In less urban areas robot vans for transit might make more sense than in urban areas. In San Francisco very few bus lines don't see full busses at several points during the day. This applies not just to the busses that run between neighborhoods and downtown along Market street, but also the many north/south arterial lines: 19th Ave in the Sunset, Divisadero and Castro Sts, Fillmore St, Van Ness Ave., the 43 between the Presideo and the south edge of SF (through my neighborhood in the Haight), and north from the Caltrain station.
It might seem to make sense in urban areas from one standpoint to introduce airport shuttle-style vans to some lower ridership "neighborhood access" routes. However, from another standpoint those vehicles can't be "load balanced" if necessary onto the other routes that make up the vast majority of the overall vehicle trips. Plus they can't be easily swapped due to maintenence considerations. I'm sure the flexibility of keeping the entire vehicle fleet as full size busses in an urban area makes up for the slight loss of efficiency in running a few busses that have oversized capacity for the routes they're on.
As for your low cost predictions for robotaxis, it's a very rosy picture, but I just can't buy in to that.
Besides, heavy users of transit get monthly passes, which drop the per-mile rider cost for transit way down. It's unlikely there'd be the same possible kinds of fare savings for robotaxis.
There are also huge road capacity considerations if robotaxis take a lot of riders away from transit: Imagine even half of the passengers of the current crowded busses during peak demand switching to robotaxis. An SF Muni bus at peak can easily be carrying 60+ or 100+ people (offical standard bus capacity is 83 passengers, or 124 in articulated busses). Putting just half of that into robotaxis means that that one vehicle on the road turns into 30+ or 50+.
It's highly questionable whether there'd even be nearly enough robotaxis available to serve so many people at times of peak capacity.
Then, what's the efficiency difference at off peak between one bus taking a route with maybe only 5 passengers versus those 5 people in individual robotaxis and 25 more robotaxis sitting idle somewhere waiting for the next period of peak capacity?
So it's not just pricing concerns that's going to keep transit systems around and worthwhile for more than "only the very poor" in and between urban areas.
Of course if robotaxis use variable, demand-based pricing then their costs will also go up significantly during peak hours, which is when a higher percentage of lower-income passengers will be travelling. Peak pricing on transit systems is uncommon, and its variability tends to be low (e.g. the London Underground has peak pricing which is only in effect on weekday mornings before 9:30am).
Further, the larger the amount of sprawl, as you suspect will continue and/or increase, the more attractive dedicated right-of-way transit (surface and subway trains, plus some bus rapid transit) is versus robotaxis to get into urban areas, for both cost and speed reasons. It will continue to be significantly cheaper per mile over longer trips, and faster due to avoiding surface streets and highway(/bridge/tunnel) bottlenecks.
The same road capacity and vehicle availability reasons also reduce the viability of using robocars for last-mile travel inside a city for people who use first-mile robocar-to-rapid transit. Imagine the demand and local traffic spikes at transit stations as a train arrives and a significant fraction of its passengers all make requests for robotaxis at the same time.
So, transit for urban intra-city passengers is still not just highly desirable but actually essential for both road capacity and vehicle availability. For travel into and between cities it would appear that again for the same road capacity and vehicle availability reasons taking a robocar for last-mile travel inside a city has more drawbacks than I first thought. The likliest scenario for commuting into a city then appears to be first-mile robocar to rapid transit, and then robobusses and walking plus a smaller amount of robotaxis for last-mile point-to-point travel inside the city. During the day robotaxis then fill in for any need of having a personal vehicle. On the return trip overall demand might be spread out enough for robotaxis to be available enough and cheap enough for a first mile return trip to a transit station. Then it's rapid transit again and a robocar from the local station near your house to home.