Mission-Aware UAV Deployment for Post-Disaster Scenarios: A Worst-Case SAC-Based Approach

In remote areas, natural disasters owing to extreme weather or other causes can easily spread regionally. Unmanned aerial vehicles (UAVs) are regarded as pioneers for exploration and sensing in post-disaster scenarios due to their characteristics of flexible deployment and low cost. Considering the network disconnection caused by the long distances between the sensing area, UAVs, and the ground station (GS), this article aims to maximize the spatial exploration ratio (SER) while minimizing energy consumption with connectivity maintenance. To ensure that the UAV swarm successfully transmits as much post-disaster data as possible back, we develop a worst-case soft actor critic (WCSAC) based deployment algorithm with role management for UAVs. First, we formulate the three-dimensional (3D) UAV deployment problem as a real-time single-step Markov decision process (CMDP), together with role management that includes role assignment and role switching for each UAV. Next, to avoid unlimited energy consumption for complete coverage of remote mission areas, we separate the reward and cost functions and translate the energy consumption minimization to the residual energy constraints with a conditional value-at-risk (CVaR). Finally, visualizations and numerical results prove that the proposed WCSAC-cntRM can strike a better trade-off among the valid SER, energy consumption, and network connectivity than other benchmarks, especially when the emergency mission is dynamically triggered in the corner area.