Hybrid Neural Control for Autonomous Navigation of Uncrewed Surface Vessels

1. Background In marine robotics, Autonomous Surface Vessels (ASVs) are essential for a wide range of applications, from environmental monitoring to commercial and Navy operations. Ensuring accurate path-following in real-world environments is a complex challenge due to environmental disturbances such as wind, waves, and currents. Traditional control methods, including Proportional-Integral (PI) controllers, are commonly used in path-following

tasks

however, these methods often struggle with the inherent nonlinearity and dynamic uncertainty of marine environments. Recently, neural networks and deep learning have shown promise in control and estimation tasks due to their ability to approximate complex functions and handle nonlinearities.

Leveraging these capabilities, this project aims to explore the potential of neural networks to enhance the robustness and adaptiveness of ASV control

through integration into classic techniques such as observer-based controllers.

2. Research Gap Current control systems for ASVs generally utilize linear

or simplified nonlinear methods, which are often unable to adapt adequately to nonlinearities and rapidly changing conditions induced by environmental

forces. While neural networks have been implemented in various control contexts, there remains a significant gap in integration and real-time

implementation to guarantee both convergence of the state estimates and stable path-following performance in ASVs. Furthermore, limited work has

focused on validating these methods on physical models of marine vehicles. These observed gaps present an opportunity to innovate a control approach

that combines the adaptiveness of neural networks with the stability of traditional observer and controller designs.

3. Objectives The primary objectives of

this study are twofold:

1. Nonlinear Neural Observer Development: Develop a structured neural observer that ensures convergence of the state estimate

for an ASV in the presence of environmental disturbances. This observer will utilize neural network-based estimations to accurately track the ASV’s

states (e.g., position, velocity) under various disturbance conditions.

2. Nonlinear Neural Control Method: Design a structured neural controller tailored

for adaptive path-following control of an ASV. This neural controller will incorporate nonlinear dynamics and adapt to disturbances, aiming to maintain

accurate path-following performance and minimize tracking errors. The overall objectives of the proposed project are not limited to these two. The

potential candidate is encouraged to extend these objectives through performing a critical literature review.

3. Scope This study encompasses both

theoretical and practical work, aiming for a comprehensive understanding and validation of the proposed neural observer and controller methods. The

scope includes: - Theoretical Analysis: Derivation and proof of stability for both the neural network-based observer and controller. This analysis will

leverage Lyapunov-based methods to ensure convergence and stability under nonlinear dynamics. - Simulation and Testing: Implementation and testing

of the observer and controller in a simulation environment, using models of the ONR Tumblehome and KVLCC2 to simulate realistic ASV dynamics and

environmental disturbances. - Practical Implementation: Verification of the neural observer and controller on physical scaled models of the ONR

Tumblehome and KVLCC2. Experiments will be conducted to assess the real-world performance of the proposed methods under varying conditions of

wind, waves, and currents. By addressing these components, this research will provide new insights into adaptive neural-based control systems for

marine surface vehicles. The research outcomes will contribute to the robustness and reliability of ASVs in real-world marine environments