Distributed Switching MAAC-Based Energy-Efficient Underwater Data Collection for Multi-AUV System

Internet of Underwater Things (IoUT) has emerged as a critical infrastructure for marine exploration and resource utilization, yet faces inherent challenges in data collection due to the limitations of multi-hop transmission. Multiple Autonomous Underwater Vehicles (AUVs) systems present a promising solution in data collection; however, conventional multi-AUV data collection schemes struggle with formation collaboration awareness, high-dimensional observation spaces, and environmental non-stationarity in underwater scenarios. To address these limitations, this paper proposes a Distributed Switching Multi-Agent Attention Actor-Critic (DS-MAAC) framework for energy-efficient underwater data collection. Specifically, we consider a scenario where the location of the sensor nodes are unknown. In this context, multi-AUV systems initially adopt a leader-follower formation for stable communication and obstacle avoidance, transitioning to a distributed system at appropriate timings to balance the effectiveness and efficiency of exploration. The fuzzy perception map is proposed to assist in distributed switching and progressively reduce environmental uncertainty. To improve the ability of AUVs to extract critical information from high-dimensional observation spaces, we introduce an attention mechanism to further enhance the collaborative collection performance. The simulation results demonstrate that the proposed DS-MAAC algorithm outperforms other baseline methods across multiple performance metrics, including reward, collection path, collection rate, collection efficiency, energy consumption, delay rate, and value of information, with performance improvements ranging from 11.07% to 56.30%. Sensitivity analyses are conducted to verify the robustness of the DS-MAAC algorithm and provide practical guidelines for parameter selection.