Pith. sign in

REVIEW 4 major objections 6 minor 43 references

Neural network based control of unknown nonlinear systems via contraction analysis

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper shows that a learned neural-ODE model, if certifiably contractive, lets a controller drive an unknown nonlinear system into a neighborhood of its unseen equilibrium, with the neighborhood size set by the approximation error.

desk verdict Useful contraction-based framework for NN control of unknown systems; main guarantee depends on an approximation-error bound that is only estimated on training data, and one simulation reports an implausible machine-precision bound. read the letter →

arxiv 2505.16511 v1 pith:AW4TVKRS submitted 2025-05-22 eess.SY cs.SYmath.OC

Hao Yin , Claudio De Persis , Bayu Jayawardhana , Santiago Sanchez Escalonilla Plaza This is my paper · ORCID
classification eess.SYcs.SYmath.OC MSC 93C1093D0593D3068T0790C22
keywords neuralordinarydifferentialequationscontractionanalysisincrementalsectorboundslinearmatrixinequalitiesunknownnonlinearsystemsHopfieldnetworknonlinearitycancellationequilibriumconvergence
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

The paper proposes a way to design controllers for nonlinear systems whose drift dynamics are unknown, using contraction analysis instead of knowing the equilibrium. The authors first train a feedforward neural network to approximate the unknown dynamics, producing a Neural ODE model. They impose incremental sector bounds on the activation functions and linear-matrix-inequality (LMI) conditions on the network weights that certify the NODE model is contractive, meaning all its trajectories converge exponentially toward each other. If the NODE model is contractive, they prove that trajectories of the original system converge to a neighborhood of the unknown equilibrium, with the neighborhood's radius proportional to the approximation error. When the NODE model fails the contraction conditions, they design a controller combining a linear gain with an NN compensation term that enforces contractivity and again guarantees convergence to such a neighborhood.

What carries the argument

The central object is the pair of incremental sector bounds $[K^i_1, K^i_2]$ on the activation functions, combined with a quadratic differential Lyapunov function $V = (x_1-x_2)^\top P(x_1-x_2)$. Lemma 2.3 turns the sector bounds into a matrix inequality on the network weights, and Theorems 3.2 and 3.4 convert that into LMI conditions certifying contraction. The Schur complement reformulations in Theorems 3.3 and 3.6 make the conditions convex and hence solvable by standard semidefinite programming. Proposition 3.7 then propagates the approximation error through the contractive comparison argument, and Proposition 3.8 uses the unknown equilibrium as a trajectory of an augmented NODE to derive the explicit neighborhood bound.

What would settle it

Train the neural network on data from a region $D$, compute the claimed error bound $\varepsilon$, and then evaluate $\sup_{x \in C\setminus D} \|\varepsilon(x)\|$ on a larger forward-invariant set $C$; if the empirical supremum exceeds the claimed $\varepsilon$, then Proposition 3.8's bound $\|x(t)-x^*\| \le (4\varepsilon/\gamma)\sqrt{\bar{p}/\underline{p}}$ fails because the derivation uses $\varepsilon = \sup_{C}\|\varepsilon(x)\|$. Alternatively, simulate a system whose drift lies outside the training region and check whether trajectories exit the predicted neighborhood of the unseen equilibrium.

Watch

Extended reading notes

Core claim

The central claim is that contractivity of the learned NODE system is a workable certificate for controlling unknown nonlinear affine systems without knowing their equilibria. Concretely, if the NODE system $\dot{x} = Ax + Z_{\mathrm{nn}}(x)$ is contractive on a forward invariant set $C$, then the original system $\dot{x} = Ax + Z_{\mathrm{nn}}(x) + \varepsilon(x)$ has every trajectory converge to a ball around the unknown equilibrium $x^*$ whose radius is $(4\varepsilon/\gamma)\sqrt{\bar{p}/\underline{p}}$, where $\varepsilon$ is the uniform bound on the approximation error and $\gamma$ is the contraction rate. This removes the need to know $x^*$ explicitly. For a single-hidden-layer network the NODE becomes a continuous-time Hopfield network, so the same certificate applies there. If the NODE is not contractive, the proposed controller $u(x) = Hx + W_{u,\min} w_k + b_u$ adds a linear state feedback and a neural-network compensation term that cancels most nonlinearities and restores contractivity, again guaranteeing convergence to a neighborhood of the unknown equilibrium.

Load-bearing premise

The approximation error $\varepsilon(x)$ must be uniformly bounded by a known $\varepsilon$ on the forward-invariant operating set $C$, even though it is only measured on the training set $D$; if the neural network's error is larger on the actual operating region than the bound used, the predicted neighborhood radius is not guaranteed.

Editorial extensions

If this is right

  • If the NODE model is contractive, the original autonomous system's trajectories converge to a neighborhood of the unknown equilibrium, and the radius shrinks as the approximation error $\varepsilon$ goes to zero or the contraction rate $\gamma$ grows.
  • The contractivity conditions are convex LMIs, so a trained network can be checked for contractivity, and the controller gain can be optimized via a convex program that minimizes the guaranteed neighborhood size.
  • When the NODE fails the contractivity test, the two-part controller (linear gain plus NN compensation) restores contractivity and still gives convergence to a specified neighborhood of the unknown equilibrium.
  • For single-layer networks the framework specializes to continuous-time Hopfield networks, giving a direct contractivity and convergence certificate for that popular recurrent architecture.
  • The bias term of the NN controller is free to choose, so the designer can shift the closed-loop equilibrium, for instance to the origin, without destroying the contraction guarantee.

Reading between the lines

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

  • The same incremental-sector-plus-LMI machinery could plausibly be adapted to discrete-time neural network dynamics or to stochastic disturbance settings, where contraction notions already exist, though the paper does not pursue those extensions.
  • The neighborhood bound $4\varepsilon/\gamma \sqrt{\bar{p}/\underline{p}}$ is global on the set $C$; a testable refinement would use localized error bounds on smaller regions of the state space to obtain tighter radii than a uniform supremum.
  • Because the equilibrium is never identified, the method offers a form of equilibrium-free stabilization that could be combined with online model adaptation, a direction the authors leave for future work.
  • A practical deployment step suggested by the results is to run the LMI certificate on a trained network before use, turning the data-fitting step into a verifiable contractivity guarantee.
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

4 major / 6 minor

Summary. The paper studies NN-based control of unknown affine nonlinear systems using contraction analysis. A NODE model is fitted to data, LMI-type sufficient conditions are proposed under which the NODE system is contractive, and it is shown that, if the approximation error is uniformly bounded by a known constant on a forward invariant set, trajectories of the original system converge to a neighborhood of the unknown equilibrium whose size is proportional to that bound. For NODE models that do not satisfy the contraction conditions, the paper proposes a two-part controller with a linear gain and an NN cancellation term, again with LMI reformulations. Two numerical examples, a pendulum and a wheeled-vehicle path-following model, illustrate the procedure.

Significance. If the stated assumptions were fully verified, the paper would provide a useful design workflow for unknown nonlinear systems when the equilibrium is not known a priori. The LMI conditions for contractivity of multi-layer NODEs with incremental sector bounds are a practical contribution, and the convergence-to-neighborhood result is a clean ISS-type Lyapunov argument. The examples are simple but demonstrate the overall architecture. On the other hand, the paper does not provide code or machine-checked proofs, and the central quantity used in the main guarantee, the uniform approximation-error bound, is only estimated on a finite training set rather than certified.

major comments (4)
  1. [Section III, Proposition 3.7-3.8, Algorithms 1-2] The main convergence guarantee requires a known uniform bound sup_{x in C} ||epsilon(x)|| <= epsilon, but the only numerical mechanism in the paper is the training loss (2) over the finite data set D. The values 0.002302 (Example 4.1) and 0.008532 (Example 4.2) are reported as 'upper bounds' after 5000 epochs, yet no dense verification, Lipschitz/generalization argument, or other certificate is provided that they dominate the error on the forward invariant set C. Because the terminal neighborhood in Eq. (37), Algorithm 1, and Algorithm 2 scales linearly in epsilon, an underestimated epsilon invalidates the headline guarantee. The forward invariance of C for the perturbed closed-loop system is also assumed, not checked. Please either certify the bound on C, restrict all claims to a known epsilon and remove 'yielding an upper bound epsilon' from Algorithm 1 Step 1, or explicitly state that the result is conditional on an unverified uniform error bound.
  2. [Theorem 3.2, Eq. (17)] The displayed Lyapunov derivative in the proof of Theorem 3.2 contains W^1 in the cross terms, but the derivative of V along the single-layer NODE system (12) contains W^2, because Znn(x) = W^2 phi(W^1 x + b^1) + b^2. The correct cross term appears implicitly in Eq. (15) through L = P W^2 + (alpha1+beta1) W^{1T}. As printed, the proof does not establish the claimed bound; replacing the two W^1 factors by W^2 makes the argument consistent. Please correct the proof.
  3. [Example 4.1, Eq. (68)] The reported incremental sector bound 0.9978 <= (phi1(x)-phi1(y))/(x-y) <= 1 for all x,y in [-2,2] is false for tanh, whose derivative at x=2 is sech^2(2) approximately 0.07065. The bound can only be true on a much smaller interval, presumably the range of the pre-activation values W^1 x + b^1, but the text states x in [-2,2]. In addition, the coefficient -102850 in the reported matrix bar(R) is not consistent with the coefficient -2 alpha beta/(beta-alpha)^2 from Theorem 3.12, which for alpha=0.9978, beta=1 is approximately -4.12e5. Please reconcile the numerical values with the formulas in the paper.
  4. [Theorem 3.15 and Example 4.2, Eq. (56), (70), (72)] The LMI status of the multilayer controller design is unclear. The term -2 s / lambda S in bar(X) of Eq. (56) is bilinear in the decision variables if s is a decision variable, while the theorem states 'there exist a fixed positive constant s' and the optimization (66) appears to optimize over s. The approximation bar(R) = (AS+gY)^T + AS + gY in Example 4.2, made 'provided that lambda is sufficiently small,' is not justified and the value of lambda is not reported. Since the feasibility of the LMI is the basis for the controller guarantee in the multilayer case, this needs to be clarified and made rigorous.
minor comments (6)
  1. [Section II, Notation] The sentence 'The symbols R, R+ denote the set of real nonnegative real numbers, respectively' should read that R denotes the real numbers and R+ denotes the nonnegative real numbers.
  2. [Section II, Definition 2.1 and Eq. (11)] The symbol D is used both for the finite training data set and for the domain on which the incremental sector bound and approximation error are defined. Please introduce a separate symbol for the domain, for example Omega, and state explicitly how C relates to it.
  3. [Proposition 3.7, Eq. (34)] The notation p and p (underline and overline) is easy to confuse; in Eq. (34) the coefficient sqrt(p) should be the upper bound overline(p) from Eq. (14). Please use a more distinct notation, for example p_min and p_max.
  4. [Algorithms 1 and 2] Step 3 in both algorithms solves the unconstrained least-squares problem min ||W^{k+1} + g W_u^{k+1}||; the closed-form solution should be stated, and the algorithm should clarify what happens if g is not injective.
  5. [Example 4.2] Please state the numerical value of lambda used in Eq. (70) and explain why the approximation bar(R) ≈ (AS+gY)^T + AS + gY is valid for that value.
  6. [Introduction] There are several typographical and grammatical errors; for example, 'the author analyze' should be 'the authors analyze' and 'non-convex conditions can complicate the search' should be 'non-convex conditions can complicate the search for feasible solutions'. A careful proofreading pass is recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the convergence guarantee is a conditional contraction-theoretic bound with the approximation error as a stated input, not as a fitted output.

full rationale

The paper's central result, Proposition 3.8, derives the bound (37) by comparing the perturbed system (32) with the contractive system (38) that has the constant input epsilon(x*), using the same Lyapunov function as in Theorem 3.4. The bound is a standard ISS-type estimate: if the NODE system (12) is contractive and the error epsilon is uniformly bounded by a known constant, then trajectories converge to a ball of radius proportional to that bound. The approximation error epsilon enters as a hypothesis, not as a quantity manufactured to make (37) hold. The NN fitting problem (2) produces the matrices and weights, and the LMI conditions (15), (23), (48), (55) are sufficient conditions checked after training; they are not constructed so that the conclusion holds by definition. The self-citations, [13] and [36], are used only for context and inspiration ('Nonlinearity Cancellation' and a comparison with offset sector bounds), and neither is load-bearing for the proof. The skeptical concern that the reported epsilon is only an empirical training loss rather than a certified uniform bound on the forward-invariant set C is a correctness and verification gap, not a circularity, because the theorem does not define its conclusion in terms of the fitted loss. Accordingly, the derivation is self-contained relative to its stated assumptions, and no circular step can be exhibited with a specific reduction of a conclusion to an input.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the trained network parameters and the approximation error bound, which are fit to data, plus assumptions of forward invariance, existence of an unknown equilibrium, uniform error bounds on C, and LMI feasibility. No new physical entities are introduced.

free parameters (3)
  • Trained linear matrix A and NN weights W^i, b^i = A = [[-0.0000,0.9996],[-1.7654,0.0012]], W1, W2, b1, b2 in Example 4.1; similar in Example 4.2
    Fitted to data by solving (2); all subsequent contraction conditions and convergence bounds depend on these values.
  • Approximation error bound epsilon = 0.002302 (Example 4.1), 0.008532 (Example 4.2)
    Upper bound of the residual on the training set, not on the forward invariant set C; the convergence neighborhood size is proportional to this value.
  • LMI decision variables (S, Y, mu, gamma, H) = Example 4.1: S, Y, H, mu=0.61168; Example 4.2: mu=8.0543e-15
    Decision variables chosen to satisfy the LMI; the extremely small mu in Example 4.2 drives the claimed neighborhood size to near zero.
assumptions (6)
  • standard math Universal approximation theorem (Hornik et al. 1989) guarantees the existence of a NN that approximates f to arbitrary accuracy, so the representation (11) is valid.
    Invoked in Section I to justify rewriting (1) as (11).
  • domain assumption The drift dynamics f(x) is unknown, and only sampled data (x(t_k), xdot(t_k)) from D are available; the control effectiveness matrix g is known.
    Problem formulation in Section II.
  • domain assumption There exists a forward invariant set C containing the trajectories of both the NODE and the original system, and the original system has an unknown equilibrium x* in C where Ax* + Znn(x*) + epsilon(x*) = 0.
    Stated before Proposition 3.8 and used in its proof.
  • ad hoc to paper The approximation error epsilon(x) is uniformly bounded by a known constant epsilon on the relevant set C, although epsilon is only computed on the training set D.
    Required for Prop. 3.7 and 3.8; no procedure is given to certify this bound beyond the training data.
  • ad hoc to paper The LMI conditions (15), (23), (48), (55) are feasible for the trained network and chosen control structure.
    The controller design and the convergence claim depend on the existence of feasible solutions; no feasibility guarantee is given.
  • domain assumption Activation functions (tanh) are differentiable and satisfy incremental sector bounds on the domain D.
    Used to derive Fact 2.2 and Lemma 2.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neural network based control of unknown nonlinear systems via contraction analysis." pith.science (2026). https://pith.science/paper/AW4TVKRS

@misc{pith2026250516511,
  author       = {Pith},
  title        = {Pith review of: Neural network based control of unknown nonlinear systems via contraction analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AW4TVKRS}},
  note         = {Machine review of arXiv:2505.16511}
}
read the original abstract

This paper studies the design of neural network (NN)-based controllers for unknown nonlinear systems, using contraction analysis. A Neural Ordinary Differential Equation (NODE) system is constructed by approximating the unknown draft dynamics with a feedforward NN. Incremental sector bounds and contraction theory are applied to the activation functions and the weights of the NN, respectively. It is demonstrated that if the incremental sector bounds and the weights satisfy some non-convex conditions, the NODE system is contractive. To improve computational efficiency, these non-convex conditions are reformulated as convex LMI conditions. Additionally, it is proven that when the NODE system is contractive, the trajectories of the original autonomous system converge to a neighborhood of the unknown equilibrium, with the size of this neighborhood determined by the approximation error. For a single-layer NN, the NODE system is simplified to a continuous-time Hopfield NN. If the NODE system does not satisfy the contraction conditions, an NN-based controller is designed to enforce contractivity. This controller integrates a linear component, which ensures contraction through suitable control gains, and an NN component, which compensates for the NODE system's nonlinearities. This integrated controller guarantees that the trajectories of the original affine system converge to a neighborhood of the unknown equilibrium. The effectiveness of the proposed approach is demonstrated through two illustrative examples.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 42 canonical work pages

  1. [1]

    Thi s property makes the single-hidden layer NN ( k = 1) commonly used for system representation in control theory [11], [22] , [29]

    Hopfield (Single-layer) neural network case : It has been demonstrated in [26] that a single-hidden-layer neural net work can approximate any continuous nonlinear function with ar- bitrary precision, given a sufficient number of neurons. Thi s property makes the single-hidden layer NN ( k = 1) commonly used for system representation in control theory [11], ...

    work page

  2. [2]

    Nonlinearity Ca n- cellation

    Multilayer neural network case : For the multilayer NN, we present the following theorem outlining the constructio n of a contractive NODE system. Theorem 3.4: Suppose there exist a positive definite matrix P , positive constants p, p, and γ such that (14) and the following condition U ⊤ 1 [ A⊤P + P A + γP P W k+1 ∗ 0 ] U1+ U ⊤ 2 [ −2Q⊤ 1 Q2 Q⊤ 1 + Q⊤ 2 ∗ ...

    work page

  3. [3]

    Hopfield (Single-layer) neural network case : For single- hidden layer NN, in the light of Lemma 3.2, we can derive the following result. Lemma 3.9: Suppose there exist a positive definite matrix P , positive constants p , p, γ, and a matrix H, such that (14) and the following condition [ L′ P Wmin + (α1 + β1)W 1⊤ ∗ − 2I ] ≤ 0, (48) hold, where Wmin = W 2 +...

    work page

  4. [4]

    Multilayer neural network case : By applying Theorem 3.4, the following result establishes the contractivity of the previously non-contractive NODE system (12). Theorem 3.14: Suppose there exist a positive definite ma- trix P , positive constants p, p, γ, and a matrix H, such that AUTHOR et al.: TITLE 9 (14) and the following condition U ⊤ 1 [ (A + gH )⊤P ...

    work page

  5. [5]

    The states x1, x2 are the angular position and velocity, respectively, u(x) is the applied torque

    Hopfield (Single-layer) neural network case : Example 4.1: Consider the following pendulum dynamics { ˙x1 = x2, ˙x2 = − g l sin(x1) − k m x2 + 1 ml2 u(x), (67) where m is the mass to be balanced, l is the distance from the base to the center of mass of the balanced body, k is the coefficient of rotational friction, and g is the acceleration due to gravity. ...

    work page

  6. [6]

    2] Consider a wheeled vehicle path following system { ˙de = ν sin(θe) ˙θe = ω − νκ(s) cos(θe) 1−κ(s)de

    Multilayer neural network case : Example 4.2: [29, Ex. 2] Consider a wheeled vehicle path following system { ˙de = ν sin(θe) ˙θe = ω − νκ(s) cos(θe) 1−κ(s)de . (69) The variables are defined as follows: θe represents the heading error, de denotes the lateral error, ν is the linear velocity, κ(s) describes the curvature of the reference path, and ω represen...

    work page

  7. [7]

    Alpcan T, Pavel L, Stefanovic N. A control theoretic appr oach to non- cooperative game design[C]//Proceedings of the 48h IEEE Co nference on Decision and Control (CDC) held jointly with 2009 28th Chi nese Control Conference. IEEE, 2009: 8575-8580

    work page 2009

  8. [8]

    Contraction methods for nonlinea r systems: A brief introduction and some open problems[C]//53rd IEEE Co nference on Decision and Control

    Aminzare Z, Sontag E D. Contraction methods for nonlinea r systems: A brief introduction and some open problems[C]//53rd IEEE Co nference on Decision and Control. IEEE, 2014: 3835-3847

    work page 2014

Show all 43 references
  1. [9]

    A Lyapunov approach to incremental stability p roperties[J]

    Angeli D. A Lyapunov approach to incremental stability p roperties[J]. IEEE Transactions on Automatic Control, 2002, 47(3): 410-4 21

    2002

  2. [10]

    On contraction of time-v arying port- Hamiltonian systems[J]

    Barabanov N, Ortega R, Pyrkin A. On contraction of time-v arying port- Hamiltonian systems[J]. Systems & Control Letters, 2019, 133: 104545

    2019

  3. [11]

    Convex optimization[M]

    Boyd S P , V andenberghe L. Convex optimization[M]. Cambr idge uni- versity press, 2004

    2004

  4. [12]

    Modeling and contractivi ty of neural- synaptic networks with Hebbian learning[J]

    Centorrino V , Bullo F, Russo G. Modeling and contractivi ty of neural- synaptic networks with Hebbian learning[J]. Automatica, 2 024, 164: 111636

  5. [13]

    Euclidean cont ractivity of neural networks with symmetric weights[J]

    Centorrino V , Gokhale A, Davydov A, et al. Euclidean cont ractivity of neural networks with symmetric weights[J]. IEEE Control Systems Letters, 2023, 7: 1724-1729

    2023

  6. [14]

    Adaptive optimal control of unknown nonlinear systems via homotopy-based policy iteration[J]

    Chen C, Lewis F L, Xie K, et al. Adaptive optimal control of unknown nonlinear systems via homotopy-based policy iteration[J] . IEEE Trans- actions on Automatic Control, 2023, 69(5): 3396-3403

    2023

  7. [15]

    Neural ordina ry differential equations[J]

    Chen R T Q, Rubanova Y , Bettencourt J, et al. Neural ordina ry differential equations[J]. Advances in neural informatio n processing systems, 2018, 31

    2018

  8. [16]

    Non-linear system identi fication using neural networks[J]

    Chen S, Billings S A, Grant P M. Non-linear system identi fication using neural networks[J]. International journal of control, 199 0, 51(6): 1191- 1214

  9. [17]

    An incremental input-to -state stability condition for a class of recurrent neural network s[J]

    D’Amico W, La Bella A, Farina M. An incremental input-to -state stability condition for a class of recurrent neural network s[J]. IEEE Transactions on Automatic Control, 2023, 69(4): 2221-2236

    2023

  10. [18]

    Non-Euclidean con traction analysis of continuous-time neural networks[J]

    Davydov A, Proskurnikov A V , Bullo F. Non-Euclidean con traction analysis of continuous-time neural networks[J]. IEEE Tran sactions on Automatic Control, 2024

    2024

  11. [19]

    Learning controllers from data via ap- proximate nonlinearity cancellation[J]

    De Persis C, Rotulo M, Tesi P . Learning controllers from data via ap- proximate nonlinearity cancellation[J]. IEEE Transactio ns on Automatic Control, 2023, 68(10): 6082-6097

    2023

  12. [20]

    Cautious optimization via data info rmativity[J]

    Eising J, Cortes J. Cautious optimization via data info rmativity[J]. arXiv preprint arXiv:2307.10232, 2023

    2023 arXiv

  13. [21]

    Safety verification and r obustness analysis of neural networks via quadratic constraints and s emidefinite programming[J]

    Fazlyab M, Morari M, Pappas G J. Safety verification and r obustness analysis of neural networks via quadratic constraints and s emidefinite programming[J]. IEEE Transactions on Automatic Control, 2 020, 67(1): 1-15

  14. [22]

    A differential Lyapunov framewor k for contraction analysis[J]

    Forni F, Sepulchre R. A differential Lyapunov framewor k for contraction analysis[J]. IEEE transactions on automatic control, 2013 , 59(3): 614- 628

    2013

  15. [23]

    Approximate opti mal trajectory tracking with sparse bellman error extrapolation[J]

    Greene M L, Deptula P , Nivison S, et al. Approximate opti mal trajectory tracking with sparse bellman error extrapolation[J]. IEEE Transactions on Automatic Control, 2022, 68(6): 3618-3624

    2022

  16. [24]

    Deep neural network -based approximate optimal tracking for unknown nonlinear system s[J]

    Greene M L, Bell Z I, Nivison S, et al. Deep neural network -based approximate optimal tracking for unknown nonlinear system s[J]. IEEE Transactions on Automatic Control, 2023, 68(5): 3171-3177

    2023

  17. [25]

    Multilayer feedforwa rd networks are universal approximators[J]

    Hornik K, Stinchcombe M, White H. Multilayer feedforwa rd networks are universal approximators[J]. Neural networks, 1989, 2( 5): 359-366

    1989

  18. [26]

    Enforcing contraction via data [J]

    Hu Z, De Persis C, Tesi P . Enforcing contraction via data [J]. arXiv preprint arXiv:2401.07819, 2024

    2024 arXiv

  19. [27]

    A tutorial on incremental stabil ity analysis using contraction theory[J]

    Jouffroy J, Fossen T I. A tutorial on incremental stabil ity analysis using contraction theory[J]. 2010. AUTHOR et al.: TITLE 13

    2010

  20. [28]

    Model-base d reinforce- ment learning for infinite-horizon approximate optimal tra cking[J]

    Kamalapurkar R, Andrews L, Walters P , et al. Model-base d reinforce- ment learning for infinite-horizon approximate optimal tra cking[J]. IEEE transactions on neural networks and learning systems, 2016 , 28(3): 753- 758

    2016

  21. [29]

    Stable neural ode with lyapu nov-stable equilibrium points for defending against adversarial atta cks[J]

    Kang Q, Song Y , Ding Q, et al. Stable neural ode with lyapu nov-stable equilibrium points for defending against adversarial atta cks[J]. Advances in Neural Information Processing Systems, 2021, 34: 14925- 14937

    2021

  22. [30]

    Nonlinear systems[M]

    Khalil H K, Grizzle J W. Nonlinear systems[M]. Upper Sad dle River, NJ: Prentice hall, 2002

    2002

  23. [31]

    On contraction analysis for n on-linear systems[J]

    Lohmiller W, Slotine J J E. On contraction analysis for n on-linear systems[J]. Automatica, 1998, 34(6): 683-696

    1998

  24. [32]

    Learning nonlinear operators v ia DeepONet based on the universal approximation theorem of operators[ J]

    Lu L, Jin P , Pang G, et al. Learning nonlinear operators v ia DeepONet based on the universal approximation theorem of operators[ J]. Nature machine intelligence, 2021, 3(3): 218-229

    2021

  25. [33]

    Unco nstrained parametrization of dissipative and contracting neural ord inary differential equations[C]//2023 62nd IEEE Conference on Decision and Co ntrol (CDC)

    Martinelli D, Galimberti C L, Manchester I R, et al. Unco nstrained parametrization of dissipative and contracting neural ord inary differential equations[C]//2023 62nd IEEE Conference on Decision and Co ntrol (CDC). IEEE, 2023: 3043-3048

    2023

  26. [34]

    Control barrier function -based quadratic programs introduce undesirable asymptotically stable equilib- ria[J]

    Reis M F, Aguiar A P , Tabuada P . Control barrier function -based quadratic programs introduce undesirable asymptotically stable equilib- ria[J]. IEEE Control Systems Letters, 2020, 5(2): 731-736

    2020

  27. [35]

    Stability analysis and con troller synthe- sis using single-hidden-layer relu neural networks[J]

    Samanipour P , Poonawala H A. Stability analysis and con troller synthe- sis using single-hidden-layer relu neural networks[J]. IE EE Transactions on Automatic Control, 2023, 69(1): 202-213

    2023

  28. [36]

    Koopman-ba sed feedback design with stability guarantees[J]

    Str¨ asser R, Schaller M, Worthmann K, et al. Koopman-ba sed feedback design with stability guarantees[J]. IEEE Transactions on Automatic Control, 2024

    2024

  29. [37]

    Learning certified control using cont raction metric[C]//conference on Robot Learning

    Sun D, Jha S, Fan C. Learning certified control using cont raction metric[C]//conference on Robot Learning. PMLR, 2021: 1519 -1539

    2021

  30. [38]

    Contraction theory for nonlinear stability analysis and learning-based control: A tutorial overview[J]

    Tsukamoto H, Chung S J, Slotine J J E. Contraction theory for nonlinear stability analysis and learning-based control: A tutorial overview[J]. Annual Reviews in Control, 2021, 52: 135-169

    2021

  31. [39]

    Reprojection met hods for Koopman-based modelling and prediction[C]//2023 62nd IEE E Confer- ence on Decision and Control (CDC)

    V an Goor P , Mahony R, Schaller M, et al. Reprojection met hods for Koopman-based modelling and prediction[C]//2023 62nd IEE E Confer- ence on Decision and Control (CDC). IEEE, 2023: 315-321

    2023

  32. [40]

    Model-free verification fo r neural network controlled systems[J]

    Wang H, Xiong Z, Zhao L, et al. Model-free verification fo r neural network controlled systems[J]. arXiv preprint arXiv:2312 .08293, 2023

    2023

  33. [41]

    A new concept using LSTM Neural Networks for dyna mic system identification[C]//2017 American control conferen ce (ACC)

    Wang Y . A new concept using LSTM Neural Networks for dyna mic system identification[C]//2017 American control conferen ce (ACC). IEEE, 2017: 5324-5329

    2017

  34. [42]

    Stability analysis using quadr atic constraints for systems with neural network controllers[J]

    Yin H, Seiler P , Arcak M. Stability analysis using quadr atic constraints for systems with neural network controllers[J]. IEEE Trans actions on Automatic Control, 2021, 67(4): 1980-1987

    2021

  35. [43]

    Neural Lyapunov cont rol of unknown nonlinear systems with stability guarantees[J]

    Zhou R, Quartz T, De Sterck H, et al. Neural Lyapunov cont rol of unknown nonlinear systems with stability guarantees[J]. A dvances in Neural Information Processing Systems, 2022, 35: 29113-29 125

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.