I did my own calculations last week and came to about 2x his numbers (https://www.linkedin.com/pulse/how-cheap-can-uber-self-drivi...). He makes a few faulty assumptions (ex. maintenance costs and depreciation are tied to miles, not years) and in my opinion underestimates costs significantly in areas like insurance, and over-estimates billable miles that will be driven per day. Still amazingly cheap, though, especially if carpooling and/or using public transit part of the time. For comparison, CalTrain runs between 17 and 25 cents per mile if you take it more than one zone.

Sharing rides with strangers is precisely the kind of thing that I'd hope to avoid by replacing public transport with self-driving cars.

But anyway, yes, it does appear that the economics of the self-driving taxi beat the economics of the bus or the train, especially given that buses and trains generally operate at a large loss and still charge people, on average, a higher fare than what they'd pay for a self-driving taxi.

A self-driving bus that operates a route could be even cheaper, but why bother when you can hail a cab to take you straight to your destination without stops for just a few tens of cents more?