AI Agent-Driven Closed-Loop Control for Integrated Sensing and Communication Systems

In this paper, we propose an AI agent-driven closedloop control framework for integrated sensing and communication (ISAC) systems, which aims to maximize the system utility under dynamic user intents and wireless conditions. Specifically, the proposed framework operates in a closed loop: 1) in the large timescale, two AI agents empowered by large language models cooperatively translate high-level user intents into executable ISAC system parameters; and 2) in the small timescale, a successive convex approximation-based optimization solver determines the local beamforming solution subject to the predetermined system parameters. Moreover, an evaluation module checks whether the requested communication and sensing parameters are feasible and sends feedback to the AI agents for parameter adjustment as needed. Preliminary results show that the proposed framework improves overall utility by 8.6% as compared with benchmarks, demonstrating its effectiveness under time-varying user intents and channel conditions.