Deep Reinforcement Learning-Based AUV Path Planning for Energy-Efficient Data Collection in Underwater Wireless Sensor Networks

Abstract

Underwater Wireless Sensor Networks face unique challenges including severe acoustic channel impairments, high propagation delays and node energy constraints. Traditional Autonomous Underwater Vehicle data collection schemes rely on static waypoint traversal, which fails to adapt to dynamic ocean environments and varying channel conditions. In this paper, we propose a deep reinforcement learning based AUV path planning framework that leverages Proximal Policy Optimization and Soft Actor-Critic algorithms to learn energy-optimal, channel-aware collection trajectories. The AUV agent observes real-time acoustic Signal-to-Noise Ratio, residual energy of sensor nodes, and ocean current vectors to make intelligent navigation decisions. Simulation results demonstrate that the proposed HPSA-AUV achieves a 94.2% data collection rate while reducing energy consumption by 25.5% and improving average acoustic SNR by 29.9% compared to greedy waypoint baselines. These results validate the efficacy of DRL for adaptive AUV mission planning in complex underwater environments.