Underwater Target Tracking with Unknown Maneuver by Remotely Operated Vehicles: A Digital Twin-Driven Strategy

Abstract

Underwater target tracking is a critical challenge in marine exploration and defense applications due to the unknown maneuvers of target and the complex marine environment. To overcome the above challenge, this paper develops a digital twin (DT)-driven unknown maneuver target tracking strategy via remotely operated vehicles (ROVs). In order to capture the maneuver characteristics of target, a state prediction-based DT framework is constructed, where the neural network learning strategy is designed to estimate the unknown state transition matrix of target. Based on the predicted target state, a reinforcement learning (RL)-based tracking controller is designed for the virtual ROVs in DT model, such that the optimal tracking policy from DT model can be implemented to physical ROVs. To reduce the matching error between virtual and physical ROVs, an RL-based optimization algorithm is conducted by using the data interaction between DT model and ROVs. Note that the DT-driven target tracking strategy not only can reduce the communication energy consumption by periodically feeding back the real-data of ROVs to the DT model, but also can relax the dependence of target maneuver model via the state prediction method. Finally, experimental results are provided to verify the effectiveness of our strategy.