Multivariable Generalized STA Control

• MGSTA is a robust sliding mode control algorithm for uncertain MIMO systems that generalizes the classical super-twisting approach.
• It employs Lyapunov-based analysis and LMI design to achieve global finite-time stability, fault tolerance, and effective disturbance rejection.
• Practical applications include omnidirectional robots and multi-trailer vehicles, demonstrating precise tracking and resilience to system uncertainties.

The Multivariable Generalized Super-Twisting Algorithm (MGSTA) is a class of robust continuous-time sliding mode control algorithms for multiple-input, multiple-output (MIMO) dynamical systems with matched perturbations and significant parametric uncertainty, particularly in the input gain matrix. MGSTA generalizes the classical super-twisting algorithm (STA) to the multivariable regime, incorporating key stability and robustness results via Lyapunov and piecewise continuous max-type Lyapunov functions amenable to both nonlinear and linear time-invariant (LTI) settings. Applications include robust tracking, fault-tolerant control, and disturbance compensation in systems with nonlinearities and time-varying, uncertain parameters (Moreno et al., 2021, Geromel et al., 26 Feb 2025).

1. System Modeling and Problem Structure

MGSTA addresses control of uncertain systems of relative degree one modeled as

$\dot x = f(t,x) + G(t,x)\,u$

where $x \in \mathbb{R}^n$ is the state, $u \in \mathbb{R}^n$ the control input, $f(t,x)$ denotes matched perturbations and uncertainties, and $G(t,x)$ is an uncertain, state- and time-dependent input matrix. The gain matrix is represented as

$G(t,x) = (I + \Delta_G(t,x))\,G_0(t,x)$

with known nominal $G_0(t,x)$ and unknown perturbation $\Delta_G(t,x)$. The disturbance term $f(t,x)$ is decomposed into structured and unstructured parts, with $\phi_1, \phi_2$ denoting nonlinear functions associated with the sliding variable. Boundedness and definiteness assumptions are imposed on all uncertainties, described by explicit constants $x \in \mathbb{R}^n$0.

In the LTI setting, the plant is expressed in regular form as

$x \in \mathbb{R}^n$1

with matrices subject to polytopic uncertainty, i.e., each system matrix lies in the convex hull of known vertices (Geromel et al., 26 Feb 2025). The disturbance $x \in \mathbb{R}^n$2 is Lipschitz, with $x \in \mathbb{R}^n$3.

2. MGSTA Law and Algorithmic Structure

The MGSTA control law is defined using a sliding variable $x \in \mathbb{R}^n$4 (with typically full-rank $x \in \mathbb{R}^n$5), designed as a multivariable extension of STA:

$x \in \mathbb{R}^n$6

where $x \in \mathbb{R}^n$7, $x \in \mathbb{R}^n$8, and the Jacobian $x \in \mathbb{R}^n$9 and scalar function $u \in \mathbb{R}^n$0 are given by

$u \in \mathbb{R}^n$1

with design parameters $u \in \mathbb{R}^n$2, $u \in \mathbb{R}^n$3 gains $u \in \mathbb{R}^n$4, and tuning coefficient $u \in \mathbb{R}^n$5 (Moreno et al., 2021).

In the LTI polytope case, the super-twisting law is formulated as

$u \in \mathbb{R}^n$6

with nonlinear gain modulation via $u \in \mathbb{R}^n$7, and full-matrix gains $u \in \mathbb{R}^n$8 extracted from LMIs (Geromel et al., 26 Feb 2025).

3. Lyapunov Framework and Finite-Time Stability

MGSTA’s global robust finite-time stability is established via Lyapunov constructions tailored to the multivariable, nonlinear, and uncertainty-rich context.

For the general nonlinear setting, the Lyapunov candidate is

$u \in \mathbb{R}^n$9

with $f(t,x)$0, $f(t,x)$1, and $f(t,x)$2. Properly chosen gains ensure that, under the uncertainty bounds,

$f(t,x)$3

with $f(t,x)$4 positive-definite via satisfaction of Schur complement and quadratic inequalities in the gains.

For uncertain LTI systems, a max-type piecewise-continuous Lyapunov function

$f(t,x)$5

is introduced, where individual Lyapunov candidates $f(t,x)$6 (for zero-dynamics) and $f(t,x)$7 (for the nonlinear block) admit rates of decrease linked by LMI-solvable design conditions. This structure ensures all system trajectories reach the sliding manifold in finite time, after which exponential decay ensues (Geromel et al., 26 Feb 2025).

4. Gain Selection and LMI Design

Gain selection procedures are anchored in explicit inequality and LMI constraints. In nonlinear uncertain systems, selection proceeds by:

1. Fixing design parameters $f(t,x)$8.
2. Extracting constants from uncertainty bounds.
3. Solving polynomial and scalar inequalities:

$f(t,x)$9

and

$G(t,x)$0

for suitable parameters $G(t,x)$1, then setting

$G(t,x)$2

(Moreno et al., 2021).

For LTI uncertain plants, all robust stability and performance criteria reduce to LMI feasibility at all vertices of the uncertainty polytope. The design parameters $G(t,x)$3 are fixed, and the gains are recovered from the SDP solution’s variables for

$G(t,x)$4

subject to strict matrix inequalities which encode (i) nonlinear block stability, (ii) linear performance including $G(t,x)$5-gain, and (iii) initial condition requirements. Additional LMIs can constrain control effort and condition number, ensuring suitability for practical implementation with standard SDP solvers (Geromel et al., 26 Feb 2025).

5. Theoretical Guarantees

Under the stated assumptions on system uncertainties and disturbance bounds, MGSTA guarantees:

• Global robust finite-time stability: For suitable constant gains, the output sliding variable and auxiliary state converge to the origin in finite time (Moreno et al., 2021).
• Asymptotic stability: Composite Lyapunov arguments ensure that after the finite-time phase, states exhibit exponential decay (Geromel et al., 26 Feb 2025).
• Disturbance rejection: The auxiliary variable within MGSTA (e.g., $G(t,x)$6 or $G(t,x)$7) dynamically estimates and compensates matched perturbations after the finite-time convergence.
• Fault tolerance: Full loss of actuators or severe parametric uncertainty in the input matrix is systematically addressed within the LMI framework.
• Explicit convergence rate and performance: The time $G(t,x)$8 to reach a residual set of Lyapunov function value $G(t,x)$9 is upper bounded by

$G(t,x) = (I + \Delta_G(t,x))\,G_0(t,x)$0

(Geromel et al., 26 Feb 2025).

6. Applications and Practical Examples

MGSTA has been validated in simulation and application studies for complex robotic and mechanical systems:

• Omnidirectional mobile robots: MGSTA was applied to a four-wheeled robot with uncertain mass, friction, and input gains. The sliding variable and tracking errors converged to zero in under 0.2 seconds, with bounded, chattering-free control inputs (Moreno et al., 2021).
• Three-DOF mechanical systems (e.g., multi-trailer vehicles): Fault-tolerant MGSTA controllers retained stability, reference tracking, and bounded control despite full actuator loss and mass uncertainties. All theoretical LMIs were verified numerically (Geromel et al., 26 Feb 2025).

These case studies underscore the algorithm’s relevance for practical high-dimensional plants with severe uncertainties, needing rigorous performance and robustness guarantees.

7. Robustness Margins and Design Trade-offs

MGSTA robustness margins are explicitly quantified via the feasibility conditions in the uncertainty parameters and the LMI-verified domains:

• Parameter margins: All vertices of the matrix uncertainty polytope are covered, so any convex combination is admissible (Geromel et al., 26 Feb 2025).
• Disturbance bounds: The resilience to matched disturbance $G(t,x) = (I + \Delta_G(t,x))\,G_0(t,x)$1 is tied to the minimum singular value of the input gain matrix after compensation.
• Gain selection trade-offs: Larger nonlinear gain regularizer $G(t,x) = (I + \Delta_G(t,x))\,G_0(t,x)$2 increases smoothness of the input, reducing chattering but slowing convergence. Increasing the performance gain $G(t,x) = (I + \Delta_G(t,x))\,G_0(t,x)$3 enhances stability at the expense of greater control effort. Control magnitude can be bounded at the cost of transient performance via dedicated LMI constraints.

This framework offers systematic quantification of trade-offs and performance guarantees, suitable for control design in highly uncertain, safety-critical, or constrained environments.

---

References:

• "Multivariable Super-Twisting Algorithm for Systems with Uncertain Input Matrix and Perturbations" (Moreno et al., 2021)
• "Multivariable Generalized Super-Twisting Algorithm Robust Control of Linear Time-Invariant Systems" (Geromel et al., 26 Feb 2025)