The vehicle would likely do what a human would do: slam on the breaks.
Also, it has no way of knowing whether anyone would die.
And, finally, a good-enough car should see people from much farther away than a human, react faster, and drive slower (at the speed limit). That means someone would almost literally have to jump in front of the car to trigger this scenario.
All wishful thinking. A self-driving car has no better brakes than an ordinary one. The problem statement says they run into the street too close to stop, so seeing them from further away is irrelevant. Lastly the whole issue is, should a self-driving car ever leave the roadway to preserve a life, precisely because it doesn't know if that could be worse.
This software choice, again, has already been made in existing self-driving vehicles. It remains to be seen if the choice holds up in court, since the first lawsuit will set a precedent for all others.
The brakes themselves are no better, you can save time by removing the input lag between the driver and the controls. There's 20 feet of braking saved by not having to move your foot from the floor to the pedal.
Also the car can signal it's intent to the steering, braking, and throttle instead of the car having to guess at what the driver is doing from their inputs. A car knowing it has to swerve around something can react sooner and differently than if it has to guess that it's swerving around something.
> A self-driving car has no better brakes than an ordinary one
I didn't say that it does. It has better reaction time and drives more slowly. That makes its stopping distance much shorter than a human's.
> The problem statement says they run into the street too close to stop, so seeing them from further away is irrelevant.
No, it's not. Seeing them from further away means this problem happens less frequently. If this problem happens so infrequently that overall deaths go down, that's a huge improvement.
> should a self-driving car ever leave the roadway to preserve a life, precisely because it doesn't know if that could be worse.
This is still better than a human, because it could actually make a decision. A human doesn't have time to consciously decide what s/he wants to do.
Also, it has no way of knowing whether anyone would die.
And, finally, a good-enough car should see people from much farther away than a human, react faster, and drive slower (at the speed limit). That means someone would almost literally have to jump in front of the car to trigger this scenario.