Even that paper demonstrates that if you weaken their assumptions about taxi-car occupancy being exactly the same as Uber/Lyft and don't assume that you will have a remote monitor per every 10 cars (unsure why they would include this assumption, even the current Waymo setup doesn't have this), self-driving actually does come out as priced less.

They also don't source their cost for CDV, which is substantially lower than any other number I can find cited in paper. Further, the current regulatory framework includes huge wealth transfers/incentives for car ownership, so it's not a fair comparison. For instance, the $130/yr cost of a parking permit in SF is a huge wealth transfer to car owners as that is substantially below what the market-price would be for that permit.