A Second-Generation Agentic Framework for Generative Ai-Driven Augmented Reality Educational Games

This paper presents a multi-agent collaborative framework with human-in-the-loop integration for creating educational augmented reality (AR) serious games. It provides a detailed analysis of the system's internal modules, agents, and tools, highlighting its complexity. This work focuses on the design phase, while implementation of the framework is yet to be developed. The framework consists of four main modules: Data Management,Narrative Crafting, Content Generation, and Game Creation. It uses the Scratch-pad as a shared memory to manage agent communication and long-context data, guiding agents step-by-step in the game-building process. The framework also incorporates short- and long-term memory to improve agent performance over time. Additionally, it integrates Generative AI (Gen-AI) models for asset creation in immersive games. The analysis reveals the system's complexity and the numerous hyper-parameters that need further study at both technical and user experience levels. This study invites other researchers to contribute to this framework, aiming to develop a collaborative educational tool that removes technical barriers and empowers teachers and students to create and engage with immersive interactive learning experiences.