We'll never get self-driving, at-grade public buses unless and until the technology is commoditized through private investment in self-driving cars.
Here in SF the Geary BRT line has yet to break ground on fully dedicated lanes despite the original timeline showing (IIRC) 2011 as the completion date. After the debacle with the Van Ness BRT, which only opened last year, it seems most politicians and city officials assume Geary BRT is dead. From a technology perspective, and after untold millions, the best the city has been able to accomplish on major bus lines is specialized signaling at a few intersections, saving a couple of minutes on a typically 45+ minute run--a run that takes 20-25 minutes max in a car.
For the past 3 years, since COVID-19, I've been leasing a parking spot downtown in the Financial District and driving to work every day. But I received my Cruise invite last month. I cancelled my parking rental as of July 1 in favor of taking the bus downtown, and either the bus or Cruise home. (I often work late.) Even if a Cruise ride was $20 and I took it every night, I'd still come out even with the price of the parking spot, but currently Cruise is free so I come out way ahead.
The Cruise ride is slow as molasses--about 30-35 minutes, which is the fastest any 38 Geary bus (i.e. 38AX--still suspended--or a very early morning 38R or 38) can do on my route, bus stop to bus stop (i.e. not including walking to the stop or waiting), in even the most optimal conditions. But Cruise is clean and I don't have to worry about standing out on Market St at midnight to catch the 38 Geary, or squeezing onto a packed 1 California--the much slower but "safe" bus. Also, interestingly, at least judging by the few Ubers I took in 2022 and my few Cruise rides so far, Cruise has similar or better availability from the downtown Financial District after midnight. (Years ago I often frequently caught a traditional cab at the taxi stand in front of the BofA building, as until about 2AM I could reliably walk up and jump into a waiting cab. But by 2020 Uber had killed that taxi stand, and presumably all others. So ironically it takes longer to catch a late-night, middle-of-the-week cab in the FiDi now than it did 10 years ago.)
Anyhow, my point is that self-driving cars seem like the only realistic path to better transit in San Francisco, if not the entire United States. San Francisco (and most American cities) is demonstrably incapable of adequately improving public transit by building fixed infrastructure. The way forward is almost certainly through automation; SFMTA labor costs already exceed capital expenditures, and the gap is growing. Despite self-driving cars being the worst option imaginable when judged in isolation, they're the key toward realizing the best option--better intracity mass transit.