arxiv: 2511.05757 · v3 · submitted 2025-11-07 · 📡 eess.SY · cs.LG· cs.SY

Zero-Shot Function Encoder-Based Differentiable Predictive Control

Hassan Iqbal , Xingjian Li , Tyler Ingebrand , Adam Thorpe , Krishna Kumar , Ufuk Topcu , J\'an Drgo\v{n}a This is my paper

Pith reviewed 2026-05-17 23:16 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SY

keywords zero-shot adaptive controlfunction encoderneural ODEdifferentiable predictive controlnonlinear dynamical systemsmodel predictive controlparametric families

0 comments

The pith

A function encoder neural ODE plus differentiable predictive control enables zero-shot adaptation across parametric nonlinear systems.

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

The paper develops a framework that uses a function encoder-based neural ODE to represent system dynamics and pairs it with differentiable predictive control to learn explicit policies offline. The goal is zero-shot adaptation to new parameter values in a family of nonlinear systems, without retraining the dynamics model or running online optimization at deployment time. A reader would care because classical model predictive control solves an optimization problem anew for every new system or parameter set, which is computationally heavy; this approach trains once on a collection of systems and then applies the policy immediately to unseen ones. Demonstrations on several nonlinear examples are used to illustrate accuracy and speed of adaptation.

Core claim

The authors claim that integrating a function encoder-based neural ODE for modeling nonlinear state transitions with differentiable predictive control for offline self-supervised policy learning yields a differentiable framework that achieves zero-shot adaptive control over parametric families of nonlinear dynamical systems while removing the need for costly online optimization.

What carries the argument

Function encoder-based neural ODE (FE-NODE) for dynamics modeling combined with differentiable predictive control (DPC) for learning explicit policies, where the encoder allows zero-shot generalization to new parameters and DPC removes real-time solving.

If this is right

Control policies become learnable once across an entire parametric family rather than per instance.
Deployment requires only forward passes of the learned policy, with no online optimization.
The same trained components handle multiple nonlinear systems that differ only in parameters.
Adaptation to a new system occurs immediately upon receiving its state measurements.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The approach could be useful in robotics or process control where physical parameters drift slowly over time.
Stability analysis of the learned policies under parameter mismatch would be a natural next verification step.
Pairing the framework with online fine-tuning could handle cases where zero-shot performance degrades.

Load-bearing premise

That a dynamics model trained on some parameter values will generalize accurately to new unseen parameterizations and that the resulting policies will remain stable and effective without further adjustment.

What would settle it

Measure closed-loop tracking error and stability margins on a test system whose parameters lie well outside the training distribution; a sharp rise in error or loss of stability compared with in-distribution cases would falsify the zero-shot claim.

Figures

Figures reproduced from arXiv: 2511.05757 by Adam Thorpe, Hassan Iqbal, J\'an Drgo\v{n}a, Krishna Kumar, Tyler Ingebrand, Ufuk Topcu, Xingjian Li.

**Figure 1.** Figure 1: Conceptual diagram of the proposed Function-Encoder Differentiable Predictive Control. 2. Problem Formulation We consider a general class of parametric optimal control problems (pOCP) that take the following continuous-time form min π∈Π Ex0∼Px0 ,ξ∼Pξ,ν∼Pν Z T 0 ℓ(x(t), u(t); ξ)dt + pT (x(T)) (1a) s.t. dx(t) dt = f(x(t), π(x(t); ξ, ν); ν), (1b) h(x(t); ξ) ≤ 0, g(u(t); ξ) ≤ 0, (1c) where ξ is a set of par… view at source ↗

**Figure 2.** Figure 2: Van der Pol oscillator dynamics under both controlled and uncontrolled scenarios using FE-DPC. Left: Uncontrolled and stabilizing results under a fixed dynamics setting. Right: Uncontrolled and stabilizing results with changing dynamics; in the example the dynamics switch after 25 steps into the simulation. determine the dynamics. The objective is to “stabilize” the system, that is, pN (xN ) = ∥xN ∥ 2 and… view at source ↗

**Figure 3.** Figure 3: Two-tank system reference tracking under multiple system switches using FE-DPC. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: True and predicted uncontrolled GO system dynamics. System parameterizations change every 500 time steps, and predictions are calibrated against the true states every 50 steps. Reference Tracking of a Glycolytic Oscillator (GO). We consider a highly nonlinear and stiff ODE system (Daniels and Nemenman, 2015), which models the yeast glycolysis dynamics as:    x˙1 = J0 − k1x1x6 1+(… view at source ↗

**Figure 5.** Figure 5: WB-MPC (left) and FE-DPC (right) based reference tracking of [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

**Figure 6.** Figure 6: 20 quadrotor models with distinct dynamics parameterization are randomly initialized within state bounds. Each experiences a random dynamics switch between 2-20 s. FEDPC successfully stabilizes all models at the reference height of h = 0.4m. to its access to exact system dynamics, FE-DPC delivers comparable accuracy across all examples despite relying on learned system identification and control approxima… view at source ↗

read the original abstract

We introduce a differentiable framework for zero-shot adaptive control over parametric families of nonlinear dynamical systems. Our approach integrates a function encoder-based neural ODE (FE-NODE) for modeling system dynamics with a differentiable predictive control (DPC) for offline self-supervised learning of explicit control policies. The FE-NODE captures nonlinear behaviors in state transitions and enables zero-shot adaptation to new systems without retraining, while the DPC efficiently learns control policies across system parameterizations, thus eliminating costly online optimization common in classical model predictive control. We demonstrate the efficiency, accuracy, and online adaptability of the proposed method across a range of nonlinear systems with varying parametric scenarios, highlighting its potential as a general-purpose tool for fast zero-shot adaptive control.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper pairs function encoders with differentiable predictive control to learn explicit policies that adapt zero-shot across parametric nonlinear systems, but the extrapolation step for unseen parameters is the weakest link.

read the letter

The main point is a framework that trains a function encoder neural ODE on multiple system parameterizations to model dynamics, then uses differentiable predictive control to learn an explicit policy offline that runs without retraining or solving an optimization problem at runtime. This targets the usual cost of model predictive control on nonlinear plants that change with parameters. The setup is a reasonable way to move computation to training time while keeping the policy explicit and fast online. The abstract reports demonstrations across nonlinear systems with varying parametric scenarios, which at least shows they tried more than one example. That part is useful for readers who want to reduce online compute in adaptive control. The soft spot is the zero-shot claim itself. The function encoder must produce usable dynamics predictions for parameter values never seen during training, and the fixed policy must stay stable on those predictions. Function encoders typically interpolate inside the convex hull of training data; outside it, error can grow without warning. The paper does not appear to supply error bounds, stability certificates, or ablation results that isolate performance on truly out-of-distribution parameters. If the experiments only test mild variations around the training set, the central practical advantage remains unproven. This is aimed at people working on learning-based control who already know neural ODEs and differentiable optimization. A reader looking for a concrete recipe to cut MPC solve time would get something to try, though they would need to verify the generalization range themselves. It deserves peer review because the combination is coherent, the goal is practical, and the empirical claims are checkable with the right additional tests on extrapolation.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces a differentiable framework for zero-shot adaptive control of parametric families of nonlinear dynamical systems. It combines a function encoder-based neural ODE (FE-NODE) to model system dynamics with differentiable predictive control (DPC) to learn explicit control policies via offline self-supervised training. The approach is claimed to enable adaptation to unseen system parameterizations without model retraining or online optimization, while eliminating the computational cost of classical MPC; results are demonstrated across several nonlinear systems with varying parametric scenarios.

Significance. If the zero-shot generalization and closed-loop stability claims hold with sufficient accuracy, the work could meaningfully advance practical adaptive control by shifting MPC-like performance to a fixed explicit policy learned offline. The integration of function encoders for parametric dynamics modeling with DPC is a coherent technical direction that addresses real-time constraints in uncertain nonlinear systems.

major comments (2)

[Abstract and §3] Abstract and §3 (FE-NODE formulation): the central claim of accurate zero-shot adaptation to unseen parameterizations without retraining rests on the unverified extrapolation accuracy of the learned function encoder embeddings. No error bounds, Lipschitz constants, or out-of-distribution prediction metrics are provided to quantify how dynamics approximation error grows outside the convex hull of training parameters; this directly undermines the assertion that the fixed DPC policy remains stable and performant.
[§5] §5 (experimental validation): the reported demonstrations do not include a clear ablation or test set with parameter values deliberately placed outside the training distribution (e.g., extrapolation distances quantified by parameter-space distance or prediction RMSE). Without such evidence, it is impossible to distinguish true zero-shot transfer from interpolation within a dense training grid, which is load-bearing for the zero-shot adaptive control claim.

minor comments (2)

[§2 and §4] Notation for the function encoder embedding dimension and the DPC cost function weighting matrices should be introduced with explicit definitions or a summary table to improve readability.
[Figures 2-4] Figure captions would benefit from stating the specific parameter values used in each subplot and whether they are in-distribution or out-of-distribution.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback, which helps clarify the validation requirements for our zero-shot adaptation claims. We address each major comment below and indicate the corresponding revisions to the manuscript.

read point-by-point responses

Referee: [Abstract and §3] Abstract and §3 (FE-NODE formulation): the central claim of accurate zero-shot adaptation to unseen parameterizations without retraining rests on the unverified extrapolation accuracy of the learned function encoder embeddings. No error bounds, Lipschitz constants, or out-of-distribution prediction metrics are provided to quantify how dynamics approximation error grows outside the convex hull of training parameters; this directly undermines the assertion that the fixed DPC policy remains stable and performant.

Authors: We agree that explicit quantification of extrapolation behavior would strengthen the central claims. The current manuscript supports zero-shot adaptation through empirical demonstrations on multiple nonlinear systems, but does not include dedicated out-of-distribution metrics or error growth analysis. In the revised version we will add an analysis of FE-NODE prediction errors for parameter values outside the training convex hull, together with their effect on closed-loop DPC performance. While deriving general Lipschitz bounds for the learned embeddings is beyond the scope of this data-driven approach, the added empirical metrics will directly address the concern about stability and performance of the fixed policy. revision: yes
Referee: [§5] §5 (experimental validation): the reported demonstrations do not include a clear ablation or test set with parameter values deliberately placed outside the training distribution (e.g., extrapolation distances quantified by parameter-space distance or prediction RMSE). Without such evidence, it is impossible to distinguish true zero-shot transfer from interpolation within a dense training grid, which is load-bearing for the zero-shot adaptive control claim.

Authors: We concur that a dedicated extrapolation test set is essential to substantiate the zero-shot claim. The existing experiments cover a range of parametric scenarios, yet do not explicitly isolate out-of-distribution cases with quantified distances. We will revise §5 to include a new subsection presenting results on deliberately chosen extrapolation parameters, reporting parameter-space distances, dynamics prediction RMSE, and closed-loop tracking performance. This will allow readers to distinguish interpolation from true zero-shot transfer and directly support the adaptive control assertions. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper presents a framework combining a function encoder-based neural ODE (FE-NODE) for dynamics modeling with differentiable predictive control (DPC) for policy learning. The central claims rest on standard supervised training of neural components on parametric system families followed by empirical demonstration of zero-shot generalization and control performance. No load-bearing step reduces a prediction or result to a fitted quantity by construction, nor does any uniqueness theorem or ansatz trace exclusively to self-citation chains within the provided text. The derivation remains self-contained against external benchmarks of neural ODE training and DPC optimization, with generalization treated as an empirical outcome rather than a definitional identity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit equations or training details, so no specific free parameters, axioms, or invented entities can be identified beyond standard neural network assumptions.

pith-pipeline@v0.9.0 · 5438 in / 1012 out tokens · 28721 ms · 2026-05-17T23:16:31.069098+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We model the family of system dynamics using a function encoder (FE) that learns basis functions parameterized by neural ordinary differential equations (NODEs)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

57 extracted references · 57 canonical work pages · 3 internal anchors

[1]

Mpcgpu: Real-time nonlinear model predictive control through preconditioned conjugate gradient on the gpu

Emre Adabag, Miloni Atal, William Gerard, and Brian Plancher. Mpcgpu: Real-time nonlinear model predictive control through preconditioned conjugate gradient on the gpu. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 9787--9794. IEEE, 2024

work page 2024
[2]

Al Seyab and Y

R.K. Al Seyab and Y. Cao. Nonlinear system identification for predictive control using continuous time recurrent neural networks and automatic differentiation. Journal of Process Control, 18 0 (6): 0 568--581, 2008. ISSN 0959-1524. doi:https://doi.org/10.1016/j.jprocont.2007.10.012

work page doi:10.1016/j.jprocont.2007.10.012 2008
[3]

CasADi -- A software framework for nonlinear optimization and optimal control

Joel A E Andersson, Joris Gillis, Greg Horn, James B Rawlings, and Moritz Diehl. CasADi -- A software framework for nonlinear optimization and optimal control. Mathematical Programming Computation, 11 0 (1): 0 1--36, 2019. doi:10.1007/s12532-018-0139-4

work page doi:10.1007/s12532-018-0139-4 2019
[4]

Automatic differentiation in machine learning: a survey

Atilim Gunes Baydin, Barak A Pearlmutter, Alexey Andreyevich Radul, and Jeffrey Mark Siskind. Automatic differentiation in machine learning: a survey. Journal of machine learning research, 18 0 (153): 0 1--43, 2018

work page 2018
[5]

Quadrotor dynamics and control

Randal W Beard. Quadrotor dynamics and control. Brigham Young University, 19 0 (3): 0 46--56, 2008

work page 2008
[6]

Development of adaptive control system for aerial vehicles

Vladimir Beliaev, Nadezhda Kunicina, Anastasija Ziravecka, Martins Bisenieks, Roberts Grants, and Antons Patlins. Development of adaptive control system for aerial vehicles. Applied Sciences, 13 0 (23): 0 12940, 2023

work page 2023
[7]

Müller, and Frank Allgöwer

Julian Berberich, Johannes Köhler, Matthias A. Müller, and Frank Allgöwer. Data-driven model predictive control with stability and robustness guarantees. IEEE Transactions on Automatic Control, 66 0 (4): 0 1702--1717, 2021. doi:10.1109/TAC.2020.3000182

work page doi:10.1109/tac.2020.3000182 2021
[8]

Predictive Control for Linear and Hybrid Systems

Francesco Borrelli, Alberto Bemporad, and Manfred Morari. Predictive Control for Linear and Hybrid Systems. Cambridge University Press, USA, 1st edition, 2017. ISBN 1107652871

work page 2017
[9]

Discovering governing equations from data by sparse identification of nonlinear dynamical systems

Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the national academy of sciences, 113 0 (15): 0 3932--3937, 2016

work page 2016
[10]

Neural ordinary differential equations

Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. Advances in neural information processing systems, 31, 2018 a

work page 2018
[11]

Approximating explicit model predictive control using constrained neural networks

Steven Chen, Kelsey Saulnier, Nikolay Atanasov, Daniel D Lee, Vijay Kumar, George J Pappas, and Manfred Morari. Approximating explicit model predictive control using constrained neural networks. In 2018 Annual American control conference (ACC), pages 1520--1527. IEEE, 2018 b

work page 2018
[12]

Efficient inference of parsimonious phenomenological models of cellular dynamics using s-systems and alternating regression

Bryan C Daniels and Ilya Nemenman. Efficient inference of parsimonious phenomenological models of cellular dynamics using s-systems and alternating regression. PloS one, 10 0 (3): 0 e0119821, 2015

work page 2015
[13]

Differentiable predictive control: Deep learning alternative to explicit model predictive control for unknown nonlinear systems

J \'a n Drgo n a, Karol Ki s , Aaron Tuor, Draguna Vrabie, and Martin Klau c o. Differentiable predictive control: Deep learning alternative to explicit model predictive control for unknown nonlinear systems. Journal of Process Control, 116: 0 80--92, 2022

work page 2022
[14]

NeuroMANCER: Neural Modules with Adaptive Nonlinear Constraints and Efficient Regularizations

J \'a n Drgo n a, Aaron Tuor, James Koch, Madelyn Shapiro, Bruno Jacob, and Draguna Vrabie. NeuroMANCER: Neural Modules with Adaptive Nonlinear Constraints and Efficient Regularizations . 2023. URL https://github.com/pnnl/neuromancer

work page 2023
[15]

Learning constrained parametric differentiable predictive control policies with guarantees

J \'a n Drgo n a, Aaron Tuor, and Draguna Vrabie. Learning constrained parametric differentiable predictive control policies with guarantees. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 54 0 (6): 0 3596--3607, 2024

work page 2024
[16]

Sindy with control: A tutorial

Urban Fasel, Eurika Kaiser, J Nathan Kutz, Bingni W Brunton, and Steven L Brunton. Sindy with control: A tutorial. In 2021 60th IEEE conference on decision and control (CDC), pages 16--21. IEEE, 2021

work page 2021
[17]

HPIPM : a high-performance quadratic programming framework for model predictive control

Gianluca Frison and Moritz Diehl. HPIPM : a high-performance quadratic programming framework for model predictive control. IFAC-PapersOnLine, 53 0 (2): 0 6563--6569, 2020

work page 2020
[18]

ANODE: Unconditionally Accurate Memory-Efficient Gradients for Neural ODEs

Amir Gholami, Kurt Keutzer, and George Biros. Anode: Unconditionally accurate memory-efficient gradients for neural odes. arXiv preprint arXiv:1902.10298, 2019

work page internal anchor Pith review Pith/arXiv arXiv 1902
[19]

Deep sparse rectifier neural networks

Xavier Glorot, Antoine Bordes, and Yoshua Bengio. Deep sparse rectifier neural networks. In Proceedings of the fourteenth international conference on artificial intelligence and statistics, pages 315--323. JMLR Workshop and Conference Proceedings, 2011

work page 2011
[20]

Tikhonov regularization and total least squares

Gene H Golub, Per Christian Hansen, and Dianne P O'Leary. Tikhonov regularization and total least squares. SIAM Journal on Matrix Analysis and Applications, 21 0 (1): 0 185--194, 1999

work page 1999
[21]

o hler, Sebastian Trimpe, and Frank Allg \

Michael Hertneck, Johannes K \"o hler, Sebastian Trimpe, and Frank Allg \"o wer. Learning an approximate model predictive controller with guarantees. IEEE Control Systems Letters, 2 0 (3): 0 543--548, 2018

work page 2018
[22]

Learning-based model predictive control: Toward safe learning in control

Lukas Hewing, Kim P Wabersich, Marcel Menner, and Melanie N Zeilinger. Learning-based model predictive control: Toward safe learning in control. Annual Review of Control, Robotics, and Autonomous Systems, 3 0 (1): 0 269--296, 2020

work page 2020
[23]

Acado toolkit—an open-source framework for automatic control and dynamic optimization

Boris Houska, Hans Joachim Ferreau, and Moritz Diehl. Acado toolkit—an open-source framework for automatic control and dynamic optimization. Optimal control applications and methods, 32 0 (3): 0 298--312, 2011

work page 2011
[24]

Zero-shot transfer of neural odes

Tyler Ingebrand, Adam Thorpe, and Ufuk Topcu. Zero-shot transfer of neural odes. Advances in Neural Information Processing Systems, 37: 0 67604--67626, 2024 a

work page 2024
[25]

Zero-shot reinforcement learning via function encoders

Tyler Ingebrand, Amy Zhang, and Ufuk Topcu. Zero-shot reinforcement learning via function encoders. In International Conference on Machine Learning, pages 21007--21019. PMLR, 2024 b

work page 2024
[26]

Function encoders: A principled approach to transfer learning in hilbert spaces

Tyler Ingebrand, Adam J Thorpe, and Ufuk Topcu. Function encoders: A principled approach to transfer learning in hilbert spaces. arXiv preprint arXiv:2501.18373, 2025

work page arXiv 2025
[27]

Identification of dynamic systems: an introduction with applications, volume 85

Rolf Isermann and Marco M \"u nchhof. Identification of dynamic systems: an introduction with applications, volume 85. Springer, 2011

work page 2011
[28]

Learning-based control: A tutorial and some recent results

Zhong-Ping Jiang, Tao Bian, Weinan Gao, et al. Learning-based control: A tutorial and some recent results. Foundations and Trends in Systems and Control , 8 0 (3): 0 176--284, 2020

work page 2020
[29]

Pontryagin differentiable programming: An end-to-end learning and control framework

Wanxin Jin, Zhaoran Wang, Zhuoran Yang, and Shaoshuai Mou. Pontryagin differentiable programming: An end-to-end learning and control framework. Advances in Neural Information Processing Systems, 33: 0 7979--7992, 2020

work page 2020
[30]

Efficient representation and approximation of model predictive control laws via deep learning

Benjamin Karg and Sergio Lucia. Efficient representation and approximation of model predictive control laws via deep learning. IEEE transactions on cybernetics, 50 0 (9): 0 3866--3878, 2020

work page 2020
[31]

Stiff neural ordinary differential equations

Suyong Kim, Weiqi Ji, Sili Deng, Yingbo Ma, and Christopher Rackauckas. Stiff neural ordinary differential equations. Chaos: An Interdisciplinary Journal of Nonlinear Science, 31 0 (9), 2021

work page 2021
[32]

Adam: A Method for Stochastic Optimization

Diederik P Kingma. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

work page internal anchor Pith review Pith/arXiv arXiv 2014
[33]

u ske, Sebastian Peitz, Jan-Hendrik Niemann, Cecilia Clementi, and Christof Sch \

Stefan Klus, Feliks N \"u ske, Sebastian Peitz, Jan-Hendrik Niemann, Cecilia Clementi, and Christof Sch \"u tte. Data-driven approximation of the koopman generator: Model reduction, system identification, and control. Physica D: Nonlinear Phenomena, 406: 0 132416, 2020

work page 2020
[34]

Learning-based model predictive control for safe exploration

Torsten Koller, Felix Berkenkamp, Matteo Turchetta, and Andreas Krause. Learning-based model predictive control for safe exploration. In 2018 IEEE conference on decision and control (CDC), pages 6059--6066. IEEE, 2018

work page 2018
[35]

On convergence of extended dynamic mode decomposition to the K oopman operator

Milan Korda and Igor Mezi \'c . On convergence of extended dynamic mode decomposition to the K oopman operator. Journal of Nonlinear Science, 28 0 (2): 0 687--710, 2018

work page 2018
[36]

Contact models in robotics: a comparative analysis

Quentin Le Lidec, Wilson Jallet, Louis Montaut, Ivan Laptev, Cordelia Schmid, and Justin Carpentier. Contact models in robotics: a comparative analysis. IEEE Transactions on Robotics, 2024

work page 2024
[37]

Napi-mpc: Neural accelerated physics-informed mpc for nonlinear pde systems

Peilun Li, Kaiyuan Tan, and Thomas Beckers. Napi-mpc: Neural accelerated physics-informed mpc for nonlinear pde systems. In Proceedings of the 7th Annual Learning for Dynamics & Control Conference, volume 283 of Proceedings of Machine Learning Research, pages 1230--1242. PMLR, 04--06 Jun 2025 a

work page 2025
[38]

Zero-shot transferable solution method for parametric optimal control problems

Xingjian Li, Kelvin Kan, Deepanshu Verma, Krishna Kumar, Stanley Osher, and J \'a n Drgo n a. Zero-shot transferable solution method for parametric optimal control problems. arXiv preprint arXiv:2509.18404, 2025 b

work page arXiv 2025
[39]

Stabilized neural ordinary differential equations for long-time forecasting of dynamical systems

Alec J Linot, Joshua W Burby, Qi Tang, Prasanna Balaprakash, Michael D Graham, and Romit Maulik. Stabilized neural ordinary differential equations for long-time forecasting of dynamical systems. Journal of Computational Physics, 474: 0 111838, 2023

work page 2023
[40]

Differentiability in unrolled training of neural physics simulators on transient dynamics

Bjoern List, Li-Wei Chen, Kartik Bali, and Nils Thuerey. Differentiability in unrolled training of neural physics simulators on transient dynamics. Computer Methods in Applied Mechanics and Engineering, 433: 0 117441, 2025

work page 2025
[41]

Arch-comp22 category report: Artificial intelligence and neural network control systems (ainncs) for continuous and hybrid systems plants

Diego Manzanas Lopez, Matthias Althoff, Luis Benet, Xin Chen, Jiameng Fan, Marcelo Forets, Chao Huang, Taylor T Johnson, Tobias Ladner, Wenchao Li, et al. Arch-comp22 category report: Artificial intelligence and neural network control systems (ainncs) for continuous and hybrid systems plants. In 9th International Workshop on Applied Verification of Contin...

work page 2022
[42]

Function spaces without kernels: Learning compact hilbert space representations

Su Ann Low, Quentin Rommel, Kevin S Miller, Adam J Thorpe, and Ufuk Topcu. Function spaces without kernels: Learning compact hilbert space representations. arXiv preprint arXiv:2509.20605, 2025

work page arXiv 2025
[43]

Combining system identification with reinforcement learning-based mpc

Andreas B Martinsen, Anastasios M Lekkas, and S \'e bastien Gros. Combining system identification with reinforcement learning-based mpc. IFAC-PapersOnLine, 53 0 (2): 0 8130--8135, 2020

work page 2020
[44]

A neural network approach for high-dimensional optimal control applied to multiagent path finding

Derek Onken, Levon Nurbekyan, Xingjian Li, Samy Wu Fung, Stanley Osher, and Lars Ruthotto. A neural network approach for high-dimensional optimal control applied to multiagent path finding. IEEE Transactions on Control Systems Technology, 31 0 (1): 0 235--251, 2022

work page 2022
[45]

Adaptive control of a quadrotor with dynamic changes in the center of gravity

Ivana Palunko and Rafael Fierro. Adaptive control of a quadrotor with dynamic changes in the center of gravity. IFAC Proceedings Volumes, 44 0 (1): 0 2626--2631, 2011

work page 2011
[46]

Pytorch: An imperative style, high-performance deep learning library

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32, 2019

work page 2019
[47]

G. Pin, M. Filippo, F.A. Pellegrino, G. Fenu, and T. Parisini. Approximate model predictive control laws for constrained nonlinear discrete-time systems: analysis and offline design. International Journal of Control, 86 0 (5): 0 804--820, 2013. doi:10.1080/00207179.2012.762121

work page doi:10.1080/00207179.2012.762121 2013
[48]

System identification: a frequency domain approach

Rik Pintelon and Johan Schoukens. System identification: a frequency domain approach. John Wiley & Sons, 2012

work page 2012
[49]

Neural ordinary differential equations for nonlinear system identification

Aowabin Rahman, J \'a n Drgo n a, Aaron Tuor, and Jan Strube. Neural ordinary differential equations for nonlinear system identification. In 2022 American control conference (ACC), pages 3979--3984. IEEE, 2022

work page 2022
[50]

Model predictive control: theory, computation, and design, volume 2

James Blake Rawlings, David Q Mayne, Moritz Diehl, et al. Model predictive control: theory, computation, and design, volume 2. Nob Hill Publishing Madison, WI, 2020

work page 2020
[51]

Industry engagement with control research: Perspective and messages

Tariq Samad, Margret Bauer, Scott Bortoff, Stefano Di Cairano, Lorenzo Fagiano, Peter Fogh Odgaard, R Russell Rhinehart, Ricardo S \'a nchez-Pe \ n a, Atanas Serbezov, Finn Ankersen, et al. Industry engagement with control research: Perspective and messages. Annual Reviews in Control, 49: 0 1--14, 2020

work page 2020
[52]

Review on model predictive control: An engineering perspective

Max Schwenzer, Muzaffer Ay, Thomas Bergs, and Dirk Abel. Review on model predictive control: An engineering perspective. The International Journal of Advanced Manufacturing Technology, 117 0 (5): 0 1327--1349, 2021

work page 2021
[53]

acados -- a modular open-source framework for fast embedded optimal control

Robin Verschueren, Gianluca Frison, Dimitris Kouzoupis, Jonathan Frey, Niels van Duijkeren, Andrea Zanelli, Branimir Novoselnik, Thivaharan Albin, Rien Quirynen, and Moritz Diehl. acados -- a modular open-source framework for fast embedded optimal control. Mathematical Programming Computation, 2021

work page 2021
[54]

Zero to Autonomy in Real-Time: Online Adaptation of Dynamics in Unstructured Environments

William Ward, Sarah Etter, Jesse Quattrociocchi, Christian Ellis, Adam J Thorpe, and Ufuk Topcu. Zero to autonomy in real-time: Online adaptation of dynamics in unstructured environments. arXiv preprint arXiv:2509.12516, 2025

work page internal anchor Pith review Pith/arXiv arXiv 2025
[55]

A S imple and F ast C oordinate- D escent A ugmented- L agrangian S olver for M odel P redictive control

Liang Wu and Alberto Bemporad. A S imple and F ast C oordinate- D escent A ugmented- L agrangian S olver for M odel P redictive control. IEEE Transactions on Automatic Control, 68 0 (11): 0 6860--6866, 2023

work page 2023
[56]

FORCES NLP : an efficient implementation of interior-point methods for multistage nonlinear nonconvex programs

Andrea Zanelli, Alexander Domahidi, Juan Jerez, and Manfred Morari. FORCES NLP : an efficient implementation of interior-point methods for multistage nonlinear nonconvex programs. International Journal of Control, 93 0 (1): 0 13--29, 2020

work page 2020
[57]

Adaptive checkpoint adjoint method for gradient estimation in neural ode

Juntang Zhuang, Nicha Dvornek, Xiaoxiao Li, Sekhar Tatikonda, Xenophon Papademetris, and James Duncan. Adaptive checkpoint adjoint method for gradient estimation in neural ode. In International Conference on Machine Learning, pages 11639--11649. PMLR, 2020

work page 2020