Effective environmental data collection is pivotal for successful disaster response operations. Mobile crowdsensing (MCS), which leverages unmanned aerial vehicles (UAVs) for coarse-grained data acquisition and human participants for finegrained data gathering, presents a viable solution to enhance disaster rescue efforts. However, integrating human and UAV resources for large-scale spatiotemporal crowdsensing tasks remains a significant challenge, particularly in complex urban disaster environments characterized by intricate road networks and densely distributed points of interest (POIs). This paper proposes Hi-HUTA, a dynamic hierarchical cooperative framework that simultaneously optimizes data freshness, human-UAV cooperation relationships, and adaptive allocation procedures in dynamically evolving disaster scenarios. At the first layer, we propose a multi-agent deep reinforcement learning (MADRL) algorithm enhanced with laziness dilemma detection and elimination mechanisms to facilitate distributed UAV scheduling. This approach ensures efficient resource utilization while maintaining comprehensive environmental perception. At the second layer, we introduce TKBF, a dynamic task priority matching algorithm, to optimize UAV-human cooperative task allocation. By evaluating bilateral preferences between UAVs and humans and designing a dynamic priority-based double-ended queue, TKBF optimizes allocation strategies in evolving environments. Extensive experiments, including simulation-based evaluations and a real-world case study, demonstrate that Hi-HUTA significantly outperforms seven baseline methods in effectiveness, scalability, and robustness.
Li et al. (Tue,) studied this question.