Trajectory Planning for Cooperative Coverage Missions with Multiple Stratospheric Airships

This paper presents a trajectory planning method for cooperative regional coverage missions involving multiple stratospheric airships under dynamic wind field conditions. We propose a De-Identified Centralized Architecture based on the Soft Actor-Critic algorithm (DICA-SAC), which uses a global module to aggregate agent states, extract compact global features, and broadcast them to all airships. By incorporating mean-field theory, the method ensures scalability to varying fleet sizes while maintaining global situational awareness. A convolutional neural network is employed to process high-resolution wind field data, compressing the raw grid into a compact set of features. A task-specific reward function is designed to maximize coverage while accelerating convergence. Simulations using historical wind field data over the South China Sea show that the proposed method achieves over 97 percent coverage in scenarios matching training conditions and maintains above 95 percent when the number of airships differs. These results demonstrate the robustness, adaptability, and scalability of DICA-SAC for large-scale multi-airship regional coverage planning in realistic environments.