사용자 가치 기반의 행동 선택을 위한 돌봄 로봇의 윤리적 의사결정 프레임워크

Abstract

This study presents an Ethical Interactive Partially Observable Markov Decision Process (Ethical I-POMDP) framework for ethical decision-making in care robots, developed based on insights from user studies. To identify ethical dilemmas in human-robot interaction, we conducted two surveys: one validating the defined ethical dilemma and another assessing human expectations for robotic decision-making. The results indicate that user’s ethical value is placed on social encouragement rather than direct assistance, confirming the existence of a genuine ethical conflict. The proposed I-POMDP model dynamically balances short-term care needs with long-term autonomy promotion through probabilistic decision-making. Using the penalty-based reward, our approach adjusts intervention levels over time based on user engagement patterns. We highlighted the limitations of predefined reward functions and propose adaptive learning mechanisms to personalize ethical policies. The future work will integrate long-term user data and reinforcement learning techniques to enhance real-world applicability.