A Conceptual Framework for a Hybrid ML-Agents LLM Architecture for Adaptive Serious Game Scenarios

DOI

10.2991/978-94-6239-817-7_16How to use a DOI?

Keywords

Machine Learning; Agents; AI; LLMs; architecture; framework; serious games

Abstract

Serious games tend to be used for training, scenario-based exercises and simulation learning, but their scripted nature limits their adaptability. Existing research explores reinforcement learning agents in simulation environments, LLM-based game agents, and LLM-powered virtual entities for serious games. However, there is still a lack of modular conceptual architecture that integrates ML-Agents as a behavioral execution layer with an LLM-based planning and communication layer for adaptive serious games and AI validation scenarios. This work aims to suggest the concept of a hybrid ML-Agent and LLM architecture, dividing behavioral execution, LLM planning, validation and action translation into individual layers.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Jakub Zagora
AU  - Peter Sykora
AU  - Roberta Hlavata
AU  - Miroslav Benco
PY  - 2026
DA  - 2026/10/29
TI  - A Conceptual Framework for a Hybrid ML-Agents LLM Architecture for Adaptive Serious Game Scenarios
BT  - Proceedings of the SAFAR Scientific Conference 2026 – Building Awareness through Immersive Learning (SSC-BAIL 2026)
PB  - Atlantis Press
SP  - 181
EP  - 194
SN  - 2667-128X
UR  - https://doi.org/10.2991/978-94-6239-817-7_16
DO  - 10.2991/978-94-6239-817-7_16
ID  - Zagora2026
ER  -