decision nor that of any of my neighbors AFAIK.
So to sum it all up: An SDC service as opposed to a privately owned car (which, b/t/w, could also be an SDC if they ever actually produce one) will always end up driving more miles simply because there is always at least some deadheading. Because of that and the nature of a for-profit operation where there is always at least some overhead, staff, etc, the SDC service will always have a higher per-mile cost unless you try to add back in things like the cost of parking at home. Thus your viable cases for an SDC service are where parking is unavailable or prohibitively expensive (on an incremental basis, not paid-in like mine) or the paid-in parking owner can somehow recover value from the unused parking and there are not good public transportation options already and the users don't mind other aspects of shared usage, like not being able to keep their beach chairs in the trunk.
We can't really predict human behaviour here, there's a big attraction to cars as an aspect of class. But I don't see the per-mile cost as being more. A typical private car spends 95% of its time parked up, slowly deprecating. An SDC can use a lot more of that time, say, pretty much all the time between 6am and midnight, with maybe one or two charging stops depending on mileage. Deadheading would assume that the car returns to the same location to pick up passengers, e.g. going between a city and an airport. But why would an SDC not just pick up the next car-passenger pair? If cars are less common anywhere, there will be a journey. It's just taken you to work, now it picks up Betty to go to the shops, and collects Dave from the shops to go back home, and then takes John to the airport... The connectivity is complete with very limited void periods; and if there are any void periods, these can be used to charge, for cleaning, or just idling (but at very little cost, as the driver's not hourly!)
The only reason (imo) Uber is as expensive as it is, is because there are limited numbers of drivers who need to be paid a wage. When you change that to an allocation of SDC vehicles then it becomes feasible to offer the service at a much lower cost.
Think about it this way, if say the average CA car is $40,000 and lasts 15 years and does the average of 12k miles per year, then per mile depreciation alone is 22 cents per mile (that's more than nct's quoted cost on depreciation alone, but this is a new car.) Plus you need parking, insurance, fuel, maintenance, have breakdowns and, if you really want to look at it from an economists lens, there's the opportunity-cost of buying that $40k car over putting that in your pension or going on holiday more often. Even if you don't allocate your driveway's cost, what about the places you go? Even if the parking is free, someone is paying for that, and that *is* reflected in whatever you buy. That could be better allocated if cars didn't need to be stationary for so much of their lives.
OK, so what if you buy a car half-way through its life but because it's a little undesirable it only costs $10k? The maths on depreciation looks better, as you buy at the more linear part of the deprecation curve, but you then have the unknown of prior service history (I've bought a car before that later transpired to have mostly fraudulent history.) You have to spend e.g. that $10k twice as often because the car lasts half as long on average, and you have potentially higher risks of getting a proper lemon (like I did!) Of course, the used car market is funny now, and demand has shifted more into used cars, raising their price. It's still probably cheaper, but harder to work out. Still, about 40% of cars on the road are leased in the UK, suggesting that there's a strong bias towards 4-year and newer cars.
An SDC may, say, with sensors and car computer cost, say, $50k (I figure you save a lot of money by making it not compatible for human drivers), but it could do 500k miles before being scrapped. (I think this is feasible with general maintenance; age is worse than mileage, in general.) Half the cost per mile, but with NONE of the other mentioned costs, just "tap on an app and go". Instead of spending 95% of its time parked up, the SDC is spending only 33% of its time parked up. It's doing perhaps 13x more mileage per year, so it reaches its mileage limit in about 3 years. At which point, it may get a new battery pack, or it might be allocated into different service e.g. shorter routes, or be recycled. Since the parts only need to last, say up to 5 years at a time, it may be possible to optimise them for higher mileage instead of age. Certainly, it's clear that buses can do many years of service with just general maintenance.