> First, this particular crash is an egregious counter-example. The car doesn't even seem to slow down when it first sees the pedestrian's foot. Nor does it try to swerve. This is basic stuff for a human driver, never mind more complex avoidance and risk mitigation a human driver can perform.

I'm not sure which side of this I fall on just yet, but something strikes me here: you're assuming a human driver with no impairment (e.g. eyesight or fatigue) paying complete attention to what they're doing. We know that this isn's always the case (and could even be a minority!), so this doesn't seem to be a great argument.