Cooperative Graph-based Predictive Collision Avoidance (CGPCA): A Decentralized Framework for Safe Drone Traffic Management

Cooperative Graph-based Predictive Collision Avoidance (CGPCA) is a novel, decentralized approach designed to manage and control drone traffic while preventing collisions with both other drones and surrounding obstacles. It combines real-time communication, Graph Neural Networks (GNNs), and Decentralized Model Predictive Control (DMPC) to achieve this goal. Each drone continuously broadcasts its state—such as position, velocity, and intended path—to nearby units, dynamically constructing a "traffic graph" that includes both drones and detected obstacles. Unlike traditional methods that rely on fixed geometric rules or purely reactive algorithms, CGPCA leverages predictive learning. A GNN, trained on a wide range of simulated traffic scenarios, processes the traffic graph to forecast near-future positions and assign collision risk scores. These predictions then feed into a DMPC module onboard each drone, which optimizes safe, efficient trajectories in real time to mitigate potential conflicts. By integrating these advanced technologies, CGPCA provides a scalable and robust solution that adapts dynamically to rapid changes in drone behavior and environmental conditions, paving theway for safer and more efficient U-space operations.