Toward Self-Evolving Agentic AI for ISAC-Enabled Low-Altitude Wireless Networks

Abstract

This paper proposes a self-evolving agentic artificial intelligence (AI) framework for low-altitude wireless networks (LAWNs), introducing integrated sensing and communication (ISAC) into a unified self-evolution paradigm that transforms static foundation models with passive perception into fully autonomous, self-evolving agentic systems. The framework integrates ISAC-enabled perception, reasoning, evolution, task-specific decision-making, and knowledge memory access into a unified self-evolving agentic architecture, enabling iterative improvement of cognition, task understanding, and decision evaluation capabilities. We further introduce a self-evolving deliberative agentic AI mechanism, in which agents generate multiple candidate actions and perform consistency-based evaluation before execution. This reason-before-act approach shifts decision-making from reactive responses to rational deliberation, enabling proactive decision-making and long-horizon optimization. A case study on the integrating sensing, communication and control task in LAWN demonstrates that the proposed framework significantly enhances data rate and sensing error performances.