MORAL DECISION-MAKING FOR AUTONOMOUS DRIVING BASED ON MULTI-OBJECTIVE REINFORCEMENT LEARNING

Vehicles must ensure safety and efficiency and deal with complex ethical dilemmas in autonomous driving. In order to deal with these ethical dilemmas effectively, moral decision-making models based on multi-objective reinforcement learning (MORL) provide a technical path to resolve such ethical dilemmas. Unlike traditional reinforcement learning (RL), MORL can generate more socially moral decision-making strategies in conflict scenarios by simultaneously optimizing multiple objectives. Of course, significant challenges remain in this research path. Assigning reward function weights is highly dependent on subjective judgement and cultural context; the dynamic environment is not adaptable enough, and the scarcity of ethical dilemma data limits model training. To address these issues, this paper points out that future research needs to focus on the dynamic weight adjustment mechanism, the construction of cross-cultural ethical frameworks, and large-scale real-world validation.