Elastic Tube MPC for Robust Control

• Elastic Tube MPC is a robust control method that dynamically adjusts trajectory tubes to handle model uncertainties and disturbances in discrete-time systems.
• It employs flexible set parameterizations—such as ellipsoidal, zonotopic, or polytopic—to reduce conservativeness while ensuring constraint satisfaction and closed-loop stability.
• Adaptive online optimization and learning mechanisms in ET-MPC enhance performance in challenging applications like obstacle avoidance and autonomous vehicle control.

Elastic Tube Model Predictive Control (ET-MPC) is a robust control methodology for enforcing state and input constraints in the presence of dynamic model uncertainty and exogenous disturbances. It extends classical Tube Model Predictive Control (TMPC) by parameterizing and optimizing a "tube"—an over-approximating set containing all possible closed-loop trajectories—whose shape and size adjust elastically in response to evolving uncertainty, disturbances, and online parameter learning. By employing flexible set parameterizations (ellipsoidal, polytopic, or zonotopic), and enabling the tube to shrink or expand adaptively, ET-MPC achieves improved performance, reduced conservativeness, and scalability for both linear and nonlinear, deterministic and uncertain systems (Parsi et al., 2022, Lopez et al., 2019, Diaconescu et al., 24 Sep 2025, Ghiasi et al., 24 Dec 2025, Morozov et al., 2020, Buerger et al., 5 Mar 2026, Dey et al., 16 Mar 2026, Badalamenti et al., 2024).

1. Problem Setting and Tube Parameterizations

ET-MPC is applicable to discrete-time systems subject to parametric and additive disturbances, frequently modeled as: $x_{k+1} = A x_k + B u_k + w_k$ where $x_k\in \mathbb{R}^{n_x}$ is the state, $u_k\in \mathbb{R}^{n_u}$ the control input, and $w_k$ is a disturbance with known or learned set bounds. Extensions introduce model uncertainty represented by:

• Linear fractional structures (dynamic, time-varying block-diagonal uncertainty) (Parsi et al., 2022)
• Polytopic, affine, or zonotopic uncertainty sets over $(A, B)$ (Badalamenti et al., 2024, Ghiasi et al., 24 Dec 2025, Dey et al., 16 Mar 2026)
• Nonlinear, parameter-affine or state-dependent uncertainty (Buerger et al., 5 Mar 2026, Lopez et al., 2019, Morozov et al., 2020)

The tube is parameterized by a time-varying family of sets, elegant choices being:

• Ellipsoidal tubes: $$\mathcal{X}_{l|k} = \mathcal{E}(z_{l|k}, P, \alpha_{l|k}) = \{ x : (x-z_{l|k})^T P^{-1} (x-z_{l|k}) \leq \alpha_{l|k}^2 \}$$ (Parsi et al., 2022, Buerger et al., 5 Mar 2026)
• Scaled zonotopic tubes: $$\mathcal{Z}_k(\delta_k) = \langle c_k, G \mathrm{diag}(\delta_k) \rangle$$ (Diaconescu et al., 24 Sep 2025, Ghiasi et al., 24 Dec 2025)
• Configuration-constrained polytopes: $X_t = \{ x : F x \leq y_t \}$ (Badalamenti et al., 2024)
• Homothetic polytopes: $$\mathcal{E}_{i|t} = \alpha_{i|t} \oplus \beta_{i|t} S_t$$ (Dey et al., 16 Mar 2026)
• Nonlinearly-parameterized tubes (e.g., boundary-layer thickness): $\Omega_i(t) = |\tilde{z}_i(t)| \leq$ function$x_k\in \mathbb{R}^{n_x}$0 (Lopez et al., 2019, Morozov et al., 2020)

Elasticity refers to online optimization or adaptation of shape parameters ($x_k\in \mathbb{R}^{n_x}$1, $x_k\in \mathbb{R}^{n_x}$2, $x_k\in \mathbb{R}^{n_x}$3, etc.), with a fixed, precomputed, or adaptively learned set template.

2. Control Law and Tube Dynamics

ET-MPC proceeds by splitting the actual system trajectory into a nominal portion and an error or deviation: $x_k\in \mathbb{R}^{n_x}$4 The nominal trajectory $x_k\in \mathbb{R}^{n_x}$5 is generated via a disturbance-free prediction (with updated parameter estimates in adaptive variants), while the ancillary control law (usually linear state feedback $x_k\in \mathbb{R}^{n_x}$6) robustifies against uncertainty and disturbances: $x_k\in \mathbb{R}^{n_x}$7 The tube error dynamics propagate as

$x_k\in \mathbb{R}^{n_x}$8

The tube cross-section at each $x_k\in \mathbb{R}^{n_x}$9 satisfies a reachability or invariance recursion: $u_k\in \mathbb{R}^{n_u}$0 for ellipsoids (Parsi et al., 2022, Buerger et al., 5 Mar 2026), or respective polytopic/zonotopic propagations.

Elasticity arises by making the tube shape a decision variable at each time, e.g., the scaling $u_k\in \mathbb{R}^{n_u}$1 of an ellipsoidal tube (Parsi et al., 2022), the scaling vector $u_k\in \mathbb{R}^{n_u}$2 in a zonotopic tube (Diaconescu et al., 24 Sep 2025), or the cross-section parameters $u_k\in \mathbb{R}^{n_u}$3 of a configuration-constrained polytope (Badalamenti et al., 2024). These are subject to set-inclusion constraints that guarantee the next tube cross-section contains all possible propagated errors.

3. Optimization Frameworks and Scalability

Elastic tube MPC problems are formulated as tractable convex programs (SDPs, SOCPs, or linear programs) with complexity that scales favorably compared to classical approaches:

Table: Complexity Scaling (Sample Cases)

Method	Scaling wrt. Dimension	Typical Tube Set Type	Reference
Ellipsoidal SDP	$u_k\in \mathbb{R}^{n_u}$4	Ellipsoid	(Parsi et al., 2022)
Zonotopic LP	Linear or quadratic in $u_k\in \mathbb{R}^{n_u}$5	Zonotope	(Diaconescu et al., 24 Sep 2025)
Polyhedral	Combinatorial in $u_k\in \mathbb{R}^{n_u}$6	Polytope	(Diaconescu et al., 24 Sep 2025)

Ellipsoidal and zonotopic elastic tubes alleviate combinatorial growth in constraint count and variables inherent to general polytopic tube parameterizations, thereby enabling medium- to high-dimensional systems (Parsi et al., 2022, Diaconescu et al., 24 Sep 2025, Ghiasi et al., 24 Dec 2025).

Dynamic tube adaptation is enabled via offline computation of invariant or contractive sets (for terminal ingredients and feedback gain synthesis), followed by online optimization of the tube parameters, nominal states/inputs, and constraint tightening. Adaptive frameworks further update parametric uncertainty sets and their tight corresponding tubes as new data are observed (set-membership identification) (Ghiasi et al., 24 Dec 2025, Dey et al., 16 Mar 2026).

4. Robustness, Constraint Satisfaction, and Stability Guarantees

Robust constraint satisfaction is ensured by tightened nominal constraints: $u_k\in \mathbb{R}^{n_u}$7 where the margin is a function of the direction and size of the tube cross-section (e.g., the maximal support function over cross-section, as in $u_k\in \mathbb{R}^{n_u}$8) (Parsi et al., 2022).

Under feasible initializations (offline and terminal set synthesis), ET-MPC schemes ensure:

• Constraint satisfaction for all uncertainty/disturbance realizations (Parsi et al., 2022, Dey et al., 16 Mar 2026, Ghiasi et al., 24 Dec 2025)
• Recursive feasibility: subsequent optimizations remain feasible indefinitely, even as tubes adapt or shrink (Parsi et al., 2022, Diaconescu et al., 24 Sep 2025, Dey et al., 16 Mar 2026)
• Closed-loop stability: regional input-to-state practical stability (ISpS), Lyapunov-based or polyhedral function decrease, exponential convergence in the absence of disturbances or after finite time (Parsi et al., 2022, Ghiasi et al., 24 Dec 2025, Badalamenti et al., 2024)

Adaptation and online learning components guarantee that as parameter uncertainty contracts, the tube geometry shrinks, reducing conservativeness without sacrificing robust stability (Morozov et al., 2020, Dey et al., 16 Mar 2026, Ghiasi et al., 24 Dec 2025).

5. Parameter Learning and Adaptive Elastic Tubes

Adaptive ET-MPC incorporates parameter learning by recursive set-membership updates: $u_k\in \mathbb{R}^{n_u}$9 This iterative refinement enables reduction of model uncertainty sets, which in turn allows for smaller tubes and reduced constraint tightening (Ghiasi et al., 24 Dec 2025, Dey et al., 16 Mar 2026). Online updates of feedback gain $w_k$0, invariant sets, and cost matrices are recomputed as the uncertainty set shrinks, an approach that does not require a common quadratically stabilizing gain for the initial set (Dey et al., 16 Mar 2026).

Backtracking line searches and feasibility recovery mechanisms are often employed to ensure recursive feasibility when aggressive tube tightening could render the online convex program infeasible (Buerger et al., 5 Mar 2026).

6. Set Representations and Computational Tradeoffs

Different tube cross-section parameterizations lead to distinct trade-offs:

• Ellipsoidal tubes provide analytical parameterizations with favorable scaling and can be efficiently handled via LMIs/SDPs (Parsi et al., 2022, Buerger et al., 5 Mar 2026).
• Scaled zonotopic tubes support linear program (LP) formulations with scalable complexity and moderate conservativeness, outperforming polytopic elastic tubes in high dimensions (Diaconescu et al., 24 Sep 2025, Ghiasi et al., 24 Dec 2025).
• Configuration-constrained polytopes allow maximum flexibility but require careful facet-configuration management to avoid combinatorial explosion (Badalamenti et al., 2024).
• Homothetic polytopes and other structured representations remain tractable for low to moderate dimensions, but rapidly lose scalability as state dimension rises (Diaconescu et al., 24 Sep 2025).

Extensive complexity analysis has demonstrated that zonotopic and ellipsoidal elastic tubes scale linearly or quadratically with problem dimension, while combinatorial scaling of polyhedral tubes restricts their practical use to small-scale problems (Diaconescu et al., 24 Sep 2025, Parsi et al., 2022).

7. Applications and Empirical Performance

Elastic Tube MPC has been successfully demonstrated on both simulated and real systems:

• Obstacle avoidance and state-dependent uncertainty handling: ET-MPC dynamically shrinks tubes near obstacles and relaxes them in open spaces (Lopez et al., 2019, Morozov et al., 2020), yielding up to 30–35% reductions in control effort and dramatic increases in allowable closed-loop speeds in nonlinear pendulum and vehicle models (Morozov et al., 2020).
• Learning-based and adaptive robust control: Incorporating set-membership learning, tubes contract as parameter uncertainty decreases, generating less conservative and more aggressive controllers (Ghiasi et al., 24 Dec 2025, Buerger et al., 5 Mar 2026, Dey et al., 16 Mar 2026).
• Scalability in high-dimensional systems: Zonotopic elastic tubes have achieved feasible MPC operation in state spaces up to $w_k$1, a regime intractable for classical polytopic tubes (Diaconescu et al., 24 Sep 2025).
• Tracking and switching references: Configuration-constrained elastic tubes reshape in real time to accommodate reference jumps and online quadratic program (QP) solvers support implementation in autonomous vehicles and other applications (Badalamenti et al., 2024).

Empirical results confirm significant expansion of robust feasibility regions and maintenance of robust invariance in the face of model mismatch and disturbance uncertainty (Ghiasi et al., 24 Dec 2025, Diaconescu et al., 24 Sep 2025).

---

References:

• (Parsi et al., 2022) Scalable tube model predictive control of uncertain linear systems using ellipsoidal sets
• (Lopez et al., 2019) Dynamic Tube MPC for Nonlinear Systems
• (Morozov et al., 2020) Performance Analysis of Adaptive Dynamic Tube MPC
• (Diaconescu et al., 24 Sep 2025) Zonotope-Based Elastic Tube Model Predictive Control
• (Ghiasi et al., 24 Dec 2025) Safe Navigation with Zonotopic Tubes: An Elastic Tube-based MPC Framework
• (Buerger et al., 5 Mar 2026) Robust adaptive NMPC using ellipsoidal tubes
• (Dey et al., 16 Mar 2026) Adaptive Tube MPC: Beyond a Common Quadratically Stabilizing Feedback Gain
• (Badalamenti et al., 2024) Configuration-Constrained Tube MPC for Tracking