IEEE Communications Magazine · Journal article · 2026
Authors: Jiahao You, Ziye Jia, Chao Dong, Qihui Wu
Published in: IEEE Communications Magazine, vol. 64, no. 5, pp. 156–162, May 2026.
Authorship: First author / 第一作者
Research overview
This article presents a prompt-to-agent framework for adapting reasoning workflows to resource-constrained low-altitude edge networks.
The problem
Fixed task-to-model bindings limit flexibility, while large models can exceed edge memory and compute budgets. Static inference pipelines also struggle to respond to changing tasks and operating conditions.
The approach
The framework combines prompt-defined cognition, reusable modular agents, and diffusion-controlled inference planning. Task intent is translated into abstract representations, instantiated with available agents, and refined using runtime feedback.
Evaluation
A representative low-altitude networking use case illustrates modular execution and adaptive planning. The article discusses the framework and its operating mechanisms; readers can consult the original for the case configuration and results.
提示驱动的低空边缘智能：模块化智能体与生成式推理
本文提出从提示到智能体的边缘认知框架，使推理流程能够适应资源受限的低空边缘网络。
研究问题
固定的任务与模型绑定限制了灵活性，大模型的存储和计算需求又可能超出边缘预算。静态推理流程也难以适应任务与运行环境变化。
研究方法
该框架结合提示定义的认知、可复用模块化智能体和扩散控制的推理规划。任务意图先转为抽象表示，再根据资源条件选择智能体执行，并结合运行反馈调整推理流程。
研究评估
论文通过低空网络用例展示模块化执行与自适应规划，并讨论框架的运行机制。具体用例配置与结果见原文。
中文题名为便于阅读所作的译名，正式引用请使用英文原题。
Citation
[1] J. You, Z. Jia, C. Dong, and Q. Wu, “Prompt-driven low-altitude edge intelligence: Modular agents and generative reasoning,” IEEE Commun. Mag., vol. 64, no. 5, pp. 156–162, May 2026, doi: 10.1109/MCOM.001.2500472.
