Toward Agentic Intelligence in Non-Terrestrial Networks: A Roadmap for Federated and Quantum-Enhanced Learning

Abstract

Non-terrestrial networks (NTNs), comprising satellites, uncrewed aerial vehicles (UAVs), and high-altitude platform stations (HAPS), are key enablers of sixth-generation (6G) wireless systems, providing global seamless connectivity. However, NTN environments present challenges, including high propagation delays, intermittent connectivity, limited backhaul capacity, and heterogeneous resources, which constrain the effectiveness of existing centralized optimization and standalone learning approaches. To address these limitations, we propose a unified decision-making framework that integrates agent-based control, hierarchical federated learning (FL), and quantum-enhanced machine learning (QML) across a multi-tier space–air–ground architecture. In the proposed framework, each network node is an autonomous agent performing decentralized task offloading and resource allocation. FL enables distributed model training across network tiers without sharing raw data. QML is incorporated to support optimization in high-dimensional and dynamic decision spaces. Unlike existing approaches, this framework jointly coordinates decision-making, learning, and optimization to enable scalable, adaptive operation under NTN-specific constraints. This roadmap outlines a pathway towards real-time, intelligent NTN systems and identifies key challenges for future research.