Pith. sign in

REVIEW 2 major objections 4 minor 23 references

From Privileged Control to Deployable Adaptation:Fusing Mechanism-Guided Task Reduction with Learned Behavior

T0 review · 2 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper shows that a deployable controller learning only the inverse input gain from causal state history can reproduce a privileged expert's actions and remain uniformly exponentially stable, with about 69% lower tracking error than a…

desk verdict A careful, honest paper with a genuinely clean sampled-data reduction; the stability certificate is conditional, but the paper says so and the limitations are scoped. read the letter →

arxiv 2608.10453 v1 pith:F75NUT5P submitted 2026-08-11 eess.SY cs.SY

Xitong Niu , Peifeng Hui , Zheyong Jiang , Yuan Gao , Chuanlin Zhang This is my paper · ORCID
classification eess.SYcs.SY
keywords privilegedlearningdisturbance-observer-basedcontrolinversegainidentificationsampled-datastabilitycommonLyapunovcertificateimitationadaptivemechanism-guided
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper addresses the training–deployment asymmetry in control: during development an expert controller may use the true input gain and disturbance, but the deployed controller sees only reference and measured states. The authors claim that for a second-order plant with simultaneously varying gain and additive disturbance, the expert action can be decomposed exactly into a measurable task coordinate and a single latent factor, the inverse input gain. Learning only that factor from causal state history yields a deployable controller that closely tracks the privileged expert and provably remains uniformly exponentially stable as long as the learned relative gain stays in $[0.1,4]$. The authors' experiments support the claim: on unseen 60-second trials the structured student reduces tracking RMSE by about 69% relative to a tuned disturbance observer, while direct action-imitation networks fail in closed loop despite moderate offline error.

What carries the argument

The load-bearing object is the exact sampled-data identity for the expert action, $u_{p,k}=k_2[\alpha_k q_k+\Delta x_{2,k}]$, where $q_k=k_1 e_k-\delta x_{1,k}$ is the measurable task coordinate and $\alpha_k=1/a_k$ is the inverse input gain. This identity removes the additive disturbance from the learning problem, so the student network only estimates $\alpha$ from a sparse history of delayed states and derivatives. The stability argument runs through the augmented sampled recursion with state $\xi_k=[e_k,w_k,w_{k-1}]^T$ and matrix $A_d(\rho)$, where $\rho=a\hat{\alpha}$ is the relative inverse gain; matrix convexity of $A_d(\rho)^T P A_d(\rho)$ reduces the certificate to checking two endpoint LMIs, yielding uniform exponential stability for every time-varying $\rho$ in $[0.1,4]$. The same recursion is used to bound residual, saturation, noise, and switching effects.

What would settle it

Run the deployed student on a trajectory where the learned relative gain $\rho=a\hat{\alpha}$ leaves $[0.1,4]$ and check whether the loop diverges; alternatively, re-implement the plant discretization with $x_{2,k}$ in place of $x_{2,k-1}$ and verify that the factorization $u_{p,k}=k_2[\alpha_k q_k+\Delta x_{2,k}]$ no longer holds, which would break the task reduction.

Watch

Extended reading notes

Core claim

The central discovery is an exact sampled-data factorization of the privileged expert law. With the plant discretized as $x_{1,k}=x_{1,k-1}+T_c(a_k x_{2,k-1}+d_k)$ and $x_{2,k}=x_{2,k-1}+T_c u_{k-1}$, the expert action $u_{p,k}=k_2(x_{2r,k}-x_{2,k})$ can be rewritten as $u_{p,k}=k_2[\alpha_k q_k+\Delta x_{2,k}]$, where $\alpha_k=1/a_k$, $q_k=k_1 e_k-\delta x_{1,k}$, and $\Delta x_{2,k}=x_{2,k-1}-x_{2,k}$. The additive disturbance $d_k$ cancels algebraically, leaving only the inverse input gain to be inferred from causal history. The paper then shows that if the estimated relative gain $\rho=a\hat{\alpha}$ remains in $[0.1,4]$, the actual three-state augmented sampled recursion admits a common quadratic Lyapunov function, so the closed loop is uniformly exponentially stable for every time-varying $\rho$ in that interval. The empirical counterpart is that the structured student reproduces the privileged expert's tracking closely (RMSE 0.00802 vs 0.00720 on unseen 60-s trials) and beats a tuned disturbance observer by about 69%.

Load-bearing premise

The argument stands or falls on the exact sample timing (the $x_1$ update uses $x_{2,k-1}$, not $x_{2,k}$) and on the learned relative gain $\rho=a\hat{\alpha}$ staying inside $[0.1,4]$ on deployment trajectories.

Editorial extensions

If this is right

  • A deployable controller can be built that never receives the true gain or disturbance, yet reproduces expert-level tracking once the inverse gain is inferred from measured history.
  • Direct action imitation—static or with history—is not a reliable route: both direct-action networks failed closed-loop trials despite moderate offline error, confirming the realizability and distribution-shift diagnosis.
  • The stability certificate converts a learned-latent error bound into a closed-loop guarantee: if the uniform error satisfies $[1-a_U\varepsilon_\alpha,1+a_U\varepsilon_\alpha]\subseteq[0.1,4]$, exponential stability follows for constant $a,d$ and unsaturated actuation.
  • The observer gain cannot be pushed arbitrarily high to close the gap: the next tested gain above $L=19000$ diverges for every $a<1$, while the structured student remains stable and close to the expert.
  • The method's advantage holds over simultaneous gain–disturbance transitions: across a 25-case grid the student had no task failure and outperformed the observer in 22 cases.

Reading between the lines

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

  • Implicit extension: if the exact sampled identity carries over to plants of higher relative degree or non-matching disturbances, the same 'eliminate the matched disturbance, learn the residual inverse gain' reduction would apply there, but this is not demonstrated in the paper.
  • Editorial: since the paper identifies the unfiltered backward difference $\delta x_{1,k}$ as the dominant noise-sensitive component, introducing a filtered derivative would likely push the reported noise boundary well beyond $\sigma_{x_1}=10^{-4}$; the paper only notes this as a future direction.
  • Editorial: incorporating the $\rho\in[0.1,4]$ interval directly into the training loss could enlarge the coverage fraction beyond the reported 99.9312% and make the certificate apply on nearly all trajectories; the paper lists this as an open question, not a result.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper studies a second-order sampled plant with simultaneous unknown input gain a(t) and additive disturbance d(t). During training, a privileged expert uses (a,d); at deployment a student sees only the reference and measured states. The central theoretical results are (i) an exact sampled-data identity (Proposition 2, Eq. (16)) that eliminates the additive disturbance and factorizes the expert action into a measurable task coordinate q_k times an inverse gain alpha_k, and (ii) a common-quadratic stability certificate (Theorem 1, Eq. (35)) for the delayed augmented sampled recursion, valid for every time-varying relative gain rho_k = a_k*hat(alpha)_k in [0.1,4]. The paper also gives residual, switching, noise, and saturation qualifications, and an action-derived label that avoids using the true plant parameter as supervision. Empirically, a structured student with a 64-32 history-based estimator and a fixed physical action layer is compared with the privileged expert and a tuned disturbance observer over several regimes and unseen 60-s trials; the student attains mean RMSE 0.00802 versus 0.00720 for the expert and 0.02587 for the observer, with no divergence in the tested trials.

Significance. If the conditional gaps are closed, the paper is a valuable contribution to privileged-control transfer: it gives an interpretable, mechanism-guided route rather than a new neural architecture, obtains an exact disturbance-free action factorization, derives a stability certificate for the actual delayed sampled recursion, and provides reproducible scripts and audit instructions. The explicit realizability diagnosis and the ablation showing that direct action imitation fails despite moderate offline error are useful and well presented. The paper is also unusually candid about its limitations, including the statement in §6.7 that the empirical 99.9312% rho coverage 'does not replace the explicit assumption.' These strengths are real; however, the certificate's central premise, rho_k in [0.1,4], is not enforced by the deployed estimator, so the deployability claim remains conditional.

major comments (2)
  1. [§6.7, Theorem 1, Eq. (26)] The stability certificate assumes rho_k = a_k*hat(alpha)_k in [0.1,4], but the deployed estimator (26) clips only hat(alpha) to [0.5,12], and no mechanism enforces the rho interval. Concretely, for a=0.1 with hat(alpha)=0.5 one obtains rho=0.05<0.1, and for a=0.864 with hat(alpha)=12 one obtains rho≈10.37>4, both within the declared clip range. The 99.9312% empirical coverage reported in §6.7 is a single-dataset statistic and, as the authors themselves note, does not replace the explicit assumption. Because the central claim is that the deployed structured controller is stabilizable with a certificate, the theorem currently covers the actual learned controller only conditionally on a premise the controller does not enforce. The revision should either redesign the clip/projection so that rho is guaranteed inside [rho_L,rho_U] for the declared a-domain (e.g., alpha_L = rho_L/a_U and alpha_U = rho_U/a_L), or restrict the operating domain and require online reporting of certificate violations.
  2. [Appendix A.3, Eq. (54)] The bridge from the learned estimator to the certificate is stated only as a hypothetical implication: if a uniform bound |hat(alpha)-alpha| <= epsilon_alpha held, then the certified interval would be guaranteed provided [1-a_U*epsilon_alpha, 1+a_U*epsilon_alpha] is contained in [rho_L,rho_U]. The paper does not establish or estimate such a uniform bound for the trained 64-32 network; the reported offline NRMSE values in Table 4 are averages, not sup-norm errors. Without a measurable uniform error bound (or a Lipschitz-plus-covering-radius argument like Eq. (53) instantiated on the actual test set), the sentence 'Therefore the certified interval is guaranteed if...' remains a conditional statement about an unverified property of the learned map. The authors should either provide a concrete sup-norm estimate on the test data or explicitly state that Theorem 1 is not asserted for the implemented student unless an additional verification step is performed.
minor comments (4)
  1. [§6.7] The phrase 'valid histories' used for the 99.9312% coverage figure is not defined; specify whether it includes saturated intervals, transient windows, or only samples with |q_k| >= q_min, and state how the fraction was computed.
  2. [Table 4 and §5.5] The task-weighted alpha variant has the lowest action NRMSE (0.1611) but is not the main method; the text gives no explicit justification for choosing the action-derived variant (0.1746) over it in terms of closed-loop or task metrics.
  3. [§6.7, Eq. (45)] For reproducibility of the numerical certificate, report the Q matrix used in Eq. (35) or the formula by which the contraction factor gamma = 0.99983295 is computed from P and Q; currently the reader cannot verify the endpoint LMIs without running the released code.
  4. [§7.5] The text states that the training-covered dataset has a in [0.1,0.864] while the grid in §7.5 reaches a=0.9; clarify how the a=0.9 cases are covered by the training distribution and whether the rho-coverage statistic extends to them.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the core derivations are algebraic identities, the stability certificate is conditional, and the empirical evaluation is out-of-sample.

full rationale

The derivation chain is self-contained at every load-bearing point. Proposition 2 / Eq. (16) follows by direct substitution of Eq. (4) into the expert law (5)-(6); the disturbance is eliminated algebraically, and the identity is not assumed as a conclusion. The action-derived label in Eq. (24) is a rearrangement of that same identity, and the paper explicitly states that it "uses the expert's action and deployment-visible trajectory only" and does not claim that the true parameter was unknown; this is ordinary supervised regression, not a fitted quantity renamed as a prediction, and the claims are evaluated on trajectory-disjoint unseen seeds. Theorem 1 is a genuine conditional certificate: it assumes ρ in [0.1,4] and derives uniform exponential stability for the exact three-state recursion using matrix convexity and endpoint LMI checks; it is not fitted to the closed-loop successes, and the paper explicitly disclaims the coverage statistic, saying that "empirical coverage supports relevance but does not replace the explicit assumption." The gap flagged by a reviewer—that α clipping does not enforce ρ in [0.1,4]—is an honest limitation, explicitly listed in Section 8.3, not a circular step. Finally, the reference list contains no prior work by the present authors, so no self-citation is load-bearing. Thus no step reduces, by the paper's own equations or references, to its own input.

Assumptions & free parameters 7 free parameters · 7 assumptions · 0 invented entities

The method does not introduce new physical entities; its free parameters are the controller gains, sampling period, clip range, label-conditioning threshold, and task weights, several of which are not given numeric values. The main axioms are the sampled timing convention, the specific expert law, the constancy of a and d on the certified interval, the rho-interval assumption, and the standard matrix-convexity argument. The certificate is conditional, not a global guarantee.

free parameters (7)
  • Controller gains k1, k2 = k1=10, k2=100
    Chosen by hand; the expert law, the task coordinate q, and the stability certificate all depend on them.
  • Sampling period Tc = 1e-4 s
    Chosen; the sampled-data identity and the LMI certificate are computed for this Tc.
  • Inverse-gain clip range (alpha_L, alpha_U) = (0.5, 12)
    Chosen to bound the estimator; the certificate interval rho in [0.1,4] does not cover the full implied rho range for the tested a values, so this choice widens the gap between theorem and experiment.
  • Label conditioning threshold qmin = not reported
    Samples with |q_k| < qmin are discarded (Section 5.3); without the value the effective training set is not fully specified.
  • Task-weight parameters qcap and c0 = not reported
    The weighting in Eq. (25) shapes the loss, but no numeric values are given.
  • Estimator architecture and regularization lambda = 64-32 ReLU; lambda not reported
    The latent estimator is a 64-32 MLP with L2 regularization; lambda is not stated.
  • History tap set L = {0,1,5,10,50,100,200,500,1000}
    Defines the student's information window; chosen by hand.
assumptions (7)
  • domain assumption The sampled plant obeys x1,k = x1,k-1 + Tc(ak x2,k-1 + dk) and x2,k = x2,k-1 + Tc uk-1.
    Eqs. (2)-(3). The exact cancellation of d and the action-derived label hinge on this timing; the paper warns that using x2,k destroys the identity.
  • domain assumption The privileged expert uses x2r,k = (k1 ek - dk)/ak and up,k = k2(x2r,k - x2,k).
    Eqs. (5)-(6). The factorization targets this specific expert law; a different teacher would not yield the same reduction.
  • domain assumption a and d are constant on the analyzed interval and the actuator is unsaturated.
    Section 6.1-6.2 and Theorem 1; switching and saturation are explicitly outside the theorem (Sections 6.6 and 8.3).
  • ad hoc to paper The learned relative gain rho_k = a hat(alpha)_k lies in [rho_L, rho_U] = [0.1, 4].
    Eq. (44) and the coverage statement in Section 6.7; this interval is chosen for LMI feasibility and is not enforced by the training procedure, so the theorem is conditional on an empirical coverage claim.
  • standard math Matrix convexity of A_d(rho)^T P A_d(rho) and the endpoint LMI condition imply interval stability.
    Lemma 1 and Theorem 1; the proof is a direct expansion and is valid for affine A_d(rho).
  • standard math The ideal latent map and learned estimator are Lipschitz with known constants.
    Appendix A.3, Eqs. (51)-(53); this is a regularity assumption for the coverage-gap bound, not verified for the trained network.
  • domain assumption The label denominator qk is bounded away from zero and derivative noise is below qmin.
    Section 5.3 and Appendix A.2; the bound (50) requires |qk| >= qmin and |nu_k| <= bar_nu < qmin.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Privileged Control to Deployable Adaptation:Fusing Mechanism-Guided Task Reduction with Learned Behavior." pith.science (2026). https://pith.science/paper/F75NUT5P

@misc{pith2026260810453,
  author       = {Pith},
  title        = {Pith review of: From Privileged Control to Deployable Adaptation:Fusing Mechanism-Guided Task Reduction with Learned Behavior},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F75NUT5P}},
  note         = {Machine review of arXiv:2608.10453}
}
abstract

Simultaneous input-gain variation and large additive disturbance create a control problem in which a fixed observer or nominal controller may be unable to reproduce the performance of a regime-aware design. We study a training--deployment asymmetry: during simulation or commissioning, an expert controller is allowed to use the known gain and disturbance, whereas the deployed controller can use only the reference and measured states. Directly imitating expert actions is generally unsafe because the same instantaneous student observation may correspond to different privileged regimes and hence different expert actions. We propose a mechanism-guided transfer route rather than a new neural architecture. An exact sampled-data identity removes the additive disturbance from the expert law and reduces learning to a task-relevant inverse input gain inferred from causal state history. The latent target is reconstructed from expert actions and deployment-visible trajectories, so the true plant parameter is not required as a student label. A common-quadratic certificate is derived for the actual augmented sampled recursion, followed by explicit residual, coverage, switching, noise, and saturation qualifications. A parameter-regime scan shows that the nominal observer's error grows sharply as $a$ decreases and that the next gain above the best non-failing tuning diverges for every tested $a<1$. Direct action networks also fail in closed loop despite moderate offline error, whereas the structured student remains close to the privileged expert and reduces tracking RMSE by about 69\% relative to the tuned observer in unseen 60-s trials. The contribution is an interpretable design perspective for turning privileged multi-regime control knowledge into a deployable adaptive controller, together with conditions under which the transfer is meaningful.

Figures

Figures reproduced from arXiv: 2608.10453 by the authors.

Figure 1
Figure 1. Overall information flow. The apparently simple answer—regress the expert action on measured states—contains a structural trap. If two hidden regimes produce nearly the same student observation but require different expert actions, no deterministic instantaneous student can reproduce both. A larger network cannot restore missing information. Moreover, low one-step regression error need not imply a good feedback cont… view at source ↗
Figure 2
Figure 2. The process of the proposed method. 5.2. Causal features. Let the sparse delays be (22) L = {0, 1, 5, 10, 50, 100, 200, 500, 1000}. The estimator input is (23) zk = colj∈L(ek−j , x2,k−j , δx1,k−j ) ∈ R 27 . At Tc = 10−4 s, this covers 100 ms while retaining dense recent taps. Training-set percentiles scale each coordinate to approximately [−1, 1]; values are clipped at deployment. Histories never cross trajectory bo… view at source ↗
Figure 3
Figure 3. Canonical observer scenarios. The nominal and a > 1 cases are benign, whereas a = 0.1 causes large parameter-induced startup transients that remain present when the disturbance step is added. A higher observer gain reduces but does not remove this effect. Disturbance cases use d : 0 → 2 at t = 2 s over a 4-s run. increases control RMS from 19.87 to 23.81 and peak control from 57.92 to 110.17. Third, the next tested … view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Systematic observer scan across a < 1, a = 1, and a > 1 for a unit disturbance step. Failed cases are omitted from the continuous metric curves and retained in the failure map. The high-gain boundary is confined to the weakened-channel region in this experiment. separa…
Figure 5
Figure 5. Figure 5: Complete closed-loop signals for the nominal and disturbance-only regimes. The disturbance changes from 0 to 2 at t = 2 s in the right column. 7.4. Offline fit does not predict closed-loop success [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Complete closed-loop signals for the low-gain parameter-only and combined low-gain regimes. All three controllers act on independent plant copies subject to the same a and d. 7.5. Deterministic parameter–disturbance grid. The student comparison is intentionally restric…
Figure 7
Figure 7. Figure 7: Complete closed-loop signals for the combined moderate- and high￾gain regimes. achieve mean one-second post-transition RMSE values 0.00751, 0.00889, and 0.02888. The student has no task failure and outperforms the observer in 22 cases; the observer fails at (0.1, 6) an…
Figure 8
Figure 8. Figure 8: Tracking performance over simultaneous post-transition values of a and d with frozen observer gain L = 19000. Crosses denote observer task failures. 7.6. Observer gain tradeoff. The observer gain is scanned from 103 to 105 on four unseen seeds. Mean RMSE decreases from…
Figure 9
Figure 9. Figure 9: Representative simultaneous multiplicative and additive uncer￾tainty transition. The student does not receive a or d. at the observed edge of the performance– robustness tradeoff rather than at a deliberately conservative setting. 7.7. Unseen 60-s trials and robustness…
Figure 10
Figure 10. Figure 10: Observer tracking, control effort, and tested high-gain stability boundary. 0 10-5 10-4 10-3 < x 1 (< x 2 =10< x 1 ) 10-5 100 Tracking RMSE Nominal: a=0.5, d=5 Privileged teacher Observer Proposed 0 10-5 10-4 10-3 < x 1 (< x 2 =10< x 1 ) 10-5 100 105 Control RMS Nomin…
Figure 11
Figure 11. Figure 11: Measurement-noise boundary of the structured student. 8. Discussion, Open Questions, and Conclusion 8.1. Interpretation. The final controller can legitimately be described as a conventional mechanism-based action law plus a learned inverse-gain identifier. This interp…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 20 canonical work pages

  1. [1]

    Disturbance-observer-based control and related methods—an overview.IEEE Transactions on Industrial Electronics, 63(2):1083–1095, 2016

    Wen-Hua Chen, Jun Yang, Lei Guo, and Shihua Li. Disturbance-observer-based control and related methods—an overview.IEEE Transactions on Industrial Electronics, 63(2):1083–1095, 2016

    work page 2016

  2. [2]

    A reduction of imitation learning and structured pre- diction to no-regret online learning

    Stéphane Ross, Geoffrey Gordon, and Drew Bagnell. A reduction of imitation learning and structured pre- diction to no-regret online learning. InProceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, volume 15 ofProceedings of Machine Learning Research, pages 627–635, 2011

    work page 2011

  3. [3]

    A new learning paradigm: Learning using privileged information

    Vladimir Vapnik and Akshay Vashist. A new learning paradigm: Learning using privileged information. Neural Networks, 22(5–6):544–557, 2009

    work page 2009

  4. [4]

    Learning by cheating

    Dian Chen, Brady Zhou, Vladlen Koltun, and Philipp Krähenbühl. Learning by cheating. InProceedings of the Conference on Robot Learning, volume 100 ofProceedings of Machine Learning Research, pages 66–75, 2020

    work page 2020

  5. [5]

    RMA: Rapid motor adaptation for legged robots

    Ashish Kumar, Zipeng Fu, Deepak Pathak, and Jitendra Malik. RMA: Rapid motor adaptation for legged robots. InProceedings of Robotics: Science and Systems, 2021

    work page 2021

  6. [6]

    Deep whole-body control: Learning a unified policy for manipulation and locomotion

    Zipeng Fu, Xuxin Cheng, and Deepak Pathak. Deep whole-body control: Learning a unified policy for manipulation and locomotion. InProceedings of the 6th Conference on Robot Learning, volume 205 of Proceedings of Machine Learning Research, pages 138–149, 2023

    work page 2023

  7. [7]

    A-NC: Adaptive neural control with implicit online inference of privileged parameters

    Marcin Paluch, Florian Bolli, Pehuen Moure, Xiang Deng, and Tobi Delbruck. A-NC: Adaptive neural control with implicit online inference of privileged parameters. InProceedings of the 7th Annual Learning for Dynamics and Control Conference, volume 283 ofProceedings of Machine Learning Research, pages 987–998, 2025

    work page 2025

  8. [8]

    SLR: Learning quadruped locomotion without privileged information

    Shiyi Chen, Zeyu Wan, Shiyang Yan, Chun Zhang, Weiyi Zhang, Qiang Li, Debing Zhang, and Fasih Ud Din Farrukh. SLR: Learning quadruped locomotion without privileged information. InProceedings of the 8th Conference on Robot Learning, volume 270 ofProceedings of Machine Learning Research, pages 3212–3224, 2025

    work page 2025

Show all 23 references
  1. [9]

    TGRL: An algorithm for teacher guided reinforcement learning

    Idan Shenfeld, Zhang-Wei Hong, Aviv Tamar, and Pulkit Agrawal. TGRL: An algorithm for teacher guided reinforcement learning. InProceedings of the 40th International Conference on Machine Learning, volume 202 ofProceedings of Machine Learning Research, pages 31077–31093, 2023

    2023

  2. [10]

    Provable partially observable reinforce- ment learning with privileged information.arXiv preprint arXiv:2412.00985, 2024

    Yang Cai, Xiangyu Liu, Argyris Oikonomou, and Kaiqing Zhang. Provable partially observable reinforce- ment learning with privileged information.arXiv preprint arXiv:2412.00985, 2024

    2024 arXiv

  3. [11]

    Distilling realizable students from unrealizable teachers.arXiv preprint arXiv:2505.09546, 2025

    Yujin Kim, Nathaniel Chin, Arnav Vasudev, and Sanjiban Choudhury. Distilling realizable students from unrealizable teachers.arXiv preprint arXiv:2505.09546, 2025

    2025 arXiv

  4. [12]

    Rusu, Sergio Gomez Colmenarejo, Caglar Gulcehre, Guillaume Desjardins, James Kirkpatrick, Razvan Pascanu, Volodymyr Mnih, Koray Kavukcuoglu, and Raia Hadsell

    Andrei A. Rusu, Sergio Gomez Colmenarejo, Caglar Gulcehre, Guillaume Desjardins, James Kirkpatrick, Razvan Pascanu, Volodymyr Mnih, Koray Kavukcuoglu, and Raia Hadsell. Policy distillation.arXiv preprint arXiv:1511.06295, 2015

    2015 arXiv

  5. [13]

    Andrew Bagnell, Pieter Abbeel, and Jan Peters

    Takayuki Osa, Joni Pajarinen, Gerhard Neumann, J. Andrew Bagnell, Pieter Abbeel, and Jan Peters. An algorithmic perspective on imitation learning.Foundations and Trends in Robotics, 7(1–2):1–179, 2018

    2018

  6. [14]

    Data-driven model-free adaptive control for a class of MIMO nonlinear discrete-time systems.IEEE Transactions on Neural Networks and Learning Systems, 24(6):887–901, 2013

    Zhongsheng Hou and Shangtai Jin. Data-driven model-free adaptive control for a class of MIMO nonlinear discrete-time systems.IEEE Transactions on Neural Networks and Learning Systems, 24(6):887–901, 2013

    2013

  7. [15]

    Formulas for data-driven control: Stabilization, optimality, and robust- ness.IEEE Transactions on Automatic Control, 65(3):909–924, 2020

    Claudio De Persis and Pietro Tesi. Formulas for data-driven control: Stabilization, optimality, and robust- ness.IEEE Transactions on Automatic Control, 65(3):909–924, 2020

    2020

  8. [16]

    Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods.IEEE Transactions on Robotics, 39(3):1749–1767, 2023

    Charles Dawson, Sicun Gao, and Chuchu Fan. Safe control with learned certificates: A survey of neural lyapunov, barrier, and contraction methods.IEEE Transactions on Robotics, 39(3):1749–1767, 2023. PRIVILEGED CONTROL TO DEPLOYABLE ADAPTATION 23

    2023

  9. [17]

    On the equivalence of model-based and data-driven approaches to the design of unknown-input observers.IEEE Transactions on Automatic Control, 70(3):2074–2081, 2025

    Giorgia Disarò and Maria Elena Valcher. On the equivalence of model-based and data-driven approaches to the design of unknown-input observers.IEEE Transactions on Automatic Control, 70(3):2074–2081, 2025

    2025

  10. [18]

    Stochastic neural simulation relations for control transfer

    Alireza Nadali, Ashutosh Trivedi, and Majid Zamani. Stochastic neural simulation relations for control transfer. InProceedings of the International Conference on Neuro-symbolic Systems, volume 288 ofPro- ceedings of Machine Learning Research, pages 597–620, 2025

    2025

  11. [19]

    Narendra and Kannan Parthasarathy

    Kumpati S. Narendra and Kannan Parthasarathy. Identification and control of dynamical systems using neural networks.IEEE Transactions on Neural Networks, 1(1):4–27, 1990

    1990

  12. [20]

    Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang

    George Em Karniadakis, Ioannis G. Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics-informed machine learning.Nature Reviews Physics, 3:422–440, 2021

    2021

  13. [21]

    Hall, Zhaocong Yuan, Siqi Zhou, Jacopo Panerati, and Angela P

    Lukas Brunke, Melissa Greeff, Adam W. Hall, Zhaocong Yuan, Siqi Zhou, Jacopo Panerati, and Angela P. Schoellig. Safe learning in robotics: From learning-based control to safe reinforcement learning.Annual Review of Control, Robotics, and Autonomous Systems, 5:411–444, 2022

    2022

  14. [22]

    Domain randomization for transferring deep neural networks from simulation to the real world

    Josh Tobin, Rachel Fong, Alex Ray, Jonas Schneider, Wojciech Zaremba, and Pieter Abbeel. Domain randomization for transferring deep neural networks from simulation to the real world. In2017 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 23–30, 2017

    2017

  15. [23]

    Learning agile and dynamic motor skills for legged robots.Science Robotics, 4(26):eaau5872, 2019

    Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. Learning agile and dynamic motor skills for legged robots.Science Robotics, 4(26):eaau5872, 2019. Shanghai University of Electric Power, Shanghai, China Email a...

    2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.