Super-Twisting Algorithm (STA) Overview

Updated 9 May 2026

Super-Twisting Algorithm (STA) is a continuous second-order sliding-mode control law achieving finite-time convergence of both the sliding variable and its derivative, effectively eliminating chattering.
Adaptive and barrier-based gain designs, supported by Lyapunov stability proofs, enable STA to dynamically adjust to uncertainties while minimizing control energy.
STA extends to higher-order, MIMO, discrete-time, fractional-order, and time-delay systems, proving its scalability and robustness in practical applications such as UAVs and precision stages.

The Super-Twisting Algorithm (STA) is a continuous, dynamic, second-order sliding-mode control law that achieves finite-time convergence of both a sliding variable and its derivative, guaranteeing robust performance in the presence of bounded perturbations with unknown but bounded derivative ("Lipschitz" disturbances). STA is the canonical representative of the second-order sliding-mode (SOSM) control family and is widely recognized for its ability to eliminate high-frequency chattering endemic to classical first-order sliding modes by providing a continuous control law. Modern extensions address challenges including gain adaptation, system uncertainty, time delay, and networked/discretized implementations.

1. Fundamentals of the Super-Twisting Algorithm

STA addresses control of first-order plants of the form

$\dot s = u + d(t), \qquad |\dot d(t)| \le L,$

with $s$ the sliding variable, $u$ the control, and $d(t)$ the (unknown) perturbation. The classical continuous-time STA is given by

$\begin{aligned} u(t) &= -k_1 |s(t)|^{1/2} \, \mathrm{sign}(s(t)) + v(t),\ \dot v(t) &= -k_2 \, \mathrm{sign}(s(t)), \end{aligned}$

with $k_1>0$ , $k_2>L$ . Under these conditions, STA ensures that both $s$ and its time derivative converge to zero in finite time, and the equivalent control $u_{\text{eq}} = -d(t)$ is continuous for $t$ large, thus fully rejecting matched disturbances with bounded derivative (Hoang et al., 2018).

The STA's distinctive "fractional power" feedback and discontinuous integral action (applied to $s$ 0) are critical to finite-time convergence and chattering elimination. Classical Lyapunov-based proofs exploit a strict homogeneous Lyapunov function, yielding an inequality $s$ 1 that underpins finite-time convergence.

2. Adaptive and Barrier-Based Gain Designs

A key limitation of nominal STA is the need for a priori knowledge (or conservative over-estimation) of the disturbance derivative bound $s$ 2, with large gains leading to excessive control energy and possible residual chattering. Adaptive and barrier-based schemes address this:

Adaptive STA: Both gains $s$ 3 are adapted on-line based on the sliding variable's state, with prototypical adaptation laws such as

$s$ 4

where $s$ 5 are adaptation rates and $s$ 6 a small threshold. Gains grow only as necessary to combat perturbations, remaining minimal for low activity, thus substantially reducing chattering and energy consumption (Hoang et al., 2018, Xiong et al., 2018).

Barrier-function Modulated STA: A state-dependent gain employing a barrier function enforces $s$ 7 and adapts automatically as $s$ 8:

$s$ 9

for design constant $u$ 0 (Obeid et al., 2019). Gains increase as $u$ 1 nears the barrier, preventing escape, and decrease as $u$ 2 shrinks, abating chattering with minimal gain over-estimation.

Both methods guarantee, via modified Lyapunov constructions, finite-time convergence and boundedness of all internal gains—and require only minimal parameter tuning.

3. Higher-Order and Multivariable Generalizations

STA admits generalizations to both higher-order chains of integrators (HOST) and MIMO (multivariable) systems:

Higher-Order Super-Twisting (HOST): For a chain of $u$ 3 integrators (relative degree $u$ 4), HOST augments any homogeneous finite-time controller $u$ 5 with an integral action on the highest-order state:

$u$ 6

with $u$ 7 a homogeneous Lyapunov function for the pure chain (Chitour et al., 2015). HOST achieves finite-time convergence for arbitrary-order systems subjected to additive and multiplicative uncertainties, contingent on certain geometric Lyapunov compatibility conditions.

Multivariable/MIMO STA: For $u$ 8-dimensional outputs, a generalized control law enables full-matrix gain synthesis:

$u$ 9

with homogeneity parameter $d(t)$ 0, element-wise vector power $d(t)$ 1, and full matrix K, $d(t)$ 2 design (Garcia-Mathey et al., 2022). Lyapunov-based passivity and LMI-type solvability conditions facilitate robust and finite-time ISS-type stability under uncertain, time- and state-varying input matrices (Moreno et al., 2021, Geromel et al., 26 Feb 2025).

4. Discrete-Time and Digital Realizations

Discrete-time STA has garnered extensive attention for digital controllers and networked/embedded systems, where naive discretizations often introduce chattering, degrade finite-time convergence, or saturate performance.

Modified Implicit and Eigenvalue-Mapped Discretizations: Properly constructed discrete STA schemes reproduce continuous-time properties such as finite-time convergence, chattering-free behavior, and gain-robust accuracy:
- Implicit update law, tuned to reproduce the dead-beat property near the origin and explicit monotonic Lyapunov decrease, achieves steady-state tracking errors $d(t)$ 3 (sampling interval $d(t)$ 4) independent of gain overestimation (Andritsch et al., 2023, Seeber et al., 2024).
- Eigenvalue-mapping or matching–exact discretizations preserve continuous-time closed-loop poles and robustly eliminate discretization-induced chattering, including under barrier or multi-layer gain adaptation (Ding, 2022, Vié et al., 28 Apr 2026).
Multi-layer Barrier Adaptation: In digital STA, barrier adaptation can be "nested," confining $d(t)$ 5 within layered bands for superior rejection of high-frequency perturbations and inter-sample "blindness," further reducing tracking RMSE and chattering compared to scheme with a single band (Vié et al., 28 Apr 2026).

5. Extensions: Fractional, Delay, and Networked STA

Several advanced research directions systematically extend STA for modern application requirements:

Fractional-Order STA (FOSTA): Employs Caputo (or Riemann–Liouville) fractional derivatives in the sliding surface and/or control, enabling improved convergence rates, terminal dynamics, and memory effects suitable for viscoelastic or anomalous-diffusive systems. Embedded in fault estimation observers or precision motion control, FOSTA maintains strong finite-time guarantees with Lyapunov–fractional derivative analysis (Kuang et al., 2021, Mousavi et al., 2017, Kuang et al., 2021).
Time-Delay Systems: Continuous Lyapunov redesign leverages STA as part of a robust stabilization law for uncertain time-delay systems, utilizing delay-free sliding variables, Lyapunov–Krasovskii functionals, and absolutely continuous control for chattering-free, finite-time convergence despite plant or measurement delays (Gomez et al., 2021, Pinto et al., 28 Jul 2025).
Distributed Differentiators/Networked Systems: Abstracting the STA’s geometric and Lyapunov structure allows systematic design of distributed, networked, event-triggered differentiation schemes with finite-time consensus and systematic gain selection, including rigorous minimum inter-event guarantees and communication/accuracy trade-offs (Aldana-López et al., 2 Feb 2026).

6. Practical Implementations and Case Studies

Recent applied work demonstrates STA-based controllers and observers in complex, real-world systems:

Unmanned aerial vehicles (UAVs): Adaptive super-twisting controllers achieve sub-second attitude regulation and robust rejection of external torques and inertia uncertainty with automatic online gain adaptation, outperforming classical PID and SMC under simulation and experiment (Hoang et al., 2018).
Hydraulic actuation and Maglev systems: Novel STA-based integral sliding surfaces (not requiring direct velocity measurement) with LMI-based gain tuning achieve sub-1% tracking in hydraulic cylinders. Combination with higher-order sliding observers enables robust tracking and fault estimation in highly unstable, nonlinear magnetic levitation systems (Estrada et al., 2023, Ganguly et al., 2020).
Precision wafer stages: Fractional/variable-gain STA with neural network adaptation achieves nanoscale tracking accuracy, reduced chattering, and robust performance over wide acceleration and disturbance regimes, both in simulation and experiment (Kuang et al., 2021, Kuang et al., 2021).

7. Synthesis: Theoretical Guarantees, Tuning, and Limitations

Across its extensions, the Super-Twisting Algorithm delivers:

Finite-Time Convergence: Both the sliding variable and its integral achieve finite-time convergence under explicit, homogeneous Lyapunov constructions.
Continuous Control/Chattering Elimination: By integrating the discontinuity, the control input is continuous, essentially eliminating chattering.
Robustness: STA admits uncertainty in matched disturbances with unknown, bounded derivative—robust in scalar, multivariable, and higher-order dynamics, including under time-varying gain and input uncertainty.
Minimal Tuning: Modern adaptive, barrier-based, and implicit-discrete STAs can be implemented with one or two free tuning parameters (e.g., accuracy band $d(t)$ 6), with substantial automatic adaptation and no need for a-priori disturbance knowledge (Xiong et al., 2018, Obeid et al., 2019).
Scalability: HOST, MGSTA, and MIMO STA frameworks allow direct extension to chains of integrators, MIMO, time-delay, and distributed settings, with scalable LMI-based tuning (Chitour et al., 2015, Garcia-Mathey et al., 2022, Moreno et al., 2021, Geromel et al., 26 Feb 2025, Gomez et al., 2021).

Potential limitations include the need for an auxiliary integrator (or observer) in implementation, possible overestimation of sampling-induced error in digital settings, and increased complexity in precise tuning for high-order or distributed versions. However, advanced gain adaptation and proper discretization effectively mitigate these drawbacks.

Selected References:

Adaptive/attitude control: (Hoang et al., 2018, Xiong et al., 2018, Obeid et al., 2019)
Higher-order/MIMO/robust: (Chitour et al., 2015, Garcia-Mathey et al., 2022, Moreno et al., 2021, Geromel et al., 26 Feb 2025)
Discrete/digital: (Andritsch et al., 2023, Seeber et al., 2024, Ding, 2022, Vié et al., 28 Apr 2026)
Fractional/fault estimation: (Kuang et al., 2021, Kuang et al., 2021, Mousavi et al., 2017)
Delay/distributed: (Gomez et al., 2021, Pinto et al., 28 Jul 2025, Aldana-López et al., 2 Feb 2026)
Case studies: (Estrada et al., 2023, Ganguly et al., 2020)

STA is now a mature, theoretically rigorous, and practically well-validated paradigm for robust, finite-time, second-order sliding-mode control across a vast spectrum of modern control applications.