Computationally Efficient State and Model Estimation via Interval Observers for Partially Unknown Systems
Mohammad Khajenejad, Zeyuan Jin · 2025
Read Computationally Efficient State and Model Estimation via Interval Observers for Partially Unknown Systems by Mohammad Khajenejad, Zeyuan Jin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-09784.txt.
Introduction
Mohammad Khajenejad and Zeyuan Jin design observers that bracket the truth about a system they only partly understand, in this 2025 paper. An interval observer does not guess one state; it maintains upper and lower bounds guaranteed to contain the real one, even with bounded noise and unknown dynamics. Computing those bounds tightly, without drowning in computation, is the art.
Their construction leans on Jacobian sign-stable decompositions and tight decomposition functions for the nonlinear parts. A data-driven step over-approximates the unknown dynamics from current and past interval estimates, then folds those bounds back into the observer, so the model of the unknown improves as it runs. Observer gains come from semidefinite programs that enforce stability and optimality. Simulations showed the method running cheaper than the authors' own earlier approach, a practical step for control systems that must certify their estimates.
Catalog record
| CREATOR | Mohammad Khajenejad, Zeyuan Jin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-09784
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Computationally Efficient State and Model Estimation via Interval Observers for Partially Unknown Systems Thanks: M. Khajenejad is with the Department of Mechanical Engineering, The University of Tulsa, Tulsa, OK, USA (e-mail: [email protected]). Z. Jin is with SafeAI, Inc., Santa Clara, CA, USA (e-mail: [email protected]). Mohammad Khajenejad, Zeyuan Jin Abstract This paper addresses the synthesis of interval observers for partially unknown nonlinear systems subject to bounded noise, aiming to simultaneously estimate system states and learn a model of the unknown dynamics. Our approach leverages Jacobian sign-stable (JSS) decompositions, tight decomposition functions for nonlinear systems, and a data-driven over-approximation framework to construct interval estimates that provably enclose the true augmented states. By recursively computing tight and tractable bounds for the unknown dynamics based on current and past interval framers, we systematically integrate these bounds into the observer design. Additionally, we formulate semi-definite programs (SDP) for observer gain synthesis, ensuring input-to-state stability and optimality of the proposed framework. Finally, simulation results demonstrate the computational efficiency of our approach compared to a method previously proposed by the authors. I Introduction Motivated by the need to ensure safe and smooth operation under various forms of uncertainties in many safety-critical engineering applications such as fault detection, urban transportation, attack (unknown input) mitigation and detection in cyber-physical systems and aircraft tracking [ 1 , 2 , 3 ] , robust set-valued algorithms for state and input estimation have been recently developed to find compatible state estimates. In addition, dynamic models of many practical systems are often only partially known. Thus, the development of algorithms that can combine model learning and set membership estimation approaches is a critical and interesting problem. Literature review . Various model-based approaches have been proposed in the literature for designing set or interval observers across different system classes [ 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 3 , 17 , 18 , 19 ] , These include linear time-invariant (LTI) systems [ 10 ] , linear parameter-varying (LPV) systems [ 12 , 16 ] , Metzler and/or partially linearizable systems [ 9 , 11 ] , cooperative systems [ 8 , 9 ] , Lipschitz nonlinear systems [ 13 ] , monotone nonlinear systems [ 6 , 7 ] and uncertain nonlinear systems [ 14 ] systems. However, many of these methods do not account for the presence of unknown inputs—which may arise from external agents, disturbances, attacks, or simply unobserved signals—nor do they address unknown dynamics. More recent efforts have tackled the design of set-valued observers capable of simultaneously estimating both states and unknown inputs, particularly for LTI [ 3 ] , LPV [ 17 ] , switched linear [ 18 ] and nonlinear systems [ 19 , 20 ] considering bounded-norm noise. On the other hand, when the system model is not precisely known, set-valued data-driven approaches have gained increasing attention in recent years [ 21 , 22 , 23 , 24 , 25 ] . These methods leverage input-output data to abstract or over-approximate unknown dynamics or functions, aiming to identify a set of dynamics that effectively frame or bracket the true system behavior [ 21 , 22 ] . This problem has been studied under different regularity assumptions on the unknown dynamics, including univariate Lipschitz continuity [ 23 ] , multivariate Lipschitz continuity [ 24 ] , and Hölder continuity [ 25 ] . In our previous work [ 26 ] , we initiated an approach that combined model abstractions [ 27 ] with specific types of decomposition functions [ 20 ] to construct interval framers for partially known dynamics subject to bounded noise. This framework accommodated a broad class of nonlinear state and observation mappings while treating the mapping of the unknown input dynamics as an entirely unknown function. However, the method in [ 26 ] required computationally expensive online optimizations to obtain affine over-approximations/abstractions of nonlinear dynamics at each time step. Moreover, it was highly case-sensitive, as it did not incorporate any stabilizing gain synthesis. As a result, the stability of the obtained interval-valued estimates was heavily dependent on the specific system dynamics. This paper proposes an alternative approach to address these limitations and ensure a more robust and systematic design. Contributions. This work bridges the gap between model-based set membership observer design approaches and data-driven function approximation methods (i.e., model learning techniques). Specifically, our approach extends observer design methodologies from systems with fully known yet noisy dynamics to partially unknown systems. By leveraging data-driven over-approximation and abstraction of unknown dynamics [ 27 ] we develop a recursive framework that refines the estimation of the unknown system components using noisy observation data and interval estimates. Our method constructs tight and tractable bounds for the unknown dynamics as a function of current and past interval framers, allowing these bounds to be systematically integrated into the observer system. Furthermore, we prove that the proposed observer maintains correctness, ensuring that the framer property [ 11 ] holds. We also show that our estimation and abstraction of the unknown dynamics become progressively more precise and tighter over time. Finally, we exploit Jacobian sign-stable decompositions and tight mixed-monotone decomposition of nonlinear systems [ 28 ] and formulate a tractable semi-definite program (SDP) for observer gain synthesis. This ensures input-to-state stability and guarantees the optimality of the proposed design. II Preliminaries Notation. \mathbb{N}_{n} , \mathbb{N} , \mathbb{R}^{n\times p} , \mathbb{R}^{n} , and \mathbb{R}^{n}_{>0} represent, respectively, the natural numbers up to n , the set of natural numbers, matrices of size n by p , n -dimensional Euclidean space, and positive vectors of dimension n . For a vector v\in\mathbb{R}^{n} , the p -norm is defined as \|v\|_{p}\triangleq\left(\sum_{i=1}^{n}{|v_{i}|^{p}}\right)^{\frac{1}{p}} . For simplicity, the 2 -norm of v is denoted by \|v\| . For a matrix M\in\mathbb{R}^{n\times p} , the i -th row and j -th column entry is denoted by M_{ij} . We define M^{\oplus}\triangleq\max(M,\mathbf{0}_{n\times p}) , M^{\ominus}\triangleq M^{\oplus}-M , and |M|\triangleq M^{\oplus}+M^{\ominus} , which represents the element-wise absolute value of M . Additionally, we use M\succ 0 and M\prec 0 (or M\succeq 0 and M\preceq 0 ) to indicate that M is positive definite and negative definite (or positive semi-definite and negative semi-definite), respectively. All vector and matrix inequalities are element-wise inequalities. The zero vectors in \mathbb{R}^{n} , and zero matrices of size n\times p are denoted by \mathbf{0}_{n\times p} and \mathbf{0}_{n} , respectively. Moreover, a continuous function \alpha:[0,a)\to\mathbb{R}_{\geq 0} belongs to class \mathcal{K} if it is strictly increasing with \alpha(0)=0 . Lastly, a continuous function \lambda:{[0,a)\times[0,\infty)}\to\mathbb{R}_{\geq 0} is considered to be in class \mathcal{KL} if, for every fixed t\geq 0 , the function \lambda(s,t) is in class \mathcal{K} ; for each fixed s\geq 0 , \lambda(s,t) decreases with respect to t and satisfies \lim_{t\to\infty}\lambda(s,t)=0 . Definition 1 (Interval) . \mathcal{I}\triangleq[\underline{z},\overline{z}]\subset\mathbb{R}^{n} is an n -dimensional interval defined as the set of all vectors z\in\mathbb{R}^{n_{z}} such that \underline{z}\leq z\leq\overline{z} . Interval matrices can be similarly defined. Proposition 1 . [ 13 , Lemma 1] Let A\in\mathbb{R}^{m\times n} and \underline{x}\leq x\leq\overline{x}\in\mathbb{R}^{n} . Then , A^{\oplus}\underline{x}-A^{\ominus}\overline{x}\leq Ax\leq A^{\oplus}\overline{x}-A^{\ominus}\underline{x} . As a corollary, if A is non-negative, then A\underline{x}\leq Ax\leq A\overline{x} . Definition 2 (Lipschitz Continuity) . The mapping