LLM-Driven NPC Interaction for Enhanced Immersion in Videogame Environments

Abstract

This research aims to incorporate large language models (LLMs) in the real-time gaming experience, adding depth to gameplay mechanics and enhancing player immersion through rich dialogue. In existing videogames, non-player character (NPC) interactions are mostly static due to their scripting and simple state machine logic. In this research, we showcase a sample video game environment populated by three different kinds of NPCs powered by AI: (1) an NPC merchant with the ability to dynamically negotiate prices, (2) a task giver who can create contextually relevant tasks for the player, and (3) a NPC storyteller who can share lore-related stories based on a pre-made knowledge document. All NPCs communicate using predefined prompts and responses generated by a large language model. The implemented system consists of dialogue handling, user interface integration, prompt engineering, and external models communication modules. The results indicate a significant improvement in realism when using AI-powered dialogue compared to statically defined ones.

Introduction

The text discusses the use of Large Language Models (LLMs) to improve non-player character (NPC) interactions in videogames, aiming to overcome the limitations of traditional scripted dialogue systems. Conventional NPCs rely on fixed dialogue trees and prewritten scripts, which often become repetitive and reduce player immersion. LLMs offer a solution by enabling dynamic, context-aware, and more natural interactions based on player input.

The proposed system integrates LLMs into three key NPC roles: merchants (bargaining/economy), quest givers (dynamic quest generation), and storytellers (lore-based narrative generation). This is implemented through a modular architecture in Unity, consisting of an NPC interaction layer, dialogue manager, LLM processing module, knowledge retrieval system, and user interface. The system uses structured prompts, role-based outputs, and a retrieval mechanism to ensure consistency with game lore.

The methodology includes Unity-based implementation, prompt engineering tailored to different NPC roles, and evaluation through qualitative user feedback. Results indicate that LLM-driven NPCs improve immersion, creativity, and responsiveness compared to traditional scripted systems, while maintaining game consistency through structured control mechanisms.

Conclusion

This project provides a model system for integration of large language models into an interactive videogame environment in order to increase the NPC dialogue generation, procedural task generation, and storytelling based on lore. The current solution involves the use of structured prompts, contextual metadata, and modular Unity architecture that allows for dynamic interactions impossible for a traditional script-based approach for dialogue generation. The three implemented NPC roles in this project, such as merchant, quest giver, and storyteller, each involve a different application of LLM-aided generation. It has been shown through analysis of the prototype that LLM-aided dialogue generation can contribute positively to the believability of NPCs and quality of narrative when constrained and rooted into the context. Despite the existence of some problems with consistency, computational complexity, and necessity to incorporate additional APIs, the results imply that the combination of deterministic game logic and probabilistic dialogue generation in hybrid models is a viable direction to further explore. The further research may involve development of more sophisticated retrieval techniques, memory for NPCs, and integration of AI systems into games.

References

[1] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. Cambridge, MA, USA: MIT Press, 2016. [2] T. Brown et al., “Language Models are Few-Shot Learners,” in Advances in Neural Information Processing Systems (NeurIPS), 2020. [3] OpenAI, “GPT-4 Technical Report,” 2023. [Online]. Available: https://openai.com/research/gpt-4 [4] OpenAI, “Chat Completions API Documentation,” 2024. [Online]. Available: https://platform.openai.com/docs [5] Meta AI, “LLaMA 3: Open and Efficient Foundation Language Models,” 2024. [Online]. [6] R. Koster, A Theory of Fun for Game Design, 2nd ed. Sebastopol, CA, USA: O’Reilly Media, 2013. [7] J. McCoy, M. Treanor, B. Samuel, C. A. Mateas, and I. Horswill, “Prom Week: Social Physics as Gameplay,” in AIIDE, 2011. [8] Bethesda Softworks, “Radiant AI System Overview,” 2006. Official Developer Documentation (Oblivion/Skyrim). [9] C. Hecker, “Behavior Trees for Next-Gen Game AI,” Game Developers Conference (GDC), 2008. [10] M. O’Hara, “The Nemesis System: Emergent Narrative in Shadow of Mordor,” GDC, 2015 [11] H. Ryan and M. Mateas, “Interactive Story Generation for Virtual Environments,” IEEE Transactions on Computational Intelligence and AI in Games, vol. 1, no. 3, pp. 173–186, 2009. [12] N. Montfort, Procedural Generation in Games, MIT Press, 2020. [13] K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “BLEU: A Method for Automatic Evaluation of Machine Translation,” in ACL, 2002. [14] J. Devlin, M. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in NAACL-HLT, 2019. [15] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” in Advances in Neural Information Processing Systems (NeurIPS), 2020. [16] Unity Technologies, Unity Manual and Scripting API, 2024. [Online]. Available: https://docs.unity3d.com [17] J. Togelius and J. Schmidhuber, “An Experiment in Automatic Game Design,” in IEEE Conference on Computational Intelligence and Games, 2008. [18] S. Parekh, T. Smith, and W. Swartout, “Illusion of Life: A Formal Model for Narrative Variation in Interactive Storytelling,” in AIIDE, 2018.

Copyright

Copyright © 2026 Viraj Zende, Vedant Patil, Kanchan Swami, Abhishek Veer, Disha Wankhede. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.