pith. sign in

arxiv: 2604.12555 · v1 · submitted 2026-04-14 · 📡 eess.SY · cs.SY· math.OC

Distributionally Robust Stochastic MPC under Disturbance-Affine Feedback Policies

Xu Chen , Lorenz D\"orschel This is my paper

Pith reviewed 2026-05-10 14:54 UTC · model grok-4.3

classification 📡 eess.SY cs.SYmath.OC
keywords distributionally robust MPCdisturbance-affine feedbackWasserstein ambiguity setstochastic model predictive controlrecursive feasibilitytube-based MPCquadratic programming
0
0 comments X

The pith

Disturbance-affine policies enlarge initial feasible sets and improve performance in Wasserstein-robust stochastic MPC.

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

The paper introduces a distributionally robust MPC scheme for linear systems facing unknown disturbances by constructing a Wasserstein ball around recent data to capture worst-case distributions. It replaces tube-based policies with disturbance-affine feedback, which adds degrees of freedom and yields a tractable quadratic program. Recursive feasibility and stability are established, and the scheme is shown to deliver larger initial feasible sets, better average performance, and tighter state variance control than tube-based alternatives. A sympathetic reader would care because the extra policy freedom reduces the conservatism that often limits practical deployment of robust MPC.

Core claim

The DA-DR MPC framework parameterizes control inputs as affine functions of the disturbance inside the Wasserstein ambiguity set, reformulates the resulting problem as a quadratic program, and proves that the closed-loop system remains recursively feasible and stable while achieving less conservative constraint satisfaction than tube-based designs.

What carries the argument

Disturbance-affine feedback policy class, which expresses the control input as an affine function of the realized disturbance and is optimized jointly with the nominal trajectory inside the Wasserstein ball.

Load-bearing premise

The true disturbance distribution lies inside the Wasserstein ball constructed from the most recent data samples, and the affine policy family is expressive enough to realize the claimed performance gains without introducing new conservatism.

What would settle it

A concrete linear system example in which the DA-DR controller either violates state constraints for some disturbance sequence inside the ball or produces a smaller initial feasible set and higher average cost than the corresponding tube-based controller.

Figures

Figures reproduced from arXiv: 2604.12555 by Lorenz D\"orschel, Xu Chen.

Figure 2
Figure 2. Figure 2: Variation comparison in CA50 under DA-DR MPC [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 1
Figure 1. Figure 1: Historical data on the disturbances of CA50 and [PITH_FULL_IMAGE:figures/full_fig_p011_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Average performance of CA50 and IMEP un [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
read the original abstract

This study addresses the stochastic Model Predictive Control (MPC) problem for linear time-invariant systems subjected to unknown disturbance distributions. By leveraging the most recent disturbance data, we construct a set of distributions with similar statistical properties contained within a Wasserstein ball, thereby accounting for the worst-case impacts on constraint satisfaction. Numerous MPC strategies, particularly tube-based approaches, have been extensively studied under the Wasserstein ambiguity set, but these methods often introduce conservatism and can limit control performance. Unlike tube-based approaches, we adopt a disturbance-affine control strategy, which introduces additional control degrees of freedom. We begin by developing the Disturbance-Affine Distributionally Robust (DA-DR) MPC framework, subsequently reformulating the control problem into a tractable quadratic programming formulation. Furthermore, we establish the recursive feasibility and stability of the proposed MPC scheme. Finally, we present comprehensive theoretical analysis and simulation results, demonstrating the superiority of the DA-DR MPC over tube-based MPC in initial feasible sets, average performance, and state variance 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.

Referee Report

2 major / 2 minor

Summary. The paper proposes a Disturbance-Affine Distributionally Robust MPC (DA-DR MPC) for LTI systems under unknown disturbances. It constructs a Wasserstein ball from recent data to capture worst-case distributions, adopts disturbance-affine policies to add degrees of freedom over tube-based methods, derives a tractable QP reformulation, proves recursive feasibility and stability, and reports simulation advantages in initial feasible sets, average performance, and state variance.

Significance. If the recursive feasibility and stability guarantees hold for the time-varying Wasserstein sets, the work would offer a less conservative data-driven robust MPC alternative by exploiting richer policy classes. The QP reformulation and direct comparisons to tube-based MPC provide practical value; the attempt to deliver both theoretical guarantees and empirical validation is a strength.

major comments (2)
  1. [theoretical analysis section] Theoretical analysis section (recursive feasibility and stability proofs): The central claim of recursive feasibility relies on shifting the prior optimal disturbance-affine policy. However, because the Wasserstein ball is explicitly rebuilt from the most recent disturbance data at each step, the ambiguity set is time-varying. The manuscript must explicitly construct a feasible candidate sequence (or prove terminal-set invariance) that satisfies the new worst-case constraints under the updated empirical measure and radius; standard static-set shifting arguments do not automatically apply.
  2. [QP reformulation] QP reformulation (following the DA-DR framework definition): The derivation of the tractable quadratic program must be checked for correctness in handling the distributionally robust constraints under the affine policy class. Any dualization or relaxation steps that convert the inner supremum over the Wasserstein ball should be fully detailed, as gaps here would undermine both the claimed tractability and the subsequent stability arguments.
minor comments (2)
  1. [Abstract] Abstract: The superiority claims (initial feasible sets, average performance, state variance) would benefit from referencing specific simulation figures or quantitative metrics rather than remaining at a high level.
  2. [problem formulation] Notation: The definitions of the Wasserstein radius, empirical measure, and disturbance-affine policy parameters should be introduced with consistent symbols in the problem formulation section before being used in the MPC optimization.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. The two major points raised are addressed point-by-point below. We believe both can be resolved by adding explicit clarifications and expanded derivations without altering the core technical contributions.

read point-by-point responses

  1. Referee: Theoretical analysis section (recursive feasibility and stability proofs): The central claim of recursive feasibility relies on shifting the prior optimal disturbance-affine policy. However, because the Wasserstein ball is explicitly rebuilt from the most recent disturbance data at each step, the ambiguity set is time-varying. The manuscript must explicitly construct a feasible candidate sequence (or prove terminal-set invariance) that satisfies the new worst-case constraints under the updated empirical measure and radius; standard static-set shifting arguments do not automatically apply.

    Authors: We agree that the time-varying nature of the Wasserstein ambiguity set (reconstructed at each step from the latest disturbance realization) requires careful handling. In Section IV, the recursive-feasibility proof constructs a candidate disturbance-affine policy by shifting the prior optimal solution and appending the terminal controller. To accommodate the updated empirical measure and radius, the argument uses the fact that the realized disturbance lies within the support of the previous ball and that the Wasserstein distance between consecutive empirical measures is bounded by the new observation; this ensures the shifted policy satisfies the new worst-case constraints. We will add a dedicated remark and an expanded paragraph in the proof that explicitly verifies the candidate sequence remains feasible under the updated ball, thereby making the time-varying aspect transparent. revision: partial

  2. Referee: QP reformulation (following the DA-DR framework definition): The derivation of the tractable quadratic program must be checked for correctness in handling the distributionally robust constraints under the affine policy class. Any dualization or relaxation steps that convert the inner supremum over the Wasserstein ball should be fully detailed, as gaps here would undermine both the claimed tractability and the subsequent stability arguments.

    Authors: The QP reformulation is obtained by applying the dual of the Wasserstein distributionally robust optimization problem to each constraint under the disturbance-affine policy parameterization. All dualization steps, including the introduction of the dual variables for the Wasserstein distance and the handling of the affine policy coefficients, are presented in Appendix A. We will move the key intermediate equalities into the main text (Section III) and add a short derivation subsection that spells out the exact dual problem, the strong-duality conditions invoked, and any conservative relaxations employed to obtain a quadratic program. This will also confirm that the resulting QP preserves the recursive feasibility and stability properties established later. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation remains self-contained

full rationale

The paper constructs the Wasserstein ambiguity set from recent data as an explicit modeling choice, adopts the disturbance-affine policy class by design, reformulates the resulting problem as a tractable QP, and separately claims to prove recursive feasibility plus stability. No quoted equation or step reduces the feasibility/stability result to a fitted parameter, a self-citation chain, or a renaming of the input data; the central claims rest on standard MPC arguments adapted to the time-varying but explicitly defined ambiguity set. The derivation therefore does not collapse to its own inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on the existence of a Wasserstein ball containing the true distribution and on the ability to optimize over affine policies without losing tractability.

free parameters (1)
  • Wasserstein ball radius
    Chosen based on recent data; its value directly affects the size of the ambiguity set and thus the conservatism of the controller.
axioms (1)
  • domain assumption The system is linear time-invariant and disturbances are additive.
    Standard for the MPC formulation but required for the affine policy and QP reformulation to hold.

pith-pipeline@v0.9.0 · 5472 in / 1196 out tokens · 43747 ms · 2026-05-10T14:54:32.590784+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

  1. [1]

    Wasserstein tube mpc with exact uncertainty propagation

    Liviu Aolaritei, Marta Fochesato, John Lygeros, and Florian D¨ orfler. Wasserstein tube mpc with exact uncertainty propagation. In 2023 62nd IEEE Conference on Decision and Control (CDC) , pages 2036–2041. IEEE, 2023

    work page 2023

  2. [2]

    Capture, propagate, and control distribu- tional uncertainty

    Liviu Aolaritei, Nicolas Lanzetti, and Florian D¨ orfler. Capture, propagate, and control distribu- tional uncertainty. In 2023 62nd IEEE Conference on Decision and Control (CDC) , pages 3081–3086. IEEE, 2023

    work page 2023

  3. [3]

    Calafiore and Marco C

    Giuseppe C. Calafiore and Marco C. Campi. The scenario approach to robust control design. IEEE Transactions on automatic control , 51(5):742–753, 2006

    work page 2006

  4. [4]

    Calafiore and Lorenzo Fagiano

    Giuseppe C. Calafiore and Lorenzo Fagiano. Ro- bust model predictive control via scenario optimiza- tion. IEEE Transactions on Automatic Control , 58(1):219–224, 2012

    work page 2012

  5. [5]

    Calafiore and Lorenzo Fagiano

    Giuseppe C. Calafiore and Lorenzo Fagiano. Stochastic model predictive control of lpv systems via scenario optimization. Automatica, 49(6):1861– 1866, 2013

    work page 2013

  6. [6]

    Probabilistic tubes in linear stochastic model predictive control

    Mark Cannon, Basil Kouvaritakis, and Desmond Ng. Probabilistic tubes in linear stochastic model predictive control. Systems & Control Letters , 58(10):747–753, 2009

    work page 2009

  7. [7]

    Rakovi´ c, and Qifeng Cheng

    Mark Cannon, Basil Kouvaritakis, Saˇ sa V. Rakovi´ c, and Qifeng Cheng. Stochastic tubes in model pre- dictive control with probabilistic constraints. IEEE Transactions on Automatic Control, 56(1):194–200, 2011

    work page 2011

  8. [8]

    Gasoline controlled auto-ignition with learning-based uncertainty using stochastic model predictive control

    Xu Chen, Maximilian Basler, Maike Stemmler, and Dirk Abel. Gasoline controlled auto-ignition with learning-based uncertainty using stochastic model predictive control. In 2023 62nd IEEE Conference on Decision and Control (CDC) , pages 6328–6335. IEEE, 2023

    work page 2023

  9. [9]

    Data-driven robust op- timization using scenario-induced uncertainty sets

    Meysam Cheramin, Richard Li-Yang Chen, Jian- qiang Cheng, and Ali Pinar. Data-driven robust op- timization using scenario-induced uncertainty sets. arXiv preprint arXiv:2107.04977, 2021

    work page arXiv 2021

  10. [10]

    Data- driven distributionally robust mpc for constrained stochastic systems

    Peter Coppens and Panagiotis Patrinos. Data- driven distributionally robust mpc for constrained stochastic systems. IEEE Control Systems Letters , 6:1274–1279, 2021

    work page 2021

  11. [11]

    Distributionally ro- bust optimization under moment uncertainty with application to data-driven problems

    Erick Delage and Yinyu Ye. Distributionally ro- bust optimization under moment uncertainty with application to data-driven problems. Operations re- search, 58(3):595–612, 2010

    work page 2010

  12. [12]

    Stochastic linear model predictive control with chance constraints–a review

    Marcello Farina, Luca Giulioni, and Riccardo Scat- tolini. Stochastic linear model predictive control with chance constraints–a review. Journal of Pro- cess Control, 44:53–67, 2016

    work page 2016

  13. [13]

    Data-driven distributionally robust bounds for stochastic model predictive control

    Marta Fochesato and John Lygeros. Data-driven distributionally robust bounds for stochastic model predictive control. In 2022 IEEE 61st Conference on Decision and Control (CDC) , pages 3611–3616. 12 IEEE, 2022

    work page 2022

  14. [14]

    Distributionally robust stochastic optimization with wasserstein distance

    Rui Gao and Anton Kleywegt. Distributionally robust stochastic optimization with wasserstein distance. Mathematics of Operations Research , 48(2):603–655, 2023

    work page 2023

  15. [15]

    Goulart, Eric C

    Paul J. Goulart, Eric C. Kerrigan, and Jan M. Ma- ciejowski. Optimization over state feedback poli- cies for robust control with constraints. Automat- ica, 42(4):523–533, 2006

    work page 2006

  16. [16]

    Wabersich, and Melanie N

    Lukas Hewing, Kim P. Wabersich, and Melanie N. Zeilinger. Recursively feasible stochastic model pre- dictive control using indirect feedback. Automatica, 119:109095, 2020

    work page 2020

  17. [17]

    Hota, Ashish Cherukuri, and John Lygeros

    Ashish R. Hota, Ashish Cherukuri, and John Lygeros. Data-driven chance constrained optimiza- tion under wasserstein ambiguity sets. In 2019 American Control Conference (ACC), pages 1501–

    work page 2019

  18. [18]

    Direct gasoline injection in the negative valve overlap of a homogeneous charge compression ig- nition engine

    Lucien Koopmans, Roy Ogink, and Ingemar Den- bratt. Direct gasoline injection in the negative valve overlap of a homogeneous charge compression ig- nition engine. SAE transactions, pages 1365–1376, 2003

    work page 2003

  19. [19]

    Model pre- dictive control

    Basil Kouvaritakis and Mark Cannon. Model pre- dictive control. Switzerland: Springer International Publishing, 38:13–56, 2016

    work page 2016

  20. [20]

    Rakovi´ c, and Qifeng Cheng

    Basil Kouvaritakis, Mark Cannon, Saˇ sa V. Rakovi´ c, and Qifeng Cheng. Explicit use of probabilistic dis- tributions in linear predictive control. Automatica, 46(10):1719–1724, 2010

    work page 2010

  21. [21]

    Wilbur Langson, Ioannis Chryssochoos, SV Rakovi´ c, and David Q. Mayne. Robust model predictive control using tubes. Automatica, 40(1):125–133, 2004

    work page 2004

  22. [22]

    A proba- bilistically constrained model predictive controller

    Pu Li, Moritz Wendt, and G¨ unter Wozny. A proba- bilistically constrained model predictive controller. Automatica, 38(7):1171–1176, 2002

    work page 2002

  23. [23]

    Lee, and Fengqi You

    Shuwen Lu, Jay H. Lee, and Fengqi You. Soft- constrained model predictive control based on data- driven distributionally robust optimization. AIChE Journal, 66(10):e16546, 2020

    work page 2020

  24. [24]

    Stochastic mpc with distributionally robust chance constraints

    Christoph Mark and Steven Liu. Stochastic mpc with distributionally robust chance constraints. IF AC-PapersOnLine, 53(2):7136–7141, 2020

    work page 2020

  25. [25]

    Data-driven dis- tributionally robust mpc: An indirect feedback ap- proach

    Christoph Mark and Steven Liu. Data-driven dis- tributionally robust mpc: An indirect feedback ap- proach. arXiv preprint arXiv:2109.09558, 2021

    work page arXiv 2021

  26. [26]

    Robust model predictive control of con- strained linear systems with bounded disturbances

    David Q Mayne, Mar ´ ıa M Seron, and Saˇ sa V Rakovi´ c. Robust model predictive control of con- strained linear systems with bounded disturbances. Automatica, 41(2):219–224, 2005

    work page 2005

  27. [27]

    Data-driven distributionally robust optimization using the wasserstein metric: Performance guaran- tees and tractable reformulations

    Peyman Mohajerin Esfahani and Daniel Kuhn. Data-driven distributionally robust optimization using the wasserstein metric: Performance guaran- tees and tractable reformulations. Mathematical Programming, 171(1):115–166, 2018

    work page 2018

  28. [28]

    Active Flow and Com- bustion Control 2018

    Eugen Nuss, Dennis Ritter, Maximilian Wick, Jakob Andert, Dirk Abel, and Thivaharan Albin. Reduced order modeling for multi-scale control of low temperature combustion engines. In Active Flow and Combustion Control 2018: Papers Con- tributed to the Conference “Active Flow and Com- bustion Control 2018”, September 19–21, 2018, Berlin, Germany , pages 167–1...

    work page 2018

  29. [29]

    Experimental studies on controlled auto- ignition (cai) combustion of gasoline in a 4-stroke engine

    Aaron Oakley, Hua Zhao, Nicos Ladommatos, and Tom Ma. Experimental studies on controlled auto- ignition (cai) combustion of gasoline in a 4-stroke engine. SAE Transactions, pages 1062–1075, 2001

    work page 2001

  30. [30]

    Jones, and Manfred Morari

    Frauke Oldewurtel, Colin N. Jones, and Manfred Morari. A tractable approximation of chance con- strained stochastic mpc based on affine disturbance feedback. In 2008 47th IEEE Conference on Deci- sion and Control , pages 4731–4736, 2008

    work page 2008

  31. [31]

    Dey, and Laurence A

    Feng Qiu, Shabbir Ahmed, Santanu S. Dey, and Laurence A. Wolsey. Covering linear programming with violations. INFORMS Journal on Computing , 26(3):531–546, 2014

    work page 2014

  32. [32]

    Distributionally robust optimization: A review

    Hamed Rahimian and Sanjay Mehrotra. Distri- butionally robust optimization: A review. arXiv preprint arXiv:1908.05659, 2019

    work page arXiv 1908

  33. [33]

    Tyrrell Rockafellar and Roger J-B Wets

    R. Tyrrell Rockafellar and Roger J-B Wets. Vari- ational analysis , volume 317. Springer Science & Business Media, 2009

    work page 2009

  34. [34]

    Optimal transport for ap- plied mathematicians

    Filippo Santambrogio. Optimal transport for ap- plied mathematicians. Birk¨ auser, NY, 55(58- 63):94, 2015

    work page 2015

  35. [35]

    Minimax analysis of stochastic problems

    Alexander Shapiro and Anton Kleywegt. Minimax analysis of stochastic problems. Optimization Meth- ods and Software, 17(3):523–542, 2002

    work page 2002

  36. [36]

    van Hessem and Okko H

    Dennis H. van Hessem and Okko H. Bosgra. A conic reformulation of model predictive control including bounded and stochastic disturbances under state and input constraints. In Proceedings of the 41st IEEE Conference on Decision and Control, 2002. , volume 4, pages 4643–4648. IEEE, 2002

    work page 2002

  37. [37]

    Glynn, and Yinyu Ye

    Zizhuo Wang, Peter W. Glynn, and Yinyu Ye. Like- lihood robust optimization for data-driven prob- lems. Computational Management Science, 13:241– 261, 2016

    work page 2016

  38. [38]

    On distributionally robust chance con- strained programs with wasserstein distance

    Weijun Xie. On distributionally robust chance con- strained programs with wasserstein distance. Math- ematical Programming, 186(1):115–155, 2021

    work page 2021

  39. [39]

    Data-driven risk-averse stochastic optimization with wasserstein metric

    Chaoyue Zhao and Yongpei Guan. Data-driven risk-averse stochastic optimization with wasserstein metric. Operations Research Letters, 46(2):262–267, 2018

    work page 2018

  40. [40]

    Dis- tributionally robust joint chance constraints with second-order moment information

    Steve Zymler, Daniel Kuhn, and Ber¸ c Rustem. Dis- tributionally robust joint chance constraints with second-order moment information. Mathematical Programming, 137:167–198, 2013. A Appendix In this appendix, we present the previously defined ma- trices that were not clarified and some of the intermedi- ate steps involved in the proofs. 13 A.1 Matrices in E...

    work page 2013