Navigating Ethics in Autonomous Vehicles: Data-Driven Prospect Theory Approach to Decision Making

Abstract

As autonomous vehicles (AVs) increasingly interact with pedestrians and other vulnerable road users, ensuring that their decisions align with human ethical expectations has become a critical challenge. Traditional AV planning approaches prioritize safety but often neglect the behavioral and moral complexity of real-world traffic interactions. This study introduces a novel ethical evaluation framework for AV behavior that integrates prospect theory (PT), a behavioral model of decision making under risk, with two foundational ethical paradigms: utilitarianism and deontology. Using high-resolution trajectory data from the Top-View Trajectories dataset, we analyzed 714 pedestrian–vehicle interactions at an intersection. For each interaction, PT-based utility functions were computed for both the study agent and the overall system, accounting for travel time and time-to-collision as proxies for efficiency and safety. These utilities were combined into a composite ethical score, enabling the classification of behavior along a continuous ethical spectrum. Our results revealed that safety, as measured by time-to-collision, strongly influenced ethical evaluations, but that prolonged inefficiency was also penalized. Vehicle-yielded interactions consistently received higher ethical scores, supporting the potential of social AV behavior. The framework additionally identified cases of ethical divergence (where agents benefit at the system’s expense), highlighting the need for decision models that balance self-interest and collective outcomes. This work offers a scalable, data-driven approach to evaluating AV ethics and lays the groundwork for ethically aligned AV planning.