Home // AIMEDIA 2025, The First International Conference on AI-based Media Innovation // View article

LLM-based Few-shot Action System for NPCs in Virtual Reality Games

Authors:
Fan Wang
Wen Zhou
Rongze Gui
Jinqiao Li
Radoslaw Malicki
Andrey Staroseltsev

Keywords: Action Agent; Large Language Model; NPC in Virtual Reality; Agentic Workflow; Retrieval-Augmented Generation.

Abstract:
Current trends in the game industry include making games more immersive and realistic through developing games in Virtual Reality (VR) and integrating Generative Artificial intelligence (AI) in Non-Player Characters (NPCs). As Large Language Model (LLM) based conversational NPCs start to emerge and show success in traditional video game medium, we seek to answer the question of ``can we leverage LLMs to build believable NPCs that can make logical actions and interact with players naturally in VR?" In this paper, we introduce a design of an LLM-based action system with few-shot learning for NPCs in VR worlds. The use of few-shot learning would allow for rapid adaptation to new games without massive data requirement or expensive model training. We also include an evaluation plan to assess our designed system's performance.

Pages: 13 to 17

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-330-9

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025