Pith. sign in

REVIEW 3 major objections 3 minor 1 cited by

Online convex optimization for constrained control of nonlinear systems

T0 review · 3 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A modular framework combines online convex optimization with reference governors to control nonlinear systems under time-varying unknown costs, guaranteeing constraint satisfaction and a dynamic regret bound that is linear in the…

desk verdict Solid modular OCO+RG upper-bound result for nonlinear constrained control, but the claimed optimality of linear dependence on the OCO regret rests on a vacuous lower-bound proof. read the letter →

arxiv 2412.00922 v2 pith:UKKVSKZ5 submitted 2024-12-01 eess.SY cs.SYmath.OC

Marko Nonhoff , Johannes Köhler , Matthias A. Müller This is my paper · ORCID
classification eess.SYcs.SYmath.OC
keywords onlineconvexoptimizationreferencegovernornonlinearsystemsconstraintsatisfactiondynamicregretpathlengthmodularcontrol
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes a modular control framework that separates the task of minimizing time-varying, a priori unknown cost functions from the task of keeping a nonlinear dynamical system inside its state and input constraints. An online convex optimization (OCO) algorithm proposes a reference signal based on previous costs; a reference governor modifies that reference only when needed to guarantee the closed loop stays in a safe set; a stabilizing feedback tracks the modified reference. The paper proves that the resulting closed-loop dynamic regret is bounded by a constant plus the OCO algorithm's own dynamic regret plus a term proportional to how often and how far the optimizer moves its reference. It further shows that a linear dependence on the optimizer's dynamic regret is unavoidable, and that for Q-linearly convergent optimizers the regret is linear only in the variation of the cost functions, matching known optimal bounds for linear systems. If correct, the framework yields per-time-step constraint satisfaction with a performance loss that is as small as the underlying online optimizer allows.

What carries the argument

The mechanism that carries the argument is the modular composition of three components: an OCO algorithm producing a desired reference $r_t$, a reference governor that outputs $v_t = \mathrm{RG}(x_t, r_t)$ so that $(x_t, v_t)$ lies in a forward-invariant safe set $\mathcal{O}$, and a stabilizing feedback $u_t = g(x_t, v_t)$. The load-bearing identity is the regret decomposition in the proof of Theorem 1, which splits the closed-loop cost into the OCO algorithm's steady-state regret, a transient cost proportional to $\|x_t - h(v_t)\|$, and a reference-tracking cost proportional to $\|v_t - r_t\|$. The transient term is controlled by an exponential-stability Lyapunov function (Lemma 1) and the tracking term by the governor's progress condition (Assumption 5): when $r_t$ is not yet feasible, the governor must step at least distance $\nu_t$ toward $r_t$ and reduce the remaining distance by a factor $1 - \rho(\nu_t)$, with $\rho$ a class-$\mathcal{K}$ function. Lemma 3 turns that per-step progress into an average contraction over a fixed horizon $M$, which yields the linear dependence of the regret on the optimizer's path length.

What would settle it

Pick a stabilizing feedback and safe-set pair satisfying Assumptions 1–4, and a reference governor that keeps the closed loop safe but whose progress $\nu_t$ toward an infeasible desired reference is zero for some reachable state—for example, a scalar governor on a safe set whose cross-section at that state is a single point. If that governor still yields good closed-loop performance but inequality (7) fails for every class-$\mathcal{K}$ function $\rho$, then the proof mechanism of Lemma 3 and Theorem 1 is violated; conversely, exhibiting such a case with regret growing faster than the claimed linear-in-path-length bound would refute the theorem's reach.

Watch

Extended reading notes

Core claim

The central claim, formalized as Theorem 1, is that under mild regularity assumptions—a steady-state map, exponential stability of the stabilizing feedback, a forward-invariant safe set with positive margin around each admissible steady state, and a reference governor that always moves toward the desired reference when it is infeasible—the closed loop satisfies $(x_t, u_t) \in \mathcal{Z}$ for all $t$ and its dynamic regret $R_T$ satisfies $R_T \leq c_0 + R_T^{\mathrm{OCO}} + c_{PL} R_T^{PL}$. Here $R_T^{\mathrm{OCO}}$ is the dynamic regret of the OCO algorithm against the sequence of optimal steady-state references and $R_T^{PL}$ is its path length; the constants depend on the Lyapunov data, the safety margin, and the governor's progress rate. Proposition 2 proves that the linear dependence on $R_T^{\mathrm{OCO}}$ cannot be improved. For OCO algorithms that converge Q-linearly on constant costs, Corollary 1 sharpens the bound to a constant plus a linear term in the cumulative movement of the optimal references, which is known to be the best possible order. The same results recover the OCO-with-memory setting as a special case.

Load-bearing premise

The guarantees stand or fall on the requirement that whenever the optimizer's desired reference is not yet feasible, the reference governor must move a guaranteed positive step toward it and shrink the remaining gap by a factor that is bounded away from one.

Editorial extensions

If this is right

  • Every instance of the framework—any OCO algorithm and any reference governor satisfying Assumptions 4–5—guarantees $(x_t, u_t) \in \mathcal{Z}$ at each time $t$, not merely asymptotically.
  • The closed-loop dynamic regret scales at most linearly with the underlying OCO algorithm's dynamic regret and its path length; choosing a Q-linearly convergent algorithm makes the regret linear only in the cumulative variation of the optimal steady-state references.
  • No control algorithm can beat a linear dependence on the OCO algorithm's dynamic regret, so the framework's loss relative to the optimizer is the best possible order.
  • The same analysis covers online convex optimization with memory (switching costs), so regret guarantees for that setting follow as a special case.
  • Larger safe sets for the governor and faster stabilizing feedbacks shrink the constants in the regret bound, quantifying the design trade-off between safety margin and performance.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The proof identifies the governor's worst-case progress rate as the true safety–performance trade-off; measuring $\rho$ for a given safe set on a given system would let a practitioner compute the regret constants before deployment.
  • Since Assumption 3 only needs exponential stability rather than global stability, the same modular argument should extend to locally stabilizing controllers if the safe set is chosen inside the region of attraction, giving a route to model-predictive or tube-based inner loops.
  • The authors' stated limitation of exact dynamics suggests a natural next test: replace the steady-state map $h$ with a set-valued uncertainty model and re-derive the regret bound as a function of the uncertainty diameter; the framework's modularity makes that perturbation analysis independent of the OCO algorithm choice.
  • For slowly time-varying cost functions, the bound predicts that the closed-loop cost should stay within a constant of the optimal steady-state cost; this could be checked in experiments or simulations by feeding the framework a sequence of slowly varying references and recording regret per window.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 3 minor

Summary. The paper proposes a modular online convex optimization (OCO) framework for constrained nonlinear discrete-time systems with time-varying unknown costs: an OCO algorithm generates a reference, a reference governor (RG) enforces state and input constraints, and a stabilizing feedback tracks the reference. The main result (Theorem 1, Eq. (22)) bounds the closed-loop dynamic regret by a constant plus the OCO algorithm's dynamic regret and path length, under Lipschitz regularity, exponential stability, and an abstract RG progress condition (Assumption 5). Corollary 1 specializes to Q-linearly convergent OCO algorithms and gives a bound linear in the variation of the optimal references. Proposition 2 claims that the linear dependence on the OCO regret cannot be improved. A numerical CSTR example compares four OCO/RG combinations and reports normalized regret and computation times.

Significance. The modular separation of OCO from reference governing is conceptually useful and broader than existing linear-system OCO-control results: it provides per-step constraint satisfaction for nonlinear systems while retaining a regret bound expressed in terms of the chosen OCO algorithm. The proof structure in Lemmas 1-3 and Theorem 1 is transparent, and the decomposition into transient, RG-tracking, and OCO-regret terms is a genuine strength. The numerical study with code availability also helps the reader assess the design choices. However, two issues affect the advertised claims: the proof of the optimality statement (Proposition 2) is vacuous as written, and Assumption 5 is an existence assertion whose satisfaction by the exemplar RGs is not demonstrated. With Proposition 2 repaired or removed and Assumption 5 made constructive, the remaining framework is a solid contribution.

major comments (3)
  1. [Section 4, Proposition 2 (Appendix F)] The lower-bound proof is vacuous. The adversary chooses L_t(x,u)=||(x-h(r_t), u-g(h(r_t),r_t))||^2, so the steady-state cost L^s_t(v)=||(h(v)-h(r_t), g(h(v),v)-g(h(r_t),r_t))||^2 attains its minimum at v=r_t. Hence η_t=r_t by Definition 1 and R^OCO_T=Σ(L^s_t(r_t)-L^s_t(η_t))=0 for every t. The displayed inequality R_T≥R^OCO_T then reduces to R_T≥0, which is already true for these nonnegative costs. This construction supplies no lower bound with positive OCO regret, and therefore does not establish the claimed optimality of a linear dependence on R^OCO_T. A valid lower bound must exhibit an adversary with large R^OCO_T and compare R_T to a positive multiple of R^OCO_T; as written, the proof forces the benchmark to equal the algorithm's own output.
  2. [Section 3.2, Assumption 5 and Lemma 3] Assumption 5 is load-bearing: Lemma 3 and Theorem 1 require existence of a class-K function ρ, an integer M, and ε>0 satisfying (20). The paper asserts without proof that the scalar RG (8) and 'almost all RGs' satisfy the progress condition (7), but no ρ is constructed for the scalar RG, the command governor, or any concrete class of RGs. This is not a purely technical formality: Remark 1 explicitly lists command governors with weighting that violate the stated condition (7), so Assumption 5 is not automatically satisfied by standard designs. Please provide a proof for at least the scalar RG (8) and the standard command governor, or replace Assumption 5 with constructive sufficient conditions that can be checked for a given RG. Without this, the main regret bound is conditional on an unverified abstract property.
  3. [Section 5.4, numerical verification of Q-linear convergence] The text states that Q-linear convergence (14) was 'verified numerically' with κ≈0.986 for all q_t∈[50,250] and c_t∈[0.25,0.65]. A finite simulation cannot verify a property over continuous parameter ranges, and no proof is supplied. Since the numerical section uses this to claim that the assumptions of Corollary 1 are satisfied, this claim should be rephrased as empirical evidence for the specific tested instances, or supported by a theoretical verification using the structure of the example.
minor comments (3)
  1. [Lemma 3, Eq. (21)] The definition α_t=ρ^{-1}(ε) presumes that ε lies in the range of ρ, but Assumption 5 only states ρ:R_{\ge0}\to[0,1] is class K, which does not imply surjectivity. Choose ε from the range of ρ, or define α_t as any value satisfying ρ(α_t)=ε.
  2. [Remark 1] The claim that replacing (7) with the R-weighted inequality 'only leads to additional constant factors' is not demonstrated; a brief derivation or a reference would make the modification precise.
  3. [Notation, Section 3.2] The cross-section notation switches between O_x(v) and O_v(x) for the same object; standardizing the notation would improve readability.

Circularity Check

1 steps flagged · score 6.0 of 10

Proposition 2's optimality proof is vacuous: its adversarial cost functions make the OCO regret zero, so the claimed lower bound reduces to R_T ≥ 0.

  1. self definitional [Section 4, Proposition 2 (proof in Appendix F)]
    "Thus, the cost functions Lt (and, hence, also Ls_t) are a priori unknown. More specifically, at each time t, they are only revealed after the algorithm AOCO applies rt = AOCO(It). Thus, the cost functions Lt may be chosen adversarially, i.e., depending on rt. Hence, choosing Lt(x,u) = ||(x-h(rt), u-g(h(rt),rt))||^2 satisfies 0 = Ls_t(rt) ≤ Lt(xt,ut) for all t ∈ N."

    The adversary defines the cost so that rt is a minimizer of the steady-state cost Ls_t: Ls_t(v) = ||(h(v)-h(rt), g(h(v),v)-g(h(rt),rt))||^2 has minimum 0 at v = rt. Hence the benchmark η_t from (10) is rt, and by Definition 1 the OCO regret is ROCO_T = Σ_t Ls_t(rt) - Ls_t(η_t) = 0. The claimed inequality RT ≥ ROCO_T then reduces to RT ≥ 0, which is automatic because the remaining term Σ_t (Lt(xt,ut) - Ls_t(rt)) is a sum of nonnegative terms (Ls_t(rt)=0 and Lt≥0). The proof never produces an instance with positive OCO regret whose effect is shown to propagate to the closed-loop regret; the target inequality is made true by construction rather than by any dynamical argument.

full rationale

The main upper-bound derivation in Theorem 1 is self-contained and not circular: equation (E.1) decomposes the closed-loop regret into the OCO regret ROCO_T plus two terms bounded through Assumptions 5-6 and Lemma 3, and Lemma 3 itself follows from Assumptions 1-5 without assuming the target bound. The self-citations [40,41,42] are prior work that this paper extends and are not load-bearing for the central theorem. The weakness of Assumption 5 (the paper asserts, but does not construct, a class-K function ρ for the scalar RG) is a correctness/support gap rather than a circularity. The genuine circular/tautological step is Proposition 2: the adversarial cost functions force η_t = rt, so ROCO_T = 0 and the claimed lower bound RT ≥ ROCO_T reduces to RT ≥ 0. Since Proposition 2 is advertised as proving that the linear dependence on ROCO_T is optimal, this is a partial circularity in a central claim, not a minor citation issue, so the overall score is 6 out of 10.

Assumptions & free parameters 2 free parameters · 8 assumptions · 0 invented entities

The central bound rests on seven domain assumptions about the plant, controller, safe set, RG, and cost, plus one background converse Lyapunov theorem. Assumption 5 is the most paper-specific and the least verified: it postulates a progress rate that the authors claim standard RGs satisfy but do not prove for the scalar RG used in the experiment. The numerical example adds two hand-tuned quantities: an OGD step size and a numerically estimated contraction factor. No new physical entities are introduced.

free parameters (2)
  • OGD step size γ = 2.5e-4
    Chosen by hand in the numerical example (Section 5.4). Not needed for the main theorem, but it influences whether OGD satisfies the Q-linear convergence used in Corollary 1.
  • OGD Q-linear contraction κ = ≈0.986
    Reported as numerically verified for the reactor example (Section 5.4). The verification procedure and grid are not given, so the conclusion that Corollary 1 applies to OGD rests on this empirical estimate.
assumptions (8)
  • domain assumption Assumption 1: there exists a steady-state mapping h with h(v) = f(h(v), g(h(v), v))
    Section 3.1. Needed to define steady-state costs Ls_t and the OCO problem (9).
  • domain assumption Assumption 2: S_v is compact and f_g, g, h are Lipschitz on the relevant sets
    Section 3.1. Used in Lemmas 1-2 and to prove Lipschitz continuity of Ls_t.
  • domain assumption Assumption 3: the closed loop under g is exponentially stable with decay factor λ
    Section 3.1. Used to construct the Lyapunov function in Lemma 1.
  • domain assumption Assumption 4: the safe set O is invariant with respect to Z_g and contains δ-balls around h(v)
    Section 3.2. Guarantees constraint satisfaction and provides the nonempty interior needed for RG progress.
  • ad hoc to paper Assumption 5: the reference governor satisfies the progress condition (7) with some class-K ρ
    Section 3.2. This is a new abstract characterization claimed to hold for standard RGs, but no proof is given for the scalar RG. Lemma 3 and Theorem 1 depend on it.
  • domain assumption Assumption 6: the cost functions Lt are Lipschitz continuous on Z
    Section 3.3. Standard regularity needed to relate closed-loop cost to steady-state cost.
  • domain assumption Q-linear convergence condition (14) for the OCO algorithm
    Section 3.3, Proposition 1 and Corollary 1. Needed to convert OCO regret plus path length into a bound linear only in the variation of η_t. Not required for Theorem 1.
  • standard math Converse Lyapunov theorem for discrete-time systems as in [24]
    Used in Lemma 1 to derive a Lyapunov function from the exponential stability assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Online convex optimization for constrained control of nonlinear systems." pith.science (2026). https://pith.science/paper/UKKVSKZ5

@misc{pith2026241200922,
  author       = {Pith},
  title        = {Pith review of: Online convex optimization for constrained control of nonlinear systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UKKVSKZ5}},
  note         = {Machine review of arXiv:2412.00922}
}
read the original abstract

This paper proposes a modular approach that combines the online convex optimization framework and reference governors to solve a constrained control problem featuring time-varying and a priori unknown cost functions. Compared to existing results, the proposed framework is uniquely applicable to nonlinear dynamical systems subject to state and input constraints. Furthermore, our method is general in the sense that we do not limit our analysis to a specific choice of online convex optimization algorithm or reference governor. We show that the dynamic regret of the proposed framework is bounded linearly in both the dynamic regret and the path length of the chosen online convex optimization algorithm, even though the online convex optimization algorithm does not account for the underlying dynamics. We prove that a linear bound with respect to the online convex optimization algorithm's dynamic regret is optimal, i.e., cannot be improved upon. Furthermore, for a standard class of online convex optimization algorithms, our proposed framework attains a bound on its dynamic regret that is linear only in the variation of the cost functions, which is known to be an optimal bound. Finally, we demonstrate implementation and flexibility of the proposed framework by comparing different combinations of online convex optimization algorithms and reference governors to control a nonlinear chemical reactor in a numerical experiment.

Figures

Figures reproduced from arXiv: 2412.00922 by the authors.

Figure 1
Figure 1. In particular, the proposed OCO-RG frame [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 1
Figure 1. Block diagram of the proposed approach. The OCO [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the case distinction in Assumption 5. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figures from the paper (3 more)
Figure 3
Figure 3. Figure 3: Steady states of the tank reactor (blue line) together [PITH_FULL_IMAGE:figures/full_fig_p010_3.png]
Figure 4
Figure 4. Figure 4: Depiction of the normalized steady-state cost function [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Closed-loop trajectories for the four combinations of [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Self-Adaptive Learning and Model Predictive Control for Tracking Unknown Dynamics with No Regret

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A self-adaptive MPC with multiple online-learned RFF predictors and Hedge-based selection achieves O(T^{3/4}) expected regret for tracking unknown, switching target dynamics.

Reference graph

Works this paper leans on

54 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [1]

    Agarwal, B

    N. Agarwal, B. Bullins, E. Hazan, S. Kakade, and K. Singh. Online control with adversarial disturbances. InProc. of the 36th International Conf. on Machine Learning, volume 97, pages 111–119, 2019

    work page 2019

  2. [2]

    Alimisis

    F. Alimisis. Characterization of optimization problems that are solvable iteratively with linear convergence. IFAC- PapersOnLine, 58(17):280–285, 2024

    work page 2024

  3. [3]

    Anava, E

    O. Anava, E. Hazan, and S. Mannor. Online learning for adversarieswithmemory:Priceofpastmistakes. In Advances in Neural Information Processing Systems, volume 28, 2015

    work page 2015

  4. [4]

    Angeli, A

    D. Angeli, A. Casavola, and E. Mosca. Command governors for constrained nonlinear systems: direct nonlinear vs. linearization-based strategies. International Journal of Robust and Nonlinear Control, 9(10):677–699, 1999

    work page 1999

  5. [5]

    Bemporad

    A. Bemporad. Reference governor for constrained nonlinear systems. IEEE Trans. on Automatic Control, 43(3):415–419, 1998

    work page 1998

  6. [6]

    Linear tracking mpc for nonlinear systems—part i: The model-based case

    J.Berberich,J.Köhler,M.A.Müller,andF.Allgöwer. Linear tracking mpc for nonlinear systems—part i: The model-based case. IEEE Trans. on Automatic Control, 67(9):4390–4405, 2022

    work page 2022

  7. [7]

    Chang and S

    T.-J. Chang and S. Shahrampour. On online optimization: Dynamic regret analysis of strongly convex and smooth problems. Proc. of the AAAI Conf. on Artificial Intelligence, 35(8):6966–6973, 2021

    work page 2021

  8. [8]

    Y. Chen, L. Cothren, J. Cortés, and E. Dall’Anese. Online regulation of dynamical systems to solutions of constrained optimization problems. IEEE Control Systems Letters , 7:3789–3794, 2023

    work page 2023

Show all 54 references
  1. [9]

    Colombino, E

    M. Colombino, E. Dall’Anese, and A. Bernstein. Online optimization as a feedback controller: Stability and tracking. IEEE Trans. on Control of Network Systems, 7(1):422–432, 2020

    2020

  2. [10]

    Cothren, G

    L. Cothren, G. Bianchin, and E. Dall’Anese. Online optimization of dynamical systems with deep learning perception. IEEE Open Journal of Control Systems, 1:306– 321, 2022

    2022

  3. [11]

    Dvurechensky and M

    P. Dvurechensky and M. Staudigl. Barrier algorithms for constrained non-convex optimization. In Proc. of the 41st International Conf. on Machine Learning, 2024

    2024

  4. [12]

    Faulwasser, L

    T. Faulwasser, L. Grüne, and M. A. Müller. Economic nonlinear model predictive control. Foundations and Trends® in Systems and Control, 5(1):1–98, 2018

    2018

  5. [13]

    Foster and M

    D. Foster and M. Simchowitz. Logarithmic regret for adversarial online control. InProc. 37th International Conf. on Machine Learning, volume 119, pages 3211–3221. PMLR, 2020

    2020

  6. [14]

    Garone, S

    E. Garone, S. Di Cairano, and I. Kolmanovsky. Reference and command governors for systems with constraints: A survey on theory and applications.Automatica, 75:306–328, 2017

    2017

  7. [15]

    Garone and M

    E. Garone and M. M. Nicotra. Explicit reference governor for constrained nonlinear systems. IEEE Trans. on Automatic Control, 61(5):1379–1384, 2016

    2016

  8. [16]

    Gilbert and I

    E. Gilbert and I. Kolmanovsky. Nonlinear tracking control in the presence of state and control constraints: a generalized reference governor. Automatica, 38(12):2063–2073, 2002

    2002

  9. [17]

    Gilbert and K

    E. Gilbert and K. Tan. Linear systems with state and control constraints: the theory and application of maximal output admissible sets.IEEE Trans. on Automatic Control, 36(9):1008–1020, 1991

    1991

  10. [18]

    Hauswirth, S

    A. Hauswirth, S. Bolognani, G. Hug, and F. Dörfler. Timescale separation in autonomous optimization. IEEE Trans. on Automatic Control, 66(2):611–624, 2021

    2021

  11. [19]

    E. Hazan. Introduction to online convex optimization. Foundations and Trends® in Optimization, 2(3-4):157–325, 2016

    2016

  12. [20]

    Hazan and K

    E. Hazan and K. Singh. Introduction to online nonstochastic control, 2022. Available online at arXiv:2211.09619

    2022 arXiv

  13. [21]

    Z. He, S. Bolognani, J. He, F. Dörfler, and X. Guan. Model-free nonlinear feedback optimization.IEEE Trans. on Automatic Control, pages 1–16, 2023

    2023

  14. [22]

    Hirata and Y

    K. Hirata and Y. Ohta. Exact determinations of the maximal output admissible set for a class of nonlinear systems. Automatica, 44(2):526–533, 2008

    2008

  15. [23]

    Jadbabaie, A

    A. Jadbabaie, A. Rakhlin, S. Shahrampour, and K. Sridharan. Online Optimization : Competing with Dynamic Comparators. In Proc. of the 18th International Conf. on Artificial Intelligence and Statistics, volume 38 of Proc. of Machine Learning Research, pages 398–406. PMLR, 2015

    2015

  16. [24]

    Jiang and Y

    Z.-P. Jiang and Y. Wang. A converse lyapunov theorem for discrete-time systems with disturbances.Systems & Control Letters, 45(1):49–58, 2002. 16

    2002

  17. [25]

    Kalabić and I

    U. Kalabić and I. Kolmanovsky. Reference and command governorsforsystemswithslowlytime-varyingreferencesand time-dependent constraints. InProc. of the 53rd IEEE Conf. on Decision and Control, pages 6701–6706, 2014

    2014

  18. [26]

    Karapetyan, D

    A. Karapetyan, D. Bolliger, A. Tsiamis, E. C. Balta, and J. Lygeros. Online linear quadratic tracking with regret guarantees. IEEE Control Systems Letters, 7:3950–3955, 2023

    2023

  19. [27]

    Karapetyan, A

    A. Karapetyan, A. Tsiamis, E. C. Balta, A. Iannelli, and J. Lygeros. Implications of regret on stability of linear dynamical systems. IFAC-PapersOnLine, 56(2):2583–2588, 2023

    2023

  20. [28]

    H. K. Khalil. Nonlinear Systems. Prentice Hall, 3 edition, 2002

    2002

  21. [29]

    Köhler, M

    J. Köhler, M. A. Müller, and F. Allgöwer. A nonlinear model predictive control framework using reference generic terminal ingredients. IEEE Trans. on Automatic Control, 65(8):3576– 3583, 2020

    2020

  22. [30]

    Y. Li, X. Chen, and N. Li. Online optimal control with linear dynamics and predictions: Algorithms and regret analysis. In Advances in Neural Information Processing Systems, pages 14858 – 14870, 2019

    2019

  23. [31]

    Y. Li, S. Das, and N. Li. Online optimal control with affine constraints. InProc. of the AAAI Conf. on Artificial Intelligence, pages 8527–8537, 2021

    2021

  24. [32]

    Y. Li, G. Qu, and N. Li. Using predictions in online optimization with switching costs: A fast algorithm and a fundamental limit. Proc. of the Annual American Control Conf., pages 3008–3013, 2018

    2018

  25. [33]

    Limon, A

    D. Limon, A. Ferramosca, I. Alvarado, and T. Alamo. Nonlinear MPC for tracking piece-wise constant reference signals. IEEE Trans. on Automatic Control, 63(11):3735– 3750, 2018

    2018

  26. [34]

    Y. Lin, I. Shames, and D. Nešić. Online convex optimization using coordinate descent algorithms. Automatica, 165:111681, 2024

    2024

  27. [35]

    D. Q. Mayne, E. C. Kerrigan, E. J. van Wyk, and P. Falugi. Tube-based robust nonlinear model predictive control. International Journal of Robust and Nonlinear Control, 21(11):1341–1353, 2011

    2011

  28. [36]

    Mokhtari, S

    A. Mokhtari, S. Shahrampour, A. Jadbabaie, and A. Ribeiro. Online optimization in dynamic environments: Improved regret rates for strongly convex problems. In Proc. of the 55th Conf. on Decision and Control, pages 7195–7201. IEEE, 2016

    2016

  29. [37]

    Mulvaney-Kemp, S

    J. Mulvaney-Kemp, S. Park, M. Jin, and J. Lavaei. Dynamic regret bounds for constrained online nonconvex optimization based on Polyak–Lojasiewicz regions. IEEE Trans. on Control of Network Systems, 10(2):599–611, 2023

    2023

  30. [38]

    Necoara, Y

    I. Necoara, Y. Nesterov, and F. Glinearu. Linear convergence of first order methods for non-strongly convex optimization. Mathematical Programming, 175:69–107, 2019

    2019

  31. [39]

    M. M. Nicotra and E. Garone. Control of euler-lagrange systems subject to constraints: An explicit reference governor approach. In Proc. of the 54th IEEE Conf. on Decision and Control, pages 1154–1159, 2015

    2015

  32. [40]

    Nonhoff, E

    M. Nonhoff, E. Dall’Anese, and M. A. Müller. Online convex optimization for robust control of constrained dynamical systems, 2024. Available online at arXiv:2401.04487

    2024

  33. [41]

    Nonhoff, J

    M. Nonhoff, J. Köhler, and M. A. Müller. Online convex optimization for constrained control of linear systems using a reference governor.IFAC-PapersOnLine, 56(2):2570–2575, 2023

    2023

  34. [42]

    Nonhoff and M

    M. Nonhoff and M. A. Müller. On the relation between dynamic regret and closed-loop stability.Systems & Control Letters, 177:105532, 2023

    2023

  35. [43]

    Shalev-Shwartz

    S. Shalev-Shwartz. Online learning and online convex optimization. Foundations and Trends ® in Machine Learning, 4(2):107–194, 2012

    2012

  36. [44]

    Simonetto, E

    A. Simonetto, E. Dall’Anese, S. Paternain, G. Leus, and G. B. Giannakis. Time-varying convex optimization: Time- structured algorithms and applications. Proceedings of the IEEE, 108(11):2032–2048, 2020

    2020

  37. [45]

    Real-timeoptimalpower flow

    Y.Tang,K.Dvijotham,andS.Low. Real-timeoptimalpower flow. IEEE Trans. on Smart Grid, 8(6):2963–2973, 2017

    2017

  38. [46]

    T. Yang, L. Zhang, R. Jin, and J. Yi. Tracking slowly moving clairvoyant: Optimal dynamic regret of online learning with true and noisy gradient. InProc. of The 33rd International Conf. on Machine Learning, volume 48, pages 449–457, 2016

    2016

  39. [47]

    Zhang, W

    X. Zhang, W. Shi, X. Li, B. Yan, A. Malkawi, and N. Li. Decentralized temperature control via HVAC systems in energy efficient buildings: An approximate solution procedure. In Proc. of the IEEE Global Conf. on Signal and Information Processing, pages 936–940, 2016

    2016

  40. [48]

    P. Zhao, G. Wang, L. Zhang, and Z.-H. Zhou. Bandit convex optimization in non-stationary environments. Journal of Machine Learning Research, 22(125):1–45, 2021

    2021

  41. [49]

    Zhao, Y.-H

    P. Zhao, Y.-H. Yan, Y.-X. Wang, and Z.-H. Zhou. Non- stationary online learning with memory and non-stochastic control. Journal of Machine Learning Research, 24(1):9831– 9900, 2024

    2024

  42. [50]

    Zheng, J

    T. Zheng, J. Simpson-Porco, and E. Mallada. Implicit trajectory planning for feedback linearizable systems: A time-varying optimization approach. In Proc. of the 2020 American Control Conf., pages 4677–4682, 2020

    2020

  43. [51]

    H. Zhou, Y. Song, and V. Tzoumas. Safe non- stochastic control of control-affine systems: An online convex optimization approach. IEEE Robotics and Automation Letters, 8(12):7873–7880, 2023

    2023

  44. [52]

    Zhou and V

    H. Zhou and V. Tzoumas. Safe non-stochastic control of linear dynamical systems. InProc. of the 62nd IEEE Conf. on Decision and Control, pages 5033–5038, 2023

    2023

  45. [53]

    H. Zhou, Z. Xu, and V. Tzoumas. Efficient online learning with memory via frank-wolfe optimization: Algorithms with bounded dynamic regret and applications to control. InProc. of the 62nd IEEE Conf. on Decision and Control, pages 8266–8273, 2023

    2023

  46. [54]

    Zinkevich

    M. Zinkevich. Online convex programming and generalized infinitesimal gradient ascent. In Proc. of the Twentieth International Conf. on Machine Learning, pages 928–936, 2003. 17

    2003

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.