Real-Time Ethical Decision Framework for Autonomous Vehicles
Abstract
Autonomous vehicles (AVs) operate in highly dynamic conditions, where split-second decisions often involve ethical dilemmas, such as balancing the principles of pedestrian safety, passenger safety, and adherence to the law, particularly in situations where human decisions are unpredictable or pose an extreme risk to people or other vehicles. Classical rule-based and utility-based methods tend to fail because they are not adaptable, transparent, or trusted by stakeholders, and thus fail to consider these obstacles comprehensively, thereby constraining their ability to be accepted and implemented publicly in the real world. We are introducing EMPATHIC-AV (Ethical Multi-Adapter Planning and Transparent Human-Inclusive Control in Autonomous Vehicles), a new neuro-symbolic AI model that addresses this issue. EMPATHIC-AV is a real-time Ethical Ontology Engine to consider morality in a grounded manner, a Stakeholder Impact Simulator to approximate the effects on all other parties, and a Multi-Agent Reinforcement Learning Core to appeal to societal values as a reward signal. One of them is a Legal Constraint Mapper that ensures laws are followed in this region, and the other is an Explainable AI (XAI) module to explain why a decision has been made. The Human Consent Layer enables passengers to program their preferred moral settings, and a Community Feedback Loop continually optimises moral action based on societal feedback. The analysis of EMPATHIC-AV was conducted in high-risk real-world traffic conditions simulated in environments (Car-Las and SUMO), utilising unpredictable and random pedestrian and car behaviour. The findings indicate that the EMPATHIC-AV is superior to traditional methods in enhancing stakeholder safety, ethical consistency, and public trust, as well as in improving the legal process and transparency. Besides addressing the real-time ethical risks associated with AV operation, this framework also provides a scaling avenue toward socially acceptable, legally acceptable, and trustworthy AV solutions in the context of autonomous driving, offering level-headed policymakers, researchers, and business entities a platform to deploy next-generation AV systems.