Autonomous driving plays an important role in many industrial sectors. From passenger transport to automated agriculture and transportation, many non-military application areas have opened up, which are intended to save personnel, costs and resources in particular. In almost all cases, video- and sensor-based systems are used to assess the current situation of the vehicle and thus influence its behaviour. For example, high-quality lawnmower robotor have camera systems that are capable of recognizing animals and children in the vicinity and intelligently avoiding them.

In the passenger sector, self-driving cars appear to be the next stage in the development of personal transportation. Even though it is not yet permitted in many countries to drive cars completely autonomously, self-driving cars are becoming increasingly popular. Reports about fatal accidents with autonomous driving vehicles are always present in the media, but the car industry promises that in the near future autonomous driving systems will be in no way inferior to human drivers in terms of safety. 

However, even if autonomous driving systems are safe, there may be situations where accidents are unavoidable. 

Use the following website to identify thirteen extreme situations where accidents are unavoidable: https://www.moralmachine.net/

1. Describe each situation in a few words.
2. How did you react in each situations and why?
3. Try to formalize situations where accidents may be unavoidable and the decision of the autonomous driving car results in an ethical conflict.
4. Do you think there is a perfect (i.e. objective) solution to this problem? If yes write an algorithm (in pseudocode) that solves this problem. (You can define functions as scenario I/II, number of fatalities, age, gender, misbehavior, social status: For example: if  fatalities (scenario 1) < fatalities (scenario 2), then use scenario 1, else …)

In all cases, there is material or human damage and even death. Simplified, it is assumed that there are always two outputs, in which there is always different loss. (e.g. older woman vs. younger child). The losses differ in number, age, social status and the behavior of the persons involved. 

It can be assumed that a self-driving car evaluates the situation "objectively", while people would at least rather protect their own life due to the instinct of self-preservation. For an "objective" evaluation of the situation, factors such as age, social status, gender, etc. must be weighted (if these can be recorded at all). I.e. subjective definitions decide the outcome of the situation. 

Put simply, each element of a scenario is assigned a value (loss of a female aged 32, for example, gives 100 points). The mathematical evaluation of the situation determines its outcome.

For example, a situation with i dead persons could be assigned the value

   value(situation)= 1/(age_1 )+...+1/(age_i) 

to a situation with i dead persons. The instruction for the self-driving car would then be:

If

   Value(Situation 1)>Value(Situation 2)

then

   excecute (situation 1)

otherwise

  execute (situation 2)

Such an evaluation cannot be objective: First, because all factors can never be taken into account, and second, because the value of different human lives cannot be objectively compared. On the other hand, such an evaluation must also be made, because an autonomous vehicle cannot stop functioning in the middle. 

There is a comparable situation in medicine when it is decided which patient is to be operated on first.

This topic can be embedded in a teaching unit on the topic of self-driving cars, e.g. by building an autonomous driving vehicle with the help of a mini-computer (e.g. Raspberry Pi).

With the help of an integrated camera, the situation can be simulated in a simplified way. Red and green spheres represent younger and older people, respectively. The vehicle has to drive through a parkour and find the best way from a moral point of view. The red and green balls can be identified with the help of opencv.org.