MIT (and I assume others) have collected data on these types of scenarios. For example, swerve and kill a bunch of people, or stay the course and kill a mom and baby?

http://moralmachine.mit.edu/