<![CDATA[The autonomous UAV swarms have fundamental issues with strong coordination that arise under delays in communication, dynamic obstacles and noisy sensing environments, and the existing centralized or heuristic-based solutions are insufficient in addressing such issues. To cover this gap, this paper proposes a Graph Attention Network (GAT)-based Hybrid Reinforcement Learning and Swarm Intelligence Framework that can enable the communication-aware decentralized cooperation of UAVs. It is a multi-agent reinforcement learning and PSO, ACO, Differential Evolution, flocking behavior and Control Barrier Function-based safety correction, and GAT-inspired adaptive graph communication encoding. The results of the simulation of 18 episodes with 24 UAVs demonstrate that the reward, coverage, and collision were demonstrated to be improved by 32%, 27%, and 40% respectively as compared to a classical greedy baseline. The findings confirm the fact that the proposed hybrid GAT-RL architecture enables to promote significantly more scalability, safety, and real-time responsiveness of UAV swarms, which is a possibility on the path to large-scale autonomous aerial coordination.]]>
Copyrights © 2025
