An ethical decision-making framework for autonomous vehicles based on public moral preferences

The rationality of decision-making by autonomous vehicles in unavoidable ethical dilemmas significantly affects public acceptance, and this, in turn, impacts the promotion and application of the technology. Given the absence of public opinion in the current autonomous vehicle decision-making, this study proposed an ethical decision-making algorithm framework based on public moral preferences. Five moral preferences reflecting widespread social consensus were identified: reducing overall injury, reducing individual injury, ensuring equal care, protecting the compliant, and avoiding active harm. Further drawing on a large-scale online survey featuring quantified injury severity information of road users, we modeled these moral preferences into machine language that autonomous vehicles could understand and execute. On this basis, we constructed the ethical decision-making algorithm framework. Referring to literature-based AV ethical dilemmas and real-world traffic conditions, we curated 4,860 diverse traffic scenarios for algorithm validation. Owing to the integration of public moral views as a significant basis in ethical decision-making, the proposed algorithm's decisions in these scenarios closely aligned with the moral preferences we introduced. Specifically, compared with the baseline algorithm, our algorithm reduced the average individual injury severity by 52 %, the standard deviation of injury severity by 15 %, and the injury severity proportion of compliers and non-occupiers by 20 % and 24 %, respectively. The proposed framework is highly scalable and is expected to serve as a reference for future research on autonomous vehicle ethics in other regions with different cultural backgrounds. This work contributes to the public acceptance of autonomous vehicle technology and further facilitates its promotion and application.