ABSTRACT The rapid expansion of the Internet of Drones (IoD) has introduced complex security challenges, particularly the emergence of hostile, autonomous swarm attacks that can easily overwhelm traditional kinetic defenses. Existing countermeasure systems often struggle with a critical trade‐off: lightweight onboard models lack the perceptual accuracy to distinguish threats in noisy environments (e.g., smoke, glare), while heavy deep learning models induce prohibitive latency and energy drain on battery‐constrained UAVs. To bridge this gap, this paper proposes a novel framework for Real‐time UAV Countermeasure via Edge‐based Multi‐modal Transformer in IoD. We introduce a multi‐modal spatio‐temporal transformer (MST‐Trans) that leverages the computational power of IoD edge nodes to fuse heterogeneous sensor streams—including RGB visuals, Infrared (IR) thermal signatures, and radio frequency (RF) fingerprints. This fusion mechanism ensures robust target locking even under severe visual occlusion. Furthermore, we deconstruct the coordination problem using an Edge‐Assisted multi‐agent deep reinforcement learning (E‐MADRL) approach, specifically a resource‐aware dynamic role assignment (RADRA) algorithm. This mechanism dynamically offloads computationally intensive inference tasks to the edge server while optimizing the swarm's energy distribution between jamming, tracking, and recharging roles. Extensive simulations in a high‐fidelity AirSim‐ROS2 environment demonstrate that the proposed framework achieves a neutralization rate of 97.6% and reduces end‐to‐end response latency to 142.6 ms. Compared to standalone decentralized baselines, our approach extends swarm endurance by 30% and reduces false engagement rates to under 4%, validating its efficacy for securing next‐generation urban airspace.
Kunxue Zhu (Fri,) studied this question.