Abstract:

The evolution of the 6G core network toward endogenous intelligence has become a consensus within the industry. However, existing research still exhibits fragmented characteristics regarding multi-agent architectures, intent processing, and the management of Artificial Intelligence (AI) capabilities. Building upon current advancements in 6G architecture standardization research, this study proposes a multi-agent collaboration and AI capability lifecycle management architecture tailored for 6G core networks. This architecture introduces two types of network-level AI agents, namely planning agents and execution agents, with task-level collaboration among heterogeneous agents implemented through dedicated agent-specific interfaces. Simultaneously, the architecture incorporates AI control functionalities that enable unified management of AI capability registration, model training, validation and database integration, provisioning and deployment, operational monitoring, retraining, and anomaly-based rollback.

Key words: 6G core network, service-oriented architecture, AI agent, intent-driven, model lifecycle management