Work

Low-Latency and Sustainable Data Collection for Wireless-Powered Multi-UAV Mobile Crowdsensing Through Resource Allocation

Year: 2026

Type: article

Abstract: Unmanned aerial vehicles (UAVs) offer flexible deployment and cooperative capabilities for mobile crowdsensing (MCS), but their performance is limited by energy and communication resources. Meanwhile, ground users often suffer rapid battery depletion under poor channels or hardware faults. These challenges affect user selection, UAV trajectories, bandwidth and power allocation, energy-harvesting strategies, and task-time assignment, thereby determining sensing and data-delivery efficiency. Accordingly, we focus on balancing the sensing task completion time and total utility of sensed data among multiple resource-limited ground users and UAVs, by jointly designing the sensing data size, ground users' selection, UAV trajectories, and transmission strategies. We formulate this problem as a multi-objective stochastic optimization problem with the aim of obtaining a set of high-performance solutions subject to predefined latency and ground users' energy constraints. Due to the non-convexity, multi-timescale dynamics, stochastic uncertainty, and a high-dimensional, coupled decision space, conventional optimization algorithms are inadequate to solve it tractably. We then propose a lyapunov-guided transformer-enhanced multi-agent deep reinforcement learning (LyaTrans-MADRL) algorithm, which leverages historical observation sequences to capture temporal dependencies and enhance decision-making capabilities. The proposed algorithm enables UAV agents to collaborate during training and allocate network resources in a distributed manner, thereby improving cooperative decision-making and convergence efficiency. Simulation results demonstrate that our proposed algorithm outperforms existing benchmarks by approximately 30%, demonstrating superior performance in balancing sensing task completion time and total sensed data utility.

Source: IEEE Transactions on Vehicular Technology

DOI: 10.1109/tvt.2026.3728432