Memory-Augmented State Machine Prompting: A Novel LLM Agent Framework for Real-Time Strategy Games

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Qi, Runnan, Ni, Yanan, Jiang, Lumin, Li, Zongyuan, Huang, Kuihua, Guo, Xian
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911223610605568
author Qi, Runnan
Ni, Yanan
Jiang, Lumin
Li, Zongyuan
Huang, Kuihua
Guo, Xian
author_facet Qi, Runnan
Ni, Yanan
Jiang, Lumin
Li, Zongyuan
Huang, Kuihua
Guo, Xian
contents This paper proposes Memory-Augmented State Machine Prompting (MASMP), a novel framework for LLM agents in real-time strategy games. Addressing key challenges like hallucinations and fragmented decision-making in existing approaches, MASMP integrates state machine prompting with memory mechanisms to unify structured actions with long-term tactical coherence. The framework features: (1) a natural language-driven state machine architecture that guides LLMs to emulate finite state machines and behavior trees through prompts, and (2) a lightweight memory module preserving strategic variables (e.g., tactics, priority units) across decision cycles. Experiments in StarCraft II demonstrate MASMP's 60% win rate against the hardest built-in AI (Lv7), vastly outperforming baselines (0%). Case studies reveal the method retains LLMs' semantic comprehension while resolving the "Knowing-Doing Gap" through strict state-action mapping, achieving both interpretability and FSM-like reliability. This work establishes a new paradigm for combining neural and symbolic AI in complex decision-making.
format Preprint
id arxiv_https___arxiv_org_abs_2510_18395
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Memory-Augmented State Machine Prompting: A Novel LLM Agent Framework for Real-Time Strategy Games
Qi, Runnan
Ni, Yanan
Jiang, Lumin
Li, Zongyuan
Huang, Kuihua
Guo, Xian
Artificial Intelligence
68T42, 68T37, 91A35
I.2.6; I.2.11; I.2.8; K.8.0
This paper proposes Memory-Augmented State Machine Prompting (MASMP), a novel framework for LLM agents in real-time strategy games. Addressing key challenges like hallucinations and fragmented decision-making in existing approaches, MASMP integrates state machine prompting with memory mechanisms to unify structured actions with long-term tactical coherence. The framework features: (1) a natural language-driven state machine architecture that guides LLMs to emulate finite state machines and behavior trees through prompts, and (2) a lightweight memory module preserving strategic variables (e.g., tactics, priority units) across decision cycles. Experiments in StarCraft II demonstrate MASMP's 60% win rate against the hardest built-in AI (Lv7), vastly outperforming baselines (0%). Case studies reveal the method retains LLMs' semantic comprehension while resolving the "Knowing-Doing Gap" through strict state-action mapping, achieving both interpretability and FSM-like reliability. This work establishes a new paradigm for combining neural and symbolic AI in complex decision-making.
title Memory-Augmented State Machine Prompting: A Novel LLM Agent Framework for Real-Time Strategy Games
topic Artificial Intelligence
68T42, 68T37, 91A35
I.2.6; I.2.11; I.2.8; K.8.0
url https://arxiv.org/abs/2510.18395