I argue that above all else, in any situation, the car should protect its occupants. While some may say this is contrary to the goal of minimizing death, it is in fact a far better way to accomplish that goal. Consider the following statements:

We wish to minimize loss of life in car accidents
Self driving cars will cause far fewer deaths than human-controlled cars
People will likely not use self-driving cars that will potentially intentionally kill them

The first statement is obvious. While technically not thoroughly field-tested, statement two is reasonable to assume as true as the technology advances. To back up the third statement, studies have shown that people overwhelmingly value passenger safety over pedestrian safety, see (link to study)

1. If we wish to minimize loss of life in car accidents, then we should maximize the amount of self-driving cars on the road
2. If we wish to maximize the amount of self-driving cars on the road, people must trust them and want to buy/use them
3. If people are to trust self-driving cars, the cars cannot be programmed to intentionally kill the occupants.
4. Therefore, if we wish to minimize loss of life in car accidents, we cannot program self-driving cars to intentionally kill their occupants.