LAMIO-6G: Large AI Model-Empowered Cross-Layer Intent Management and Multi-Domain Policy Orchestration in 6G Terrestrial Networks

As differentiated services emerge, intent-driven management and orchestration in the 6G terrestrial networks will face two key challenges. On the one hand, the intent mapping gap, limited orchestration flexibility, and lack of policy abstraction constrain intent refinement across the Business Support System, Operation Support System, and Network Operation Provider layers. On the other hand, multi-domain orchestration among the radio access, transport, and core networks remains hard due to limited global awareness, resource conflicts, and inconsistent policy models. In this paper, we present LAMIO-6G, a large AI model-empowered framework for cross-layer intent management and multi-domain policy orchestration, which autonomously generates policies at different levels of abstraction from intents. The LAMIO-6G incorporates two key models: (i) a unified network policy model and (ii) a monitor-analyze-plan-execute-knowledge feedback closed-loop model. To address limited intent generality and scenario generalization in the Business Support System layer, we then introduce intent decomposition techniques via generic large AI models using low-rank adaptation-based fine-tuning, design of intent decomposition prompts, and few-shot learning-assisted intent decomposition. Within the Operation Support System layer, we design an intent reasoning and optimization scheme through collaboration between domain-specific large AI models guided by long-short chain-of-thought techniques and lightweight proximal policy optimization model. Finally, we present a proof-of-concept implementation of a wireless energy-saving intent. Simulation results demonstrate that the DeepSeek-R1-14B model achieves 17% to 30% gains over all baseline schemes in fine-tuned metrics, intent decomposition and reasoning accuracy, and intent optimization performance.