At the heart of (Io)RT-E ecosystems

Abstract

One of the major research areas in artificial intelligence focuses on designing and improving a robot’s cognitive capabilities. This involves enabling robots to accurately interpret human behavior and intentions based on their perception of the environment. To achieve this, it is essential not only to understand human intentions, but also to anticipate the causal effects of elementary and complex actions and their consequences within a given context. Modes of action preparation and emotions play an important theoretical role in this pro cess. Frijda explains that the way individuals perceive and appraise events triggers different modes of action preparation. Similarly, psychologist James J. Gibson describes interaction with the environment through the concept of affordances, which guide action. Consequently, the contribution of artificial intelligence to modeling contextual understanding in (Io)RT-E ecosystems is undeniable. Action recognition remains a critical research challenge, particularly in the field of hu man–robot interaction. Many questions remain unresolved, especially those related to un derstanding human behavior and anticipating future actions. This requires making spatio temporal projections, predicting multiple possible futures, and inferring the effects of actions based on the current or inferred context. When a robot lacks information, it must adapt by enriching its knowledge through the properties of observed actions. This process involves endowing robots with social cognitive capabilities that enable them to engage in joint actions with humans. In this context, we refer to joint human–robot agency. This thesis proposes a hybrid framework that combines semantic annotation, spatio temporal ontological modeling, narrative reasoning using NKRL, and reinforcement learning for adaptive human activity recognition. The proposed approach enables contextual, tempo ral, and causal interpretation of human actions. Experimental results conducted in Internet of Everything (IoE) environments demonstrate improved robustness, adaptability, and ac curacy compared to classical activity recognition approaches.