Putting the soap box aside for just a moment however, I could see this being feasible for static bus routes, assuming your city maintains the roads and signals along that route so the ML algorithms piloting the multi-ton death machine on open roads isn’t distracted by a new pot hole or missing road stripe. Since no city in America actually does this though, these things seem about as reliable as the “self-driving” Teslas under Vegas. Then you throw in the need to retrain the algorithm every time the route changes, and the millions in expenses that’ll cost versus the few thousand it takes to retrain a human driver on a new route…
Yeah, nevermind, breaking out the soap box again. Stop trying to short-change infrastructure investments and just invest in the infrastructure already. It won’t make some SV VC fund richer, sure, but it’ll at least go towards something useful and necessary for the city’s occupants.