Intent Fusion: Resolving Agent Conflicts Through Large Language Models and Digital Twins in 6G Networks

Abstract

Intent-based management in 6G networks introduces the risk of conflicting agents and overlapping intents operating simultaneously, where current solutions often rely on rigid, rule-based priority assignments. This paper introduces an intent-based management system for base stations, using large language models (LLMs) and digital twins. High-level natural language intents are parsed and translated into system actions by LLMs. In a multi-agent digital twin environment, conflicting intents are resolved via a learning-based meta-agent that prioritizes objectives dynamically. An integrated xAI module explains agent decisions, which are applied to the physical network, with interpretable visual feedback. Our work is the first to introduce agent-level intent fusion through multi-objective optimization for wireless network management, enabling adaptive and interpretable conflict resolution.