REVIEW 2 major objections 6 minor 1 cited by
Learning Robust Safety Controllers for Uncertain Input-Affine Polynomial Systems
T0 review · 2 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash
Pith's one-line read From a single finite-horizon input-state trajectory, an SOS program can synthesize a robust safety certificate and controller for an unknown discrete-time input-affine polynomial system, guaranteeing that trajectories starting from a…
desk verdict Solid extension of single-trajectory data-driven barrier synthesis, but Theorem 2 quietly assumes L†L = I, which breaks the SOS formulation for general dictionaries. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing device is a data-based reformulation of the closed loop: with $u=U_{0,T}Q(x)x$ and $Q(x)$ chosen so that $R_{0,T}Q(x)=L(x)$, where $R(x)=L(x)x$, the unknown dynamics become $x^{+}=\tilde{A}\tilde{R}(x)Q(x)x+w$, with $\tilde{A}=[A\; B]$ and $\tilde{R}(x)=[R_{0,T};\; G(x)U_{0,T}]$ built entirely from measured data. This reduces the barrier condition to a matrix inequality in $H(x)$ and $P$; block-matrix elimination and an $S$-procedure rewrite it as the SOS condition (21c), avoiding the bilinear term $H(x)PH(x)^{\top}$ by dilation. The rank condition that $R_{0,T}$ be full row rank is what allows $Q(x)$ to exist and encodes persistency of excitation.
What would settle it
Take the academic system (22), add a monomial of degree four to its true dynamics while keeping the declared maximum degree at three, collect a trajectory that is full row rank for this reduced dictionary, and run the SOS conditions (21a)--(21c). If they return a feasible $P$ and controller, simulate many disturbance realizations bounded by $\delta$ from $X_0$; any trajectory reaching $X_1$ would refute the infinite-horizon claim.
Extended reading notes
Core claim
The paper's central claim is that for an unknown system of the form $x^{+}=A R(x)+B G(x)u+w$ whose chosen dictionaries contain the true monomials, if the collected data make the matrix $R_{0,T}$ full row rank and there exist matrices $H(x)$ and $P$ satisfying $R_{0,T}H(x)=L(x)P^{-1}$, then the SOS feasibility conditions (21a)--(21c) produce a quadratic $B(x)=x^{\top}P x$ that is a robust control barrier certificate and $u(x)=U_{0,T}H(x)[L(x)^{\dagger}R_{0,T}H(x)]^{-1}x$ is its associated robust safety controller. The certificate uses a relaxed decay condition $B(x^{+})\le \lambda B(x)+\rho\|w\|^2$ with $c=\rho\delta\le \gamma_2(1-\lambda)$, so the barrier may grow by a bounded amount instead of strictly decaying. Consequently, by the safety-guarantee theorem, all trajectories from $X_0$ avoid $X_1$ for any disturbance sequence with $\|w\|^2\le \delta$ over an infinite horizon.
Load-bearing premise
The whole construction stands on knowing in advance a degree bound large enough that the true monomials of the dynamics appear in the chosen dictionaries $R(x)$ and $G(x)$, and on the collected trajectory making the monomial matrix $R_{0,T}$ full row rank with a factorization $R(x)=L(x)x$; if the bound is too low or the data are not exciting enough, the synthesized certificate need not describe the real system.
Editorial extensions
If this is right
- For any system satisfying the assumptions, the synthesized controller guarantees robust safety for all time, not only for the horizons seen in the data.
- No system identification or disturbance estimation step is needed; the same data that certify safety also define the controller.
- Because the barrier only needs to decay relative to a constant, certificates become feasible for systems where a strictly decreasing barrier would not exist.
- The framework accommodates multiple disjoint unsafe regions by repeating the barrier condition for each region, as demonstrated in both case studies.
- After fixing the scalar parameters $\lambda$ and $\pi$, the design is a convex SOS and semidefinite program, so the synthesis can be automated.
- The method extends to state-dependent control matrices $G(x)$, going beyond static input matrices treated in prior discrete-time data-driven safety work.
Reading between the lines
- Editorial inference: the same rank-conditioned data representation should apply to non-quadratic barrier candidates of the form $B(x)=R(x)^{\top}P R(x)$ whenever the dictionary is rich enough to express $R(x^{+})$ in the data coordinates; the paper's quadratic choice $B(x)=x^{\top}P x$ is a special case.
- Editorial inference: the degree-bound assumption is the practical point of failure. Before deployment, one could validate the certificate on a second independent trajectory and check that the defining barrier inequality holds on it; a violation would indicate a missing monomial in the dictionary.
- Editorial inference: the scalar parameters $\lambda$ and $\pi$ are fixed a priori, and tuning them trades feasibility against the allowable disturbance size $c= (1+1/\pi)\|\sqrt{P}\|^2\delta$; a systematic line search over these parameters is a natural extension that the paper leaves implicit.
- Editorial inference: the factorization $R(x)=L(x)x$ with $L^{\dagger}L=I$ is asserted for general monomial dictionaries, but it holds cleanly only when the dictionary contains the coordinate monomials; the paper does not list this as an explicit assumption in Theorem 2.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper develops a direct data-driven method for synthesizing robust control barrier certificates (R-CBCs) B(x) = x^T P x and associated robust safety controllers (R-SCs) for discrete-time input-affine polynomial systems (dt-IAPS) with unknown dynamics and unknown-but-bounded disturbances, using a single finite input-state trajectory satisfying a persistency-of-excitation rank condition. The main results are Lemma 1 (data-based closed-loop representation), Theorem 1 (infinite-horizon safety guarantee from an R-CBC), Theorem 2 (sufficient conditions (14a)-(14c) derived via the S-procedure and Schur complement), Lemma 2 (SOS reformulation (21a)-(21c)), Algorithm 1, and two case studies (an academic system and a Lorenz system). The central claim is that if R0,T has full row rank and conditions (13) and (21a)-(21c) are feasible, then B(x) = x^T [L(x)^dagger R0,T H(x)]^-1 x is an R-CBC and u = U0,T H(x) [L(x)^dagger R0,T H(x)]^-1 x is its R-SC, guaranteeing that all trajectories from X0 avoid X1 for all disturbances with ||w||^2 <= delta over an infinite horizon.
Significance. If the gaps identified in the major comments are fixed, this is a worthwhile contribution to direct data-driven safety synthesis for discrete-time nonlinear systems. The construction is a legitimate certificate-synthesis framework: no quantity is fitted to data and then re-used as a prediction, and the design is validated against external benchmarks (De Persis-Tesi data representations, barrier-certificate theory, the S-procedure). The S-procedure argument over the data-conformity set and the Schur-complement dilation that removes the bilinear term H(x) P H(x)^T are correct and implementable as SOS, and the relaxed barrier condition (3c) with the rho ||w||^2 term is a sensible improvement over strict-decay conditions in prior single-trajectory discrete-time results. Treating state-dependent input matrices G(x) and multiple unsafe regions also extends existing work. The two case studies, including a chaotic Lorenz system, substantiate the practical claims. The main shortcomings are a missing hypothesis in Theorem 2 and a well-posedness gap in Lemma 2, both fixable without changing the manuscript's scope, plus a numerical inconsistency in the reported case-study parameters.
major comments (2)
- [Theorem 2, Section III (Eqs. (13)-(14))] The stress-test concern is valid: Theorem 2 omits a hypothesis that is load-bearing for the identification of the certificate matrix. The conclusion and the proof set P = [L(x)^dagger R0,T H(x)]^-1, but condition (13) only implies L(x)^dagger R0,T H(x) = L(x)^dagger L(x) P^-1; the desired equality L(x)^dagger R0,T H(x) = P^-1 requires L(x)^dagger L(x) = I_n. The theorem never assumes that L(x) in (8) has a left inverse (for example, that the dictionary contains the linear monomials x1,...,xn as its first entries), and Remark 3's "without loss of generality" claim addresses only the existence of L, not its left-invertibility. Without L^dagger L = I_n, the P appearing in (14c) and in c = (1 + 1/pi) ||sqrt(P)||^2 delta is not the matrix appearing in (14a)-(14b), so the verification of (3a)-(3b) in the proof fails; moreover, for a general monomial dictionary L^dagger is a rational function of x, making (14a)-(14b) and the SOS expressions (21a)-(21b) non-polynomial and Lemma 2 ill-defined. The case studies happen to satisfy the missing hypothesis only because their dictionaries list the linear monomials first, so L = [I_n; M(x)] and L^dagger = [I_n, 0] is a constant polynomial matrix. Please add this assumption to Theorem 2 and Lemma 2, and note that it is without loss of generality because redundant monomials, including the linear ones, can always be added to the dictionary as permitted by Remark 7.
- [Lemma 2, Section III (Eqs. (21a)-(21c)); Algorithm 1] The SOS formulation in Lemma 2 is not well-posed as written and does not match the actual implementation. Conditions (21a)-(21b) declare expressions in x^T [L(x)^dagger R0,T H(x)]^-1 x to be SOS polynomials in the decision variables H(x), but for a free polynomial matrix H(x) this expression is generally a rational function of x, so the proof's assertion that B(x) is a non-negative SOS polynomial is unjustified at the point where feasibility is being decided. The intended procedure, evident from Remark 5 and both case studies ("we set Z = P^-1"), is to solve constraints (21c) and (13) for Z and H, recover P = Z^-1, and then compute the level sets gamma1 and gamma2 from (14a)-(14b), which is exactly what Algorithm 1 does in Steps 4-6. Please rewrite Lemma 2 with Z (= P^-1) and H as decision variables, impose (13) as the polynomial equality R0,T H = L Z, and state that (21a)-(21b) are level-set verifications performed after P is recovered (or impose them with the additional constraint P Z = I_n); this aligns the lemma with Algorithm 1 and makes the SOS problem polynomial.
minor comments (6)
- [Section IV (case studies)] The reported case-study parameters are inconsistent with the formula rho = (1 + 1/pi) ||sqrt(P)||^2 and the induced-2-norm definition of ||sqrt(P)||: for the academic system, lambda_max(P) is approximately 7.73 x 10^4, so with pi = 10^-5 one obtains rho approximately 7.7 x 10^9 rather than the reported rho = 1.4257 x 10^5, and the Lorenz case has the same problem (lambda_max(P) is on the order of 10^5). Please correct pi, P, or rho so that the reported values satisfy the hypotheses of Theorem 2 and the numbers are reproducible.
- [Section II-C, Theorem 1 (proof)] The induction in the proof of Theorem 1 applies condition (3c) at x = x(k), which requires x(k) to lie in X~ = X intersect {x : B(x) < gamma2}, but the induction hypothesis only establishes B(x(k)) < gamma2. Please add the assumption {x : B(x) < gamma2} is a subset of X (which holds in both case studies because the ellipsoid B < gamma2 lies inside the box X) or otherwise close the gap in the infinite-horizon argument.
- [Section III, Eqs. (20) and (14c)] The claim that inequality (20) is equivalent to (14c) is not correct as stated: the Schur complement shows that (14c) is equivalent to the (n + N + N-hat)-dimensional matrix inequality M(x) <= 0, which is only sufficient for the congruence (20) with [I_n; A~^T], not necessary. Since the proof ultimately uses (14c) as a sufficient condition, the argument is valid, but the equivalence wording should be corrected.
- [Section II-B and Section III (Eqs. (6a), (14c))] The paper states that R(x) and G(x) are unknown, yet G0,T in (6a) and G(x) in R~(x) in (14c) are used as known quantities computed from data; please clarify that the monomial dictionaries are known functions of x and only the coefficient matrices A and B are unknown (Remark 7 makes this point for R(x) only).
- [Section III, Lemma 2] The multiplier alpha(x) must be explicitly constrained to be pointwise nonnegative (for example, SOS) in the SOS program; writing alpha(x) in R_+ as in (14c) does not guarantee that the solver returns a nonnegative polynomial, and a negative multiplier would invalidate the S-procedure step in the proof of Theorem 2.
- [Section II-B and case studies] The input constraint set U is never specified; the constructed controller u = U0,T H(x) P x is assumed to be feasible for all x in X~, which implicitly requires U = R^m. Please state this assumption, or add input constraints to the SOS formulation.
Circularity Check
No significant circularity: the barrier and controller are SOS feasibility variables solved jointly from the De Persis–Tesi data representation, not fitted-then-predicted quantities; Theorem 1 is proven in full in-paper, and the only caveat (an unstated full-column-rank condition on L(x) in Theorem 2) is a soundness gap rather than a self-referential reduction.
full rationale
The derivation chain is self-contained. Lemma 1 is the standard De Persis–Tesi / Guo et al. data-based representation (Eqs. (7)–(12)), applied to an openly declared quadratic ansatz B(x)=x^T P x (Section III), and Theorem 2 reduces the R-CBC conditions (3a)–(3c) to the matrix conditions (14a)–(14c) using only Schur complements and the S-procedure over the data-conformity set (19). The infinite-horizon safety claim (Theorem 1) is an induction proven in full inside the paper, so its attribution to [32] (which shares an author with this paper) is not load-bearing; the proof does not import an unverified external result. No fitted-input-called-prediction pattern exists: P and H(x) are SOS feasibility variables solved jointly, λ and π are fixed a-priori (Remark 5), and γ1, γ2 are computed after the solve and checked against c ≤ γ2(1−λ) in Algorithm 1, so no quantity is fitted to data and then reported as a prediction of that same data. Self-citations in the introduction ([20], [26], [27], [29]) are related-work comparisons, not premises of the main theorem. The one genuine weakness is a missing hypothesis in Theorem 2 / Lemma 2: identifying P=[L(x)† R_{0,T} H(x)]^{-1} with the constant matrix P that enters (14c) and (21c) requires L(x)† L(x)=I (equivalently, L(x) full column rank for each x), and L(x)† is rational in general, so (21a)–(21b) are polynomial SOS constraints only when the dictionary contains the linear monomials x1,…,xn (as in both case studies, where L=[I; M(x)] and L†=[I, 0]). That is a soundness/generality gap, not a circular reduction: no conclusion of the paper is re-imported as a premise, and the two demonstrated certificates satisfy the stated conditions and are externally validated by simulation against the true systems.
Assumptions & free parameters
free parameters (3)
- lambda (barrier decay rate) =
0.99 in both case studies
- pi (Young's inequality weight) =
1e-5 in both case studies
- SOS degrees of H(x), alpha(x), varpi(x) =
not reported
assumptions (5)
- domain assumption The unknown system is exactly a dt-IAPS (1) with R(x), G(x) contained in dictionaries whose maximum degree is known a priori.
- domain assumption The collected data satisfy the full-row-rank condition on R0,T (persistency of excitation).
- domain assumption The dictionary is ordered so that L(x) in (8) satisfies L^dagger L = I, i.e., the first n rows of R(x) are the state variables, making P constant.
- domain assumption The disturbance bound delta is known exactly.
- standard math Standard tools: Schur complement, S-procedure, Cauchy-Schwarz and Young inequalities are valid.
Cite this review
Pith. "Pith review of Learning Robust Safety Controllers for Uncertain Input-Affine Polynomial Systems." pith.science (2026). https://pith.science/paper/LNQ4E7WZ
@misc{pith2026241203919,
author = {Pith},
title = {Pith review of: Learning Robust Safety Controllers for Uncertain Input-Affine Polynomial Systems},
year = {2026},
howpublished = {\url{https://pith.science/paper/LNQ4E7WZ}},
note = {Machine review of arXiv:2412.03919}
}
read the original abstract
This paper offers a direct data-driven approach for learning robust control barrier certificates (R-CBCs) and robust safety controllers (R-SCs) for discrete-time input-affine polynomial systems with unknown dynamics under unknown-but-bounded disturbances. The proposed method relies on data from input-state observations collected over a finite-time horizon while satisfying a specific rank condition to ensure the system is persistently excited. Our data-driven scheme enables the synthesis of R-CBCs and R-SCs directly from observed data, bypassing the need for explicit modeling of the system's dynamics and thus ensuring robust system safety against disturbances within an infinite time horizon. Our proposed approach is formulated as a sum-of-squares (SOS) optimization problem, providing a structured design framework. Two case studies showcase our method's capability to provide robust safety guarantees for unknown input-affine polynomial systems under bounded disturbances, demonstrating its practical effectiveness.
Figures
Forward citations
Cited by 1 Pith paper
-
Data-Driven Stochastic Control: Foundations and Guarantees
A data-driven framework synthesizes stochastic control barrier certificates and safety controllers for unknown polynomial stochastic systems, with probabilistic safety guarantees and certified confidence from multiple...
Reference graph
Works this paper leans on
-
[29]
B. Samari, O. Akbarzadeh, M. Zaker, and A. Lavaei, “From a Single Trajectory to Safety Controller Synthesis of Discrete-Tim e Nonlinear Polynomial Systems,” arXiv: 2409.10026 , 2024
arXiv 2024
-
[1]
Analysis and desig n of safety- critical, cyber-physical systems,
J. McGregor, D. Gluch, and P . Feiler, “Analysis and desig n of safety- critical, cyber-physical systems,” ACM SIGAda Ada Letters , vol. 36, no. 2, pp. 31–38, 2017
work page 2017
-
[2]
From model-based control to data-d riven control: Survey, classification and perspective,
Z. S. Hou and Z. Wang, “From model-based control to data-d riven control: Survey, classification and perspective,” Information Sciences , vol. 235, pp. 3–35, 2013
work page 2013
-
[3]
Formulas for data-driven contr ol: Stabi- lization, optimality, and robustness,
C. De Persis and P . Tesi, “Formulas for data-driven contr ol: Stabi- lization, optimality, and robustness,” IEEE Transactions on Automatic Control, vol. 65, no. 3, pp. 909–924, 2019
work page 2019
-
[4]
Bridging dire ct and indirect data-driven control formulations via regularizations and relaxations,
F. D¨ orfler, J. Coulson, and I. Markovsky, “Bridging dire ct and indirect data-driven control formulations via regularizations and relaxations,” IEEE Transactions on Automatic Control , vol. 68, no. 2, pp. 883–897, 2022
work page 2022
-
[5]
Z. Wang, R. Lu, F. Gao, and D. Liu, “An Indirect Data-Drive n Method for Trajectory Tracking Control of a Class of Nonlinear Disc rete-Time Systems,” IEEE Transactions on Industrial Electronics , vol. 64, no. 5, pp. 4121–4129, 2017
work page 2017
-
[6]
On Direct vs Indirect Da ta-Driven Predictive Control,
V . Krishnan and F. Pasqualetti, “On Direct vs Indirect Da ta-Driven Predictive Control,” in Proceeding of 60th IEEE Conference on Decision and Control (CDC) , pp. 736–741, 2021
work page 2021
-
[7]
Dat a-driven control of nonlinear systems: An on-line direct approach,
M. Tanaskovic, L. Fagiano, C. Novara, and M. Morari, “Dat a-driven control of nonlinear systems: An on-line direct approach,” Automatica, vol. 75, pp. 1–10, 2017
2017
Show all 40 references
-
[8]
Direct Data-Dri ven Control of Constrained Systems,
D. Piga, S. Formentin, and A. Bemporad, “Direct Data-Dri ven Control of Constrained Systems,” IEEE Transactions on Control Systems Technology, vol. 26, no. 4, 2018
2018
-
[9]
Safety V erification of Hybri d Systems Using Barrier Certificates,
S. Prajna and A. Jadbabaie, “Safety V erification of Hybri d Systems Using Barrier Certificates,” in Hybrid Systems: Computation and Control, 2004
2004
-
[10]
Constructive safety usin g control barrier functions,
P . Wieland and F. Allg¨ ower, “Constructive safety usin g control barrier functions,” IF AC Proceedings V olumes, vol. 40, no. 12, pp. 462–467, 2007
2007
-
[11]
Control barrier functions: Theory and appl ications,
A. D. Ames, S. Coogan, M. Egerstedt, G. Notomista, K. Sre enath, and P . Tabuada, “Control barrier functions: Theory and appl ications,” in Proceedings of the 18th European control conference (ECC) , pp. 3420–3431, IEEE, 2019
2019
-
[12]
Control barrier functions for stochastic sy stems,
A. Clark, “Control barrier functions for stochastic sy stems,” Automat- ica, vol. 130, 2021
2021
-
[13]
From Small-Gain The ory to Compositional Construction of Barrier Certificates for L arge-Scale Stochastic Systems,
M. Anand, A. Lavaei, and M. Zamani, “From Small-Gain The ory to Compositional Construction of Barrier Certificates for L arge-Scale Stochastic Systems,” IEEE Transactions on Automatic Control, vol. 67, no. 10, pp. 5638–5645, 2022
2022
-
[14]
Reactive synthesis of sto chastic control systems: A mode-triggered safety barrier approach,
A. Nejati and A.-K. Schmuck, “Reactive synthesis of sto chastic control systems: A mode-triggered safety barrier approach,” in 2024 IEEE 63rd Conference on Decision and Control (CDC) , pp. 5225–5230, 2024
2024
-
[15]
Scalable synthesis of safet y barrier certifi- cates for networks of stochastic switched systems,
A. Lavaei and E. Frazzoli, “Scalable synthesis of safet y barrier certifi- cates for networks of stochastic switched systems,” IEEE Transactions on Automatic Control , vol. 69, no. 11, pp. 7294–7309, 2024
2024
-
[16]
The scenario approach to robust control design,
G. C. Calafiore and M. C. Campi, “The scenario approach to robust control design,” IEEE Transactions on Automatic Control , vol. 51, no. 5, pp. 742–753, 2006
2006
-
[17]
Dat a-driven verification and synthesis of stochastic systems via barrie r certificates,
A. Salamati, A. Lavaei, S. Soudjani, and M. Zamani, “Dat a-driven verification and synthesis of stochastic systems via barrie r certificates,” Automatica, vol. 159, 2024
2024
-
[18]
Data-driven synthesis of safe ty controllers via multiple control barrier certificates,
A. Nejati and M. Zamani, “Data-driven synthesis of safe ty controllers via multiple control barrier certificates,” IEEE Control Systems Letters, vol. 7, pp. 2497–2502, 2023
2023
-
[19]
A note on persistency of excitation,
J. C. Willems, P . Rapisarda, I. Markovsky, and B. L. M. De Moor, “A note on persistency of excitation,” Systems & Control Letters , vol. 54, no. 4, pp. 325–329, 2005
2005
-
[20]
Data-driven dynamic controll er synthesis for discrete-time general nonlinear systems,
B. Samari and A. Lavaei, “Data-driven dynamic controll er synthesis for discrete-time general nonlinear systems,” in Proceedings of the 28th ACM International Conference on Hybrid Systems: Compu tation and Control, pp. 1–12, 2025
2025
-
[21]
Behavioral systems theor y in data-driven analysis, signal processing, and control,
I. Markovsky and F. D¨ orfler, “Behavioral systems theor y in data-driven analysis, signal processing, and control,” Annual Reviews in Control , vol. 52, pp. 42–64, 2021
2021
-
[22]
Data-dri ven design of safe control for polynomial systems,
A. Luppi, A. Bisoffi, C. De Persis, and P . Tesi, “Data-dri ven design of safe control for polynomial systems,” European Journal of Control , vol. 75, 2024
2024
-
[23]
Synthesizing Safe ty Con- trollers for Uncertain Linear Systems: A Direct Data-drive n Ap- proach,
B. Zhong, M. Zamani, and M. Caccamo, “Synthesizing Safe ty Con- trollers for Uncertain Linear Systems: A Direct Data-drive n Ap- proach,” in 2022 IEEE Conference on Control Technology and Ap- plications (CCTA), pp. 1278–1284, 2022
2022
-
[24]
Data-based guaran tees of set invariance properties,
A. Bisoffi, C. D. Persis, and P . Tesi, “Data-based guaran tees of set invariance properties,” IF AC-PapersOnLine, vol. 53, no. 2, pp. 3953– 3958, 2020
2020
-
[25]
Data-dr iven controller synthesis of unknown nonlinear polynomial syst ems via control barrier certificates,
A. Nejati, B. Zhong, M. Caccamo, and M. Zamani, “Data-dr iven controller synthesis of unknown nonlinear polynomial syst ems via control barrier certificates,” in Proceedings of Learning for Dynamics and Control Conference , pp. 763–776, PMLR, 2022
2022
-
[26]
From data to global a symptotic stability of unknown large-scale networks with provable gu arantees,
M. Zaker, A. Nejati, and A. Lavaei, “From data to global a symptotic stability of unknown large-scale networks with provable gu arantees,” in Proceedings of the 28th ACM International Conference on Hyb rid Systems: Computation and Control , pp. 1–14, 2025
2025
-
[27]
From data to co ntrol: A formal compositional framework for large-scale intercon nected networks,
O. Akbarzadeh, A. Nejati, and A. Lavaei, “From data to co ntrol: A formal compositional framework for large-scale intercon nected networks,” arXiv: 2409.12469 , 2024
2024 arXiv
-
[28]
Guarantees f or data- driven control of nonlinear systems using semidefinite prog ramming: A survey,
T. Martin, T. B. Sch¨ on, and F. Allg¨ ower, “Guarantees f or data- driven control of nonlinear systems using semidefinite prog ramming: A survey,” Annual Reviews in Control , 2023
2023
-
[30]
Trade-offs in lear ning controllers from noisy data,
A. Bisoffi, C. De Persis, and P . Tesi, “Trade-offs in lear ning controllers from noisy data,” Systems & Control Letters , vol. 154, 2021
2021
-
[31]
H. J. Kushner, Stochastic Stability and Control . Mathematics in Science and Engineering, Elsevier Science, 1967
1967
-
[32]
Formal V erification of Unknown Discrete- and Continuous-Time syst ems: A Data-Driven Approach,
A. Nejati, A. Lavaei, P . Jagtap, S. Soudjani, and M. Zama ni, “Formal V erification of Unknown Discrete- and Continuous-Time syst ems: A Data-Driven Approach,” IEEE Transactions on Automatic Control , vol. 68, no. 5, pp. 3011–3024, 2023
2023
-
[33]
Data-driven stabiliz ation of nonlinear polynomial systems with noisy data,
M. Guo, C. De Persis, and P . Tesi, “Data-driven stabiliz ation of nonlinear polynomial systems with noisy data,” IEEE Transactions on Automatic Control , vol. 67, no. 8, pp. 4210–4217, 2021
2021
-
[34]
A Cauchy-Schwarz inequality fo r operators with applications,
R. Bhatia and C. Davis, “A Cauchy-Schwarz inequality fo r operators with applications,” Linear Algebra and its Applications , vol. 223, pp. 119–129, 1995
1995
-
[35]
On classes of summable functions and their Fourier se- ries,
W. H. Y oung, “On classes of summable functions and their Fourier se- ries,” in Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences , vol. 87, pp. 225–229, 1912
1912
-
[36]
LMI properties and appli cations in systems, stability, and control theory,
R. J. Caverly and J. R. Forbes, “LMI properties and appli cations in systems, stability, and control theory,” arXiv: 1903.08599 , 2019
1903 arXiv
-
[37]
SOS- TOOLS: Control applications and new developments,
S. Prajna, A. Papachristodoulou, P . Seiler, and P . A. Pa rrilo, “SOS- TOOLS: Control applications and new developments,” in Proceed- ings of IEEE International Conference on Robotics and Autom ation, pp. 315–320, 2004
2004
-
[38]
Using SeDuMi 1.02, A Matlab toolbox for opt imization over symmetric cones,
J. F. Sturm, “Using SeDuMi 1.02, A Matlab toolbox for opt imization over symmetric cones,” Optimization Methods and Software , vol. 11, no. 1-4, pp. 625–653, 1999
1999
-
[39]
Design o f Polynomial Control laws for Polynomial Systems Subject to Actuator Sat uration,
G. V almorbida, S. Tarbouriech, and G. Garcia, “Design o f Polynomial Control laws for Polynomial Systems Subject to Actuator Sat uration,” IEEE Transactions on Automatic Control , vol. 58, no. 7, pp. 1758– 1770, 2013
2013
-
[40]
Synchronization of complex networks of identica l and nonidentical chaotic systems via model-matching control,
D. L ´ opez-Mancilla, G. L ´ opez-Cahuich, C. Posadas-Ca stillo, C. E. Casta˜ neda, J. H. Garc´ ıa-L ´ opez, J. L. V´ azquez-Guti´ errez, and E. Tlelo- Cuautle, “Synchronization of complex networks of identica l and nonidentical chaotic systems via model-matching control, ” Pl...
2019
Reviewed August 11, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.