I think this makes the most sense. The robot car is driving along at a certain speed V along a certain path P with path options A,B,C. The car has a reaction ti

I think this makes the most sense. The robot car is driving along at a certain speed V along a certain path P with path options A,B,C. The car has a reaction time of T, a resolving distance D, and a hazard confidence H. These are the robot car's motion constraints for that moment in time, the options available to the car are a function of these constraints. The robot car should never ever be intentionally operating in a state where solving for the constraints means the available option of one death or another. In the case of a child and a narrow tunnel: -As you enter the narrowing path P (without even considering the child at this point), your path options are reduced to 0. You cannot consider the option of a change in path any more, unless you consider death by wall an option. (reverse would first require a change in V) -The car can see and resolve objects and environment up to distance D, and should be designed to be able to stop within that distance. If it can't stop within its resolution envelope, that is bad design. Unless the child pops up from behind a boulder when the car is 10 feet from the tunnel entrance, the car should detect the child from a long way off. That scenario, however, is like someone jumping in front of a train. -The route chosen by the robot car should have a hazard level associated with it. The robot car's hazard confidence will be affected by its current position along a route and proximity to route hazards, by its proximity to other car's traveling in the same direction, and detection of other hazard signals. -Given all the constraints, and since only speed V can change, the robot should not be operating at a speed where it has an OPTION of killing the child or the car occupants. TLDR: Drive slow approaching a narrow road, otherwise you give yourself no options.