TOUCAM: Task-Oriented Uncertainty-Aware Cooperative Charging for Multiple AUVs in Underwater Wireless Sensor Networks
Abstract
Underwater wireless sensor networks (UWSNs) are increasingly deployed for long-term marine monitoring, but their sustainable operation is severely constrained by limited battery capacity and the difficulty of manual maintenance. Although autonomous underwater vehicles (AUVs) have been introduced as mobile chargers, most existing studies mainly optimize network lifetime or average energy replenishment efficiency, while overlooking the heterogeneous importance of sensing tasks and the uncertainty of underwater environments. In practical task-driven UWSNs, sensor nodes contribute unequally to different missions, and the global network state is often only partially observable due to constrained underwater communication and dynamic environmental disturbances. To address these challenges, this paper proposes a task-oriented uncertainty-aware cooperative charging scheme for multiple AUVs, termed TOUCAM. The proposed method integrates task-importance-aware modeling, shared spatiotemporal graph belief representation, latent state prediction, and constrained multi-agent reinforcement learning into a unified decision framework. In this way, multiple AUVs can adaptively allocate charging resources to mission-critical nodes while maintaining robust cooperative behavior under uncertain observations. Experimental results demonstrate that TOUCAM achieves superior task completion rate, network survival performance, and energy utilization efficiency compared with representative baseline methods.