CKM-Driven Communication-Aware UAV Intelligent Trajectory Optimization for Urban Inspection

This paper addresses the challenge of maintaining reliable communication for UAVs during urban inspection tasks, where complex building layouts and street canyons cause severe signal attenuation. The authors introduce a two-layer trajectory planning framework called GATSAC. First, they construct a Channel Knowledge Map (CKM) using a diffusion model, which reconstructs high-fidelity global channel quality distributions from sparse, time-accumulated observation data. Second, they employ a Graph Attention Network (GAT) to solve the combinatorial node-ordering problem (a variation of the Traveling Salesman Problem), followed by a Soft Actor-Critic (SAC) reinforcement learning agent to generate smooth, continuous flight paths that proactively avoid communication-degraded regions.

The proposed GATSAC framework effectively balances flight efficiency with communication reliability. By leveraging the diffusion-reconstructed CKM, the UAVs can navigate through high-quality signal regions without needing constant, real-time channel feedback. Simulation results indicate that this approach outperforms traditional distance-based path planning and baseline reinforcement learning methods by significantly reducing the time spent in poor-signal zones while maintaining optimal inspection sequences.

In urban environments, standard path planning often prioritizes energy or distance, leading to communication outages that can compromise mission success. This research provides a robust, data-driven method for autonomous systems to "sense" the radio environment through historical accumulation and intelligent planning. By decoupling the discrete task-ordering from the continuous trajectory control, the method offers a scalable solution for multi-UAV coordination in complex, signal-obstructed urban spaces.