|
Our first monthly feature paper is an invited tutorial paper co-authored by researchers from University of Chinese Academy of Science, Zhejiang University and Macau University of Science and Technology, which introduces an advanced Large Language Model (LLM) Multi-Agent Reinforcement Learning (MARL) integrated framework for UAV swarm control that combines LLM-driven semantic reasoning with MARL-based exploration.
Complementing the invited paper is a feature article coauthored by researchers from Universitat Politčcnica de Catalunya, Robert Bosch GmbH, and Aalborg University which proposed an efficient and novel data fusion solution, based on the Graph Neural Network to enhance the positioning accuracy of the target Automated Guided Vehicle in industrial environments.
We’ve provided short summaries of these feature articles, written in accessible language that we hope will make your reading experience enjoyable.
RALLY: Role-Adaptive LLM-Driven Yoked Navigation for Agentic UAV Swarms
Ziyao Wang, Rongpeng Li, Sizhao Li, Yuming Xiang, Haiping Wang, Zhifeng Zhao, and Honggang Zhang
Summary by Honggang Zhang: Imagine a scenario in communication-constrained post-disaster search and rescue or area inspection missions, where a swarm of drones must collaborate autonomously: they need to avoid obstacles and dynamic threats, efficiently cover multiple task objectives, while maintaining formation and real-time communication. Up to date, various conventional approaches often rely on rigid numerical communication and fixed roles among the drones, resulting in poor adaptability to complex environments.
On the other hand, it is widely recognized that Large Language Model (LLM)-based agentic frameworks have demonstrated strong semantic reasoning capabilities by leveraging extensive prior knowledge. Nevertheless, due to the lack of online learning and over-reliance on static priors, these frameworks often struggle with effective exploration, leading to reduced individual potential and overall system performance.
We address the above critical challenge with an innovative “two-stage semantic reasoning + dynamic role allocation” mechanism, named as RALLY (Role-Adaptive LLM-Driven Yoked navigation algorithm). Specifically, an LLM-driven semantic decision framework is established, which takes advantage of structured natural language for efficient semantic communication and collaborative reasoning. Afterward, a dynamic role-heterogeneity mechanism is introduced for adaptive role switching and personalized decision-making. Furthermore, a Role-value Mixing Network (RMIX)-based assignment strategy is developed, which integrates LLM offline priors with Multi-Agent Reinforcement Learning (MARL) online policies to enable offline training of role selection strategies.
Accordingly, by enabling drones to interpret intents and build consensus through natural language, and to dynamically switch roles (commander, coordinator, executor) in real time, the proposed approach grants each drone human-like decision-making ability, achieving efficient collaboration and robust multi-target coverage in dynamic scenarios.
Experiments show that RALLY significantly outperforms various existing methods in terms of task completion, convergence speed, system robustness, and generalization, offering a novel pathway for deploying the next-generation intelligent, collaborative UAV systems.
Full article: IEEE Open Journal of Vehicular Technology, Volume 6
PosGNN: A Graph Neural Network Based Multimodal Data Fusion for Indoor Positioning in Industrial Non-Line-of-Sight Scenarios
Karthik Muthineni, Alexander Artemenko, Daniel Abode, Josep Vidal, and Montse Nájar
Summary by Karthik Muthineni: Accurately locating mobile vehicles such as Automated Guided Vehicles (AGVs) inside factories is essential for smart manufacturing. One promising technology for this task is Ultra-Wideband (UWB), which can measure distances between the tag and the anchors very precisely. However, industrial environments often contain obstacles like machines and metallic objects that block signals. Because of this, the number of UWB anchors that can communicate with the tag (AGV) can change from one moment to the next, causing sudden drops or fluctuations in positioning accuracy.
To address this problem, we propose a new method called PosGNN. This method uses a machine learning model, a Graph Neural Network, that can naturally handle a changing number of UWB anchors. At each moment, the available UWB anchors and their distance measurements are organized in a flexible graph structure, allowing the system to work reliably even when some UWB anchors become temporarily unavailable due to Non-Line-of-Sight (NLoS) conditions. In addition, PosGNN combines the UWB measurements with motion data from the AGV’s onboard Inertial Measurement Unit (IMU) sensor. By fusing these two sources of information, the PosGNN can estimate the AGV’s position more accurately and consistently.
We tested our approach in a real industrial setting where signal blockages frequently occur. The results show that PosGNN performs better than existing traditional model-based and data-driven methods (improvement of 40%), demonstrating that it is well-suited for reliable positioning of mobile vehicles in large industrial environments.
Full article: IEEE Open Journal of Vehicular Technology, Volume 7
Open Call for Editors
The IEEE Open Journal of Vehicular Technology (OJVT) is calling for editors in any research areas that belong to the below-mentioned scopes:
- Mobile radio shall include all terrestrial mobile services
- Motor vehicles shall include the components and systems and motive power for propulsion and auxiliary functions
- Land transportation shall include the components and systems used in both automated and non-automated facets of ground transport technology
If you are interested in serving as an editor in these areas, please send your resume with the contact of two references for consideration to Editor-in-Chief Edward Au, no later than 1 May 2026.
| 
