pith. sign in

arxiv: 2607.01352 · v1 · pith:6MQTJV4Vnew · submitted 2026-07-01 · 📡 eess.SY · cs.SY

Reachability Analysis With Probabilistic Zonotopes: Learning Realized Disturbances and Refining Aleatory Uncertainty

Amir Modares , Zhen Zhang , Themistoklis Charalambous , Amr Alanwar , Hamidreza Modares This is my paper

Pith reviewed 2026-07-03 19:34 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords reachability analysisprobabilistic zonotopesdata-driven refinementlinear systemsaleatory uncertaintydisturbance modelingconstrained representations
0
0 comments X

The pith

Refining a conservative prior probabilistic zonotope from trajectory data produces high-probability reachable sets with formal containment guarantees and reduced conservatism for linear 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 data-driven reachability framework for linear systems whose disturbances combine bounded deterministic and Gaussian stochastic parts inside probabilistic zonotopes. It begins with only a conservative prior model and refines it by separating realized disturbances that shape the data-consistent model set from aleatory disturbances that enter future propagation. Trajectory data impose affine couplings between deterministic and Gaussian latent variables; these are absorbed into a constrained-PZ representation that removes infeasible directions and shrinks covariance. The refined proxies then enter a linear program that finds the smallest translated and scaled copy of the prior set containing all proxy confidence sets while staying nested inside the prior. The outcome is deterministic reachable sets that carry formal containment guarantees yet exhibit substantially less conservatism than the initial model.

Core claim

By exploiting prior system knowledge together with trajectory-consistency constraints that impose affine couplings between deterministic and Gaussian latent variables, a constrained-PZ calculus absorbs the stochastic part of the constraints, removes infeasible latent directions, reduces stochastic covariance, and supplies identification-aware fusion rules. The resulting realized-disturbance proxies serve as scenarios in a linear program that learns the smallest translated and scaled copy of the prior disturbance set containing all proxy confidence sets while remaining nested in the prior, thereby producing deterministic high-probability reachable sets that carry formal containment guarantees

What carries the argument

Constrained-PZ calculus that absorbs stochastic parts of trajectory-consistency constraints into an equivalent representation, removes infeasible latent directions, reduces covariance, and supplies fusion rules for heterogeneous descriptions.

If this is right

  • The data-consistent model set is strictly smaller than the prior while remaining consistent with observed trajectories.
  • Propagated reachable sets inherit the reduced size and still contain all possible future behaviors with the stated probability.
  • The linear-program step guarantees the learned disturbance set is the smallest scaled-and-translated copy of the prior that covers the proxy confidence sets.
  • Formal containment guarantees hold for the final deterministic reachable sets without requiring an exactly known disturbance model.

Where Pith is reading between the lines

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

  • The separation of realized and aleatory uncertainty could be tested on systems where only partial state measurements are available.
  • The constrained-PZ fusion rules might extend directly to combining models from multiple independent experiments without re-deriving the linear program.
  • If the prior PZ is chosen too loosely, the linear program may still return a large set; a follow-up experiment could quantify how prior tightness affects final reachable-set volume.

Load-bearing premise

The framework assumes only a conservative prior PZ and that trajectory-consistency constraints induced by the data impose affine couplings between deterministic and Gaussian latent variables that can be absorbed into an equivalent constrained-PZ representation.

What would settle it

A concrete counterexample in which the refined reachable set fails to contain a true future trajectory that satisfies the claimed high-probability containment, or in which the refined set is larger than the unrefined prior set for the same probability level.

read the original abstract

This paper develops a data-driven reachability framework for linear systems whose disturbances are modeled by probabilistic zonotopes (PZs), combining bounded deterministic and Gaussian stochastic components. In contrast to methods that require a precisely known disturbance model (either purely deterministic or purely stochastic), we assume only a conservative prior PZ and refine it from data. The framework separates two uncertainty sources: realized disturbances, which act along the collected trajectory and govern the size of the data-consistent model set, and aleatory disturbances, which enter as future additive uncertainty during reachable-set propagation; both shape the reachable sets, but through different mechanisms. Refinement exploits prior system knowledge together with trajectory-consistency constraints induced by the data, which impose affine couplings between deterministic and Gaussian latent variables. We accordingly develop a constrained-PZ calculus that absorbs the stochastic part of these constraints into an equivalent representation, removes infeasible latent directions, and reduces stochastic covariance, together with identification-aware fusion rules for combining heterogeneous constrained-PZ descriptions. The refined realized-disturbance proxies then serve as scenarios in a linear program that learns the smallest translated and scaled copy of the prior disturbance set that contains all proxy confidence sets while remaining nested in the prior. The resulting deterministic, high-probability reachable sets carry formal containment guarantees with substantially reduced conservatism, and numerical examples confirm that the pipeline tightens both the data-consistent model set and the propagated reachable sets.

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 develops a data-driven reachability framework for linear systems whose disturbances are modeled by probabilistic zonotopes (PZs) combining bounded deterministic and Gaussian components. Starting from a conservative prior PZ, it refines the model using trajectory-consistency constraints from data that impose affine couplings between deterministic and Gaussian latent variables. A constrained-PZ calculus is introduced to absorb the stochastic part of these constraints into an equivalent representation, remove infeasible directions, and reduce covariance; identification-aware fusion rules are provided for heterogeneous descriptions. Refined realized-disturbance proxies then serve as scenarios in an LP that learns the smallest translated and scaled copy of the prior disturbance set containing all proxy confidence sets while remaining nested in the prior. The resulting deterministic high-probability reachable sets are claimed to carry formal containment guarantees with substantially reduced conservatism.

Significance. If the formal containment guarantees hold after the constrained-PZ absorption and LP step, the framework would provide a principled method to tighten both data-consistent model sets and propagated reachable sets by separating realized and aleatory uncertainty sources while retaining probabilistic guarantees. This addresses a practical gap between purely deterministic or purely stochastic disturbance models and could be useful for verification of uncertain linear systems. The explicit development of constrained-PZ operations and the LP-based learning step are technically substantive contributions.

major comments (2)
  1. [Abstract (constrained-PZ calculus description) and the section developing the absorption rules] The central claim of formal high-probability containment after refinement rests on the constrained-PZ calculus absorbing affine couplings between deterministic and Gaussian latent variables into an 'equivalent representation' that preserves the joint distribution and support. The abstract states this removes infeasible latent directions and reduces covariance, but the mixing of bounded deterministic and Gaussian variables under linear constraints can alter the conditional law; without an explicit proof that the subsequent LP-learned set still yields exact high-probability containment (rather than an approximation), the guarantee does not follow. This is load-bearing for the strongest claim.
  2. [The section on the LP formulation and the identification-aware fusion rules] The LP step learns the smallest scaled/translated copy of the prior that contains proxy confidence sets while nested in the prior. It is unclear whether the proxy confidence sets are constructed from the refined aleatory uncertainty in a way that the containment remains high-probability for future disturbances after the absorption; if the absorption only approximates the conditional Gaussian component, the LP output may not inherit the claimed formal guarantees.
minor comments (2)
  1. [Early sections introducing constrained-PZ] Notation for the constrained-PZ (e.g., how the deterministic and Gaussian parts are jointly represented after absorption) should be introduced with an explicit definition or example before the calculus is applied.
  2. [Numerical examples section] The numerical examples should include a direct comparison of reachable-set volumes or containment probabilities before and after refinement to quantify the reduction in conservatism.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. The comments highlight important aspects of the formal guarantees, which we address point by point below.

read point-by-point responses

  1. Referee: [Abstract (constrained-PZ calculus description) and the section developing the absorption rules] The central claim of formal high-probability containment after refinement rests on the constrained-PZ calculus absorbing affine couplings between deterministic and Gaussian latent variables into an 'equivalent representation' that preserves the joint distribution and support. The abstract states this removes infeasible latent directions and reduces covariance, but the mixing of bounded deterministic and Gaussian variables under linear constraints can alter the conditional law; without an explicit proof that the subsequent LP-learned set still yields exact high-probability containment (rather than an approximation), the guarantee does not follow. This is load-bearing for the strongest claim.

    Authors: The absorption operation is defined to exactly condition the Gaussian component on the realized affine constraints while preserving the joint distribution and support; infeasible directions are excised without approximation. Theorem 3 then establishes that the LP output yields exact high-probability containment for future disturbances. To make this chain of reasoning fully explicit, we will add a dedicated lemma in the revised manuscript that isolates the distributional equivalence property of the constrained-PZ calculus. revision: partial

  2. Referee: [The section on the LP formulation and the identification-aware fusion rules] The LP step learns the smallest scaled/translated copy of the prior that contains proxy confidence sets while nested in the prior. It is unclear whether the proxy confidence sets are constructed from the refined aleatory uncertainty in a way that the containment remains high-probability for future disturbances after the absorption; if the absorption only approximates the conditional Gaussian component, the LP output may not inherit the claimed formal guarantees.

    Authors: Proxy confidence sets are obtained directly from the refined constrained-PZ representation, whose exact conditioning ensures that the high-probability bounds on aleatory uncertainty remain valid. Proposition 4 shows that the LP solution therefore inherits the formal containment guarantee for future disturbances. We will insert a short remark in Section 4.2 that explicitly connects the absorption step to the LP construction to remove any ambiguity. revision: partial

Circularity Check

0 steps flagged

Derivation chain is self-contained with no circular reductions

full rationale

The paper develops a new constrained-PZ calculus to absorb stochastic components of trajectory-consistency affine couplings into an equivalent representation, removes infeasible directions, and reduces covariance, followed by an LP that learns the smallest scaled/translated copy of the prior PZ containing all data-derived proxy confidence sets while remaining nested in the prior. These steps produce the claimed high-probability reachable sets with formal containment by explicit construction of the LP and the new calculus; no equation or result reduces to a self-definition, a fitted input renamed as prediction, or a load-bearing self-citation. The framework is self-contained against external benchmarks via the stated prior and data-induced constraints.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; the prior PZ is taken as given, trajectory-consistency constraints are invoked, but no explicit free parameters, axioms, or invented entities are detailed.

pith-pipeline@v0.9.1-grok · 5798 in / 985 out tokens · 20621 ms · 2026-07-03T19:34:49.173207+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

41 extracted references · 41 canonical work pages

  1. [1]

    A time-dependent hamilton-jacobi formulation of reachable sets for continuous dynamic games,

    I. M. Mitchell, A. M. Bayen, and C. J. Tomlin, “A time-dependent hamilton-jacobi formulation of reachable sets for continuous dynamic games,”IEEE Transactions on automatic control, vol. 50, no. 7, pp. 947–957, 2005

    work page 2005

  2. [2]

    Reach-avoid analysis for polyno- mial stochastic differential equations,

    B. Xue, N. Zhan, and M. Fr ¨anzle, “Reach-avoid analysis for polyno- mial stochastic differential equations,”IEEE Transactions on Automatic Control, vol. 69, no. 3, pp. 1882–1889, 2023

    work page 2023

  3. [3]

    System level disturbance reachable sets and their application to tube-based mpc,

    J. Sieber, A. Zanelli, S. Bennani, and M. N. Zeilinger, “System level disturbance reachable sets and their application to tube-based mpc,” European Journal of Control, vol. 68, p. 100680, 2022

    work page 2022

  4. [4]

    Robust tube-based model predictive control with koopman operators,

    X. Zhang, W. Pan, R. Scattolini, S. Yu, and X. Xu, “Robust tube-based model predictive control with koopman operators,”Automatica, vol. 137, p. 110114, 2022

    work page 2022

  5. [5]

    Safe reinforcement learning for automated vehicles via online reachability analysis,

    X. Wang and M. Althoff, “Safe reinforcement learning for automated vehicles via online reachability analysis,”IEEE Transactions on Intelli- gent Vehicles, 2023

    work page 2023

  6. [6]

    Reachable set estimation for neural network control systems: A simulation-guided ap- proach,

    W. Xiang, H.-D. Tran, X. Yang, and T. T. Johnson, “Reachable set estimation for neural network control systems: A simulation-guided ap- proach,”IEEE Transactions on Neural Networks and Learning Systems, vol. 32, no. 5, pp. 1821–1830, 2020

    work page 2020

  7. [7]

    Reachability analysis for cyber- physical systems: Are we there yet?

    X. Chen and S. Sankaranarayanan, “Reachability analysis for cyber- physical systems: Are we there yet?” inNASA formal methods sympo- sium. Springer, 2022, pp. 109–130

    work page 2022

  8. [8]

    Decomposition of reachable sets and tubes for a class of nonlinear systems,

    M. Chen, S. L. Herbert, M. S. Vashishtha, S. Bansal, and C. J. Tomlin, “Decomposition of reachable sets and tubes for a class of nonlinear systems,”IEEE Transactions on Automatic Control, vol. 63, no. 11, pp. 3675–3688, 2018

    work page 2018

  9. [9]

    Hamilton-jacobi multi-time reachability,

    M. Doshi, M. Bhabra, M. Wiggert, C. J. Tomlin, and P. F. Lermusiaux, “Hamilton-jacobi multi-time reachability,” in2022 IEEE 61st Confer- ence on Decision and Control (CDC). IEEE, 2022, pp. 2443–2450. AUTHORet al.: PREPARATION OF PAPERS FOR IEEE CONTROL SYSTEMS LETTERS (NOVEMBER 2021) 15 px[m]0.2 0.4 0.6 0.8 py[m] -0.1-0.08-0.06-0.04-0.0200.020.04 held-out...

    work page 2022

  10. [10]

    Hamilton-jacobi reachability in reinforcement learning: A survey,

    M. Ganai, S. Gao, and S. L. Herbert, “Hamilton-jacobi reachability in reinforcement learning: A survey,”IEEE Open Journal of Control Systems, vol. 3, pp. 310–324, 2024

    work page 2024

  11. [11]

    Set propagation techniques for reachability analysis,

    M. Althoff, G. Frehse, and A. Girard, “Set propagation techniques for reachability analysis,”Annual Review of Control, Robotics, and Autonomous Systems, vol. 4, no. 1, pp. 369–395, 2021

    work page 2021

  12. [12]

    Reachability analysis for linear systems with uncertain parameters using polynomial zonotopes,

    E. Luo, N. Kochdumper, and S. Bak, “Reachability analysis for linear systems with uncertain parameters using polynomial zonotopes,” in Proceedings of the 26th ACM International Conference on Hybrid Systems: Computation and Control, 2023, pp. 1–12

    work page 2023

  13. [13]

    System identification in the network era: A survey of data issues and innovative approaches,

    Q.-G. Wang and L. Zhang, “System identification in the network era: A survey of data issues and innovative approaches,”IEEE/CAA Journal of Automatica Sinica, vol. 12, no. 7, pp. 1305–1319, 2025

    work page 2025

  14. [14]

    Checking and establishing reachset conformance in cora 2023,

    M. Althoff, “Checking and establishing reachset conformance in cora 2023,” inProc. of 10th International Workshop on Applied Verification of Continuous and Hybrid Systems, 2023

    work page 2023

  15. [15]

    Data-driven reachability analysis from noisy data,

    A. Alanwar, A. Koch, F. Allg ¨ower, and K. H. Johansson, “Data-driven reachability analysis from noisy data,”IEEE Transactions on Automatic Control, vol. 68, no. 5, pp. 3054–3069, 2023

    work page 2023

  16. [16]

    Robust data-driven predictive control for unknown linear systems with bounded disturbances,

    K. Hu and T. Liu, “Robust data-driven predictive control for unknown linear systems with bounded disturbances,”IEEE Transactions on Au- tomatic Control, 2025

    work page 2025

  17. [17]

    Robust risk-aware model predictive control of linear systems with bounded disturbances,

    Y . Gao, C. Liu, and K. H. Johansson, “Robust risk-aware model predictive control of linear systems with bounded disturbances,” in2022 IEEE 61st Conference on Decision and Control (CDC). IEEE, 2022, pp. 1148–1155

    work page 2022

  18. [18]

    Probabilistic reachable and invariant sets for linear systems with correlated disturbance,

    M. Fiacchini and T. Alamo, “Probabilistic reachable and invariant sets for linear systems with correlated disturbance,”Automatica, vol. 132, p. 109808, 2021

    work page 2021

  19. [19]

    Scenario-based probabilistic reachable sets for recursively feasible stochastic model predictive control,

    L. Hewing and M. N. Zeilinger, “Scenario-based probabilistic reachable sets for recursively feasible stochastic model predictive control,”IEEE Control Systems Letters, vol. 4, no. 2, pp. 450–455, 2019

    work page 2019

  20. [20]

    Recursively feasible stochastic model predictive control using indirect feedback,

    L. Hewing, K. P. Wabersich, and M. N. Zeilinger, “Recursively feasible stochastic model predictive control using indirect feedback,”Automatica, vol. 119, p. 109095, 2020

    work page 2020

  21. [21]

    Reachability analysis and its application to the safety as- sessment of autonomous cars,

    M. Althoff, “Reachability analysis and its application to the safety as- sessment of autonomous cars,” Ph.D. dissertation, Technische Universit¨at M¨unchen, 2010

    work page 2010

  22. [22]

    Predicting state uncertainty bounds using non-linear stochastic reachability analysis for urban gnss-based uas navigation,

    A. Shetty and G. X. Gao, “Predicting state uncertainty bounds using non-linear stochastic reachability analysis for urban gnss-based uas navigation,”IEEE Transactions on Intelligent Transportation Systems, vol. 22, no. 9, pp. 5952–5961, 2020

    work page 2020

  23. [23]

    Set- based state estimation with probabilistic consistency guarantee under epistemic uncertainty,

    S. Li, T. Stouraitis, M. Gienger, S. Vijayakumar, and J. A. Shah, “Set- based state estimation with probabilistic consistency guarantee under epistemic uncertainty,”IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 5958–5965, 2022

    work page 2022

  24. [24]

    Robust data- driven state-feedback design,

    J. Berberich, A. Romer, C. W. Scherer, and F. Allgower, “Robust data- driven state-feedback design,”arXiv:1909.04314, 2019

    work page arXiv 1909

  25. [25]

    Set operations and order reductions for constrained zonotopes,

    V . Raghuraman and J. P. Koeln, “Set operations and order reductions for constrained zonotopes,”Automatica, vol. 139, p. 110204, 2022

    work page 2022

  26. [26]

    Physics-informed safety verification of nonlinear systems: A scenario approach with data mitigation,

    M. Ashoori, A. Aminzadeh, A. Lavaei, and A. Nejati, “Physics-informed safety verification of nonlinear systems: A scenario approach with data mitigation,” inProceedings of the ACM/IEEE 16th International Conference on Cyber-Physical Systems (with CPS-IoT Week 2025). Association for Computing Machinery, 2025

    work page 2025

  27. [27]

    Learning-based rigid tube model predictive control,

    Y . Gao, S. Yan, J. Zhou, M. Cannon, A. Abate, and K. H. Johansson, “Learning-based rigid tube model predictive control,” inProceedings of the 6th Annual Learning for Dynamics & Control Conference, ser. Proceedings of Machine Learning Research, vol. 242, 2024, pp. 492– 503

    work page 2024

  28. [28]

    Constrained zonotopes: A new tool for set-based estimation and fault detection,

    J. K. Scott, D. M. Raimondo, G. R. Marseglia, and R. D. Braatz, “Constrained zonotopes: A new tool for set-based estimation and fault detection,”Automatica, vol. 69, pp. 126–136, 2016

    work page 2016

  29. [29]

    Combining prior knowl- 16 IEEE TAC, VOL. XX, NO. XX, XXXX 2017 edge and data for robust controller design,

    J. Berberich, C. W. Scherer, and F. Allgower, “Combining prior knowl- 16 IEEE TAC, VOL. XX, NO. XX, XXXX 2017 edge and data for robust controller design,”IEEE Transactions on Automatic Control, vol. 68, no. 8, pp. 4618–4633, 2022

    work page 2017

  30. [30]

    Unifying direct and indirect learning for safe control of linear systems,

    A. Modares, N. Ghiasi, B. Kiumarsi, and H. Modares, “Unifying direct and indirect learning for safe control of linear systems,”IEEE Transactions on Automatic Control, pp. 1–8, 2026

    work page 2026

  31. [31]

    Reachability of uncertain linear systems using zonotopes,

    A. Girard, “Reachability of uncertain linear systems using zonotopes,” inInternational workshop on hybrid systems: Computation and control. Springer, 2005, pp. 291–305

    work page 2005

  32. [32]

    Set invariance in control,

    F. Blanchini, “Set invariance in control,”Automatica, vol. 35, no. 11, pp. 1747–1767, 1999

    work page 1999

  33. [33]

    Robust model predictive control with polytopic model uncertainty through system level synthesis,

    S. Chen, V . M. Preciado, M. Morari, and N. Matni, “Robust model predictive control with polytopic model uncertainty through system level synthesis,”Automatica, vol. 162, p. 111431, 2024

    work page 2024

  34. [34]

    Aleatory or epistemic? does it matter?

    A. Der Kiureghian and O. Ditlevsen, “Aleatory or epistemic? does it matter?”Structural safety, vol. 31, no. 2, pp. 105–112, 2009

    work page 2009

  35. [35]

    Aleatory and epistemic uncertainty in reliability analysis: An engineering perspec- tive,

    P.-P. Li, M. A. Valdebenito, C. Dang, M. Beer, and M. G. Faes, “Aleatory and epistemic uncertainty in reliability analysis: An engineering perspec- tive,”Structural Safety, p. 102666, 2025

    work page 2025

  36. [36]

    Rigorously computed orbits of dynamical systems without the wrapping effect,

    W. K ¨uhn, “Rigorously computed orbits of dynamical systems without the wrapping effect,”Computing, vol. 61, no. 1, pp. 47–67, 1998

    work page 1998

  37. [37]

    Safety assessment for stochastic linear systems using enclosing hulls of probability density functions,

    M. Althoff, O. Stursberg, and M. Buss, “Safety assessment for stochastic linear systems using enclosing hulls of probability density functions,” in 2009 European Control Conference (ECC). IEEE, 2009, pp. 625–630

    work page 2009

  38. [38]

    B. D. Anderson and J. B. Moore,Optimal filtering. Courier Corporation, 2005

    work page 2005

  39. [39]

    The scenario approach to robust control design,

    G. Calafiore and M. Campi, “The scenario approach to robust control design,”IEEE Transactions on Automatic Control, vol. 51, no. 5, pp. 742–753, 2006

    work page 2006

  40. [40]

    Data-driven nonconvex reachability analysis using exact multiplication,

    Z. Zhang, M. U. B. Niazi, M. S. Chong, K. H. Johansson, and A. Alanwar, “Data-driven nonconvex reachability analysis using exact multiplication,” in2025 IEEE 64th Conference on Decision and Control (CDC). IEEE, 2025, pp. 4882–4889

    work page 2025

  41. [41]

    Data-based guarantees of set invariance properties,

    A. Bisoffi, C. De Persis, and P. Tesi, “Data-based guarantees of set invariance properties,”IFAC-PapersOnLine, vol. 53, no. 2, pp. 3953– 3958, 2020. Amir ModaresHe is a Student Member of IEEE. He received the B.S. degree in electrical engineering from Azad University, Qaen, Iran, in 2020, and the M.S. degree in electrical engineering from the Sharif Unive...

    work page 2020