Pith. sign in

REVIEW 3 major objections 4 minor 28 references

AutoIntervene: Calibrated Intervention for Action-Chunking Imitation Learning Policies

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read AutoIntervene selectively hands control from an action-chunking robot policy to an operator when the proposed chunk loses visual-action support, then returns control after the operator recovers, and retrains the policy on the retained…

desk verdict A credible, genuinely new intervention loop for action-chunking policies, with a real but not fatal gap in evidence for the 'calibrated' threshold claim. read the letter →

arxiv 2608.07065 v1 pith:64DSWCKR submitted 2026-08-07 cs.RO cs.AIcs.CVcs.HCcs.LG

Jinhe Tang , Weiming Zhi This is my paper · ORCID
classification cs.ROcs.AIcs.CVcs.HCcs.LG
keywords action-chunkingpoliciesimitationlearninginterventionframeworkvisual-actionsupportmemorybidirectionalhandoffpolicyadaptationbimanualmanipulationruntimemonitoring
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 claims that an action-chunking visuomotor policy can be made more reliable at deployment by automatically detecting when its proposed action sequences leave the support of successful demonstrations and handing control to an operator precisely at those moments. The framework, AutoIntervene, lets the operator recover from failure, then returns control only when the policy's proposals regain demonstrated support. By retaining the operator-controlled segments as corrective training data and retraining the policy in rounds, the method claims higher post-adaptation task success and lower operator-control time than manual intervention or collecting additional full demonstrations. A sympathetic reader would care because it offers a path to robot learning that uses human time more efficiently and focuses supervision on the states that actually cause failures.

What carries the argument

The central machinery is a visual-action support memory: a set of (multi-view image embedding, action-chunk) pairs mined from every successful training trajectory, combined with a bidirectional control-authority selector. At each evaluation, a proposed action prefix and current visual embeddings form a query; the monitor computes a visual-support score (minimum cosine similarity across camera views to the best-remembered visual neighbors) and an action-risk score (mean normalized Euclidean distance to the best-matching reference actions). Phase-local retrieval from forward windows along the best-matching training trajectories governs policy-to-operator transfer, while global retrieval from the complete memory governs operator-to-policy return. Separate thresholds for the two directions are calibrated from empirical quantiles of these scores on a held-out set of successful expert demonstrations, and a persistence counter requires the same decision across consecutive evaluations to prevent chattering. Retained operator-controlled segments are mixed into the training distribution with a fixed sampling weight for the next round of behavior cloning.

What would settle it

Collect a set of deployment rollouts with a deliberately novel failure (e.g., a different object pose or lighting); if the visual-support score at the moments where the policy first leaves successful trajectories lies above the calibrated lower threshold for a substantial fraction of these rollouts, the trigger is miscalibrated.

Watch

Extended reading notes

Core claim

AutoIntervene establishes that a bidirectional, support-based control handoff can convert targeted operator corrections into effective policy improvement for action-chunking visuomotor policies. The core claim is that each proposed action chunk should be evaluated not by visual similarity alone but jointly with the action itself against a visual-action support memory built from successful task executions. Phase-local retrieval from forward windows along best-matching training trajectories governs the policy-to-operator transfer, while global retrieval from the complete memory governs the operator-to-policy return. Separate thresholds for each direction are calibrated from empirical quantiles of the evaluation scores on held-out successful demonstrations, avoiding manual threshold tuning. In real-world bimanual manipulation experiments, the method raises the seven-task mean success from 30.9% to 80.0% after two adaptation rounds, using roughly 74% less additional control time than adding ten full demonstrations, and outperforms manual switching in both success and operator-control time.

Load-bearing premise

The thresholds for triggering help are calibrated on successful demonstrations only, and these are assumed to also be correct for the unfamiliar, failure-prone states that actually trigger the help.

Editorial extensions

If this is right

  • If the calibrated thresholds transfer to deployment, action-chunking policies can be retrained from targeted operator corrections instead of full demonstrations, using roughly 74% less additional control time than collecting ten full demonstrations in the reported benchmark.
  • The bidirectional handoff with separate, mode-specific thresholds achieves reliable cut-in and cut-out transitions in perturbed rollouts while producing zero false triggers on nominal rollouts in the controlled comparison.
  • The framework is head-agnostic: ACT, Diffusion Policy, and Flow Matching action heads all improve over successive adaptation rounds without head-specific modification.
  • Iterative adaptation continues to improve over multiple rounds on longer-horizon tasks, with Two-Towel Box Packing rising from 28% to 88% success over three rounds, suggesting the loop can be run repeatedly.
  • Both visual support and action risk are necessary components: ablations removing either degrade handoff reliability, with visual support detecting when intervention is needed and action risk preventing premature return to policy control.
  • Retaining operator-controlled segments as separate intervention trajectories focuses supervision on learner-induced failure states and dilutes the corrective update less than manual switching that may extend beyond recovery.
  • The framework's persistence-based switching avoids immediate reversal on score fluctuations, unlike prior monitors that use a single threshold for both directions.
  • The support-memory design could be extended to incorporate object-state estimates or language-based progress checks, which would strengthen the visual-action pairing in tasks where visual similarity is ambiguous.

Reading between the lines

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

  • The paper leaves open whether the same calibration-set quantiles hold when deployment shifts are semantic (new object poses, lighting, tool variants) rather than positional; one could test this by recalibrating only on the new failure states and measuring threshold drift.
  • A natural extension is to make the support memory itself a learned filter that rejects memory entries incompatible with the current task phase, which could reduce false matches that require the forward-window mechanism.
  • One could run AutoIntervene without any held-out calibration set by replacing quantile thresholds with an online change-point detector on the score streams, turning the method into a fully self-calibrating monitor.
  • Because the retained interventions are conditioned on the current policy, the loop resembles a selective DAgger; a potential side effect is that early policy errors shape later data, so the method's benefits may interact with how failures are instantiated early in deployment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. AutoIntervene is an online framework for selectively transferring control between an action-chunking visuomotor policy and a human operator during deployment. It constructs a visual-action memory from successful task executions, evaluates each proposed action chunk against this memory using a visual-support score and an action-risk score, and uses mode-specific thresholds to decide when to switch from policy to operator (phase-local support) and when to return control to the policy (global support). Intervention segments from successful rollouts are retained and mixed into the training set for policy adaptation in successive rounds. Thresholds are calibrated as empirical quantiles of score distributions on a held-out set of successful expert demonstrations. The paper evaluates the framework on nine real-world bimanual manipulation tasks, comparing against manual intervention and additional full demonstrations, and also tests compatibility with ACT, Diffusion Policy, and Flow Matching action heads.

Significance. If the calibration-transfer concern is resolved, AutoIntervene is a practical deployment-time monitoring and adaptation loop: it converts targeted operator corrections into policy improvement with less operator time than manual switching and less data than collecting full demonstrations. The paper has real strengths: the central claim is tested on unseen physical rollouts rather than on the calibration data; the method is action-head agnostic and is evaluated with three different action-generation mechanisms; the controlled handoff comparison includes prior monitors and informative ablations of the two score components; and the data-efficiency comparison against additional full demonstrations is directly quantified. The main weaknesses are the thin validation of the threshold-calibration transfer assumption and the absence of uncertainty quantification in the headline empirical claims.

major comments (3)
  1. [Section III-C, Eq. (5) and Section IV-E, Table V] The load-bearing claim that AutoIntervene is 'calibrated' is not yet established. Equation (5) sets θ_s^β and θ_r^β as empirical quantiles of visual-support and action-risk score collections computed on D_cal, a held-out set of successful expert trajectories. The intervention trigger, however, is intended to fire on learner-induced states that are unsupported or failed, i.e., on states outside the D_cal score distribution. The only direct evidence for transfer of these thresholds is the single-task handoff study in Table V (Lidded Box Packing, 10 perturbed and 10 nominal rollouts, no confidence intervals), and no failure-detection metrics are reported for the other eight tasks. The main benchmark in Table II reports post-adaptation success and operator time, which are compatible with good thresholds but do not measure missed triggers or false triggers. To support the calibration claim, the authors should report per-task missed-detection and false-trigger rates on a validation set of failure rollouts and provide a sensitivity analysis over α_s^β and α_r^β (and ideally over L_pol and L_op).
  2. [Section IV, Tables II-IV and Section IV-E] All success rates in the main benchmark are point estimates over 25 physical rollouts (10 rollouts in Table V), without confidence intervals or significance tests. The abstract and Section IV-B claim higher post-adaptation success and lower operator-control time than manual intervention, but per-task comparisons sometimes go in the opposite direction (e.g., Potato Transfer R2: Human 80% vs. AutoIntervene 76%, Table II), and the operator-time reductions vary widely across tasks. The authors should report binomial confidence intervals for success rates and a bootstrap or paired comparison for operator-control time, or explicitly moderate the claim to a reported trend rather than a demonstrated improvement.
  3. [Section III-C] The phrase 'avoiding direct manual tuning of score cutoffs' is only partially supported. While Eq. (5) replaces manual raw-score cutoffs with empirical quantiles, the tail rates (α_pol_s=0.05, α_pol_r=0.05, α_op_s=0.30, α_op_r=0.30) and the persistence lengths L_pol=L_op=2 are fixed manually. Because these values fully determine the operating point of the monitor, the paper should include a sensitivity analysis over them before claiming that the thresholds are not manually tuned.
minor comments (4)
  1. [Section IV-B and IV-E] The full text contains duplicated paragraphs and captions: the text under Figure 6 is repeated under Figure 7 in Section IV-B, and the Figure 6 caption appears twice; the authors should remove the duplicates.
  2. [Table II] Table II has a formatting issue in the Peg Disassembly R1 AutoIntervene column ('49.372%'), where a value separator appears to be missing; the aligned table should be checked.
  3. [Section III-B, Eq. (3)] The symbol M is used both for the visual-action memory and for the number of action-selected entries in Eq. (3); please rename one of them to avoid confusion.
  4. [Section IV-E] In the handoff protocol, the definition of a 'valid cut-out' and the counting of 'extra transitions' should be made more precise for rollouts that terminate before a full handoff cycle.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claims are evaluated on out-of-sample physical rollouts, and thresholds calibrated on held-out expert demonstrations are used only as fixed deployment decision rules.

full rationale

The derivation chain is not circular. The claimed predictions—higher post-adaptation task success and lower operator-control time than manual intervention—are assessed on unassisted physical rollouts (Tables II–V), not on the calibration set. Thresholds (Eq. 5) are fitted to empirical quantiles of visual-support and action-risk scores on a held-out set D_cal of successful expert demonstrations, but D_cal is excluded from policy training and from the visual-action memory; the fitted thresholds are then used as fixed decision rules during deployment, and the outcomes (task success, cut-in/cut-out recall/precision, and false-trigger rate) are measured on separate physical rollouts. The visual-action memory is built from the same demonstrations used to train the policy, so the notion of 'support' is internal to the training distribution; however, the paper does not claim that these scores are calibrated probabilities of failure, only that the switching thresholds are chosen as empirical quantiles of a reference distribution. The main weakness is the distribution-transfer assumption from D_cal to learner-induced states, but that is a validation/correctness concern, not a circular reduction. Self-citations (TriPilot-FF, Rewind-IL, DOSE3, PATCH, TRACE) appear only in related-work context or implementation details and are not load-bearing; no uniqueness theorem or ansatz is imported from them. No step reduces by construction to its inputs.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central mechanism rests on several domain assumptions about embedding similarity, action-distance validity, and threshold transfer. The free parameters are all hand-chosen monitor and calibration values rather than fitted quantities. No new physical entities are introduced.

free parameters (3)
  • Quantile tail rates alpha_s^pol, alpha_r^pol, alpha_s^op, alpha_r^op = 0.05, 0.05, 0.30, 0.30
    Chosen by hand; these set the thresholds in Eq. (5). No sensitivity analysis is reported.
  • Intervention mixing weight lambda_mix = 2/3
    Chosen by hand; this sets the sampling weight in Eq. (7). No sensitivity analysis is reported.
  • Monitor hyperparameters J, K, B, H_r, M, W, L, U_win, evaluation frequencies = J=K=16, B=H_r=40, M=3, W=5, L_pol=L_op=2, U_win=6, eval 5 Hz (policy) and 30 Hz (operator), H=100
    All set by hand in the implementation details. No ablation or sensitivity analysis is reported for most of them.
assumptions (4)
  • domain assumption Cosine similarity between DINOv3 embeddings reflects task-relevant visual state similarity.
    Used in Eq. (1) for visual-neighbour retrieval; no validation that embedding distance tracks task phase or task success.
  • domain assumption Per-dimension normalized Euclidean distance between proposed and reference action chunks is a valid proxy for action correctness.
    Used in Eqs. (2) and (3) for action-risk scoring; a low distance to successful reference actions is assumed to indicate the proposed chunk will make progress.
  • domain assumption Empirical quantiles of support and risk scores on held-out successful expert demonstrations are representative of deployment score distributions at learner-induced states.
    Eq. (5) calibrates thresholds on D_cal, which contains only successful expert trajectories, while the intended trigger condition is for unsupported, often failed, states.
  • domain assumption Phase-local retrieval windows that advance monotonically along training trajectories correctly track the current task phase.
    Eq. (4) updates window positions nondecreasingly; this assumes the policy never needs to return to an earlier phase after operator recovery.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AutoIntervene: Calibrated Intervention for Action-Chunking Imitation Learning Policies." pith.science (2026). https://pith.science/paper/64DSWCKR

@misc{pith2026260807065,
  author       = {Pith},
  title        = {Pith review of: AutoIntervene: Calibrated Intervention for Action-Chunking Imitation Learning Policies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/64DSWCKR}},
  note         = {Machine review of arXiv:2608.07065}
}
read the original abstract

Action-chunking visuomotor policies learn from demonstrations and improve temporal consistency by predicting short action sequences rather than single-step commands. Yet perception errors and execution drift can move the robot outside the demonstration distribution, while the policy continues to produce smooth action chunks that are inconsistent with the observed state. We present AutoIntervene, an online framework that selectively transfers control between an action-chunking policy and an operator during deployment. AutoIntervene evaluates proposed chunks against a visual-action support memory built from successful task executions, combining visual similarity with consistency between proposed and reference actions. Phase-local support governs policy-to-operator transfer within the current task phase, whereas global support governs the return to policy control after operator recovery. We calibrate separate switching thresholds for the two directions from empirical quantiles of evaluation-level scores on held-out expert demonstrations, avoiding direct manual tuning of score cutoffs. Intervention segments retained from successful rollouts target learner-induced states and provide corrective supervision for subsequent policy updates. Experiments on real-world bimanual manipulation tasks show higher post-adaptation task success and lower operator-control time than manual intervention. Videos and additional results are available at https://aus.bot/research/autointervene/.

Figures

Figures reproduced from arXiv: 2608.07065 by the authors.

Figure 1
Figure 1. AutoIntervene transfers control between the au [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of AutoIntervene. During Stage I, the inter￾vention loop sequentially (A) constructs a visual-action query from the current visual observations and policy-proposed action chunk, (B) evaluates visual-support and action-risk scores, and (C) selects control authority using calibrated acceptance criteria. The retained intervention segments are subsequently used to adapt the policy in Stage II, after which the u… view at source ↗
Figure 3
Figure 3. Memory construction and mode-specific retrieval. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Policies adapted with AutoIntervene handle deformable objects in the Towels-and-Cable Bagging task, including [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Across-task mean success and total recorded control [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Policies adapted with AutoIntervene perform precise Fig. 6: Policies adapted with AutoIntervene perform precise [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: One AutoIntervene rollout on Peg Disassembly containing two policy–operator–policy intervention cycles. Top: the four control switches. Bottom: the corresponding timeline, with policy control in grey and operator control in red. centage points, with gains across all se…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 23 canonical work pages

  1. [1]

    An algorithmic perspective on imitation learning,

    T. Osa, J. Pajarinen, G. Neumann, J. A. Bagnell, P. Abbeel, and J. Pe- ters, “An algorithmic perspective on imitation learning,”Foundations and Trends in Robotics, 2018

    work page 2018

  2. [2]

    Diffeomorphic transforms for generalised imitation learning,

    W. Zhi, T. Lai, L. Ott, and F. Ramos, “Diffeomorphic transforms for generalised imitation learning,” inProceedings of the 4th Annual Learning for Dynamics and Control Conference, ser. Proceedings of Machine Learning Research, vol. 168, 2022, pp. 508–519

    work page 2022

  3. [3]

    What matters in learning from offline human demonstrations for robot manipulation,

    A. Mandlekaret al., “What matters in learning from offline human demonstrations for robot manipulation,” inProceedings of the 5th Conference on Robot Learning, 2022

    work page 2022

  4. [4]

    GTI: Learning to generalize across long-horizon tasks from human demonstrations,

    A. Mandlekar, D. Xu, R. Martín-Martín, S. Savarese, and L. Fei-Fei, “GTI: Learning to generalize across long-horizon tasks from human demonstrations,” inRobotics: Science and Systems, 2020

    work page 2020

  5. [5]

    Learning fine-grained bimanual manipulation with low-cost hardware,

    T. Z. Zhao, V . Kumar, S. Levine, and C. Finn, “Learning fine-grained bimanual manipulation with low-cost hardware,” inProceedings of Robotics: Science and Systems, 2023

    2023

  6. [6]

    A reduction of imitation learning and structured prediction to no-regret online learning,

    S. Ross, G. J. Gordon, and J. A. Bagnell, “A reduction of imitation learning and structured prediction to no-regret online learning,” in Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, 2011

    work page 2011

  7. [7]

    DART: Noise injection for robust imitation learning,

    M. Laskey, J. Lee, R. Fox, A. Dragan, and K. Goldberg, “DART: Noise injection for robust imitation learning,” inProceedings of the 1st Annual Conference on Robot Learning, 2017

    work page 2017

  8. [8]

    Unpacking failure modes of generative policies: Runtime monitoring of consistency and progress,

    C. Agiaet al., “Unpacking failure modes of generative policies: Runtime monitoring of consistency and progress,” inProceedings of the 8th Conference on Robot Learning, 2025

    work page 2025

Show all 28 references
  1. [9]

    Rewind- IL: Online failure detection and state respawning for imitation learn- ing,

    G. Zheng, S. Seenivasan, M. Johnson-Roberson, and W. Zhi, “Rewind- IL: Online failure detection and state respawning for imitation learn- ing,” arXiv preprint arXiv:2604.16683, 2026

    2026 arXiv

  2. [10]

    HG-DAgger: Interactive imitation learning with human experts,

    M. Kelly, C. Sidrane, K. Driggs-Campbell, and M. J. Kochenderfer, “HG-DAgger: Interactive imitation learning with human experts,” in IEEE International Conference on Robotics and Automation, 2019

    2019

  3. [11]

    RaC: Robot learning for long-horizon tasks by scaling recovery and correction,

    Z. Huet al., “RaC: Robot learning for long-horizon tasks by scaling recovery and correction,” arXiv preprint arXiv:2509.07953, 2025

    2025 arXiv

  4. [12]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    C. Chiet al., “Diffusion policy: Visuomotor policy learning via action diffusion,” inProceedings of Robotics: Science and Systems, 2023

    2023

  5. [13]

    FlowPolicy: Enabling fast and robust 3D flow-based policy via consistency flow matching for robot manipulation,

    Q. Zhang, Z. Liu, H. Fan, G. Liu, B. Zeng, and S. Liu, “FlowPolicy: Enabling fast and robust 3D flow-based policy via consistency flow matching for robot manipulation,” inAAAI Conference on Artificial Intelligence, 2025

    2025

  6. [14]

    Safe policies post-training: Constraining streaming flow models for adapting learned robot trajectory distributions,

    J. Long, D. Liu, W. Cai, I. Manchester, and W. Zhi, “Safe policies post-training: Constraining streaming flow models for adapting learned robot trajectory distributions,”IEEE Robotics and Automation Letters, vol. 11, no. 9, pp. 10 656–10 663, 2026

    2026

  7. [15]

    LazyDAgger: Reducing context switching in in- teractive imitation learning,

    R. Hoqueet al., “LazyDAgger: Reducing context switching in in- teractive imitation learning,” inIEEE International Conference on Automation Science and Engineering, 2021

    2021

  8. [16]

    Efficient active imitation learning with random network distillation,

    E. Biré, A. Kobanda, L. Denoyer, and R. Portelas, “Efficient active imitation learning with random network distillation,” inInternational Conference on Learning Representations, 2025

    2025

  9. [17]

    Error-aware imitation learning from teleoperation data for mobile manipulation,

    J. Wonget al., “Error-aware imitation learning from teleoperation data for mobile manipulation,” inProceedings of the 5th Conference on Robot Learning, 2022

    2022

  10. [18]

    ConditionNET: Learning preconditions and effects for execution monitoring,

    D. Sliwowski and D. Lee, “ConditionNET: Learning preconditions and effects for execution monitoring,”IEEE Robotics and Automation Letters, 2025

    2025

  11. [19]

    Can we detect failures without failure data? Uncertainty- Aware runtime failure detection for imitation learning policies,

    C. Xuet al., “Can we detect failures without failure data? Uncertainty- Aware runtime failure detection for imitation learning policies,” in Proceedings of Robotics: Science and Systems, 2025

    2025

  12. [20]

    DOSE3: Diffusion-based out-of-distribution detection onSE(3)tra- jectories,

    H. Cheng, T. Zheng, T. Zhang, M. Johnson-Roberson, and W. Zhi, “DOSE3: Diffusion-based out-of-distribution detection onSE(3)tra- jectories,” arXiv preprint arXiv:2502.16725, 2025

    2025 arXiv

  13. [21]

    PATCH: Action- chunk-conditioned latent patch innovation monitoring for robot ma- nipulation,

    Y . Zhou, R. Qiu, Y . Chen, J. Cui, and W. Zhi, “PATCH: Action- chunk-conditioned latent patch innovation monitoring for robot ma- nipulation,” arXiv preprint arXiv:2606.16690, 2026

    2026

  14. [22]

    TriPilot-FF: Coordinated whole-body teleoperation with force feedback,

    Z. Li, Y . Zhou, R. Qiu, H. Wu, G. Ren, and W. Zhi, “TriPilot-FF: Coordinated whole-body teleoperation with force feedback,” arXiv preprint arXiv:2602.09888, 2026

    2026

  15. [23]

    Time-contrastive networks: Self-supervised learn- ing from video,

    P. Sermanetet al., “Time-contrastive networks: Self-supervised learn- ing from video,” inIEEE International Conference on Robotics and Automation, 2018

    2018

  16. [24]

    SCIL: Stage- conditioned imitation learning for multi-stage manipulation,

    Z. Ouyang, K. Wang, J. Liu, H. Lu, and W. Zhang, “SCIL: Stage- conditioned imitation learning for multi-stage manipulation,”IEEE Control Systems Letters, 2025

    2025

  17. [25]

    TRACE: Trajectory- routed causal memory for delayed-evidence visuomotor imitation,

    Z. Li, R. Qiu, Y . Chen, G. Ren, and W. Zhi, “TRACE: Trajectory- routed causal memory for delayed-evidence visuomotor imitation,” arXiv preprint arXiv:2606.14551, 2026

    2026 arXiv

  18. [26]

    Human-in-the-loop imitation learning using remote tele- operation,

    A. Mandlekar, D. Xu, R. Martín-Martín, Y . Zhu, L. Fei-Fei, and S. Savarese, “Human-in-the-loop imitation learning using remote tele- operation,”arXiv preprint arXiv:2012.06733, 2020

    2012 arXiv

  19. [27]

    Experience replay for continual learning,

    D. Rolnick, A. Ahuja, J. Schwarz, T. P. Lillicrap, and G. Wayne, “Experience replay for continual learning,” inAdvances in Neural Information Processing Systems, vol. 32, 2019

    2019

  20. [28]

    Siméoniet al., “DINOv3,”Transactions on Machine Learning Research, 2026

    O. Siméoniet al., “DINOv3,”Transactions on Machine Learning Research, 2026

    2026

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.