As far as I can tell, none of the reported incidents in SF involved a "computer crash". Further, none have been as 'catastrophic' as other very-common accidents

As far as I can tell, none of the reported incidents in SF involved a "computer crash". Further, none have been as 'catastrophic' as other very-common accidents with human drivers. So yes, the details of recovery are important – but only in comparison to the human baseline. If incidents happen less often than with humans, and/or recovery happens as fast or faster than with humans, that's the relevant criteria. Also deserving weight: might autonomous cars avoid dangerous panics in some situations where humans sometimes panic & overreact? I've often seen human drivers make things worse by making an ill-advised impatient navigate-around. I've seen people trying, earnestly, to avoid a block or prior accident who – distracted, anxious – do something else wrong, causing another collision. Weigh all the rates & severities. Don't set policy based on selective worst-cases, including many imaginary/theoretical contrived cases.