Abstract

Embodied learning in Extended Reality (XR) can enhance educational outcomes by integrating physical action, emotion, and cognition in immersive environments. Within these settings, Embodied Pedagogical Agents (EPAs) powered by AI and natural language processing enable empathic, lifelike interactions, enriching the learning experience through personalised guidance. However, designing effective EPAs requires a comprehensive framework that addresses embodiment fidelity, emotional engagement, and motivation. To bridge this gap, we introduce and validate the EMPOWER framework (Embodied, Motivational, Pedagogical, and XR-enhanced), which integrates embodied learning, affective computing, motivational strategies, and XR affordances. We validated the framework using an agent-based simulator that models the interactions of autonomous learners and EPAs across three immersive scenarios (AR/VR/MR). The simulator implements the pillars with advanced AI features, including a novel emotion recognition system that infers learner emotions from physiological signals (e.g., heart rate, GSR, EEG). Simulation results show that all five pillars contribute positively to knowledge outcomes, with embodiment, motivation, and pedagogical alignment showing the strongest effects. EMPOWER offers actionable, data-driven guidance for designing effective EPAs and a practical pre-deployment testbed for immersive learning experiences, paving the way for rigorous field evaluations and scalable real-world adoption.