Artificial Intelligence (AI) is rapidly evolving in educational contexts, offering unprecedented opportunities for personalised learning. However, most AI-based educational systems still lack the capacity to recognise and respond appropriately to students' emotional states, creating a significant gap in their effectiveness as learning facilitators. This research introduces a theoretical multi-agent framework designed to enhance the educational experience by integrating affective computing with adaptive pedagogical strategies.

The proposed system, "Affective Educational Agent Network" (AEAN), builds upon our previous work on classroom multi-agent systems, incorporating specialised AI agents that collaborate to detect, interpret, and respond to students' emotional states.

The framework includes:
(1) an Emotion Recognition Agent utilising computer vision and natural language processing;
(2) a Learning Profile Agent that maps emotional patterns to learning preferences;
(3) an Adaptive Content Agent that modifies educational materials; and
(4) a Pedagogical Strategy Agent that adjusts teaching approaches.

An innovative feature of the framework is the integration of realistic virtual avatars simulating various educational figures: teachers conducting assessments in a protected environment to reduce evaluation anxiety; historical figures or authors directly explaining their works; laboratory experts guiding virtual experiments. Using advanced image and video generation technologies like Sora, these avatars can simulate complex scientific experiments, making abstract concepts visible and enhancing comprehension through immersive experiences.

The proposed methodology combines multimodal emotion detection (facial expressions, voice analysis, and interaction patterns) with machine learning algorithms trained to recognise subtle indicators of engagement, confusion, frustration, and satisfaction. The system employs reinforcement learning to continuously refine its emotional intelligence, creating increasingly accurate personalised responses.

In the near future, the framework is designed to integrate emerging technologies such as Google's Project Gemini and OpenAI's GPT-4 Vision, which will enable educational agents to "see" and interact with the student's physical environment. This functionality will allow advanced contextual assistance, such as recognition of physical teaching materials, identification of errors in laboratory experiments in real time, and the ability to provide direct visual support in learning spaces, both virtual and physical.

The AEAN framework is currently under review by the university ethics committee, with the aim of implementing a rigorous development protocol that ensures the protection of student data and the ethical use of AI technology in educational contexts. The proposed pilot study envisages gradual implementation of the system, beginning with basic emotional recognition modules and progressing towards more complex interactions based on avatars and artificial vision technologies.

The integration of emotional intelligence into AI educational agents represents a crucial advancement in educational technology, addressing the often-overlooked affective dimension of learning. This approach bridges the gap between technological innovation and human-centred pedagogy, acknowledging that effective learning is as much an emotional process as it is cognitive.

Keywords: Affective Computing, Educational AI Agents, Emotional Intelligence, Personalised Learning, Multi-Agent Systems, Virtual Avatars, Simulated Experiments, Vision-Enabled AI.