Towards Real Autonomy in Robots: From Internal Regulation to Adaptive Cognition

A key distinction between biological and robotic systems lies in real autonomy. Biological systems derive their behavior from internal drives shaped by the interaction between physiology and cognition, enabling self-maintenance, emotion, and decision-making. In contrast, robots rely on external rewards and task-based learning, lacking self-referential drives and self-defined goals. Bridging this gap requires biologically inspired mechanisms that integrate internal variables, cognition, and external interactions. This thesis seeks to present a holistic framework for understanding autonomy by grounding it within foundational biological principles. Drawing on insights from biology, cognitive science, and neuroscience, it explores the evolutionary and bodily roots of autonomy, emphasizing the interconnected roles of internal physiology, and cognition in creating a self-referential systems and shaping behavior. Current robotic methods for implementing autonomy are evaluated through this lens, revealing their limitations and providing a perspective for advancing the field. This theoretical foundation addresses key questions about goal setting, decision-making, and the emergence of emotions, offering a more comprehensive approach to robotic autonomy. The second part of this work explores practical implementations through two approaches: (1) modeling the effects of artificial stress (inspired by cortisol) within the CLARION cognitive architecture, demonstrating how stress modulates decision-making by shifting from habitual to safety-focused behaviors, mirroring human responses; and (2) leveraging Active Inference principles with a predictive coding-inspired model (PV-RNN) to enable a robot to balance internal states and external goals through dynamic self-regulation. While not achieving full autonomy, the results confirm that integrating internal variables improves adaptability and self-regulatory behaviors, fundamental for context-sensitive decision-making. This work demonstrates how biologically inspired mechanisms can guide robots toward human-like, self-sustaining autonomous behavior, with potential applications in healthcare, assistive robotics, and search-and-rescue operations. Ultimately, it lays the groundwork for advancing robotic self-determinism and the future development of truly autonomous systems.