VoI-Driven and Online Learning-Aided 3D Path Planning for UAV-Assisted Contingency Monitoring

Uncrewed Aerial Vehicle (UAV)-assisted data gathering has been emerging as a compelling solution to contingency event monitoring where Internet of Things devices (IoTDs) are communication-limited, such as industrial accidents and disasters in remote areas. Nevertheless, due to the UAV energy constraint, it is almost impossible for one or a limited number of UAVs to achieve complete visitation of monitoring regions in a single mission for a large-scale monitoring area with numerous IoTDs, even with energy-optimal three-dimensional (3D) path planning, which can further impact the unpredictable Value of Information (VoI) of data gathering because of geospatial heterogeneity. In this paper, we present a long-term mission framework to ensure complete region visitation by sharing the visitation opportunities across regions. Specifically, we formulate a multi-mission round 3D UAV path planning problem aiming at maximizing the long-term expected VoI, subject to region visitation fairness constraints. By combining bandit learning and virtual queue techniques, we propose a VoI-driven and Online Learning-aided 3D Path Planning (VOLPP) algorithm without requiring prior knowledge of VoI. We demonstrate the feasibility and asymptotic optimality of the proposed algorithm by rigorous performance analysis. Extensive simulation results show that our proposed algorithm achieves an asymptotically optimal trade-off between maximizing the objective and satisfying constraints and effectively guides the UAV to fully explore the interplay between them, outperforming comparison algorithms in enhancing VoI and balancing visitation frequency. To broaden its applicability, the VOLPP algorithm is further extended to a multi-UAV version to accommodate more general monitoring scenarios.