Decision modeling for automated driving in dilemmas based on bidirectional value alignment of moral theory values and fair human moral values

Abstract

The decision-making issues in autonomous driving dilemmas, namely, the inevitable personal injury accidents, have become a key factor affecting the future development of autonomous driving. Currently, the main difficulty in dilemma decision-making problems lies in the moral controversies involved. To address this challenge, this paper proposes a dilemma decision model for autonomous driving with bidirectional value alignment of moral theory values and human moral values. First, this paper utilizes the counterfactual fairness to eliminate personal bias and learn fair human values from human feedback. Then, the dilemma decision problem for autonomous driving is solved by bidirectional value alignment of moral values fair human values. Through real human decision-making data from the MM experiment, the necessity of establishing this model is demonstrated, and its effectiveness and stability are verified in a virtual environment. The experimental results are shown that compared to other ethical decision-making methods, this model is more in line with ethical morals and human values. It also avoids introducing discriminatory elements from human values into the model’s dilemma decisions, providing better interpretability for decisions and offering superior solutions for dilemmas.