⚡ electrical engineering

Zonotope-Based Elastic Tube Model Predictive Control

This paper proposes a novel Zonotope-Based Elastic Tube Model Predictive Control framework that reduces numerical complexity by introducing new scaled-zonotope inclusion conditions, thereby eliminating the need for a priori set-containment constraints while balancing computational effort with an enlarged domain of attraction.

Original authors: Sabin Diaconescu, Florin Stoican, Bogdan D. Ciubotaru, Sorin Olaru

Published 2026-05-19

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sabin Diaconescu, Florin Stoican, Bogdan D. Ciubotaru, Sorin Olaru

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Driving a Car in a Storm

Imagine you are driving a car (the system) and you want to get to a specific destination. You have a GPS (the controller) that plans your route. However, there are two problems:

The Road is Slippery: Wind gusts and rain (disturbances) might push your car off course.
The Car is Imperfect: Your GPS doesn't know the exact weight of the car or the exact friction of the tires (model uncertainty).

If you just plan a single, perfect line on the map, a single gust of wind could make you crash into a wall (violate constraints).

The Solution: The "Elastic Tube"
Instead of planning a single line, this paper proposes planning a tube around your path. As long as your car stays inside this tube, you are safe, even if the wind pushes you around.

Rigid Tube: Imagine a steel pipe. It's the same width everywhere. It's safe, but if the road curves sharply, the pipe might be too wide to fit, or too narrow to be useful.
Homothetic Tube: Imagine a rubber hose that can stretch or shrink, but it must stretch evenly (like blowing up a balloon). It gets bigger or smaller, but keeps the same shape.
Elastic Tube (This Paper): Imagine a super-flexible, stretchy fabric tube. It can stretch more in one direction and less in another, adapting perfectly to the shape of the road and the wind. This allows for a much wider "safe zone" (Domain of Attraction), meaning you can start from further away and still make it safely.

The Problem: The "Math Soup"

While the "stretchy fabric" idea is great, calculating exactly how to stretch it is a nightmare for computers.

In the past, to describe this tube, engineers used complex shapes called polyhedra (like a 3D diamond with many flat faces). To make the math work, they had to add extra "helper variables" (like extra strings holding the fabric) that made the calculation slow and heavy.
As the system gets bigger (more dimensions), the number of faces on these shapes explodes, and the computer gets stuck trying to solve the math.

The Innovation: The "Zonotope" Shortcut

This paper introduces a new way to describe the tube using shapes called Zonotopes.

The Analogy: Think of a zonotope as a shape built by stacking many flat, straight sticks (generators) together. It's like a bundle of straws.
The Magic Trick: The authors found a new mathematical rule (a "linear encoding") that allows the computer to check if one of these stick-bundles fits inside another without needing to count every single corner or add those messy "helper strings."

Previously, checking if a stretched shape fits inside another required solving a complicated, non-linear puzzle (like trying to fit a square peg in a round hole while the peg is changing shape). The authors' new method turns this into a simple, straight-line equation (a linear problem).

What They Actually Did (The Results)

The paper doesn't just talk theory; they tested it with two specific examples:

The Double Integrator: A simple physics model of a moving object (like a sliding block).
- Result: The new "elastic zonotope" method created a much larger safe zone than the old "rigid" or "polyhedral" methods. It allowed the system to start from further away and still be safe.
- Trade-off: It was slightly slower to calculate than the simplest method, but much faster than the complex old methods.
The Coupled Spring Experiment (CSE): A more complex system with many connected springs and masses (simulating a chain of objects).
- Result: When they increased the complexity (added more springs), the old methods (using polyhedra) crashed. They ran out of time or memory.
- Success: The new zonotope method kept working perfectly, even with 20 different moving parts. It proved that this approach scales up to handle much bigger, more complex systems without breaking the computer.

Key Takeaways

Flexibility: The new method allows the "safety tube" to stretch and shrink in specific directions, giving the controller more freedom to maneuver.
Simplicity: They replaced a complex, non-linear math problem with a simple, linear one. This means the computer doesn't have to work as hard to find a solution.
Scalability: The method works for systems with many variables (high dimensions) where older methods fail completely.
No "Magic" Variables: They removed the need to pre-select certain "helper" variables, making the system more automatic and less prone to human error in setup.

Summary

Think of this paper as inventing a smart, stretchy safety net for robots and machines. Instead of using a heavy, rigid cage or a complicated, hard-to-calculate net, they created a flexible net made of "stick-bundles" (zonotopes) that is easy for computers to calculate. This allows machines to operate safely in messy, unpredictable environments, even when the machines are very complex and have many moving parts.

Technical Summary: Zonotope-Based Elastic Tube Model Predictive Control

Problem Statement
Tube-based Model Predictive Control (MPC) is a robust framework for constrained linear systems subject to additive disturbances, ensuring all future trajectories remain within a "tube" of sets. While effective, the numerical tractability of tube-based MPC is often limited by the complexity of set parameterizations. Traditional polyhedral approaches suffer from a quadratic growth in decision variables relative to the horizon length or require complex descriptions that increase problem size. Homothetic (scaled) and elastic (offset-controlled) tubes offer enlarged domains of attraction but often introduce bilinearities or require pre-selecting auxiliary variables to handle set inclusions, leading to suboptimal parameter selection or high computational burdens, particularly in higher-dimensional systems.

Methodology
The paper proposes a novel Elastic Tube MPC framework utilizing scaled zonotopes to parameterize the tube cross-sections. The core methodology involves:

Scaled Zonotope Representation: The tube is defined as a sequence of zonotopes $\langle c_k, G\Delta_k \rangle$ , where $c_k$ is the center, $G$ is a fixed generator matrix, and $\Delta_k = \text{diag}(\delta_k)$ represents time-varying scaling factors. This allows the tube to stretch and shrink elastically along specific directions.
Linear Inclusion Conditions: A new class of scaled-zonotope inclusion conditions is derived. Unlike previous methods that rely on bilinear terms or require the pre-selection of auxiliary variables (like $\Gamma$ $Γ$ ) to handle scaling, the proposed approach uses a matrix $\Phi$ $Φ$ to linearly encode the containment of Minkowski sums of zonotopes.
- Theorem 1 formulates the MPC optimization as a Quadratic Program (QP) where the one-step reachability and state/input admissibility constraints are expressed as linear equalities and inequalities involving the scaling factors $\delta_k$ and the auxiliary matrix $\Phi_k$ .
- Crucially, the zonotope centers $c_k$ are treated as independent decision variables, enhancing flexibility without introducing non-convexities.
Pre-computation Alternatives: To further reduce runtime complexity, the paper introduces two variants:
- $\Phi_0$ -based: A constant matrix $\Phi_0$ is pre-computed offline, replacing the optimization of $\Phi_k$ at each step with a linear constraint on scaling factors.
- $\Gamma$ -based: A pre-computed pair $(\Gamma, \gamma)$ is used, similar to earlier polyhedral approaches but adapted for zonotopes.
Robust Positively Invariant (RPI) Sets: The generator matrix $G$ is derived from an RPI set approximation for the error dynamics. The paper enhances this by treating the RPI set center as a decision variable in the minimization problem, reducing the number of decision variables compared to prior work.
Stability and Feasibility: Recursive feasibility and asymptotic stability are guaranteed by constructing a terminal set via set recurrence and enforcing Lyapunov-like contraction conditions on the terminal cost and scaling factors.

Key Contributions

Novel Linear Encoding: The paper introduces a linear encoding for zonotope containment (Corollary 1) that generalizes standard conditions. This eliminates the need for pre-selecting auxiliary variables to circumvent nonlinear optimization, a common issue in polyhedral scaling.
Computational Efficiency: The proposed formulation remains linear in both auxiliary variables and scaling factors, facilitating integration into large-scale optimization frameworks. It avoids the bilinearities typically associated with scaling factors in other tube MPC methods.
Complexity Reduction: By leveraging zonotopes, the method significantly reduces the number of constraints and scaling factors required compared to polyhedral elastic tubes, especially as system dimension increases.
Explicit Trade-off Analysis: The paper provides a comprehensive complexity analysis comparing rigid, homothetic, and elastic tubes across polyhedral and zonotopic settings, illustrating the trade-off between the enlarged domain of attraction and computational effort.

Results
The proposed approach was validated through numerical experiments on two benchmarks:

Double Integrator: Comparisons between polyhedral elastic tubes, zonotopic elastic tubes with pre-computed $\Phi_0$ , and zonotopic elastic tubes with optimized $\Phi$ showed that zonotopic variants yield larger domains of attraction. The fully optimized $\Phi$ variant offered a slight advantage over the pre-computed $\Phi_0$ variant but at a higher computational cost.
Coupled Spring Experiment (CSE): This high-dimensional system (up to 20 states) demonstrated the scalability of the method. Polyhedral implementations failed to find feasible solutions for systems with 3 or more springs due to computational limits. In contrast, all zonotopic variants successfully solved the problem across the entire range of parameters.
- Runtime: The zonotopic methods exhibited linear computational complexity (or quadratic for the $\Phi$ variant), whereas polyhedral methods became intractable.
- Centers: Treating zonotope centers as decision variables resulted in only marginal computational overhead while providing crucial robustness guarantees.

Significance
The paper claims that integrating elastically-scaled zonotopes into tube MPC significantly reduces runtime and improves scalability for high-dimensional systems compared to traditional polyhedral methods. By alleviating the need for a priori specification of set-containment constraints and avoiding bilinearities, the approach achieves a more efficient balance between the domain of attraction and computational effort. The work extends preliminary research on elastic tube MPC by providing a computationally efficient, precomputation-based alternative and establishing rigorous conditions for recursive feasibility and stability. The authors note that while the method is robust, future work could address the sensitivity to the initial RPI set and potential extensions to data-driven MPC.