Enhanced situational awareness for AUV's stochastic model by multirate neural control

This paper focuses on a critical component of the situational awareness (SA), the neural control of depth flight of an autonomous underwater vehicle (AUV). Constant depth flight is a challenging but important task for AUVs to achieve high level of autonomy under adverse conditions. The fundamental requirement for constant depth flight is the knowledge of the depth, and a properly designed controller to govern the process. With the SA strategy, we proposed a multirate depth control procedure to address the dynamics variation and performance requirement difference in various stages of AUV's trajectory for a nontrivial mid-small size AUV “r2D4” stochastic model. One neural network controller, named NARMA-L2 controller, is designed for fast and stable diving maneuvers of this AUV model. This control strategy for chosen AUV model has been verified by simulation of diving maneuvers using software package Simulink and demonstrated good performance for fast SA in real-time search-and-rescue operations.