LEHiD: LLM-Empowered Hierarchical DRL Framework for Path Planning in UAV Networks
IEEE Transactions on Mobile Computing · 2026-12 · 25 (12) · 22793-22807
DOI: 10.1109/tmc.2026.3725374 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Mobile Computing, 2026-12. Verfasst von Jie Zhang und 3 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Jiangsu Province, Natural Science Research Start-up Foundation of Recruiting Talents of Nanjing University of Posts and Telecommunications.
Abstract (Originalsprache)
In the era of 6G and the massive Internet of Things (IoT), Uncrewed Aerial Vehicles (UAVs) are a promising option for autonomous data collection under limited mission budget. The core difficulty is to plan trajectories that collect useful data while preserving safe return in environments with structured obstacles, communication-dependent service quality, and changing mission priorities. Classical planners such as A* provide geometric feasibility, but their objectives must often be redesigned when the mission changes. Deep reinforcement learning (DRL), on the other hand, suffers from sparse rewards and weak sample efficiency in these settings. To address this gap, we propose LEHiD, a hierarchical framework that combines a Large Language Model (LLM) for high-level target sequencing, an A*-based layer for safety validation and planner-facing cost estimation, and a DRL policy for low-level reactive control. The A* layer also provides heuristic state augmentation and A*-guided replay-buffer warm starts that improve low-level learning. We evaluate the framework under static structured routing, where each learned-control hierarchy is trained with its corresponding high-level planner, and under semantic-adaptation settings, where the trained D3QN controller is frozen and only the planner-side interface changes after online priority updates. The results indicate that LEHiD remains competitive with strong non-LLM hierarchical training baselines under static routing, exceeds planner-only route-ordering references on landing-aware collection, improves semantic target redirection in a controlled exact priority-shift experiment, and yields the strongest results among the tested planners in the richer structured regional-priority experiment.
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Jiangsu Province
- Natural Science Research Start-up Foundation of Recruiting Talents of Nanjing University of Posts and Telecommunications
Zitieren
APA
Zhang, J., Huang, H., Li, D., & Zhao, H. (2026). LEHiD: LLM-Empowered Hierarchical DRL Framework for Path Planning in UAV Networks. IEEE Transactions on Mobile Computing, 25(12), 22793-22807. https://doi.org/10.1109/tmc.2026.3725374
Vancouver
Zhang J, Huang H, Li D, Zhao H. LEHiD: LLM-Empowered Hierarchical DRL Framework for Path Planning in UAV Networks. IEEE Transactions on Mobile Computing. 2026;25(12):22793-22807. doi:10.1109/tmc.2026.3725374
BibTeX
@article{Zhang2026,
author = {Zhang, Jie and Huang, Hao and Li, Dapeng and Zhao, Haitao},
title = {LEHiD: LLM-Empowered Hierarchical DRL Framework for Path Planning in UAV Networks},
journal = {IEEE Transactions on Mobile Computing},
year = {2026},
volume = {25},
number = {12},
pages = {22793--22807},
doi = {10.1109/tmc.2026.3725374}
}
EUVIMED