Hierarchical and Collaborative LLM-Based Control for Multi-UAV Motion and Communication in Integrated Terrestrial and Non-Terrestrial Networks

Zijiang Yan, Hao Zhou, Jianhua Pei, Hina Tabassum · 2025

Read Hierarchical and Collaborative LLM-Based Control for Multi-UAV Motion and Communication in Integrated Terrestrial and Non-Terrestrial Networks by Zijiang Yan, Hao Zhou, Jianhua Pei, Hina Tabassum (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-06532.txt.

Introduction

Zijiang Yan and co-authors put large language models in charge of drone fleets that share ground and sky networks. Fleets of drones must dodge each other, hold traffic flow, and keep radio links alive at once. The team tests a layered control scheme on an aerial highway scene.

One model rides on a high platform station and handles which drone joins which link. A second model on each drone plans its own moves, such as speeding up, slowing down, or changing lane. This split lets high-level plans guide quick local choices in real time. Trials show higher task rewards, lower running costs, and far fewer crashes than baseline methods. The work points toward managed three-dimensional drone highways for future cities.

Catalog record

CREATORZijiang Yan, Hao Zhou, Jianhua Pei, Hina Tabassum
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-06532

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Hierarchical and Collaborative LLM-Based Control for Multi-UAV Motion and Communication in Integrated Terrestrial and Non-Terrestrial Networks Zijiang Yan, Hao Zhou, Jianhua Pei, Hina Tabassum Abstract Unmanned aerial vehicles (UAVs) have been widely adopted in various real-world applications. However, the control and optimization of multi-UAV systems remain a significant challenge, particularly in dynamic and constrained environments. This work explores the joint motion and communication control of multiple UAVs operating within integrated terrestrial and non-terrestrial networks that include high-altitude platform stations (HAPS). Specifically, we consider an aerial highway scenario in which UAVs must accelerate, decelerate, and change lanes to avoid collisions and maintain overall traffic flow. Different from existing studies, we propose a novel hierarchical and collaborative method based on large language models (LLMs). In our approach, an LLM deployed on the HAPS performs UAV access control, while another LLM onboard each UAV handles motion planning and control. This LLM-based framework leverages the rich knowledge embedded in pre-trained models to enable both high-level strategic planning and low-level tactical decisions. This knowledge-driven paradigm holds great potential for the development of next-generation 3D aerial highway systems. Experimental results demonstrate that our proposed collaborative LLM-based method achieves higher system rewards, lower operational costs, and significantly reduced UAV collision rates compared to baseline approaches. † † affiliationnotice: Equal contribution 1 Introduction Unmanned aerial vehicles (UAVs) have seen widespread adoption across numerous application areas due to the ease of deployment, steadily reduced production costs, and the availability of line-of-sight (LOS) channels ( Yu et al. 2022 ) . Depending on the mission, a UAV may act as user equipment requiring cellular access or operate as an aerial base station (BS) to extend network coverage. Nonetheless, for UAVs flying beyond visual line-of-sight, robust command-and-control links are indispensable to guarantee both the operation reliability and security. Meanwhile, High Altitude Platform Stations (HAPS) has recently emerged as a crucial part of 6G Non-Terrestrial Networks (NTNs). HAPS, known for its extensive propagation coverage, long-distance reach, and utilization of eco-friendly energy sources like solar and wind power, is progressively becoming a fundamental component of 6G networks ( Kurt et al. 2021 ) . Compared to the control of an individual UAV, the joint coordination of multiple UAVs inherently incurs substantially greater complexity, since it must address not only the increased state-space dimensionality but also the inter-UAV synchronization required to maintain formation and collaborative mission objectives ( Cherif et al. 2021 ; Kotarski et al. 2020 ) . In addition, when these UAV swarms are integrated with a HAPS within a NTN, additional challenges may arise from the heterogeneity of air-to-air and air-to-ground links. Resource allocation, network topology management, and control-plane algorithms will jointly contribute to the overall complexity across multi-UAV control, further complicating system design, real-time decision making, and end-to-end quality-of-service (QoS) guarantees. Existing research primarily focuses on optimizing cellular link availability and QoS using reinforcement learning (RL) algorithms, without considering multi-UAV aerial traffic flow and motion dynamics of UAVs. In ( Cherif 2022 ) , the authors proposed a RL algorithm that considers disconnectivity, handovers, and energy consumption for trajectory planning and cell association in cargo UAVs. Nesrine et al. leverage RL and propose an algorithm for joint trajectory planning and cell association in ( Cherif et al. 2023 ) , to minimize energy consumption and number of handoff events with QoS constraints. The authors in ( Chen et al. 2020 ) present strategies based on deep learning to predict handovers in mmWave communications and optimize handover (HO) rates and radio link quality for known UAV trajectories. However, most existing algorithm’s actions only consider the direction of motion with no speed and lane considerations ( Li et al. 2020 ) . These works have not considered the motion dynamics factors, such as acceleration, deceleration, and lane changes of multi-UAVs ( Zhang et al. 2022 ) . For instance, collision avoidance is crucial for guaranteeing the performance for multi-UAV traffic flows. Specifically, in densely populated aerial corridors, each UAV must continuously monitor the positions and velocities of its neighbors and execute timely trajectory adjustments to preserve safe separation distances. Failure to do so can lead to emergency maneuvers that disrupt coordinated flight patterns, introduce significant latency in UAV task execution. Different from existing studies, this work considers a more challenging scenario of a 3D aerial highway, in which multiple UAVs have to jointly share the highway lanes by acceleration/deceleration, lane changes, etc. Meanwhile, from the communication perspective, the UAVs are expected to achieve higher data rates and minimal handover losses. The transportation decisions are closely coupled with the wireless communication system performance. For instance, increasing speed can increase transportation traffic flow, but it will also lead to frequent handovers, which can negatively impact the UAV communications. To better handle such complexity, this work employs a hierarchical and collaborative control method based on large language models (LLMs). Leveraging the intrinsic strengths of LLMs, our approach harnesses their extensive pre-trained knowledge base to support both high-level planning and fine-grained decision-making. Through in-context learning, the LLM dynamically interprets real-time traffic patterns, channel state information, and mission objectives, allowing it to infer optimal acceleration, deceleration, and lane-change policies without requiring exhaustive task-specific retraining. In addition, the instruction-following capabilities further enable seamless incorporation of safety constraints, such as minimum inter-UAV spacing and maximum allowable handover rates, directly into the decision pipeline, ensuring both transportation and communication objectives are satisfied. Existing studies have explored various application scenarios of LLM-enabled 6G networks, including grounding and alignment ( Xu et al. 2024 ) , prompt engineering ( Zhou et al. 2025 ) , network design ( Qiu et al. 2024 ) , power control ( Zhou et al. 2024 ) , etc. However, this work is different by including hierarchical and collaborative multiple LLM interactions. In particular, we consider one LLM deployed in the HAPS to manage the UAV-HAPS connections, and another LLM for UAV action control such as acceleration, deceleration, and lane change. By unifying transportation and communication control within an LLM‐driven architecture, our method can significantly reduce the design complexity associated with traditional model‐based controllers. It also accelerates adaptation to novel operational scenarios, such as varying UAV densities or spectrum availability, through prompt and instruction‐guided updates. This end‐to‐end, knowledge‐rich paradigm holds significant promise for next‐generation 3D aerial highways, where the seamless interplay of mobility and connectivity is paramount. Finally, the simulations show that our proposed technique can achieve higher rewards, lower transportation and communication costs, and lower collision rates for multi-UAV scenarios. 2 System Model Figure 1 : Illustration of the proposed aerial network model. Fig. 1 shows our considered aerial highway scenario, where N_{\mathrm{UAV}} UAVs are flying along the highway lanes, and UAVs can connect HAPS or terrestrial BSs. The HAPS and terrestrial BSs form a set \mathcal{B}=\{b_{1},\ldots,b_{N_{R}},b_{\mathrm{HAPS}}\} , where each BS is denoted by i for \forall i\in\mathcal{B} . To simulate the UAVs’ movements on a given aerial highway, we consider the continuous intelligent driver model that models acceleration as in ( Treiber & Kesting 2013 ) . 2.1 G2A Channel Model Firstly, the ground-to-air (G2A) channel model mainly depends on the BS’s antenna gain, and the experienced path loss, and line-of-sight (LoS) probability ( 3GPP 2018 ) . In cellular‐connected aerial systems, UAVs depend on

Read the complete text