Real-Time Tracking Control Using Adaptive Dynamic Programming for Underwater Vehicles

This manuscript presents the real-time experiments of a model-free nonlinear controller based on Adaptive Dynamic Programming (ADP) for an Autonomous Underwater Vehicle (AUV). These types of vehicles exhibit significant coupled nonlinearities and parameter uncertainties. To address this, our approach introduces a data-driven scheme, where open - loop data are collected and used to train a differential neural network for identifying the vehicle's dynamics. Based on the identified model, an ADP scheme with an actor-critic structure is implemented to conduct trajectory tracking experiments for yaw and depth dynamics. Additionally, this study examines the implications of using the ADP algorithm and offers insights for necessary adaptations to achieve successful implementation on an open-source commercial underwater vehicle within the ROS environment, thereby bridging the gap between theoretical concepts and practical experimentation.