Autonomous systems, whether AI-enabled or not, increasingly operate in ethically sensitive contexts such as healthcare, assistive robotics, and smart environments. These systems are progressively acting on behalf of humans implicitly or explicitly, being humans proactive, reactive, or passive. However, ethical aspects, including both regulatory constraints (hard ethics) and human-dependent ethical preferences (soft ethics), are still insufficiently considered in their design and operation. This leads to a lack of trust and misalignment between system behavior and individuals and organizational values, as trust in autonomous systems requires the integration of ethical considerations alongside technical and legal requirements. In particular, current autonomous systems are typically unable to incorporate humans’ ethical preferences into their decision-making process, or to adapt their behavior at runtime based on evolving contextual and human preferences. This calls for architectural mechanisms that enable systems to monitor contextual dynamics, embed ethical considerations, and continuously align their behavior with both hard and soft ethics. Existing works, including laws, regulatory frameworks, ethical guidelines for AI, and international standards and recommendations, primarily ensure compliance with legal, safety, and societal principles. While they highlight the importance of ethical considerations in autonomous systems, they do not provide concrete guidance on how to operationalise and implement ethical preferences within the software architecture of such systems. Similarly, prior research in machine ethics often focuses on embedding predefined ethical rules or specific ethical theories, which fail to address the variability of human values and the need for systems to adapt to different humans and contexts. This gap between high-level ethical principles and practical system design motivates the need for a reference architecture that helps engineers integrate both regulatory requirements and human-centered ethical adaptation in the development of ethical-aware autonomous systems. In this direction, this paper proposes a \emph{reference architecture for ethical-aware autonomous systems} that explicitly supports runtime adaptation to both hard and soft ethics. The proposed architecture extends the classical MAPE-K feedback loop (Monitor-Analyze-Plan-Execute-Knowledge) into an \emph{E-MAPE-K} loop (\emph{Ethical}-Monitor-Analyze-Plan-Execute-Knowledge), introducing an \emph{Ethic Connector} component responsible for managing ethical profiles, mediating ethical decisions through ethics-driven negotiation, and explainability. Additionally, ethical awareness is embedded across all feedback-loop phases, enabling the Monitor, Analyze, Plan, and Execute components to incorporate ethical reasoning, compliance checking, ethics-driven negotiation, and autonomy adjustment as first-class architectural concerns.

The main contributions of the paper are: - Definition of ethical-aware autonomous systems: We provide a precise conceptual definition of the concept of ethical-aware autonomous systems. - Requirements for ethical-aware autonomous systems: We derive the main requirements useful to identify the main functionalities that ethical-aware autonomous systems should support. The identified requirements are validated with domain experts with background knowledge of autonomous systems, digital ethics, trust, and privacy. - Reference architecture for ethical-aware autonomous systems: We propose a reusable reference architecture for ethical-aware autonomous systems, built on top of the MAPE-K feedback loop. The reference architecture is validated through a scenario-based evaluation in two ethically sensitive domains from the literature, i.e., body sensor networks in healthcare and assistive robotics.

The proposed reference architecture aims to help system and software architects and engineers in designing and developing ethical-aware autonomous systems that should interact and collaborate with humans while respecting values important to individuals, society, and the environment.