You are comparing with the least efficient form of transit, and looking at cities that are not really designed for transit anyway. So yes, bad transit in cities not designed for it isn't that efficient. Note, on top of this, that you are purely looking at energy efficiency, when the key problem of individual robotaxis is space efficiency.
Realistically, what breaks efficiencies in the US is city design: Places of work mostly detached from activities and housing, pretty much guaranteeing inefficiency. The robotaxis don't make a difference here: The max number of people looking to travel at once doesn't change, and neither does their destination. This makes the number of vehicles actually on the road not change very much at all: The best you can do is hope for people being comfortable (and not inconvenienced) by carpooling with strangers, not unlike what lyft and uber offer already. In the middle of the day, there's no new demand for robotaxis: They'd be parked somewhere, and every mile they move is a car/mile of congestion they create. So you either park the cars where the jobs are, getting, in practice, the same results as private vehicles, or you park them somewhere else, increasing total congestion. It's not just robotaxis that would park: See what happens in all the train commuter lines in American cities, where most trains just get parked in a yard.
The miracle of cities with top transit (See, for instance, Madrid), is that a relatively high percentage of the network is about equally busy in both directions. Trains might stop in rush hour, but route 6 will be useful in either direction most of the time, with little waste compared to US cities. This is the real weakness of the argument of just adding more transit: What we need to make it work is basically urban rebuilds. IMO still a good idea, bit it's a far slower, and more expensive problem than it might appear.