Flexible Resource Allocation for UAV-Assisted Distributed IoT Data Collection

UAV-assisted distributed IoT data collection plays a vital role in scenarios ranging from post-disaster response to large-scale temporary events. This paper investigates dynamic resource allocation for UAV-assisted IoT uplink transmission under partial channel state information (CSI). Considering the impracticality of acquiring full CSI due to pilot overhead and limited device computing capabilities, the problem is modeled as a Partially Observable Markov Decision Process (POMDP). A dynamic scheduling strategy is proposed, jointly optimizing node selection, beamforming weights, and UAV trajectory. By leveraging belief updates to integrate noisy channel observations with historical information, the proposed approach enhances decision-making under uncertainty. Furthermore, we introduce a novel channel-aware belief-space rollout (CABR) algorithm, which combines reliability-driven action candidate generation, a weighted multi-factor reward function, and adaptive planning depth based on task process to efficiently allocate resources under partial CSI. Simulation results demonstrate that the proposed method significantly improves throughput and reduces latency compared to the state of the art.