Ethical decision system for autonomous vehicles in unavoidable accident scenarios

The automotive industry is heading towards the introduction of fully autonomous vehicles. However, before this type of vehicles are commercially available at mass scale, some issues need to be solved. A major issue is the ethics involved in the decision-making during an accident; this research presents an analysis of how to approach this issue and a way to implement a solution based on concepts from Belief-Desire-Intention agent modelling. The first part of this research identifies and defines a pre-programmed system with different ethical settings based on five formal ethical theories. For each, eight ethical concerns are defined and ordered accordingly. These concerns are defined in terms of harm to self and harm to others. The ethical concerns are used as a guideline to define the level of importance of each person or object in an accident scenario. The resulting rank of concerns is novel in the field of ethical decision-making for autonomous vehicles, and serves as the basis for the implementation of an ethical decision-making agent for unavoidable collisions. The second part of this thesis focuses on the design, derivation and implementation of the decision-making system in a BDI agent framework. With the proposed system, the vehicle is partially tailored to the ethical preferences of different users while still being bounded by legal requirements to avoid any misuse. The resulting outcomes of the decision-making system under different scenarios are shown and discussed. In these discussions it is clear that the proposed system successfully captures ethical concerns, priorities and behaves in accordance to the ethical theories.