pith. sign in

arxiv: 2603.28758 · v2 · submitted 2026-03-30 · 📡 eess.SY · cs.SY

Distributionally Robust Planning with mathcal{L}₁ Adaptive Control

Astghik Hakobyan , Amaras Nazarians , Aditya Gahlawat , Naira Hovakimyan , Ilya Kolmanovsky This is my paper

Pith reviewed 2026-05-14 21:15 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords distributionally robust MPCL1 adaptive controlWasserstein ambiguity setscertifiable safetystochastic nonlinear systemsmodel predictive controladaptive controlrobust planning
0
0 comments X

The pith

A hierarchical framework uses L1-adaptive control certificates to set ambiguity-set radii for distributionally robust MPC, ensuring safety under simultaneous system and environmental uncertainties.

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

This paper presents DRP-L1AC, which combines distributionally robust model predictive control with L1-adaptive control for stochastic nonlinear systems. The adaptive controller supplies online bounds on the Wasserstein distance between nominal and actual state distributions. These bounds directly define the ambiguity sets used by the planner to enforce robust chance constraints over a receding horizon. Environmental uncertainty is handled separately through data-driven sets built from finite samples. The result is a planning method that certifies safety without requiring additional samples to characterize system uncertainty.

Core claim

The paper establishes that the L1-adaptive controller's online distributional certificates can be used directly as radii for Wasserstein ambiguity sets in a DR-MPC planner. This integration produces tractable reformulations via Wasserstein duality and guarantees that closed-loop trajectories satisfy distributionally robust chance constraints even when both the system model and the environment deviate from nominal conditions.

What carries the argument

The L1-adaptive controller's online distributional certificates that bound the Wasserstein distance between nominal and true state distributions, serving as certified radii for the ambiguity sets in the DR-MPC planner.

If this is right

  • The combined planner admits tractable reformulations and a sample-based implementation for the environmental ambiguity sets.
  • Safety guarantees hold for stochastic nonlinear systems without needing separate sampling to characterize system dynamics uncertainty.
  • Receding-horizon optimization enforces distributionally robust chance constraints on both types of uncertainty at each step.
  • The framework separates system-level adaptation from environmental uncertainty handling while keeping the overall problem computationally feasible.

Where Pith is reading between the lines

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

  • The method could lower planning conservatism in settings where full distributional sampling for dynamics is expensive or impossible.
  • Similar certificate-driven ambiguity sets might be constructed from other adaptive controllers that track distributional mismatch.
  • The separation of system and environmental uncertainty sources suggests a template for modular robust control in domains like robotics or vehicle navigation.
  • Numerical results could be used to test whether the certified radii remain valid when the system is deployed on hardware with unmodeled delays.

Load-bearing premise

The L1-adaptive controller's online bounds on the Wasserstein distance between nominal and true distributions are tight enough to be used directly as ambiguity-set radii without extra conservatism.

What would settle it

A simulation or experiment in which the measured Wasserstein distance between the true state distribution and the nominal one exceeds the certificate bound, yet the planner still produces a trajectory that violates the intended safety probability.

Figures

Figures reproduced from arXiv: 2603.28758 by Aditya Gahlawat, Amaras Nazarians, Astghik Hakobyan, Ilya Kolmanovsky, Naira Hovakimyan.

Figure 1
Figure 1. Figure 1: Overview of the proposed DRP-L1AC framework. A DR-MPC planner generates reference trajectories satisfying safety constraints, which are tracked by a baseline controller augmented with an L1-AC term. The L1-AC certifies that the true state distribution remains within a prescribed Wasserstein tube around the nominal distribution. The distributional perspective, which operates directly over probability measur… view at source ↗
Figure 2
Figure 2. Figure 2: Closed-loop trajectories for the figure-eight scenario at different times. Without adapta [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
read the original abstract

Safe operation of autonomous systems requires robustness to both model uncertainty and uncertainty in the environment. We propose DRP-$\mathcal{L}_1$AC, a hierarchical framework for stochastic nonlinear systems that integrates distributionally robust model predictive control (DR-MPC) with $\mathcal{L}_1$-adaptive control. The key idea is to use the $\mathcal{L}_1$-adaptive controller's online distributional certificates that bound the Wasserstein distance between nominal and true state distributions, thereby certifying the ambiguity sets used for planning without requiring distribution samples. Environmental uncertainty is captured via data-driven ambiguity sets constructed from finite samples. These are incorporated into a DR-MPC planner enforcing distributionally robust chance constraints over a receding horizon. Using Wasserstein duality, the resulting problem admits tractable reformulations and a sample-based implementation. We show theoretically and via numerical experimentation that our framework ensures certifiable safety in the presence of simultaneous system and environmental uncertainties.

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 / 1 minor

Summary. The manuscript proposes DRP-ℒ₁AC, a hierarchical framework for stochastic nonlinear systems that combines distributionally robust model predictive control (DR-MPC) with ℒ₁-adaptive control. Online distributional certificates from the ℒ₁ controller are used to bound the Wasserstein distance between nominal and true state distributions, thereby defining ambiguity-set radii for the planner without requiring distribution samples. Environmental uncertainty is handled via separate data-driven Wasserstein ambiguity sets. The resulting distributionally robust chance constraints are reformulated tractably via Wasserstein duality, and the paper claims that the overall framework certifies safety both theoretically and in numerical experiments under simultaneous system and environmental uncertainties.

Significance. If the safety claims hold, the work would offer a practical route to certifiable robustness for autonomous systems facing both model mismatch and exogenous disturbances. The integration of ℒ₁-AC online certificates with DR-MPC is a distinctive technical contribution, and the emphasis on sample-free ambiguity-set construction for the system uncertainty component is a clear strength relative to purely data-driven approaches.

major comments (2)
  1. [Theoretical safety analysis] The central safety guarantee rests on the claim that ℒ₁-AC certificates directly supply valid Wasserstein radii for the DR-MPC ambiguity sets. The manuscript does not derive the required Lipschitz constant of the state map or moment bounds on the disturbance process that would convert the standard ℒ₁ pointwise tracking-error bound ||x(t)−x_m(t)||_∞≤γ(t) into a Wasserstein distance W_p(μ_nom,μ_true). This step is load-bearing for the simultaneous-uncertainty claim and must be supplied explicitly.
  2. [Numerical results] The numerical experiments must report the precise rule used to select the Wasserstein radii from the ℒ₁ certificates, together with error-bar statistics and a statement confirming that the certificate data are disjoint from the safety-metric evaluation data. Without these details it is impossible to verify that the reported safety margins are not inflated by post-hoc tuning.
minor comments (1)
  1. [Problem formulation] Clarify the notation for the Wasserstein order p and the precise definition of the ambiguity-set radius in the DR-MPC formulation; inconsistent usage appears between the abstract and the problem statement.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and will revise the manuscript to incorporate the requested clarifications and additions.

read point-by-point responses

  1. Referee: [Theoretical safety analysis] The central safety guarantee rests on the claim that ℒ₁-AC certificates directly supply valid Wasserstein radii for the DR-MPC ambiguity sets. The manuscript does not derive the required Lipschitz constant of the state map or moment bounds on the disturbance process that would convert the standard ℒ₁ pointwise tracking-error bound ||x(t)−x_m(t)||_∞≤γ(t) into a Wasserstein distance W_p(μ_nom,μ_true). This step is load-bearing for the simultaneous-uncertainty claim and must be supplied explicitly.

    Authors: We agree that the conversion from the L1-AC pointwise tracking-error bound to a valid Wasserstein radius requires an explicit derivation involving the Lipschitz constant of the state map and moment bounds on the disturbance. The current manuscript invokes the distributional certificates but does not spell out these intermediate steps. In the revised version we will insert a new lemma (Lemma 3) that derives the Wasserstein radius explicitly: under the assumption that the nonlinear state transition map is L-Lipschitz and the disturbance process has bounded second moment M, we obtain W_p(μ_nom, μ_true) ≤ L·γ(t) + C·M^{1/2}·Δt, where C is a universal constant. This supplies the missing link and makes the simultaneous-uncertainty safety claim fully rigorous. revision: yes

  2. Referee: [Numerical results] The numerical experiments must report the precise rule used to select the Wasserstein radii from the ℒ₁ certificates, together with error-bar statistics and a statement confirming that the certificate data are disjoint from the safety-metric evaluation data. Without these details it is impossible to verify that the reported safety margins are not inflated by post-hoc tuning.

    Authors: We will revise the numerical-results section to state the exact selection rule: the Wasserstein radius for each planning step is set to the supremum of the online L1 certificate γ(t) over the prediction horizon, scaled by the Lipschitz constant L derived in the new Lemma 3. We will also report mean and standard-deviation safety margins computed over 50 independent Monte-Carlo trials and will add an explicit statement that the L1-certificate trajectories used for radius selection were generated from separate simulation runs whose state sequences are disjoint from the evaluation trajectories used to compute the reported safety metrics. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's derivation integrates established L1-adaptive control tracking-error certificates with standard Wasserstein-based DR-MPC reformulations. The L1 certificates are invoked to define ambiguity-set radii for system uncertainty while environmental uncertainty uses separate finite-sample sets; the combination yields distributionally robust chance constraints via duality. No step reduces a claimed prediction or safety certificate to a fitted parameter or self-citation by construction, and no uniqueness theorem or ansatz is smuggled in from overlapping prior work. The central safety claim remains independently supported by the cited adaptive-control bounds and convex optimization results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Because only the abstract is available, the ledger is populated at the level of high-level assumptions stated in the abstract. No explicit free parameters, axioms, or invented entities are named; the framework implicitly assumes that Wasserstein bounds from L1 control are valid certificates and that environmental samples define valid ambiguity sets.

axioms (2)
  • domain assumption L1-adaptive controller produces online bounds on Wasserstein distance between nominal and true distributions
    Abstract states these certificates are used to certify ambiguity sets without requiring distribution samples.
  • standard math Wasserstein duality yields tractable reformulations of the distributionally robust chance-constrained problem
    Abstract invokes this duality to claim tractability and sample-based implementation.

pith-pipeline@v0.9.0 · 5476 in / 1495 out tokens · 49964 ms · 2026-05-14T21:15:34.988788+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

29 extracted references · 29 canonical work pages

  1. [1]

    Zhou and J

    K. Zhou and J. C. Doyle,Essentials of robust control. Prentice Hall Upper Saddle River, NJ, 1998, vol. 104

    work page 1998

  2. [2]

    FaSTrack: A modular framework for fast and guaranteed safe motion planning,

    S. L. Herbert, M. Chen, S. Han, S. Bansal, J. F. Fisac, and C. J. Tomlin, “FaSTrack: A modular framework for fast and guaranteed safe motion planning,” in2017 IEEE 56th Annual Conference on Decision and Control (CDC). IEEE, 2017, pp. 1517–1522

    work page 2017

  3. [3]

    Robust feedback motion planning via contraction theory,

    S. Singh, B. Landry, A. Majumdar, J.-J. Slotine, and M. Pavone, “Robust feedback motion planning via contraction theory,”The International Journal of Robotics Research, 2019

    work page 2019

  4. [4]

    Constrained policy optimization,

    J. Achiam, D. Held, A. Tamar, and P. Abbeel, “Constrained policy optimization,” inProc. Int. Conf. Mach. Learn.PMLR, 2017, pp. 22–31

    work page 2017

  5. [5]

    Constrained variational policy optimization for safe reinforcement learning,

    Z. Liu, Z. Cen, V. Isenbaev, W. Liu, S. Wu, B. Li, and D. Zhao, “Constrained variational policy optimization for safe reinforcement learning,” inProc. Int. Conf. Mach. Learn.PMLR, 2022, pp. 13 644–13 668

    work page 2022

  6. [6]

    Distributionally robust joint chance constraints with second-order moment information,

    S. Zymler, D. Kuhn, and B. Rustem, “Distributionally robust joint chance constraints with second-order moment information,”Math. Program., vol. 137, no. 1, pp. 167–198, 2013

    work page 2013

  7. [7]

    On distributionally robust chance constrained programs with Wasserstein distance,

    W. Xie, “On distributionally robust chance constrained programs with Wasserstein distance,” Math. Program., vol. 186, no. 1, pp. 115–155, 2021

    work page 2021

  8. [8]

    Data-driven distributionally robust chance-constrained optimization with Wasserstein metric,

    R. Ji and M. A. Lejeune, “Data-driven distributionally robust chance-constrained optimization with Wasserstein metric,”J. Global Optim., vol. 79, no. 4, pp. 779–811, 2021

    work page 2021

  9. [9]

    Distributional robustness in minimax linear quadratic control with Wasserstein distance,

    K. Kim and I. Yang, “Distributional robustness in minimax linear quadratic control with Wasserstein distance,”SIAM J. Control Optim., vol. 61, no. 2, pp. 458–483, 2023

    work page 2023

  10. [10]

    Distributionally robust linear quadratic control,

    B. Taskesen, D. Iancu, C ¸ . Ko¸ cyi˘ git, and D. Kuhn, “Distributionally robust linear quadratic control,”Adv. Neural Inf. Process. Syst., vol. 36, pp. 18 613–18 632, 2023

    work page 2023

  11. [11]

    Risk-averse model uncertainty for distributionally robust safe reinforcement learning,

    J. Queeney and M. Benosman, “Risk-averse model uncertainty for distributionally robust safe reinforcement learning,”Adv. Neural Inf. Process. Syst., vol. 36, pp. 1659–1680, 2023

    work page 2023

  12. [12]

    Distributionally robust chance constrained data- enabled predictive control,

    J. Coulson, J. Lygeros, and F. D¨ orfler, “Distributionally robust chance constrained data- enabled predictive control,”IEEE Trans. Autom. Control, vol. 67, no. 7, pp. 3289–3304, 2021

    work page 2021

  13. [13]

    Distributionally robust model predictive control: Closed-loop guarantees and scalable algorithms,

    R. D. McAllister and P. M. Esfahani, “Distributionally robust model predictive control: Closed-loop guarantees and scalable algorithms,”IEEE Trans. Autom. Control, vol. 70, no. 5, pp. 2963–2978, 2024

    work page 2024

  14. [14]

    Wasserstein distributionally robust motion control for collision avoidance using conditional value-at-risk,

    A. Hakobyan and I. Yang, “Wasserstein distributionally robust motion control for collision avoidance using conditional value-at-risk,”IEEE Trans. Robot., vol. 38, no. 2, pp. 939–957, 2021

    work page 2021

  15. [15]

    Iterative risk-constrained model predictive control: A data- driven distributionally robust approach,

    A. Zolanvari and A. Cherukuri, “Iterative risk-constrained model predictive control: A data- driven distributionally robust approach,”IEEE Trans. Autom. Control, 2025

    work page 2025

  16. [16]

    Flight test of anL 1 adaptive controller on the NASA AirSTAR flight test vehicle,

    I. Gregory, E. Xargay, C. Cao, and N. Hovakimyan, “Flight test of anL 1 adaptive controller on the NASA AirSTAR flight test vehicle,” inProceedings of AIAA Guidance, Navigation, and Control Conference, Toronto, Ontario, Canada, August 2010, AIAA 2010-8015

    work page 2010

  17. [17]

    Evaluation of anL1 adaptive flight control law on Calspan’s variable-stability Learjet,

    K. A. Ackerman, E. Xargay, R. Choe, N. Hovakimyan, M. C. Cotting, R. B. Jeffrey, M. P. Blackstun, T. P. Fulkerson, T. R. Lau, and S. S. Stephens, “Evaluation of anL1 adaptive flight control law on Calspan’s variable-stability Learjet,”AIAA Journal of Guidance, Control, and Dynamics, vol. 40, no. 4, pp. 1051–1060, 2017

    work page 2017

  18. [18]

    Robust model based reinforcement learning usingL 1 adaptive control,

    M. Sung, S. H. Karumanchi, A. Gahlawat, and N. Hovakimyan, “Robust model based reinforcement learning usingL 1 adaptive control,” inThe Twelfth International Conference on Learning Representations (ICLR), 2024. [Online]. Available: https: //openreview.net/forum?id=GaLCLvJaoF

    work page 2024

  19. [19]

    L1-GP:L 1 adaptive control with bayesian learning,

    A. Gahlawat, P. Zhao, A. Patterson, N. Hovakimyan, and E. Theodorou, “L1-GP:L 1 adaptive control with bayesian learning,” inLearning for dynamics and control. PMLR, 2020, pp. 826– 837

    work page 2020

  20. [20]

    Contraction l1-adaptive control using gaussian processes,

    A. Gahlawat, A. Lakshmanan, L. Song, A. Patterson, Z. Wu, N. Hovakimyan, and E. A. Theodorou, “Contraction l1-adaptive control using gaussian processes,” inProceedings of the 3rd Conference on Learning for Dynamics and Control. PMLR, 2021, pp. 1027–1040

    work page 2021

  21. [21]

    Robust adaptive model predictive control for guaranteed fast and accurate stabilization in the presence of model errors,

    K. Pereida, L. Brunke, and A. P. Schoellig, “Robust adaptive model predictive control for guaranteed fast and accurate stabilization in the presence of model errors,”Int. J. Robust Nonlinear Control, vol. 31, no. 18, pp. 8750–8784, 2021

    work page 2021

  22. [22]

    L 1-adaptive mppi architecture for robust and agile control of multirotors,

    J. Pravitra, K. A. Ackerman, C. Cao, N. Hovakimyan, and E. A. Theodorou, “L 1-adaptive mppi architecture for robust and agile control of multirotors,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst., 2020, pp. 7661–7666

    work page 2020

  23. [23]

    L 1-DRAC: Distributionally robust adaptive control,

    A. Gahlawat, S. H. Karumanchi, and N. Hovakimyan, “L 1-DRAC: Distributionally robust adaptive control,”arXiv preprint arXiv:2509.04619, 2025

    work page arXiv 2025

  24. [24]

    Wasserstein distributionally robust adaptive covariance steering,

    A. Gahlawat, V. Khatana, D. Wang, S. H. Karumanchi, N. Hovakimyan, and P. Voulgaris, “Wasserstein distributionally robust adaptive covariance steering,” inProc. IEEE Conf. Decis. Control, 2025, pp. 4845–4850

    work page 2025

  25. [25]

    On the rate of convergence in Wasserstein distance of the empirical measure,

    N. Fournier and A. Guillin, “On the rate of convergence in Wasserstein distance of the empirical measure,”Probab. Theory Relat. Fields, vol. 162, no. 3, pp. 707–738, 2015

    work page 2015

  26. [26]

    Data-driven distributionally robust optimization using the Wasserstein metric: Performance guarantees and tractable reformulations,

    P. Mohajerin Esfahani and D. Kuhn, “Data-driven distributionally robust optimization using the Wasserstein metric: Performance guarantees and tractable reformulations,”Math. Pro- gram., vol. 171, no. 1, pp. 115–166, 2018

    work page 2018

  27. [27]

    Distributionally robust stochastic optimization with Wasserstein distance,

    R. Gao and A. Kleywegt, “Distributionally robust stochastic optimization with Wasserstein distance,”Math. Oper. Res., vol. 48, no. 2, pp. 603–655, 2023

    work page 2023

  28. [28]

    A short and general duality proof for Wasserstein distribu- tionally robust optimization,

    L. Zhang, J. Yang, and R. Gao, “A short and general duality proof for Wasserstein distribu- tionally robust optimization,”Oper. Res., vol. 73, no. 4, pp. 2146–2155, 2025

    work page 2025

  29. [29]

    Conditional value-at-risk for general loss distributions,

    R. T. Rockafellar and S. Uryasev, “Conditional value-at-risk for general loss distributions,”J. Bank. Finance, vol. 26, no. 7, pp. 1443–1471, 2002

    work page 2002