본문 바로가기
  • Home

Implementation of a Mobile RPG NPC System Based on Distributed Hybrid AI

  • Journal of Internet of Things and Convergence
  • Abbr : JKIOTS
  • 2025, 11(6), 9
  • Publisher : The Korea Internet of Things Society
  • Research Area : Engineering > Computer Science > Internet Information Processing
  • Received : November 4, 2025
  • Accepted : December 8, 2025
  • Published : December 31, 2025

Kim, Dong-Ju 1 KOH, SEOK JOO ORD ID 1

1경북대학교

Accredited

ABSTRACT

Prior research optimized individual NPC combat performance by combining Finite State Machines (FSM) and reinforcement learning but faced limitations regarding cooperative interactions in multi-agent environments and constrained mobile resources. To overcome these issues, this study implements and optimizes a distributed hybrid AI system—integrating FSM/BT with PPO/MARL as proposed in a preceding design study—within a mobile environment. The system adopts a decoupled client-server architecture to efficiently distribute computational loads. The client handles real-time state management and immediate tactical actions using FSM and Behavior Trees (BT), while the server manages individual policy optimization and learns high-level team strategies utilizing PPO and MARL. Both modules execute organic cooperative decision-making through real-time data synchronization. Experimental results demonstrate that the system effectively overcomes multi-agent cooperation limitations by significantly improving metrics, such as concentrated attacks and supportive actions, compared to previous models. Furthermore, by offloading high-load computations to the server, the system significantly reduced client CPU usage and secured stable response times. This study validates the feasibility of operating high-performance cooperative AI in real-time within resource-constrained mobile environments and provides concrete implementation guidelines.

Citation status

* References for papers published after 2024 are currently being built.

This paper was written with support from the National Research Foundation of Korea.