Data-driven path-following control for unmanned surface vehicles Authors: Je Wook Ryu, Seunghwan Han, Dowan Kim Published: 2025 ABSTRACT This paper develops a formulation of linear matrix inequalities (LMIs) for data-driven path following control of an unmanned surface vehicle (USV) subject to partially unknown dynamics. The proposed formulation is to decompose the whole system into a fully known guidance-error subsystem and sway yaw dynamics with the partially unknown components, rather than treating it as unknown. These unknown components are constrained to a data-consistent, matrix-ellipsoidal uncertainty set inferred from input state measurements, thereby avoiding explicit parameter identification. Sufficient conditions in the form of LMIs are derived to guarantee robust asymptotic stability of the associated closed-loop system under a matrix-ellipsoidal uncertainty set. An example is provided to examine that the proposed data-driven controller achieves the path-following performance comparable to that of a fully model-based controller across diverse reference paths. Introduction Path-following control of unmanned surface vehicles (USVs) refers to regulating the position and heading so that they converge to a prescribed geometric path without explicit time constraints1. A broad range of control strategies has been developed for this purpose, including adaptive, backstepping, fuzzy, and model predictive controls, most of which belong to the class of model-based approaches2–9. Moreover, advanced control methods extending model-based approaches, such as T S fuzzy model-based relaxed model predictive control10 and event-triggered adaptive global prescribed performance control11, have been proposed, thereby broadening the design space. An exact characterization of the underlying dynamics is, however, required by these methods, which is rarely attainable in practice 12. The hydrodynamic coefficients, associated with rigid-body dynamics, vary with operating conditions. Consequently, establishing a fixed parametric model of USV motion is, in general, infeasible. Data-driven control (DDC) has been increasingly established as a rigorous methodology for addressing modeling challenges, since controllers are synthesized directly from measured input output data without explicit parametric identification. Two main approaches to DDC are reinforcement learning (RL) and iterative learning control (ILC), which optimizes control policies through iterative interactions with the environment, proving effective in representing complex nonlinear dynamics, thereby allowing direct application to path-following control problems in USVs by using predictive modeling13,14. In previous work15,16, it has been shown that iterative learning control improves the tracking accuracy of USVs through iterative task execution. In addition to these approaches, a data-driven model predictive control that uses a dynamic fuzzy neural network learned from measured data has reported for path-following of USV17. Nevertheless, in these approaches, control performance cannot usually be guaranteed on untrained paths, and Lyapunov-based stability guarantees remain elusive18–21. In addition to learning-based approaches, linear matrix inequality approaches for DDC12,22,23 provide a convex framework for stability analysis. The approaches build on Willems’ fundamental lemma24 on a theoretical basis. The lemma establishes that the behavior of a linear time-invariant system can be characterized by a single persistently exciting input-output sequence. Based on this result24, De Persis and Tesi22 proposed LMI formulas that, with sufficiently rich data, guarantee stability, robustness, and optimality. Bisoffi et al.23 presented robust a synthesis for linear and polynomial systems, incorporating into the LMI conditions via Petersen’s lemma25,26 an ellipsoidal uncertainty set constructed from experimentally measured input-state data. Applying this result to a USV, Li et al.12 formulated a semidefinite programming problem for the tracking problem under simplified dynamics. Even with theoretical bases of LMI-based DDC12,22,23, inherent obstacles remain in the application to the path-following control of USVs. Its control system can be viewed as a cascade connection of the guidance subsystem and the rigid-body one. The guidance part is governed solely by path geometry and USV pose, and hence it is straightforward to model. By contrast, the rigid-body part is not easy to model due to parametric uncertainties and hydrodynamic effects. If these properties are ignored, the controller design becomes purely data-driven and unstructured, causing even the known parts of the system to be handled as uncertain. Also, for uniform path-following performance across admissible paths, the design is required to be independent of the prescribed reference path. Such a requirement is not readily achievable within the standard formulations of DDC. This paper aims to develop an LMI-based DDC for the path-following problem of USVs under partially unknown dynamics. This is achieved by extending the previous work23 from stabilization to path-following control. This extension is nontrivial. The previous work23 regarded the entire system as unknown and constructed a matrix ellipsoidal uncertainty set from input-state data. Instead, through separating known and unknown structures from mathematical models of USV, our design characterizes only the unknown part from the input-state data. By augmenting the guidance errors with the sway-yaw dynamics, we then obtain sufficient LMI conditions that guarantee robust asymptotic stabilization of the augmented system under a data-consistent, matrix ellipsoidal uncertainty set. The main contributions are summarized as follows. The proposed formulation represents only the unmodeled dynamics of USV by an ellipsoidal uncertainty set derived from input state data, while preserving all analytically known dynamics of the system. That is, it enables a data-driven representation only for the unmodeled dynamics without loss of structural information.The proposed design procedure of DDC is independent of the reference path. This independence guarantees uniform path-following performance across diverse paths, as validated through simulations. The proposed formulation represents only the unmodeled dynamics of USV by an ellipsoidal uncertainty set derived from input state data, while preserving all analytically known dynamics of the system. That is, it enables a data-driven representation only for the unmodeled dynamics without loss of structural information. The proposed formulation represents only the unmodeled dynamics of USV by an ellipsoidal uncertainty set derived from input state data, while preserving all analytically known dynamics of the system. That is, it enables a data-driven representation only for the unmodeled dynamics without loss of structural information. The proposed design procedure of DDC is independent of the reference path. This independence guarantees uniform path-following performance across diverse paths, as validated through simulations. The proposed design procedure of DDC is independent of the reference path. This independence guarantees uniform path-following performance across diverse paths, as validated through simulations. Notation: The relation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \succ Q$$\end{document} (resp. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \succeq Q$$\end{document}) indicates that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P-Q$$\end{document} is positive definite (resp. positive semidefinite). Similarly, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \prec Q$$\end{document} (resp. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P \preceq Q$$\end{document}) indicates that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Q-P$$\end{document} is positive definite (resp. positive semidefinite). For symmetric \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A = A^\top$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C = C^\top$$\end{document}, and general B, the block matrix\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{bmatrix}A & B^\top \\ B & C\end{bmatrix} \end{aligned}$$\end{document}is abbreviated using the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\star$$\end{document} notation as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{bmatrix}A & B^\top \\ \star & C\end{bmatrix}, \quad \begin{bmatrix}A & \star \\ B & C\end{bmatrix}. \end{aligned}$$\end{document}The operator \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{He}\{X\}:= X + X^\top$$\end{document} denotes the symmetric part of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X \in {\mathbb {R}}^{n\times n}$$\end{document}. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\times n$$\end{document} identity matrix is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I_n$$\end{document}, and the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p\times q$$\end{document} zero matrix by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$0_{p\times q}$$\end{document}. For index sets \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {I}},{\mathscr {J}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{({\mathscr {I}},{\mathscr {J}})}$$\end{document} denotes the submatrix of A with rows \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {I}}$$\end{document} and columns \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {J}}$$\end{document}. Problem formulation Consider an underactuated USV operating in the horizontal plane, whose motion is described by three degrees of freedom—surge u, sway v, and yaw rate r. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\eta =[x,y,\psi ]^\top \in {\mathbb {R}}^3$$\end{document} denote the earth-fixed pose and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[u,v,r]^\top \in {\mathbb {R}}^3$$\end{document} the body-fixed velocity. The kinematic equation is given by1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\dot{\eta }}} = J(\eta )\nu \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} J(\eta )= \begin{bmatrix} \cos \psi & -\sin \psi & 0\\ \sin \psi & \cos \psi & 0\\ 0 & 0 & 1 \end{bmatrix}. \end{aligned}$$\end{document}The 3-DOF equations of motion1 are2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} M{{\dot{\nu }}} + C(\nu )\nu + D(\nu )\nu = \tau \end{aligned}$$\end{document}with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau =[\tau _u,0,\tau _r]^\top \in {\mathbb {R}}^3$$\end{document}, i.e., actuation is available only in u and r, while v is unactuated, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M=M^\top \succ 0$$\end{document} denotes the inertia matrix (including added mass), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document} the Coriolis centripetal matrix, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(\nu )$$\end{document} the hydrodynamic damping. The matrices M, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(\nu )$$\end{document} are given by\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} M&= \begin{bmatrix} m_{11} & 0 & 0\\ 0 & m_{22} & m_{23}\\ 0 & m_{23} & m_{33} \end{bmatrix}, ~ C(\nu )= \begin{bmatrix} 0 & 0 & -m_{22}v - m_{23}r\\ 0 & 0 & m_{11}u\\ m_{22}v + m_{23}r & -m_{11}u & 0 \end{bmatrix},~ D(\nu )= \begin{bmatrix} d_{11}(u) & 0 & 0\\ 0 & d_{22}(v,r) & d_{23}(v,r)\\ 0 & d_{32}(v,r) & d_{33}(v,r) \end{bmatrix} \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} & m_{11}=m-X_{\dot{u}}, \quad m_{22}=m-Y_{\dot{v}}, \quad m_{23}=mx_g-Y_{\dot{r}}, \quad m_{33}=I_z-N_{\dot{r}} \\ & d_{11}(u) = -(X_u + X_{|u|u}|u|), \quad d_{22}(v,r) = -(Y_v + Y_{|v|v}|v| + Y_{|r|v}|r|), \quad d_{23}(v,r) = -(Y_r + Y_{|v|r}|v|)\\ & d_{32}(v,r) = -(N_v + N_{|v|v}|v| + N_{|r|v}|r|), \quad d_{33}(v,r) = -(N_r + N_{|v|r}|v| + N_{|r|r}|r|). \end{aligned}$$\end{document}Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {W}}:=\{p_k=(x_k,y_k)\}_{k=0}^{N}$$\end{document} denote a finite sequence of waypoints in the North East Down (NED) frame, with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_k\in {\mathbb {R}}^2$$\end{document}. The reference path is defined as the polyline obtained by connecting consecutive waypoints. Within each straight segment, the path tangential angle is constant,\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \gamma _p:= \text {atan2}(y_{k+1}-y_k, x_{k+1}-x_k) \end{aligned}$$\end{document}and the active segment index is updated upon the vessel’s entrance into the capture region of radius \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{\textrm{wp}}>0$$\end{document} centered at the corresponding waypoint. Assumption 1 The following conditions are imposed: The vessel operates in a compact set where M is constant and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu ),D(\nu )$$\end{document} are well-defined and locally Lipschitz in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu$$\end{document}.A surge inner loop enforces \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u(t)={{\bar{u}}}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{u}}}>0$$\end{document}. Thus the surge input is regarded as regulated and excluded from the subsequent design; only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r$$\end{document} is the control variable.During a data-collection experiment, the signals \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}(t)=[v(t),r(t)]^\top$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r(t)$$\end{document}, and either \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\nu }}}_{vr}(t)$$\end{document} or its differentiator estimate are available.The data collected under (A3) are persistently exciting, so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{bmatrix}X_0^\top&U_0^\top \end{bmatrix}^\top$$\end{document} has full row rank.Segment transitions are separated by a uniform dwell time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_d>0$$\end{document} (known), so that each segment remains active long enough for the Lyapunov decrease to accumulate across transitions.At segment switches, the guidance-error coordinates undergo a bounded reset \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^+=S_ke^-$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert S_k\Vert \le \rho$$\end{document} for some known \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \ge 1$$\end{document}, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}$$\end{document} remains continuous. The vessel operates in a compact set where M is constant and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu ),D(\nu )$$\end{document} are well-defined and locally Lipschitz in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu$$\end{document}. The vessel operates in a compact set where M is constant and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu ),D(\nu )$$\end{document} are well-defined and locally Lipschitz in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu$$\end{document}. A surge inner loop enforces \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u(t)={{\bar{u}}}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{u}}}>0$$\end{document}. Thus the surge input is regarded as regulated and excluded from the subsequent design; only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r$$\end{document} is the control variable. A surge inner loop enforces \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u(t)={{\bar{u}}}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{u}}}>0$$\end{document}. Thus the surge input is regarded as regulated and excluded from the subsequent design; only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r$$\end{document} is the control variable. During a data-collection experiment, the signals \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}(t)=[v(t),r(t)]^\top$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r(t)$$\end{document}, and either \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\nu }}}_{vr}(t)$$\end{document} or its differentiator estimate are available. During a data-collection experiment, the signals \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}(t)=[v(t),r(t)]^\top$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r(t)$$\end{document}, and either \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\nu }}}_{vr}(t)$$\end{document} or its differentiator estimate are available. The data collected under (A3) are persistently exciting, so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{bmatrix}X_0^\top&U_0^\top \end{bmatrix}^\top$$\end{document} has full row rank. The data collected under (A3) are persistently exciting, so that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{bmatrix}X_0^\top&U_0^\top \end{bmatrix}^\top$$\end{document} has full row rank. Segment transitions are separated by a uniform dwell time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_d>0$$\end{document} (known), so that each segment remains active long enough for the Lyapunov decrease to accumulate across transitions. Segment transitions are separated by a uniform dwell time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T_d>0$$\end{document} (known), so that each segment remains active long enough for the Lyapunov decrease to accumulate across transitions. At segment switches, the guidance-error coordinates undergo a bounded reset \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^+=S_ke^-$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert S_k\Vert \le \rho$$\end{document} for some known \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \ge 1$$\end{document}, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}$$\end{document} remains continuous. At segment switches, the guidance-error coordinates undergo a bounded reset \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^+=S_ke^-$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert S_k\Vert \le \rho$$\end{document} for some known \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \ge 1$$\end{document}, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}$$\end{document} remains continuous. Define the guidance error vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e:=[y_{e},e_{\psi }]^{\top }\in {\mathbb {R}}^{2}$$\end{document}, where the cross-track error is given by\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} y_{e}:=-(x-x_{k})\sin \gamma _{p}+(y-y_{k})\cos \gamma _{p} \end{aligned}$$\end{document}and the heading error by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi :=\psi -\gamma _p$$\end{document}. On straight segments (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\gamma }}}_p=0$$\end{document}), the exact kinematics of e are3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{bmatrix} \dot{y}_e\\ \dot{e}_\psi \end{bmatrix} = \begin{bmatrix} u\sin e_\psi + v\cos e_\psi \\ r \end{bmatrix}. \end{aligned}$$\end{document}In a neighborhood \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{|e_\psi |\le \epsilon _\psi \}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon _\psi >0$$\end{document}, a first-order Taylor expansion of (3), together with (A2), yields the local approximation4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\dot{e}}=A_{g}e+I_{2}\nu _{vr} \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{g}= \begin{bmatrix} 0& {\bar{u}} \\ 0& 0 \end{bmatrix}, \quad \nu _{vr}= \begin{bmatrix} v\\ r \end{bmatrix}. \end{aligned}$$\end{document}The sway yaw dynamics, retaining their nonlinear dependence rather than being locally linearized, are expressed in the form5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\dot{\nu }}_{vr} = \big (A_{vr}+{{\tilde{A}}}_{vr}\big )\nu _{vr} + \big (B_{vr}+{{\tilde{B}}}_{vr}\big )\tau _r \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(A_{vr},B_{vr})$$\end{document} collect the nominal contributions retained from M, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(\nu )$$\end{document} in (2), which may vanish if no such terms are trusted, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({{\tilde{A}}}_{vr},\tilde{B}_{vr})$$\end{document} aggregate the residual nonlinear and unmodeled effects, including u-dependent terms. Contributions involving the surge velocity u (e.g., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m_{11}u$$\end{document} in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document}) can either be evaluated at the regulated surge \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u={{\bar{u}}}$$\end{document} and absorbed into the nominal matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(A_{vr},B_{vr})$$\end{document}, or alternatively treated as part of the residual terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({{\tilde{A}}}_{vr},{{\tilde{B}}}_{vr})$$\end{document}. Both conventions are equivalently accommodated within the proposed data-driven synthesis framework. We define the augmented error state \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi := \begin{bmatrix}e^\top&\nu _{vr}^\top \end{bmatrix}^\top \in {\mathbb {R}}^4$$\end{document}, which combines the guidance error subsystem and the sway yaw subsystem into a unified representation suitable for robust stability analysis, and let the structured uncertainty block\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Delta := \begin{bmatrix}{{\tilde{A}}}_{vr}&{{\tilde{B}}}_{vr}\end{bmatrix}. \end{aligned}$$\end{document}From (4) and (5), the augmented dynamics becomes6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\dot{\chi }}} = A_{\textrm{ol}}\chi + B_{\textrm{ol}}\tau _r + E\Delta \big (F_{1}\chi +F_{2}\tau _{r}\big ) \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{\textrm{ol}}= \begin{bmatrix} A_g & I_{2}\\ 0_{2\times 2} & A_{vr} \end{bmatrix},\quad B_{\textrm{ol}}=\begin{bmatrix}0_{2\times 1}\\ B_{vr}\end{bmatrix},\quad E=\begin{bmatrix}0_{2\times 2}\\ I_{2}\end{bmatrix},\quad F_{1}=\begin{bmatrix}0_{2\times 2} & I_{2}\\ 0_{1\times 2} & 0_{1\times 2}\end{bmatrix},\quad F_{2}=\begin{bmatrix}0_{2\times 1}\\ 1\end{bmatrix}. \end{aligned}$$\end{document}Consider the linear state-feedback law7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tau _r = K\chi . \end{aligned}$$\end{document}Substituting this control law (7) into (6) yields the closed-loop augmented dynamics8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\dot{\chi }}} = \left( A_{\textrm{cl}}(K) + E\Delta F_{\textrm{cl}}(K)\right) \chi \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{\textrm{cl}}(K) = A_{\textrm{ol}}+B_{\textrm{ol}}K,\qquad F_\textrm{cl}(K)=F_{1}+F_{2}K. \end{aligned}$$\end{document}The representation (8) explicitly separates the nominal closed-loop dynamics from the structured uncertainty \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}, which is constrained to a data-consistent ellipsoidal set. Figure 1 provides an overview of the data driven control framework. The data matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(X_0,U_0,X_1)$$\end{document} are generated under a persistently exciting input \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r$$\end{document}, and the residual term is computed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\tilde{X}}}_1 = X_1 - A_{vr}X_0 - B_{vr}U_0$$\end{document}. Under Assumptions (A3)–(A4), the identified residual block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document} belongs to the data consistent ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} constructed from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(X_0,U_0,X_1)$$\end{document}. Incorporating \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in \mathscr {E}$$\end{document} into the augmented dynamics \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\chi }}} = A_{\textrm{ol}}\chi + B_{\textrm{ol}}\tau _r + E\Delta (F_1\chi + F_2\tau _r)$$\end{document}, a Lyapunov function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V(\chi )=\chi ^\top P^{-1}\chi$$\end{document} will later be shown to satisfy a uniform decay condition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}(\chi )\le -\alpha V(\chi )$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi \in {\mathscr {D}}$$\end{document} and all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document} through a single LMI constraint. Fig. 1Conceptual overview of the data-driven control framework. Conceptual overview of the data-driven control framework. Problem 1 Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {W}}=\{p_k\}_{k=0}^{N}$$\end{document} be a finite sequence of waypoints with associated path–tangential angles \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _p$$\end{document}. Consider the uncertain closed-loop system (8) under Assumptions 1. From a dataset collected according to Assumption (A3), a data-consistent ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} is obtained, which contains all admissible realizations of the residual block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}. Define \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}\subset {\mathbb {R}}^4$$\end{document} as a prescribed compact forward–invariant set guaranteed by Assumptions 1, within which the local guidance model (4) and sway–yaw dynamics (5) remain valid. The robust path-following control problem is to synthesize a linear state-feedback law (7) such that: The origin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =0$$\end{document} of the closed-loop dynamics (8) is robustly asymptotically stable on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}.Consequently, on each path segment the guidance error satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e(t)\rightarrow 0$$\end{document} as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow \infty$$\end{document}, and the convergence is preserved across segment transitions. The origin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =0$$\end{document} of the closed-loop dynamics (8) is robustly asymptotically stable on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. The origin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =0$$\end{document} of the closed-loop dynamics (8) is robustly asymptotically stable on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. Consequently, on each path segment the guidance error satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e(t)\rightarrow 0$$\end{document} as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow \infty$$\end{document}, and the convergence is preserved across segment transitions. Consequently, on each path segment the guidance error satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e(t)\rightarrow 0$$\end{document} as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow \infty$$\end{document}, and the convergence is preserved across segment transitions. Main result We proceed under Assumptions 1, which ensure admissible paths, well-posed closed-loop dynamics, and data consistency. The following lemma constructs the ellipsoidal uncertainty set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} employed in the main theorem, based on the data-driven framework of Bisoffi et al.23. Lemma 1 Consider the sway yaw dynamics (5), and suppose that the trajectories \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\{\nu _{vr}(t_i),\tau _r(t_i),{{\dot{\nu }}}_{vr}(t_i)\}_{i=0}^{T-1}$$\end{document} are collected over T samples. Define the data matrices\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} X_0&:= \begin{bmatrix}\nu _{vr}(t_0)&\cdots&\nu _{vr}(t_{T-1})\end{bmatrix}, \quad U_0 := \begin{bmatrix}\tau _r(t_0)&\cdots&\tau _r(t_{T-1})\end{bmatrix}, \quad X_1 := \begin{bmatrix}{{\dot{\nu }}}_{vr}(t_0)&\cdots&{{\dot{\nu }}}_{vr}(t_{T-1})\end{bmatrix}. \end{aligned}$$\end{document}Subtracting the nominal contribution yields\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\widetilde{X}}}_1:= X_1 - A_{vr}X_0 - B_{vr}U_0. \end{aligned}$$\end{document} Then the following statements hold: (i)There exists a disturbance matrix W with bounded energy, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$WW^\top \preceq \Sigma \Sigma ^{\top }$$\end{document} such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\widetilde{X}}}_1 = \Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix} + W \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta = \begin{bmatrix}{{\tilde{A}}}_{vr}&\tilde{B}_{vr}\end{bmatrix}$$\end{document} is the uncertainty block.(ii)If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {rank}\begin{bmatrix}X_0\\ U_0\end{bmatrix} = \dim (\nu _{vr}) + \dim (\tau _r) \end{aligned}$$\end{document} then every admissible realization of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document} consistent with the data belongs to the ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathscr {E}} = \Big \{ (\zeta + {\textbf{A}}^{-1/2}\Gamma \textbf{Q}^{1/2})^\top : \Vert \Gamma \Vert \le 1 \Big \}, \end{aligned}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\textbf{A}} = \begin{bmatrix}X_0\\ U_0\end{bmatrix}\begin{bmatrix}X_0\\ U_0\end{bmatrix}^\top \succ 0, \quad {\textbf{B}} = -\begin{bmatrix}X_0\\ U_0\end{bmatrix}\widetilde{X}_1^\top , \quad {\textbf{C}} = {{\widetilde{X}}}_1{{\widetilde{X}}}_1^\top - \Sigma \Sigma ^{\top } \end{aligned}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \zeta = -{\textbf{A}}^{-1}{\textbf{B}},\qquad {\textbf{Q}} = \textbf{B}^\top {\textbf{A}}^{-1}{\textbf{B}} - {\textbf{C}} \succeq 0. \end{aligned}$$\end{document} (i)There exists a disturbance matrix W with bounded energy, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$WW^\top \preceq \Sigma \Sigma ^{\top }$$\end{document} such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\widetilde{X}}}_1 = \Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix} + W \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta = \begin{bmatrix}{{\tilde{A}}}_{vr}&\tilde{B}_{vr}\end{bmatrix}$$\end{document} is the uncertainty block. There exists a disturbance matrix W with bounded energy, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$WW^\top \preceq \Sigma \Sigma ^{\top }$$\end{document} such that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\widetilde{X}}}_1 = \Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix} + W \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta = \begin{bmatrix}{{\tilde{A}}}_{vr}&\tilde{B}_{vr}\end{bmatrix}$$\end{document} is the uncertainty block. (ii)If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {rank}\begin{bmatrix}X_0\\ U_0\end{bmatrix} = \dim (\nu _{vr}) + \dim (\tau _r) \end{aligned}$$\end{document} then every admissible realization of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document} consistent with the data belongs to the ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathscr {E}} = \Big \{ (\zeta + {\textbf{A}}^{-1/2}\Gamma \textbf{Q}^{1/2})^\top : \Vert \Gamma \Vert \le 1 \Big \}, \end{aligned}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\textbf{A}} = \begin{bmatrix}X_0\\ U_0\end{bmatrix}\begin{bmatrix}X_0\\ U_0\end{bmatrix}^\top \succ 0, \quad {\textbf{B}} = -\begin{bmatrix}X_0\\ U_0\end{bmatrix}\widetilde{X}_1^\top , \quad {\textbf{C}} = {{\widetilde{X}}}_1{{\widetilde{X}}}_1^\top - \Sigma \Sigma ^{\top } \end{aligned}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \zeta = -{\textbf{A}}^{-1}{\textbf{B}},\qquad {\textbf{Q}} = \textbf{B}^\top {\textbf{A}}^{-1}{\textbf{B}} - {\textbf{C}} \succeq 0. \end{aligned}$$\end{document} If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {rank}\begin{bmatrix}X_0\\ U_0\end{bmatrix} = \dim (\nu _{vr}) + \dim (\tau _r) \end{aligned}$$\end{document} then every admissible realization of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document} consistent with the data belongs to the ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\mathscr {E}} = \Big \{ (\zeta + {\textbf{A}}^{-1/2}\Gamma \textbf{Q}^{1/2})^\top : \Vert \Gamma \Vert \le 1 \Big \}, \end{aligned}$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\textbf{A}} = \begin{bmatrix}X_0\\ U_0\end{bmatrix}\begin{bmatrix}X_0\\ U_0\end{bmatrix}^\top \succ 0, \quad {\textbf{B}} = -\begin{bmatrix}X_0\\ U_0\end{bmatrix}\widetilde{X}_1^\top , \quad {\textbf{C}} = {{\widetilde{X}}}_1{{\widetilde{X}}}_1^\top - \Sigma \Sigma ^{\top } \end{aligned}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \zeta = -{\textbf{A}}^{-1}{\textbf{B}},\qquad {\textbf{Q}} = \textbf{B}^\top {\textbf{A}}^{-1}{\textbf{B}} - {\textbf{C}} \succeq 0. \end{aligned}$$\end{document} Proof The result follows by applying the quadratic characterization of data-consistent matrices in23 to the residual dynamics (5), with the modeled part \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(A_{vr},B_{vr})$$\end{document} subtracted. Full details are given in23. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square$$\end{document} Remark 1 The matrix inequality in Lemma 1 directly follows from the data equation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widetilde{X}}}_1=\Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix}+W$$\end{document} with bounded residual \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$WW^\top \preceq \Sigma \Sigma ^\top$$\end{document}. By expressing \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W=\Sigma F$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert F\Vert \le 1$$\end{document}, we obtain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma ^{-1}\left( {{\widetilde{X}}}_1-\Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix}\right) =F$$\end{document}, which implies the quadratic constraint\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Sigma ^{-1} \left( {{\widetilde{X}}}_1-\Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix}\right) \left( \widetilde{X}_1-\Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix}\right) ^{ \top } \Sigma ^{-\top }\preceq I. \end{aligned}$$\end{document}This relation ensures that every \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document} consistent with the data satisfies the inequality \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {\textbf{A}}\Delta ^\top +\Delta \textbf{B}^\top +{\textbf{B}}\Delta ^\top +{\textbf{C}}\preceq 0$$\end{document}, and therefore lies within the ellipsoidal uncertainty set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document}. Remark 2 Although the main results of Bisoffi et al.23 are stated in discrete time, a continuous-time counterpart is also discussed therein. In the present setting, part of the dynamics \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(A_{vr},B_{vr})$$\end{document} is retained explicitly from the physical model. Discretization of these known terms would introduce additional approximation errors, which are avoided by adopting a continuous-time formulation. Accordingly, Lemma 1 is stated in continuous time based on measurements or estimates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\nu }}}_{vr}$$\end{document}, with differentiation noise absorbed into the disturbance matrix W through \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$WW^\top \preceq \Sigma \Sigma ^\top$$\end{document}. This choice ensures a consistent data-driven uncertainty description while preserving the accuracy of the known nominal dynamics. Remark 3 Persistent excitation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{bmatrix}X_0^\top&U_0^\top \end{bmatrix}^\top$$\end{document} ensures \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\textbf{A}}\succ 0$$\end{document}. Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\widetilde{X}}}_1 = X_1 - A_{vr}X_0 - B_{vr}U_0$$\end{document}, one must either measure or estimate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\nu }}}_{vr}=[\dot{v},\dot{r}]^\top$$\end{document} reliably; only the residual part is then attributed to the uncertainty block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}. In practice, such derivative signals can be readily obtained from standard USV sensor suites. The yaw rate r is directly measured by an IMU, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{r}$$\end{document} (yaw acceleration) is typically available from the same unit. The sway velocity v can be estimated by GPS/IMU fusion or Doppler log sensors, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{v}$$\end{document} can be obtained by numerical differentiation with appropriate filtering. Hence, the continuous-time formulation of Lemma 1 is practically feasible with existing sensor technology. We now derive a convex condition that guarantees robust asymptotic stability of the closed-loop system (8) for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. Theorem 1 Consider the uncertain closed-loop system (8) under Assumptions 1. Let the state-feedback law be given by (7), and assume that the residual block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document} belongs to the ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} of Lemma 1. If there exist matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P=P^\top \succ 0$$\end{document} and Y such that9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \begin{bmatrix} \operatorname {He}\left\{ A_{\textrm{ol}}P+B_{\textrm{ol}}Y - E{\textbf{B}}^{\top }{\textbf{A}}^{-1}(F_{1}P+F_{2}Y)\right\} + E {\textbf{Q}} E^\top & \star \\ F_{1}P+F_{2}Y & - {\textbf{A}} \end{bmatrix} \prec 0, \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({\textbf{A}},{\textbf{B}},{\textbf{C}})$$\end{document} are the data-dependent matrices of Lemma 1, then the feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K=YP^{-1}$$\end{document} renders the origin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi =0$$\end{document} robustly asymptotically stable on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. Consequently, the guidance error e converges to zero on each path segment, and this convergence is preserved across segment transitions, thereby ensuring global path-following stability for all admissible uncertainties. Proof Consider the uncertain closed-loop system (8) on the forward invariant set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} of Problem 1, under the feedback law (7) and any admissible \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. Considier\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} V(\chi )=\chi ^\top P^{-1}\chi \end{aligned}$$\end{document}as a Lyapunov function candidate for (8). Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P=P^{T}\succ 0$$\end{document}, V is positive definite on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document}. By Assumption 1, the closed-loop system is locally Lipschitz on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document}, ensuring existence and uniqueness of solutions and differentiability of V along trajectories. Differentiating V along (8) yields10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \dot{V}(\chi )=\chi ^\top \operatorname {He}\left\{ P^{-1}A_{\textrm{cl}}(K)+P^{-1}E\Delta F_\textrm{cl}(K)\right\} \chi . \end{aligned}$$\end{document}Hence \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}<0$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi \ne 0$$\end{document} and all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document} is guaranteed if11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {He}\left\{ A_{\textrm{cl}}(K)P+E\Delta F_{\textrm{cl}}(K)P\right\} \prec 0. \end{aligned}$$\end{document}Introducing the variable change \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y=KP$$\end{document}, and recalling that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_\textrm{cl}(K)=A_{\textrm{ol}}+B_{\textrm{ol}}K$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{\textrm{cl}}(K)=F_1+F_2K$$\end{document}, condition (11) becomes12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {He}\left\{ A_{\textrm{ol}}P+B_{\textrm{ol}}Y+E\Delta (F_1P+F_2Y)\right\} \prec 0. \end{aligned}$$\end{document}This inequality is affine in (P, Y) and linear in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}. By Lemma 1, each admissible uncertainty admits the decomposition\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \Delta =(\zeta +{\textbf{A}}^{-1/2}\Gamma {\textbf{Q}}^{1/2})^\top =\zeta ^\top +{\textbf{Q}}^{1/2}\Gamma ^\top {\textbf{A}}^{-1/2}, \qquad \Vert \Gamma \Vert \le 1. \end{aligned}$$\end{document}Substitution into (12) gives13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {He}\left\{ A_{\textrm{ol}}P+B_{\textrm{ol}}Y +E\zeta ^\top (F_1P+F_2Y) +E\textbf{Q}^{1/2}\Gamma ^\top {\textbf{A}}^{-1/2}(F_1P+F_2Y)\right\} \prec 0 . \end{aligned}$$\end{document}By Petersen’s lemma25,26, for any \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda >0$$\end{document},\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {He}\left\{ (E{\textbf{A}}^{-1/2})\Gamma ({\textbf{Q}}^{1/2}(F_1P+F_2Y))\right\} \preceq \lambda E{\textbf{Q}} E^\top +\lambda ^{-1}(F_1P+F_2Y)^\top {\textbf{A}}^{-1}(F_1P+F_2Y). \end{aligned}$$\end{document}Therefore, a sufficient condition for (13) is14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \operatorname {He}\left\{ A_{\textrm{ol}}P+B_{\textrm{ol}}Y+E\zeta ^\top (F_1P+F_2Y)\right\} +\lambda E{\textbf{Q}} E^\top +\lambda ^{-1}(F_1P+F_2Y)^\top \textbf{A}^{-1}(F_1P+F_2Y)\prec 0. \end{aligned}$$\end{document}Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda >0$$\end{document} is a free scalar, the LMI is positively homogeneous in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(P,Y,\lambda )$$\end{document}. By an appropriate scaling of the decision variables, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda$$\end{document} can be fixed to unity without loss of generality. Substituting \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\zeta ^{\top }=-{\textbf{B}}^{\top } \textbf{A}^{-1}$$\end{document} from Lemma 1, and applying the Schur complement, (14) is equivalently expressed as the LMI condition (9). Feasibility of (9) then guarantees that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}(\chi )<0$$\end{document} on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi \ne 0$$\end{document} and all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}, so V is a strict Lyapunov function on each fixed path segment, implying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi (t)\rightarrow 0$$\end{document} as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t\rightarrow \infty$$\end{document} in the absence of segment switching. At each segment transition \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_k$$\end{document}, corresponding to a change in the path tangential angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma _p$$\end{document}, the guidance-error coordinates reset as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e^+=S_ke^-$$\end{document} by condition (A6) of Assumption 1, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e=[y_e,e_\psi ]^\top$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_k$$\end{document} is the coordinate transformation induced by the new segment. Since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$S_k$$\end{document} is nonsingular with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Vert S_k\Vert \le \rho$$\end{document} for some \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \ge 1$$\end{document}, one has \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V(t_k^+)\le \rho V(t_k^-)$$\end{document}, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}$$\end{document} remains continuous. By condition (A5), the dwell time satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_{k+1}-t_k\ge T_d>0$$\end{document}, ensuring strict decrease of V on each segment:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} V(t_{k+1}^-)\le e^{-\alpha (t_{k+1}-t_k)}V(t_k^+)\le e^{-\alpha T_d}V(t_k^+). \end{aligned}$$\end{document}Combining the flow and reset inequalities yields\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} V(t_{k+1}^+)\le \rho \textrm{e}^{-\alpha T_d}V(t_k^+). \end{aligned}$$\end{document}If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \textrm{e}^{-\alpha T_d}<1$$\end{document}, repeated application yields \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$V(t_k^+)\rightarrow 0$$\end{document} as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k\rightarrow \infty$$\end{document}, and consequently \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi (t)\rightarrow 0$$\end{document} globally along the waypoint sequence. Therefore, the closed-loop system (8) is robustly asymptotically stable on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document}, and the guidance error e converges globally to zero for all admissible uncertainties \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. Furthermore, the LMI condition (9) can be enforced with a strict margin, for instance by requiring the upper-left block to satisfy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\preceq -\epsilon I$$\end{document} for some \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon >0$$\end{document}. This ensures \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}(\chi )\le -\epsilon \Vert \chi \Vert ^2$$\end{document}, and by the quadratic bounds on V it follows that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V} \le -\alpha V$$\end{document} with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \ge \epsilon /\kappa (P)$$\end{document}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\kappa (P)=\lambda _{\max }(P)/\lambda _{\min }(P)$$\end{document}. Hence the dwell-time requirement \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\rho \textrm{e}^{-\alpha T_d}<1$$\end{document} admits a direct interpretation in terms of the design margin \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon$$\end{document} and the Lyapunov matrix P. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\square$$\end{document} Remark 4 The set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}\subset {\mathbb {R}}^4$$\end{document} denotes a compact operating region in which (i) the local guidance model (4) and the sway yaw split (5) are valid, and (ii) the data-driven uncertainty description \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document} from Lemma 1 applies. A convenient choice is a Lyapunov sublevel set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}=\{\chi :V(\chi )\le c\}$$\end{document}, where feasibility of the robust LMI (9) yields a decay estimate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}(\chi )\le -\alpha V(\chi )$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi \in \mathscr {D}$$\end{document} and all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}, with some \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha >0$$\end{document}. Alternatively, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} may be defined as the intersection between the admissible bounds on the guidance error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e = [\,y_e,\, e_\psi \,]^\top$$\end{document}, which are determined from the path geometry and waypoint capture radii, and the validity domain of the measured datasets \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(X_0,U_0,X_1)$$\end{document} employed to construct the data–consistent ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document}. If, due to modeling or measurement errors, the trajectory approaches \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\partial {\mathscr {D}}$$\end{document}, one may either switch to a conservative safe mode (heading hold with speed reduction) or re-estimate a larger feasible region by enlarging the data-consistent set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} and re-solving (9). Remark 5 Assume that the measured velocity derivative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\dot{\nu }}_{vr}^{\,\mathrm m}(t_k)$$\end{document} contains bounded additive noise \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_d(t_k)\in {\mathbb {R}}^2$$\end{document}, so that\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\dot{\nu }}_{vr}^{\,\mathrm m}(t_k) = {\dot{\nu }}_{vr}(t_k) + w_d(t_k), \qquad w_d(t_k) w_d(t_k)^\top \preceq \Sigma _d \Sigma _d^\top \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma _d=\sigma _d I$$\end{document} represents the differentiator accuracy bound. Collecting N samples yields the data matrix\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} X_1^{\mathrm m} = X_1 + W_d \end{aligned}$$\end{document}where\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} X_1^{\mathrm m} =\begin{bmatrix} {\dot{\nu }}_{vr}^{\,\mathrm m}(t_0)&{\dot{\nu }}_{vr}^{\,\mathrm m}(t_1)&\cdots&{\dot{\nu }}_{vr}^{\,\mathrm m}(t_{N-1}) \end{bmatrix}, \quad X_1=\begin{bmatrix} {\dot{\nu }}_{vr}(t_0)&{\dot{\nu }}_{vr}(t_1)&\cdots&{\dot{\nu }}_{vr}(t_{N-1}) \end{bmatrix} \end{aligned}$$\end{document}and\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} W_d=[\,w_d(t_0)\;w_d(t_1)\;\cdots \;w_d(t_{N-1})\,] \end{aligned}$$\end{document}collects the differentiation errors over the data horizon and satisfies \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W_d W_d^\top \preceq \Sigma _d \Sigma _d^\top .$$\end{document} The corresponding noisy data relation then takes the form\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\widetilde{X}}}_1^{\mathrm m}&= X_1^{\mathrm m} - A_{vr}X_0 - B_{vr}U_0 \nonumber \\&= \Delta \begin{bmatrix} X_0\\ U_0 \end{bmatrix} + W_d, \qquad W_dW_d^\top \preceq \Sigma _d\Sigma _d^\top . \end{aligned}$$\end{document}This is identical to the data equation of Lemma 1 with the identification \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W:=W_d$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma :=\Sigma _d$$\end{document}, namely\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\widetilde{X}}}_1 = \Delta \begin{bmatrix}X_0\\ U_0\end{bmatrix} + W, \qquad WW^\top \preceq \Sigma \Sigma ^\top . \end{aligned}$$\end{document}Hence, the measurement noise \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$w_d(t_k)$$\end{document} in the differentiated velocity is rigorously embedded in the bounded disturbance matrix W of Lemma 1. Consequently, the resulting data consistent ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} and the LMI condition (9) of Theorem 1 remain valid without modification. In particular, bounded differentiation noise merely enlarges \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} monotonically while preserving the negativity of the Lyapunov derivative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\dot{V}(\chi )$$\end{document}, which ensures robust asymptotic stability on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document}. Remark 6 The proposed data–driven controller requires no online optimization or data storage. The LMI (9) in Theorem 1, with decision variables \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P=P^{\top } \succ 0$$\end{document} and Y, is solved offline once to obtain the constant feedback gain K. The experimental data \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(X_0,U_0,X_1)$$\end{document} and the derived matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({\textbf{A}},{\textbf{B}},{\textbf{Q}})$$\end{document} are used only offline during the synthesis of K and are not stored or updated on the embedded platform. During operation, the control input is computed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r=K\chi$$\end{document}, which involves only a few arithmetic operations for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi \in {\mathbb {R}}^4$$\end{document}. The computation is independent of the data length T, and only \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K \in {\mathbb {R}}^{1\times 4}$$\end{document} is stored. As a result, the runtime cost and memory demand are negligible. Example: path-following control of CyberShip II27 We illustrate the proposed LMI-based data-driven controller on the nonlinear dynamics of the CyberShip \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{II}$$\end{document}27. The objective is to quantify path-following performance under fully known, partially unknown, and fully unknown error dynamics, while explicitly demonstrating consistency with the Lyapunov-based stability guarantees established in Theorem 1. The vehicle is characterized by the coefficients27\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} & m = 23.800, \quad I_z = 1.760, \quad x_g = 0.046, \quad X_{\dot{u}} = -2, \quad X_u = -0.72253, \quad X_{|u|u} = -1.32742,\\ & Y_{\dot{v}} = -10, \quad Y_v = -0.88965, \quad Y_{{\dot{r}}} = 0,\quad Y_r = -7.25,\quad Y_{|v|v} = -36.47287, \quad Y_{|r|v} = -0.805, \quad Y_{|v|r} = -0.845,\\ & N_{\dot{v}} = 0, \quad N_{\dot{r}} = -1, \quad N_v = 0.0313, \quad N_r = -1.9,\quad N_{|v|v} = 3.95645, \quad N_{|r|v} = 0.13, \quad N_{|v|r} = 0.08, \quad N_{|r|r} = -0.75. \end{aligned}$$\end{document}Two reference paths are considered: (i)A waypoint path defined by six prescribed waypoints \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (x_{1},y_{1})=(50,50),\; (x_{2},y_{2})=(0,0),\; (x_{3},y_{3})=(50,0),\; (x_{4},y_{4})=(0,0),\; (x_{5},y_{5})=(0,50),\; (x_{6},y_{6})=(0,0). \end{aligned}$$\end{document}(ii)A parametric path generated by six uniformly spaced waypoints defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (x_k,y_k)=\big (6\sin (0.4\theta _k)+0.9\theta _k,\;\theta _k\big ), \quad k=1,\dots ,6 \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _k\in \{0,10,20,30,40,50\}$$\end{document}.In both scenarios, the vessel maintains a constant surge velocity of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\bar{u}}}=0.4~\mathrm {m/s}$$\end{document} under the feedback linearization controller\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tau _u = -K_u(u-{\bar{u}}) - (X_u + X_{|u|u}|u|)u - (m_{22}v+m_{23}r)r, \quad K_u=11.6465 \end{aligned}$$\end{document}which isolates the sway yaw subsystem for analysis and control synthesis. A waypoint is considered reached whenever the vessel enters the capture region of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r_{\textrm{wp}}=1~\textrm{m}$$\end{document}. Upon reaching a waypoint, the next one on the path is activated as the new target. Initial states are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(u,v,r)=(0.4,1,1)$$\end{document} in both scenarios. (i)A waypoint path defined by six prescribed waypoints \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (x_{1},y_{1})=(50,50),\; (x_{2},y_{2})=(0,0),\; (x_{3},y_{3})=(50,0),\; (x_{4},y_{4})=(0,0),\; (x_{5},y_{5})=(0,50),\; (x_{6},y_{6})=(0,0). \end{aligned}$$\end{document} A waypoint path defined by six prescribed waypoints \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (x_{1},y_{1})=(50,50),\; (x_{2},y_{2})=(0,0),\; (x_{3},y_{3})=(50,0),\; (x_{4},y_{4})=(0,0),\; (x_{5},y_{5})=(0,50),\; (x_{6},y_{6})=(0,0). \end{aligned}$$\end{document} (ii)A parametric path generated by six uniformly spaced waypoints defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (x_k,y_k)=\big (6\sin (0.4\theta _k)+0.9\theta _k,\;\theta _k\big ), \quad k=1,\dots ,6 \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _k\in \{0,10,20,30,40,50\}$$\end{document}. A parametric path generated by six uniformly spaced waypoints defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} (x_k,y_k)=\big (6\sin (0.4\theta _k)+0.9\theta _k,\;\theta _k\big ), \quad k=1,\dots ,6 \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta _k\in \{0,10,20,30,40,50\}$$\end{document}. To construct the data-driven uncertainty set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} of Lemma 1, we collected \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T=2000$$\end{document} samples of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\nu _{vr},\tau _r,{{\dot{\nu }}}_{vr})$$\end{document} with sampling step \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta t=0.01~\textrm{s}$$\end{document}. In the present work, these trajectories are generated by simulating the nonlinear dynamics (2) using forward Euler integration:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \nu _{vr,i+1} = \nu _{vr,i} + {{\dot{\nu }}}_{vr,i}\Delta t, \quad i=0,\dots ,T-1 \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr,i}=[v_i,r_i]^\top$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\dot{\nu }}}_{vr,i}$$\end{document} is obtained from (2). It is important to note that the same procedure can be applied in experiments, since only input and velocity measurements are required, together with either direct or estimated derivatives of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}$$\end{document}. During data collection, the surge velocity u was regulated to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bar{u}}=0.4~\mathrm {m/s}$$\end{document} by the feedback linearization controller. so that the sway–yaw subsystem could be isolated for identification. To guarantee persistence of excitation, the yaw control input \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _r$$\end{document} was assigned the test signal\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \tau _r(t) = 1.37\sin (0.3t + \pi /7) + 0.3\sin (0.8t + \pi /3) + 0.4\sin (1.3t + \pi /5). \end{aligned}$$\end{document}The collected trajectories yield the following data matrices:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} X_{0}&= \begin{bmatrix} 1.0000 & 0.9831 & 0.9666 & \cdots \\ 1.0000 & 1.0049 & 1.0082 & \cdots \end{bmatrix} \in {\mathbb {R}}^{2\times T}\\ U_0&= \begin{bmatrix} 1.5960&1.2953&0.8529&\cdots \end{bmatrix}\in {\mathbb {R}}^{1\times T}\\ X_1&= \begin{bmatrix} -1.6900 & -1.6501 & -1.6092 & \cdots \\ 0.4913 & 0.3301 & 0.1208 & \cdots \end{bmatrix}\in {\mathbb {R}}^{2\times T}. \end{aligned}$$\end{document}which correspond to the definition in Lemma 1. The values shown above are illustrative excerpts of the data matrices. The complete matrices can be regenerated using the Euler scheme and the input signal specified earlier. From these data, the residual matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\widetilde{X}_1 = X_1 - A_{vr}X_0 - B_{vr}U_0$$\end{document} is computed, and the data-dependent matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({\textbf{A}},{\textbf{B}},{\textbf{C}})$$\end{document} of Lemma 1 are formed. This yields the ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} containing all admissible realizations of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta =[{{\tilde{A}}}_{vr}\;{{\tilde{B}}}_{vr}]$$\end{document}, which is then employed in Theorem 1 to synthesize a robust stabilizing gain K. For comparison with the proposed data-driven controller, we evaluate two linear model-based designs on the same linearized model: (i) a nominal design that assumes all coefficients of the dynamic model are exactly known, and (ii) a robust design that models the nonlinear damping coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{r}$$\end{document} as unknown . In these cases, nonlinear dependence \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu _{vr}$$\end{document} in the error dynamics is replaced by its linearized form, and the corresponding linearized system is expressed as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {\dot{\chi }} = A_{L}\chi + B_{L}\tau _{r} \end{aligned}$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau _{r}=K_{L}\chi$$\end{document}. The matrices \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{L}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$B_{L}$$\end{document} are defined as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{L} = \begin{bmatrix} 0 & {\bar{u}} & 1 & 0\\ 0 & 0 & 0 & 1\\ 0 & 0 & \tfrac{1}{\Delta _{L}}(m_{33}Y_{v} + (m_{22} - m_{11})m_{23}{\bar{u}} - m_{23}N_{v}) & \tfrac{1}{\Delta _{L}}(m_{33}Y_{r} + (m_{23}^{2} - m_{11}m_{33}){\bar{u}} - m_{23}N_{r})\\ 0 & 0 & \tfrac{1}{\Delta _{L}}(-m_{23}Y_{v} + (m_{11} - m_{22})m_{22}{\bar{u}} + m_{22}N_{v}) & \tfrac{1}{\Delta _{L}}(-m_{23}Y_{r} + (m_{11} - m_{22})m_{23}{\bar{u}} + m_{22}N_{r}) \end{bmatrix},~ B_{L} = \begin{bmatrix} 0\\ 0\\ -m_{23}/\Delta _{L}\\ m_{22}/\Delta _{L} \end{bmatrix} \end{aligned}$$\end{document}with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta _{L} = m_{22}m_{33} - m_{23}^{2}$$\end{document}. (i)In the nominal design case, the feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document} derived via the LMI formulation is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_{L} = \begin{bmatrix} -1.8675&-25.5499&-37.7847&-2.8618 \end{bmatrix} \end{aligned}$$\end{document} which assigns the closed-loop eigenvalues as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{1,2} = -0.5653 \pm 1.0699i,\qquad \lambda _{3,4} =-0.0543 \pm 0.0445i \end{aligned}$$\end{document} thereby confirming asymptotic stability.(ii)In the robust design case, only the nonlinear damping coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_r$$\end{document} is regarded as uncertain. Accordingly, the matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{L}$$\end{document} in the linearized system is rewritten as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{L} = A_{u} + \Delta A_{u} \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&A_{u} = \begin{bmatrix} 0 & {\bar{u}} & 1 & 0\\ 0 & 0 & 0 & 1\\ 0 & 0 & \frac{1}{\Delta _{L}}(m_{33}Y_{v} + (m_{22} - m_{11})m_{23}{\bar{u}} - m_{23}N_{v}) & \frac{1}{\Delta _{L}}(m_{33}Y_{r} + (m_{23}^{2} - m_{11}m_{33}){\bar{u}})\\ 0 & 0 & \frac{1}{\Delta _{L}}(-m_{23}Y_{v} + (m_{11} - m_{22})m_{22}{\bar{u}} + m_{22}N_{v}) & \frac{1}{\Delta _{L}}(-m_{23}Y_{r} + (m_{11} - m_{22})m_{23}{\bar{u}}) \end{bmatrix} \end{aligned}$$\end{document} is the known dynamics matrix, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta A_u = DG(t)H$$\end{document} is a structured perturbation matrix, and G(t) is the uncertain block satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$||G(t)||<1$$\end{document}. The matrices D and H are given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} D=nI_{4\times 4},\quad H=\frac{1}{\Delta _L}\begin{bmatrix} 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & -m_{23}\\ 0 & 0 & 0 & m_{22}\\ \end{bmatrix}, \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n \in {\mathbb {R}}_{>0}$$\end{document} scales the uncertainty in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta A_u$$\end{document}. We conducted a parametric feasibility analysis for n, and found that LMIs are feasible for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\le 0.7$$\end{document}. For comparison of robust design, we adopt the largest feasible value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=0.7$$\end{document}. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n<0.7$$\end{document} is adopted, the bound becomes less conservative and the performance improves. However, this corresponds to assuming a smaller admissible uncertainty set. The feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document} derived via the LMI formulation is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_{L} = \begin{bmatrix}-3.9353&-150.2981&-129.8873&-146.7861\end{bmatrix} \end{aligned}$$\end{document} which assigns the closed-loop eigenvalues as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{1} = -51.6698 + 0.0000i, \quad \lambda _{2} = -0.5591 + 0.0000i, \quad \lambda _{3,4} = -0.0217 \pm 0.0076i \end{aligned}$$\end{document} thereby confirming asymptotic stability. (i)In the nominal design case, the feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document} derived via the LMI formulation is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_{L} = \begin{bmatrix} -1.8675&-25.5499&-37.7847&-2.8618 \end{bmatrix} \end{aligned}$$\end{document} which assigns the closed-loop eigenvalues as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{1,2} = -0.5653 \pm 1.0699i,\qquad \lambda _{3,4} =-0.0543 \pm 0.0445i \end{aligned}$$\end{document} thereby confirming asymptotic stability. In the nominal design case, the feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document} derived via the LMI formulation is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_{L} = \begin{bmatrix} -1.8675&-25.5499&-37.7847&-2.8618 \end{bmatrix} \end{aligned}$$\end{document} which assigns the closed-loop eigenvalues as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{1,2} = -0.5653 \pm 1.0699i,\qquad \lambda _{3,4} =-0.0543 \pm 0.0445i \end{aligned}$$\end{document} thereby confirming asymptotic stability. (ii)In the robust design case, only the nonlinear damping coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_r$$\end{document} is regarded as uncertain. Accordingly, the matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{L}$$\end{document} in the linearized system is rewritten as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{L} = A_{u} + \Delta A_{u} \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&A_{u} = \begin{bmatrix} 0 & {\bar{u}} & 1 & 0\\ 0 & 0 & 0 & 1\\ 0 & 0 & \frac{1}{\Delta _{L}}(m_{33}Y_{v} + (m_{22} - m_{11})m_{23}{\bar{u}} - m_{23}N_{v}) & \frac{1}{\Delta _{L}}(m_{33}Y_{r} + (m_{23}^{2} - m_{11}m_{33}){\bar{u}})\\ 0 & 0 & \frac{1}{\Delta _{L}}(-m_{23}Y_{v} + (m_{11} - m_{22})m_{22}{\bar{u}} + m_{22}N_{v}) & \frac{1}{\Delta _{L}}(-m_{23}Y_{r} + (m_{11} - m_{22})m_{23}{\bar{u}}) \end{bmatrix} \end{aligned}$$\end{document} is the known dynamics matrix, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta A_u = DG(t)H$$\end{document} is a structured perturbation matrix, and G(t) is the uncertain block satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$||G(t)||<1$$\end{document}. The matrices D and H are given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} D=nI_{4\times 4},\quad H=\frac{1}{\Delta _L}\begin{bmatrix} 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & -m_{23}\\ 0 & 0 & 0 & m_{22}\\ \end{bmatrix}, \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n \in {\mathbb {R}}_{>0}$$\end{document} scales the uncertainty in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta A_u$$\end{document}. We conducted a parametric feasibility analysis for n, and found that LMIs are feasible for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\le 0.7$$\end{document}. For comparison of robust design, we adopt the largest feasible value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=0.7$$\end{document}. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n<0.7$$\end{document} is adopted, the bound becomes less conservative and the performance improves. However, this corresponds to assuming a smaller admissible uncertainty set. The feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document} derived via the LMI formulation is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_{L} = \begin{bmatrix}-3.9353&-150.2981&-129.8873&-146.7861\end{bmatrix} \end{aligned}$$\end{document} which assigns the closed-loop eigenvalues as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{1} = -51.6698 + 0.0000i, \quad \lambda _{2} = -0.5591 + 0.0000i, \quad \lambda _{3,4} = -0.0217 \pm 0.0076i \end{aligned}$$\end{document} thereby confirming asymptotic stability. In the robust design case, only the nonlinear damping coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_r$$\end{document} is regarded as uncertain. Accordingly, the matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$A_{L}$$\end{document} in the linearized system is rewritten as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{L} = A_{u} + \Delta A_{u} \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&A_{u} = \begin{bmatrix} 0 & {\bar{u}} & 1 & 0\\ 0 & 0 & 0 & 1\\ 0 & 0 & \frac{1}{\Delta _{L}}(m_{33}Y_{v} + (m_{22} - m_{11})m_{23}{\bar{u}} - m_{23}N_{v}) & \frac{1}{\Delta _{L}}(m_{33}Y_{r} + (m_{23}^{2} - m_{11}m_{33}){\bar{u}})\\ 0 & 0 & \frac{1}{\Delta _{L}}(-m_{23}Y_{v} + (m_{11} - m_{22})m_{22}{\bar{u}} + m_{22}N_{v}) & \frac{1}{\Delta _{L}}(-m_{23}Y_{r} + (m_{11} - m_{22})m_{23}{\bar{u}}) \end{bmatrix} \end{aligned}$$\end{document} is the known dynamics matrix, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta A_u = DG(t)H$$\end{document} is a structured perturbation matrix, and G(t) is the uncertain block satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$||G(t)||<1$$\end{document}. The matrices D and H are given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} D=nI_{4\times 4},\quad H=\frac{1}{\Delta _L}\begin{bmatrix} 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0\\ 0 & 0 & 0 & -m_{23}\\ 0 & 0 & 0 & m_{22}\\ \end{bmatrix}, \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n \in {\mathbb {R}}_{>0}$$\end{document} scales the uncertainty in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta A_u$$\end{document}. We conducted a parametric feasibility analysis for n, and found that LMIs are feasible for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\le 0.7$$\end{document}. For comparison of robust design, we adopt the largest feasible value \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n=0.7$$\end{document}. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n<0.7$$\end{document} is adopted, the bound becomes less conservative and the performance improves. However, this corresponds to assuming a smaller admissible uncertainty set. The feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document} derived via the LMI formulation is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K_{L} = \begin{bmatrix}-3.9353&-150.2981&-129.8873&-146.7861\end{bmatrix} \end{aligned}$$\end{document} which assigns the closed-loop eigenvalues as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \lambda _{1} = -51.6698 + 0.0000i, \quad \lambda _{2} = -0.5591 + 0.0000i, \quad \lambda _{3,4} = -0.0217 \pm 0.0076i \end{aligned}$$\end{document} thereby confirming asymptotic stability. Simulation results for Scenario (i) and Scenario (ii) are presented in Figs. 2, 3, 4, 5, 6, 7, 8, 9, 10 and 11, respectively. Figures 2 and 7 depict the path-following trajectories with respect to the waypoint-defined reference paths. Figures 3, 4, 5, 6 and 8, 9, 10 and 11 display the time responses of the sway velocity v, yaw rate r, cross-track error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document}, and heading error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document}. As expected under full model knowledge, all states converge to the origin, showing that asymptotic path-following can be readily achieved by a linear state-feedback design. While the robust design achieves asymptotic path following for all admissible uncertainties, its following accuracy is lower than that of the nominal design. When the sway yaw dynamics are only partially known, conventional linearization-based designs cannot be applied directly. We therefore consider two representative uncertainty configurations for (5) and apply the robust synthesis result of Theorem 1. (i)In the case of partially unknown dynamics, the rigid-body inertia M and Coriolis terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document} in (2) are retained in the nominal model, while only the nonlinear damping coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Y_{|v|v},\; Y_{|r|v},\; Y_{|v|r},\; N_{|v|v},\; N_{|r|v},\; N_{r},\; N_{|v|r},\; N_{|r|r} \end{aligned}$$\end{document} are regarded as uncertain. Accordingly, in the decomposition (5), the residual block is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\tilde{A}}}_{vr}(v,r) = -\big (M^{-1}\big )_{(2:3,2:3)} \begin{bmatrix} Y_{|v|v}|v| + Y_{|r|v}|r| & Y_{|v|r}|v|\\ N_{|v|v}|v| + N_{|r|v}|r| & N_{r} + N_{|v|r}|v| + N_{|r|r}|r| \end{bmatrix}, \qquad {{\tilde{B}}}_{vr} = 0_{2\times 1} \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\big (M^{-1}\big )_{(2:3,2:3)}$$\end{document} denotes the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\times 2$$\end{document} submatrix of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M^{-1}$$\end{document} formed by its second and third rows and columns.(ii)In the case of fully unknown dynamics, the entire sway yaw subsystem in (5) is regarded as uncertain. Accordingly, we set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{vr}=0_{2\times 2}, \qquad B_{vr}=0_{2\times 1} \end{aligned}$$\end{document} so that all contributions from M, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(\nu )$$\end{document} in (2) are absorbed into the residual block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\tilde{A}_{vr},{{\tilde{B}}}_{vr})$$\end{document}. (i)In the case of partially unknown dynamics, the rigid-body inertia M and Coriolis terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document} in (2) are retained in the nominal model, while only the nonlinear damping coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Y_{|v|v},\; Y_{|r|v},\; Y_{|v|r},\; N_{|v|v},\; N_{|r|v},\; N_{r},\; N_{|v|r},\; N_{|r|r} \end{aligned}$$\end{document} are regarded as uncertain. Accordingly, in the decomposition (5), the residual block is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\tilde{A}}}_{vr}(v,r) = -\big (M^{-1}\big )_{(2:3,2:3)} \begin{bmatrix} Y_{|v|v}|v| + Y_{|r|v}|r| & Y_{|v|r}|v|\\ N_{|v|v}|v| + N_{|r|v}|r| & N_{r} + N_{|v|r}|v| + N_{|r|r}|r| \end{bmatrix}, \qquad {{\tilde{B}}}_{vr} = 0_{2\times 1} \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\big (M^{-1}\big )_{(2:3,2:3)}$$\end{document} denotes the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\times 2$$\end{document} submatrix of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M^{-1}$$\end{document} formed by its second and third rows and columns. In the case of partially unknown dynamics, the rigid-body inertia M and Coriolis terms \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document} in (2) are retained in the nominal model, while only the nonlinear damping coefficients \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Y_{|v|v},\; Y_{|r|v},\; Y_{|v|r},\; N_{|v|v},\; N_{|r|v},\; N_{r},\; N_{|v|r},\; N_{|r|r} \end{aligned}$$\end{document} are regarded as uncertain. Accordingly, in the decomposition (5), the residual block is given by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\tilde{A}}}_{vr}(v,r) = -\big (M^{-1}\big )_{(2:3,2:3)} \begin{bmatrix} Y_{|v|v}|v| + Y_{|r|v}|r| & Y_{|v|r}|v|\\ N_{|v|v}|v| + N_{|r|v}|r| & N_{r} + N_{|v|r}|v| + N_{|r|r}|r| \end{bmatrix}, \qquad {{\tilde{B}}}_{vr} = 0_{2\times 1} \end{aligned}$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\big (M^{-1}\big )_{(2:3,2:3)}$$\end{document} denotes the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$2\times 2$$\end{document} submatrix of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$M^{-1}$$\end{document} formed by its second and third rows and columns. (ii)In the case of fully unknown dynamics, the entire sway yaw subsystem in (5) is regarded as uncertain. Accordingly, we set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{vr}=0_{2\times 2}, \qquad B_{vr}=0_{2\times 1} \end{aligned}$$\end{document} so that all contributions from M, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(\nu )$$\end{document} in (2) are absorbed into the residual block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\tilde{A}_{vr},{{\tilde{B}}}_{vr})$$\end{document}. In the case of fully unknown dynamics, the entire sway yaw subsystem in (5) is regarded as uncertain. Accordingly, we set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} A_{vr}=0_{2\times 2}, \qquad B_{vr}=0_{2\times 1} \end{aligned}$$\end{document} so that all contributions from M, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C(\nu )$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D(\nu )$$\end{document} in (2) are absorbed into the residual block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(\tilde{A}_{vr},{{\tilde{B}}}_{vr})$$\end{document}. In both uncertainty configurations, the structured block \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta := \begin{bmatrix}{{\tilde{A}}}_{vr}&{{\tilde{B}}}_{vr}\end{bmatrix}\in \mathbb R^{2\times 3}$$\end{document} is over-approximated by the data-driven ellipsoidal set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document} constructed from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$(X_0,U_0,X_1)$$\end{document} according to Lemma 1. In the construction of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document}, the weighting matrix in the quadratic form of Lemma 1 was chosen as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Sigma = I$$\end{document} so that the energy-bounded measurement and differentiation errors are consistently absorbed into the uncertainty description. Applying Theorem 1, the LMIs admit feasible solutions for both cases. In the partially unknown case, the LMI synthesis yields the matrix\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} P=\begin{bmatrix} 10368 & -799 & -5 & 25\\ -799 & 162 & -32 & -31\\ -5 & -32 & 30 & -9\\ 25 & -31 & -9 & 194 \end{bmatrix}\succ 0 \end{aligned}$$\end{document}together with the state-feedback gain\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K=\begin{bmatrix}-0.1492&-2.0548&-2.5982&-4.3653\end{bmatrix}. \end{aligned}$$\end{document}In the fully unknown case, the solution is\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} P=\begin{bmatrix} 3095.3 & -398.9 & -96.8 & 43\\ -398.9 & 140.8 & -36 & -110.5\\ -96.8 & -36 & 111.4 & 47.5\\ 43 & -110.5 & 47.5 & 151.4\\ \end{bmatrix}\succ 0 \end{aligned}$$\end{document}with the corresponding feedback gain\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} K=\begin{bmatrix}-3.8783&-32.3859&-2.1390&-26.2958\end{bmatrix}. \end{aligned}$$\end{document}The feasibility of the LMIs confirms that the proposed data-driven controller guarantees robust asymptotic stability of the closed-loop system (8) across all admissible uncertainties in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {E}}$$\end{document}. To assess the effectiveness of the proposed design, simulations were conducted for Scenarios (i) and (ii). The corresponding results are shown in Figs. 12, 13, 14, 15, 16, 17, 18, 19, 20 and 21, respectively. Figures 12 and 17 confirm that the vessel successfully follows the prescribed paths and passes all six waypoints under both partially (Scenarios (i)) and fully unknown dynamics (Scenarios (ii)). The time responses of v, r, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} are provided in Figs. 13, 14, 15, 16 and 18, 19, 20 and 21, showing that v and r converge asymptotically to zero while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} vanish over time. For comparison, the proposed approach is compared with the aforementioned linearized model-based controllers with the feedback gain \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K_{L}$$\end{document}. Compared to the model-based results in Figs. 2 and 7, the proposed data-driven controller in Figs. 12 and 17 achieves path-following accuracy similar to that of the model-based results, despite the inclusion of sharp turns. Furthermore, to quantitatively evaluate path-following accuracy, the time-averaged absolute cross-track error\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} {{\bar{y}}}_e = \frac{1}{t_f}\int _0^{t_f} |y_e(t)|\,dt \end{aligned}$$\end{document}is computed for both scenarios, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t_f$$\end{document} denotes the total simulation time. Table 1 summarizes the time-averaged absolute cross-track error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_e|$$\end{document} across both scenarios: the nominal deign yields 0.0861 and 0.2049, the robust design 0.6687 and 0.4982, the partially unknown case 0.4343 and 0.3532, and the fully unknown case 0.4790 and 0.6863, for Scenarios (i) and (ii), respectively. As expected, the model-based nominal controller designed under the assumption of complete knowledge of the mathematical model achieved the lowest \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_e|$$\end{document} in both scenarios (i) and (ii). Nevertheless, the proposed controller, despite being designed under partial or complete model uncertainty, guarantees stability and attains a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_e|$$\end{document} comparable to that of the model-based counterpart. Moreover, the partially unknown case yields lower \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_{e}|$$\end{document} than the robust design, indicating that the proposed controller provides a less conservative bound while preserving stability. Importantly, the same feedback gain K was applied to different reference paths, Scenarios (i) and (ii) without retraining. Our resulting consistency of performance shows the characteristics of the proposed path-independent design, unlike existing DDC approaches18–21. Overall, the simulations validate that the proposed design scheme, even under partial or completely unknown dynamics, achieves robust asymptotic stability and accurate path following. Table 1Time-averaged cross-track error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_e|$$\end{document} for Scenarios (i)–(ii).ScenarioLinearized model-based (nominal)Linearized model-based (robust)Partially unknown(Theorem 1)Fully unknown(Theorem 1)(i)0.08610.66870.43430.4790(ii)0.20490.49820.35320.6863 Table 1Time-averaged cross-track error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_e|$$\end{document} for Scenarios (i)–(ii).ScenarioLinearized model-based (nominal)Linearized model-based (robust)Partially unknown(Theorem 1)Fully unknown(Theorem 1)(i)0.08610.66870.43430.4790(ii)0.20490.49820.35320.6863 Time-averaged cross-track error \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$|y_e|$$\end{document} for Scenarios (i)–(ii). Fig. 2Scenario i: Path-following trajectory under the linear model-based controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid lines: nominal case, dashed: robust case). Scenario i: Path-following trajectory under the linear model-based controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid lines: nominal case, dashed: robust case). Fig. 3Scenario i: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based nominal controller. Scenario i: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based nominal controller. Fig. 4Scenario i: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based robust controller. Scenario i: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based robust controller. Fig. 5Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based nominal controller. Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based nominal controller. Fig. 6Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based robust controller. Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based robust controller. Fig. 7Scenario ii: Path-following trajectory under the linear model-based controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid lines: nominal case, dashed: robust case). Scenario ii: Path-following trajectory under the linear model-based controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid lines: nominal case, dashed: robust case). Fig. 8Scenario ii: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based nominal controller. Scenario ii: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based nominal controller. Fig. 9Scenario ii: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based robust controller. Scenario ii: Time responses of sway yaw velocities \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under the linear model-based robust controller. Fig. 10Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based nominal controller. Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based nominal controller. Fig. 11Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based robust controller. Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e_\psi$$\end{document} under the linear model-based robust controller. Fig. 12Scenario i: Path-following trajectories under the proposed data-driven controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid: partially unknown case, dashed: fully unknown case). Scenario i: Path-following trajectories under the proposed data-driven controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid: partially unknown case, dashed: fully unknown case). Fig. 13Scenario i: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under partially unknown dynamics. Scenario i: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under partially unknown dynamics. Fig. 14Scenario i: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under fully unknown dynamics. Scenario i: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under fully unknown dynamics. Fig. 15Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under partially unknown dynamics. Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under partially unknown dynamics. Fig. 16Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under fully unknown dynamics. Scenario i: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under fully unknown dynamics. Fig. 17Scenario ii: Path-following trajectories under the proposed data-driven controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid: partially unknown case, dashed: fully unknown case). Scenario ii: Path-following trajectories under the proposed data-driven controller (‘o’: waypoints, dotted lines: path connecting waypoints, solid: partially unknown case, dashed: fully unknown case). Fig. 18Scenario ii: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under partially unknown dynamics. Scenario ii: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under partially unknown dynamics. Fig. 19Scenario ii: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under fully unknown dynamics. Scenario ii: Time responses of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\nu =[v,r]^\top$$\end{document} under fully unknown dynamics. Fig. 20Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under partially unknown dynamics. Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under partially unknown dynamics. Fig. 21Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under fully unknown dynamics. Scenario ii: Time responses of guidance errors \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$y_e,e_\psi$$\end{document} under fully unknown dynamics. Conclusions We addressed the path following control problem of underactuated USVs with partially known sway yaw dynamics. The guidance error subsystem was modeled exactly, whereas the unmodeled dynamics were restricted to a data consistent, matrix ellipsoidal uncertainty set from input state data. Within this framework, LMI condition was derived such that a static state feedback K robustly asymptotically stabilizes the augmented error system on a prescribed forward invariant set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathscr {D}}$$\end{document} for all \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \in {\mathscr {E}}$$\end{document}. Under the dwell time and bounded reset assumptions (A5) (A6) of Assumption 1, Lyapunov decrease is preserved across segment transitions, which guarantees \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$e(t)\rightarrow 0$$\end{document} along admissible waypoint paths. Simulations on CyberShip II demonstrate the proposed controller achieves path following accuracy comparable to that of a fully model based linear design in both the partially and fully unknown cases, and it preserves stability. These results indicate path independent synthesis and robustness to modeling deficits confined by data.