Agent-based ethical decision-making control optimization framework for self-driving vehicles

Yusha'u, S

Abstract

When driving, humans balance values like safety, legality, and mobility. When sharing the road with humans, an autonomous vehicle will likely use the same values. Incorporating human values into algorithm design is tough for self-driving vehicle engineers. To address this problem, a decision-making algorithm is designed using philosophical concepts translated into mathematical frameworks. Deontological ethics parallels rule-based mathematical concepts, whereas consequentialism parallels cost-based mathematical concepts. The virtue ethics philosophical principle is also used to motivate the different weightings of path tracking, obstacle avoidance, and traffic regulation compliance. The consequences of different design decisions made in a model predictive steering controller are highlighted by simulation results of a self-driven vehicle negotiating an obstructed two-lane route with a double yellow line.
The value-sensitive design (VSD) iterative process is used to formalize the link between human values and technological needs. A modified VSD technique was used to develop a self-driven vehicle speed control algorithm for pedestrian crossings. The VSD iterations model the problem as a partially observable Markov decision process that was used to generate an effective approach for controlling the vehicle's longitudinal acceleration based on the belief of a pedestrian crossing.
An ethical valence that characterizes self-driven vehicle decision-making as a mechanism for claim mitigation, in which various road users make varying moral claims on the vehicle's behavior, and the vehicle must neutralize these claims while making assessments about its surroundings. With self-driving vehicles, the harm produced by an action and the uncertainties connected with it are assessed and accounted for, leading to an ethical implementation that is realistic. Instead of describing how moral concepts need self-driving vehicles to behave, this approach provides a computational approach that may accommodate a variety of moral positions about what morality demands and what road users could expect.

• Thesis Type: Thesis
• Deposit Date: Jul 8, 2022
• Publicly Available Date: Jun 2, 2024
• Award Date: Mar 1, 2022