WAPPOS: A Distributed-Learning-Based Off-Board Path Planning System for AAV-Assisted Emergency Networks

Establishing reliable communication networks in post-disaster environments is essential for effective emergency response. Deploying autonomous aerial vehicless (AAVs) equipped with base stations provides a rapid and promising solution for restoring connectivity. However, onboard path planning is computationally expensive due to the constantly varying terrain, and precomputing paths for all locations is impractical. We propose waypoint assisted path planning for off-board systems (WAPPOSs), a data and knowledge-driven framework that optimizes AAV path planning through distributed off-board processing. WAPPOS integrates satellite imagery from Google Earth into its Target Region Mapping module, which employs the DeepLabV3+ model to segment buildings into no-fly zones (NFZs) and fly zones (FZs) with 92.2% accuracy. To further refine navigation, WAPPOS introduces density-based spatial clustering of applications with noise algorithm (DBSCAN)-PP (density-based spatial clustering of Applications with Noise for Path Planning). This novel clustering algorithm identifies optimal waypoints by analyzing spatial patterns and building density, which are then transmitted to the AAV for adaptive navigation. A comparative study with onboard deep-reinforcement learning (DRL)-based path planning demonstrated that WAPPOS reduced CPU, GPU & Battery usage significantly and extended flight time by 4.4 min. By leveraging off-board computation, data-driven segmentation, and knowledge-driven clustering, WAPPOS reduces onboard computation, improves flight efficiency, and enhances AAV-based network deployment in disaster-stricken regions.