Research Article
Flexible and Efficient Air-Ground Collaborative Intelligence in Low-altitude Intelligent Networking
@INPROCEEDINGS{10.1007/978-3-032-33662-0_18, author={Jiyuan Xie and Yuben Qu and Chao Dong and Lei Zhang and Qihui Wu}, title={Flexible and Efficient Air-Ground Collaborative Intelligence in Low-altitude Intelligent Networking}, proceedings={Communications and Networking. 20th EAI International Conference, ChinaCom 2025, Xi’an, China, November 8--9, 2025, Proceedings}, proceedings_a={CHINACOM}, year={2026}, month={10}, keywords={UAVs edge computing collaborative inference multi-agent reinforcement learning (MARL) air--ground coordination}, doi={10.1007/978-3-032-33662-0_18} }- Jiyuan Xie
Yuben Qu
Chao Dong
Lei Zhang
Qihui Wu
Year: 2026
Flexible and Efficient Air-Ground Collaborative Intelligence in Low-altitude Intelligent Networking
CHINACOM
Springer
DOI: 10.1007/978-3-032-33662-0_18
Abstract
The rapid development of the Low Altitude Intelligent Networking (LAIN) has accelerated the integration of deep neural networks (DNNs) with unmanned aerial vehicles (UAVs), which greatly enhances their capabilities in perception, decision-making, and autonomous operation. However, the computational demands of complex DNNs often exceed the limited onboard resources of UAVs. Traditional approaches either deploy lightweight models on UAVs or offload raw data to ground stations. The former offers low latency but suffers from limited accuracy, while the latter achieves high accuracy at the cost of significant latency. Both struggle to strike a balance between accuracy and latency. In this paper, we investigate dynamic switching between lightweight onboard inference and complex air-ground collaborative processing. We propose Flexible and Efficient Collaborative Intelligence (FECI), an adaptive solution that employs an enhanced Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm to learn optimal inference strategies. Simulation results demonstrate that FECI significantly improves inference efficiency, latency, and accuracy compared with several baseline methods, providing a robust solution for UAV intelligence in LAIN environments.