Information-Entropy-Based Trajectory Planning for AUV-Aided Network Localization: A Reinforcement Learning Approach

Accurate positioning is essential for meaningful data collection in underwater acoustic sensor networks (UASNs), and localization has become a fundamental technology that provides real-time position estimates for sensor nodes. However, due to the harsh underwater environment as well as the difficulties and high expenses in network maintenance, localization in UASNs has always been a challenging problem. Different from previous works that rely on fixed anchors (e.g., buoys), this article uses an autonomous underwater vehicle (AUV) as a mobile anchor and proposes a reinforcement-learning-based trajectory planning method that allows the AUV navigation to meet the localization requirements of all sensor nodes. Specifically, based on gridded scenarios, this work models the node position uncertainties with information entropy and formulates AUV trajectory planning as a process of reducing the entropy of the whole network. Moreover, a modified actor-critic-based deep deterministic policy gradient (DDPG) reinforcement learning algorithm is designed to shorten AUV trajectory on the premise of ensuring a certain localization accuracy for UASNs. Through various numerical comparisons, the advantages of the proposed method have been validated in terms of efficiency and localization accuracy.