Tube-Based MPC Overview

• Tube-based MPC is a robust control framework that decomposes control into a nominal trajectory and a feedback law, ensuring the system state remains within a safe tube under disturbances.
• Advanced designs such as elastic, homothetic, and system-level tubes reduce conservatism, enhance computational efficiency, and enlarge the feasible domain.
• Applications in autonomous vehicles, robotics, and networked control demonstrate its practical effectiveness with lower closed-loop costs and improved scalability.

Tube-based Model Predictive Control (MPC) is a rigorous framework for robust predictive control of constrained systems under uncertainty, combining a nominal trajectory generation with a feedback policy that constrains the system to remain within a set-valued tube around the nominal evolution. This tube is designed so that, regardless of disturbance realizations or model deviations, the real system trajectory never exits the admissible (state and input) constraint sets. Recent developments have expanded classical tube-MPC—including rigid, homothetic, elastic, system-level, distributionally robust, and adaptive forms—providing both stronger theoretical guarantees and enhanced computational tractability for large-scale and uncertain systems.

1. Fundamental Principles and Control Architecture

The classical tube-based MPC framework decomposes the control signal into two components: a nominal open-loop trajectory (computed by solving a disturbance-free optimization problem), and a feedback law designed to restrict the tracking error (between actual and nominal states) within a robust positive invariant (RPI) tube. The nominal system (with state $z_k$ and input $v_k$) evolves as: $z_{k+1} = A z_k + B v_k,$ while the real system dynamics, including additive disturbances $w_k$, are

$$x_{k+1} = A x_k + B u_k + w_k, \quad w_k \in \mathcal{W}.$$

Setting the control as

$u_k = v_k + K(x_k - z_k),$

the error evolution $e_k = x_k - z_k$ follows

$e_{k+1} = (A + B K) e_k + w_k.$

A set $\mathcal{E}$ is robust positively invariant (RPI) for this system if $$(A + B K) \mathcal{E} \oplus \mathcal{W} \subseteq \mathcal{E}$$.

Constraint satisfaction for all $w_k$ is ensured by tightening—i.e., replacing the original constraints $x_k \in \mathcal{X}$, $u_k \in \mathcal{U}$ with

$$z_k \in \mathcal{X} \ominus \mathcal{E}, \qquad v_k \in \mathcal{U} \ominus K \mathcal{E}.$$

This basic structure is universally adopted in both time-invariant linear (Sieber et al., 2021, Sieber et al., 2022) and many nonlinear (Luo et al., 2024, Luo et al., 2021) and parameter-varying (Hanema et al., 2017, Alcala et al., 2020) models.

2. Advanced Tube Parameterizations and System-Level Synthesis

Multiple research directions have emerged to reduce conservatism and computational burden of classical tube-MPC:

• Homothetic and Elastic Tubes: Allow tube cross-sections to be scaled or elongated per prediction step, rather than keeping a rigid (RPI) shape. This enables a larger domain of attraction and less conservative constraint tightening. Parameterizations include homothetic polytopes (Hanema et al., 2017), configuration-constrained polytopes (Badalamenti et al., 2024), and elastically-scaled zonotopes (Diaconescu et al., 24 Sep 2025, Ghiasi et al., 24 Dec 2025). The state is confined within $z_k \oplus \mathcal{Z}_k$ with $\mathcal{Z}_k$ scaled or shaped online, and recursive feasibility is ensured by enforcing $\mathcal{Z}_{k+1} \subseteq f(\mathcal{Z}_k)$ via inclusion conditions linear in the scaling parameters, notably using new dual linear embedding techniques for zonotopes (Diaconescu et al., 24 Sep 2025).
• System Level Parameterization (SLP): Instead of fixing a feedback gain $K$ offline, SLP jointly optimizes closed-loop responses as affine maps from disturbances to states and inputs over a prediction horizon, subject to affine and sparsity constraints (Sieber et al., 2021, Sieber et al., 2021, Sieber et al., 2024). System-level tube-MPC (SLTMPC) thus optimizes the tube shape and feedback gains online, yielding larger regions of attraction and improved closed-loop costs compared to fixed-tube approaches, with recursive feasibility guaranteed by FIR constraints and specialized terminal set selections.

3. Optimization Problem Formulations

Tube-based MPC frameworks (including classical, elastic, SLTMPC) result in convex optimization programs, typically quadratic programs (QPs) (or linear for polytope-based schemes, or SDPs for ellipsoid-based continuous-time tubes (Villanueva et al., 2016)) at each control update. The general template is: $$\min \, \ell_f(z_N) + \sum_{i=0}^{N-1} \ell(z_i, v_i)$$ subject to nominal dynamics, constraint tightening (due to tube cross-sections), tube evolution constraints (e.g., set containment or inclusion conditions), and terminal set properties. As computational efficiency is critical, recent methods focus on:

• Simultaneous joint optimization over nominal trajectory and tube parameters (Sieber et al., 2024, Sieber et al., 2021)
• Reduced-complexity inclusion constraints for scalable zonotopic tubes (Diaconescu et al., 24 Sep 2025)
• Asynchronous computation architectures, splitting tube updates and nominal control for real-time capability (Sieber et al., 2022, Sieber et al., 2024)

4. Extensions: Model Uncertainty, Nonlinearity, and Learning

• Model Uncertainty: Parametric (multiplicative) uncertainty is addressed through online over-approximation of both disturbance and model uncertainty into a unified disturbance set, often described as a time-varying polytope, with filter-based constructions for tractable online optimization (Sieber et al., 2024). Adaptive and self-tuning versions integrate online system identification (e.g., least squares), intersecting data-derived and prior model sets to construct tubes and feedback gains robust to the current feasible parameters (Tranos et al., 2022, Ghiasi et al., 24 Dec 2025).
• Nonlinear and LPV Systems: Nonlinear systems are handled via repeated local linearization and robust bounding of linearization errors in the tube design, ensuring recursive feasibility and constraint satisfaction (Luo et al., 2024, Luo et al., 2021, Nikou et al., 2018). In the linear parameter-varying (LPV) context, the tube and the feedback law are scheduled based on measurable parameters, and the tube propagation uses vertex-based or polytopic multi-model inclusion properties (Hanema et al., 2017, Alcala et al., 2020).

5. Terminal Set Construction and Stability Guarantees

Terminal sets and terminal controls are critical to guarantee closed-loop recursive feasibility and stability:

• Robust Positively Invariant Sets: Classical schemes employ invariant sets for the error dynamics under the terminal control law; contractivity is enforced over one or finite steps (Hanema et al., 2017).
• FIR/PI Terminal Constraints: For system-level and filter-based tubes, only positive invariance (not robust invariance) with respect to the nominal dynamics is required (Sieber et al., 2021).
• Scaling and Asynchronous Updates: Recent work introduces measures for online scaling of terminal sets, and asynchronous architectures that guarantee recursive feasibility via convex fusion of previously computed feasible solutions (Sieber et al., 2022, Sieber et al., 2024).

Input-to-state stability (ISS) and practical convergence are established via Lyapunov arguments and set-gauge metrics, guaranteeing bounded steady-state errors under persistent bounded disturbances.

6. Computational Scalability and Implementation Strategies

Algorithmic advances have rendered tube-MPC practical for high-dimensional, fast-sampled, and embedded applications:

• Zonotopic and Polyhedral Tubes: Use of zonotopes and scalable inclusion constraints attain order-of-magnitude reductions in variable and constraint counts compared to classic polyhedral tubes, with near-linear scaling in problem dimension (Diaconescu et al., 24 Sep 2025, Ghiasi et al., 24 Dec 2025).
• Output Feedback: Output-feedback tube design leverages constant cross-section RPI sets constructed offline through a single LP, yielding online complexity equivalent to nominal MPC with full state feedback (Lorenzetti et al., 2019).
• Networked/Remote Control: Extensions tackle lossy networks by decoupling nominal optimization (remote) and tube-based disturbance rejection (local) with strong recursive feasibility properties despite random packet drops (Umsonst et al., 2024).

7. Practical Applications and Empirical Evaluations

Tube-based MPC is experimentally validated in diverse domains:

• Autonomous Vehicles: Used for robust lane-keeping and maneuvering under LPV and nonlinear dynamic uncertainties, with low-latency, high-frequency implementations (Alcala et al., 2020, Badalamenti et al., 2024).
• Robotic Manipulation and Mobile Robots: Tube-MPC ensures constraint satisfaction and trajectory tracking under model error, actuation limits, and environmental uncertainties (Luo et al., 2024, Luo et al., 2021, Nikou et al., 2019, Nikou et al., 2018).
• Networked and Distributed Systems: Applied to platooning, mixed-traffic control, and remote tracking over unreliable channels, often reducing both computational and communication loads versus traditional receding-horizon schemes (Feng et al., 2019, Umsonst et al., 2024).

Empirical studies consistently demonstrate that advanced tube parameterizations (elastic, SLTMPC, configuration-constrained) provide significantly enlarged feasible domains, lower closed-loop costs, and reduced conservatism relative to fixed-tube baseline methods, often without commensurate increases in computation time (Sieber et al., 2021, Diaconescu et al., 24 Sep 2025, Sieber et al., 2022).

---

Key References:

• (Sieber et al., 2021) A System Level Approach to Tube-based Model Predictive Control
• (Sieber et al., 2024) Computationally Efficient System Level Tube-MPC for Uncertain Systems
• (Diaconescu et al., 24 Sep 2025) Zonotope-Based Elastic Tube Model Predictive Control
• (Sieber et al., 2022) Asynchronous Computation of Tube-based Model Predictive Control
• (Hanema et al., 2017) Stabilizing Tube-Based Model Predictive Control: Terminal Set and Cost Construction for LPV Systems
• (Luo et al., 2024, Luo et al., 2021) Nonlinear/robotic applications
• (Ghiasi et al., 24 Dec 2025) Safe Navigation with Zonotopic Tubes: An Elastic Tube-based MPC Framework
• (Tranos et al., 2022) Self-Tuning Tube-based Model Predictive Control
• (Lorenzetti et al., 2019) A Simple and Efficient Tube-based Robust Output Feedback Model Predictive Control Scheme
• (Umsonst et al., 2024) Remote Tube-based MPC for Tracking Over Lossy Networks
• (Alcala et al., 2020) Fast Zonotope-Tube-based LPV-MPC for Autonomous Vehicles
• (Badalamenti et al., 2024) Configuration-Constrained Tube MPC for Tracking