⚡ electrical engineering

Intelligent Control for Path-Following of an Unmanned Mass-Centric Surface Vehicle

This paper proposes a Lyapunov-based intelligent control scheme that utilizes thrust force and a sliding mass to enable an unmanned surface vehicle with dynamically changing mass distribution to accurately follow a desired path while compensating for unmodeled dynamics and external disturbances via artificial neural networks.

Original authors: Gabriel da Silva Lima, Tomi Westerlund, Wallace Moreira Bessa

Published 2026-07-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Gabriel da Silva Lima, Tomi Westerlund, Wallace Moreira Bessa

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a high-tech surfboard that doesn't just ride the waves but actually flies above them. These are called efoils, and they look like magic carpets powered by electric motors. Usually, a human rider steers them by shifting their weight left or right, moving the board's center of gravity to turn. But what happens when there's no human on board? How does a robot surfboard know which way to go?

This paper tackles that exact puzzle. The researchers propose a clever solution: instead of adding extra propellers or thrusters to the sides (which would be messy and heavy), they put a heavy weight inside the board that can slide back and forth. Think of it like a secret passenger who runs from the front of the boat to the back, or from left to right, to tip the board and steer it.

The Big Challenge: The "Ghost" in the Machine
The problem is that water is tricky. It pushes, pulls, and swirls in ways that are hard to predict. Plus, the sliding weight changes how the boat moves, creating a complex dance of physics. If you just program the boat to move left when it needs to turn, it might overshoot or get confused by a sudden wave.

To solve this, the team built an "intelligent" brain for the boat. They didn't just write a rigid set of rules; they gave the boat a neural network. You can think of this as a digital apprentice that learns on the job. As the boat moves, this apprentice watches for things it didn't expect—like a sudden gust of wind or a weird current—and instantly adjusts the sliding weight to keep the boat on track.

How It Works: The Line-of-Sight Guide
The boat uses a "Line-of-Sight" (LOS) strategy. Imagine you are walking down a winding path and you always look at a point a few steps ahead of you. You naturally turn your body to face that point. The boat does the same thing: it looks at a point on the desired path ahead and steers toward it.

The controller has two main jobs:

Pushing: It controls the electric motor to keep the boat moving forward.
Steering: It calculates exactly where the sliding mass needs to be to turn the boat.

To handle the unknowns, the neural network acts like a detective. It estimates the "ghost" forces (the unmodeled water friction and wind) and tells the sliding mass how to move to cancel them out. The math behind this uses something called a "Lyapunov" function, which is basically a mathematical guarantee that the boat won't go crazy, even if things get messy.

The Test Drive: Simulations Only
The researchers didn't put this on a real lake yet; they tested it in a computer simulation. They created a virtual world where the boat had to follow a giant circle (300 meters in radius) while dealing with different water conditions.

Calm Water: When the water was still, the intelligent boat followed the path almost perfectly.
Steady Current: They then added a steady stream of water pushing the boat sideways (0.3 m/s in one direction and -0.2 m/s in another). The intelligent boat adjusted the sliding mass to fight the current and stayed on course.
Wobbly Water: Finally, they made the water current change constantly, like a sine wave. Even then, the boat stayed on the path.

The Results: Smarter is Better
The team compared their "intelligent" controller against a "conventional" one (a standard controller without the learning brain). The results showed that the smart controller was significantly better.

In the calm water test, the smart controller reduced the total error by 27% compared to the standard one. Specifically, the "time-weighted absolute error" (a fancy way of measuring how far off the boat was over time) dropped from 65.9 m·s with the old method to 47.9 m·s with the new method. Even in the tricky, moving water, the smart controller kept the error low, around 50.9 m·s and 50.8 m·s for the different current scenarios.

What This Means
The paper suggests that using a sliding mass controlled by an intelligent neural network is a very effective way to steer unmanned boats, especially when the water is unpredictable. It proves that you don't need extra engines to steer; you just need a smart brain and a sliding weight.

However, it's important to remember that these results are from computer simulations. The boat hasn't actually been tested in a real lake or ocean yet. The authors show that the math works and the simulation looks great, but the real-world test is still waiting in the future. For now, this is a promising blueprint for how a robot surfboard might one day glide silently across the water, adjusting its own weight to dance with the waves.

Technical Summary: Intelligent Control for Path-Following of an Unmanned Mass-Centric Surface Vehicle

Problem Statement
The paper addresses the control and maneuverability challenges of surface vehicles with dynamically changing mass distributions, specifically focusing on autonomous electric hydrofoil boards (efoils). Unlike traditional vessels controlled by rudders or side thrusters, these mass-centric vehicles rely on shifting their center of gravity to steer. In the absence of a rider, this requires an automated mechanism to adjust the center of mass. The core challenge lies in designing a control system capable of handling the system's nonlinearities, unmodeled dynamics, and external disturbances (such as water currents) while ensuring accurate path-following.

Methodology
The authors propose an intelligent control scheme for the planar motion (surge, sway, and yaw) of the vessel. The control inputs are the propeller thrust force ( $F_p$ ) for forward motion and the position of a sliding mass ( $p$ ) to adjust the vessel's orientation.

Mathematical Modeling: The system dynamics are derived using Newton's Second Law and transformed into a body-fixed reference frame. The model accounts for hydrodynamic friction and the coupling between the sliding mass and the vessel's rotation.
Guidance Law: A Line-of-Sight (LOS) algorithm is employed to generate the desired heading angle ( $\alpha_d$ ) and path parameter dynamics based on the cross-track error and lookahead distance.
Control Architecture:
- Feedback Linearization: A Lyapunov-based nonlinear control law is designed to stabilize the surge speed and heading angle errors.
- Intelligent Compensation: To handle unmodeled dynamics and external disturbances, a single-hidden-layer Artificial Neural Network (ANN) is integrated. The ANN approximates the unknown terms in the dynamic equations.
- Adaptation: The weights of the neural network are updated online using an adaptation law derived from Lyapunov stability analysis. A projection algorithm is utilized to ensure the neural network weights remain bounded.
- Inverse Dynamics: The required torque ( $M_a$ ) calculated by the controller is converted into the physical position of the sliding mass ( $p$ ) using an inverse dynamics approach.

Key Contributions

Mass-Centric Control Strategy: The paper develops a specific control framework for a surface vehicle steered solely by a sliding mass mechanism, avoiding the need for additional thrusters.
Hybrid Control Scheme: The integration of a Lyapunov-based nonlinear controller with an ANN allows the system to compensate for uncertainties (including a 20% mass error in simulations) and external disturbances without requiring an exact hydrodynamic model.
Stability Proof: The authors provide a rigorous Lyapunov-based stability analysis, demonstrating that the tracking errors converge to a closed region and that the neural network weights remain bounded.

Simulation Results
The proposed controller was evaluated in Python simulations under three scenarios: calm water, constant water current, and time-varying water currents. The performance was compared against a "conventional" controller (the same structure but with neural network weights set to zero, i.e., $\eta_x = \eta_\alpha = 0$ ).

Path-Following Accuracy: In all scenarios, the intelligent controller maintained the vessel on the desired path with minimal error.
Performance Metrics: The Integral of Time-weighted Absolute Error (ITWAE) was used to quantify performance.
- In calm water, the intelligent controller achieved an ITWAE of 47.9 m·s, a 27% improvement over the conventional controller's 65.9 m·s.
- Under constant and time-varying current conditions, the intelligent controller maintained ITWAE values of approximately 50.9 m·s and 50.8 m·s, respectively, demonstrating robustness against environmental disturbances.
Adaptive Behavior: The results show that the intelligent controller effectively adjusts the sliding mass position and thrust force to counteract hydrodynamic forces and mass distribution changes, allowing the vessel to reach higher speeds as heading errors decrease.

Significance and Claims
The paper claims that the proposed intelligent controller effectively solves the path-following problem for unmanned surface vehicles subject to dynamic mass distribution changes. The significance lies in demonstrating that a mass-centric vehicle, which lacks traditional steering surfaces, can be autonomously controlled with high precision using a combination of Lyapunov-based control and neural network approximation. The authors assert that this approach offers a viable solution for autonomous efoils in surveillance and monitoring applications, where low noise and energy efficiency are critical, by relying on a single mechanical system (the sliding mass) powered by the existing battery. The work is presented as a simplified planar model, focusing on the feasibility of the control strategy rather than full 3D hydrodynamic complexity.

More like this