Pith. sign in

REVIEW 4 major objections 6 minor 29 references

SAFE-CHEM: Uncertainty-Aware Policy Switching for Robust Robotic Chemistry

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read SAFE-CHEM claims that ensemble action variance can be calibrated into a threshold that switches a learned robot policy to a rule-based backup before unsafe chemistry manipulations cause failures, improving success and reducing safety…

desk verdict The success-rate result is real, but the safety claim is misattributed: their own tables show the switching mechanism can increase safety violations, and the real-world section does not demonstrate closed-loop transfer. read the letter →

arxiv 2608.09303 v1 pith:YWVSGDR4 submitted 2026-08-10 cs.RO cs.LG

Laura Jones , Shazil Shahzad , Ayesha Sana , Gabriella Pizzuto This is my paper · ORCID
classification cs.ROcs.LG
keywords robotmanipulationimitationlearninguncertaintyquantificationensemblevariancesafetyswitchingchemistrylaboratoryautomationkerneldensityestimationsim-to-realtransfer
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 tries to show that a robot working in a chemistry lab can tell when its learned manipulation policy is out of its depth. It trains an ensemble of recurrent-network imitation policies, uses the spread of their action predictions as a running uncertainty score, and switches control to a deterministic rule-based backup controller whenever that score crosses a threshold calibrated only on successful demonstrations. Across three laboratory tasks—lifting glassware, pick and place, and vial insertion—the hybrid system improves success rates and reduces critical safety violations relative to single-policy baselines. The framework is policy-agnostic, needs no failure examples during training, and transfers to physical hardware without real-world fine-tuning.

What carries the argument

The load-bearing mechanism is the ensemble action variance $\sigma^2_{i,t} = E[(a_{i,t} - \mu_{i,t})^2]$ computed online across $M$ recurrent-network policies for each of the seven joint-space action components. Kernel density estimation with a Gaussian kernel (bandwidth by Scott's rule) models the distribution of variances observed on successful rollouts; the calibrated threshold $\theta_{i,q}$ is the $q$-quantile of that success-conditioned distribution, so $q\times100\%$ of successful timesteps have variance below it. A sliding-window detector marks a component when its variance stays above $\theta_{i,q}$ for at least $n_{\text{peaks}}$ steps within the last $w$ steps, and the switching logic hands control to the rule-based backup controller when more than two components are flagged. This converts a continuous uncertainty estimate into a discrete, explainable control intervention.

What would settle it

In the simulated pick-and-place task, log the ensemble variance across all rollouts that end in a dropped or tilted vial; if a sizeable fraction of those failures occur while fewer than three action components ever exceed the calibrated $\theta_{0.90}$ threshold, then the switching rule is not detecting the unsafe states the paper says it detects.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the spread of an ensemble of behavioural-cloning recurrent policies, measured as the variance of their predicted actions across all seven action components, is a signal that separates successful from failed chemistry manipulations. Using kernel density estimation on variance values from successful rollouts, the authors build a per-component quantile threshold; a sliding-window peak count flags a component as uncertain when its variance repeatedly exceeds the threshold, and the system switches to a deterministic finite-state backup controller when more than two components are flagged simultaneously. In simulation, the hybrid controller raises success rates for all three tasks relative to both a single policy and a non-intervened ensemble—for instance, insertion success climbs from 21.3% for one policy to as high as 67–68% with an ensemble plus a conservative threshold—while critical safety violations such as dropped or tilted glassware are reduced. A zero-shot deployment on a physical 7-DoF arm confirms that the ensemble and the switching logic transfer without real-world fine-tuning.

Load-bearing premise

The ensemble's disagreement over what action to take is a reliable online proxy for how likely the learned policy is to fail, so a threshold on that disagreement can catch dangerous moments before a spill or drop occurs.

Editorial extensions

If this is right

  • A learned robot chemist can carry an 'I am not sure' signal that is computed online from the ensemble spread, with no failure demonstrations required during training.
  • The switching gate operates on the variance of action predictions, so the same calibrated mechanism can be attached to any ensemblable imitation policy.
  • For high-precision tasks such as vial insertion, the conservative $\theta_{0.90}$ threshold gives the best success rates, while the relaxed $\theta_{0.99}$ threshold suits simpler tasks like lifting.
  • Safety guarantees ultimately rest on the hand-written rule-based backup controller, not on the learned policy, because the switch delegates control to that explicit state machine.

Reading between the lines

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

  • A static per-component threshold may be too blunt: the paper's own uncertainty profiles peak during grasping in some tasks and during final alignment in others, so phase-dependent thresholds could cut both false alarms and missed detections.
  • The reported misplaced-vial failure suggests that adding force-torque feedback—already named as future work—would let the same switching logic catch failures the variance signal alone misses.
  • The need for a hand-designed state machine per task could make the backup controller the main engineering bottleneck when scaling to many chemistry procedures.
  • An online re-calibrating threshold that refreshes the success-conditioned variance density as new rollouts accumulate would be a direct testable upgrade to the current offline calibration.
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

4 major / 6 minor

Summary. The paper proposes SAFE-CHEM, a hybrid control framework for imitation-learned robotic manipulation in chemistry laboratories. A 15-member ensemble of BC-RNN policies provides both an aggregated action and an online epistemic-uncertainty signal (the variance of ensemble action predictions). Using KDE on variance values from successful and failed rollouts, component-specific quantile thresholds are calibrated, and a sliding-window peak detector decides when to switch control from the learned policy to a deterministic finite-state-machine backup controller. The method is evaluated in simulation on three tasks (lift, pick and place, insertion) across ensemble sizes 1, 5, 10, 15 and three thresholds, with success rate and critical-safety-violation rate as metrics. A zero-shot sim-to-real demonstration on a Franka Production 3 arm is described. The main success-rate result is that hybrid switching improves task success over the non-intervened ensemble in nearly all configurations; the main safety result is that ensembles reduce safety violations relative to a single policy, while the additional safety benefit of switching is task-dependent and, for insertion, appears absent or negative.

Significance. If its claims are fully supported, SAFE-CHEM would be a useful contribution to safe deployment of imitation-learned policies in laboratory automation: it formalizes a concrete uncertainty signal, a calibration procedure, and an intervention mechanism, and it reports systematic comparisons over ensemble sizes and thresholds. The paper's strengths include the explicit formalization in Section III, the use of 300 evaluation episodes per condition (3x100) for the main tables, and the demonstration on two domains (simulation and hardware). However, the two most distinctive claims -- that the uncertainty-aware switching mechanism itself improves safety, and that the detector is well calibrated -- are not established by the current evidence, because the safety comparison lacks the ensemble-only control as the relevant baseline and the detector F1 is computed on what appears to be the same data used for calibration. The real-world claim is also qualitative rather than quantitative. These gaps are fixable, but they are load-bearing for the stated contributions.

major comments (4)
  1. [Section IV-D / Equations (3)-(6)] The headline safety claim is not isolated to the proposed switching mechanism. Table IV shows that, for the insertion task, hybrid switching generally increases critical safety violations relative to the ensemble-only baseline: for N=5, None is 25.3±1.2 while θ0.90 is 33.0±3.6; for N=10, None is 21.0±7.0 while θ0.90 is 24.7±2.3; for N=15 the values are equal within noise. The text itself states that 'the insertion task did not benefit from the hybrid approach in regards to the safety failure rate.' Therefore, the abstract's statement that the hybrid strategy 'reduces critical safety violations compared to traditional single-policy baselines' may be driven by ensembling rather than by the uncertainty-aware switching that is the paper's contribution. The authors should present ensemble-only (None) as the primary control for the switching mechanism, report the switching benefit for each task, and temper the safety claim accordingly.
  2. [Section IV-F] The detector performance reported in Table II appears to be in-sample. The thresholds θ_i,q are calibrated from un-intervened ensemble rollouts (Section III-C, Equations (3)-(6)), and Section IV-D says that the same type of rollout data was used to calibrate the mechanism and then to measure detection rate, false alarm rate, and F1. The text does not describe any held-out split between calibration and evaluation rollouts, nor does it report confidence intervals for the F1 values. As a result, the F1 scores in Table II (0.90, 0.91, 0.87 at n_peaks=8) are likely optimistic and do not support the claim that the chosen threshold and window parameters generalize. The authors should use a separate validation set for threshold and hyperparameter selection and then evaluate detection performance on a fresh test set.
  3. [Section III-B] The real-world demonstration does not provide quantitative evidence for the claim of 'practical viability through zero-shot sim-to-real transfer.' The section describes only the hardware setup: inferred joint targets from Isaac Lab are sent via UDP and ROS 2 to the Franka arm. No task success rates, safety incidents, number of trials, or comparison with simulation performance are reported. In addition, inference is performed inside Isaac Lab with simulated state observations, so it is unclear whether the physical robot observes its own actual state or simply tracks simulated targets. The authors should either report quantitative real-world outcomes and clarify the closed-loop sensing path, or reduce the strength of the claim to a hardware-in-the-loop feasibility demonstration.
  4. [Section IV-E] The central assumption that 'the ensemble's disagreement correlates with the expected prediction error' is not directly validated. Figure 4 shows that successful and failed rollouts have different variance distributions, but this does not establish a per-step correlation between ensemble variance and prediction error, nor does it show that the threshold detects the high-risk states that actually cause safety violations. Section IV-E acknowledges a failure case where a slightly misplaced vial does not exceed the calibrated threshold. The authors should report a quantitative correlation or calibration analysis between ensemble variance and actual action error / eventual outcome, and discuss the resulting detection limits in the safety claims.
minor comments (6)
  1. [Section II-A] There are typographical errors: 'discrete-time Markov decision processs' and 'where where' in the sentence preceding Equation (1). These should be corrected.
  2. [Section III-C] Related work contains 'training a a behaviour cloning'; this should read 'training a behaviour cloning policy.'
  3. [Section III-C] The text says a switch is triggered when 'more than two action components are flagged,' which could mean at least three components. Specify the exact integer threshold (e.g., ≥3) consistently in Section III-C and Section IV-D.
  4. [Table II] The KDE bandwidth is said to follow Scott's Rule, but no formula or numerical value is given, and it is unclear whether bandwidth is estimated per component or globally. Adding this detail would make the calibration reproducible.
  5. [Section IV-E] Table II reports F1 scores without standard deviations or the size of the evaluation set. Given that the main comparisons in Tables I, III, and IV use 3x100 episodes, reporting analogous uncertainty for detector metrics would improve comparability.
  6. [Section IV-E] The qualitative observation that 'the switching mechanism occasionally triggered at suboptimal moments, causing the backup controller to tip the object' is important; this failure mode should be quantified or at least listed explicitly in the safety results, since it directly affects the interpretation of Table IV.

Circularity Check

1 steps flagged · score 3.0 of 10

In-sample detector evaluation: thresholds and peak-detection hyperparameters are fit to the calibration rollouts and then scored on the same data; central success results are fresh-rollout measurements.

  1. fitted input called prediction [Section IV-D (Experiment II), Table II and Eqs. (3)-(7)]
    "To calibrate the mechanism, we initially collected trajectory data from ensemble policy rollouts without intervention, classifying each as completed (successful) or failed. As established in Section III-C, we applied KDE to the aggregated variance data. ... Detection performance was measured via detection rate (true positives), false alarm rate (false positives), and F1-Score. As shown in Table II, an n peaks value of 8 consistently provided the highest F1-score across lift, pick and place, and insertion tasks when using the conservative θ0.90 limit."

    The detection thresholds θ_i,q (Eq. 6) are fitted as quantiles of the success-conditioned variance aggregated from these same 'ensemble policy rollouts without intervention' (Eqs. 3-4), and the sliding-window parameters w and n_peaks (Eq. 7) are tuned to maximize F1 on the same data. Therefore Table II's F1 is an in-sample goodness-of-fit of the calibrated detector, not an independent estimate of detection performance; part of its reported accuracy is forced by the fitting procedure. The final success/safety results (Tables III-IV) are measured on fresh 3×100 rollouts per condition, so the central hybrid-claim is not itself constructed from the fit.

full rationale

The main derivation is not circular: Eq. (1) is standard behaviour cloning; the ensemble uncertainty hypothesis in Section III-B is explicitly stated as a hypothesis ('We hypothesise that the ensemble's disagreement correlates with the expected prediction error'), and the switch threshold in Eq. (6) is a calibration statistic rather than a prediction. Tables III-IV report success and safety rates on fresh 3×100 rollouts per condition, so the central claim that the hybrid strategy improves success and reduces violations is an empirical comparison, not a reconstruction of its inputs. The self-citations (Refs. [9], [25], [26]) are background examples of laboratory automation skills and are not load-bearing. The only circular element is the detector evaluation in Section IV-D/Table II: the same 'initially collected trajectory data' are used to fit θ_i,q and to tune w and n_peaks, and then F1 is reported on that same data without a described held-out split. That makes the F1 partly in-sample by construction and inflates the apparent detector quality, but it does not force the main success/safety findings. The separate concern that the safety benefit is not isolated from ensembling (the None columns) is an experimental attribution issue, not a derivation-circularity issue, so it does not raise the circularity score further.

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

The central claim depends on two domain assumptions that are stated rather than proven: ensemble variance tracks task risk, and the backup controller is a safe fallback for any switched state. The paper itself reports violations of both. Threshold calibration uses the system's own successful rollouts, which is standard but means the reported detection numbers are partly self-referential. No new physical entities are introduced.

free parameters (4)
  • Component-specific confidence threshold theta_i,q = per-task quantile of success-conditioned variance (q=0.90, 0.95, 0.99)
    Computed from Equation 6 using KDE over the ensemble's own successful rollouts for each task; it controls when the backup controller is engaged.
  • Sliding-window peak count n_peaks = 8 (selected by grid search over 1 to 10)
    Grid searched per task in Table II to maximize detection F1; it sets how many variance exceedances are needed before a component is flagged.
  • Sliding window size w = 8 to 10 (Lift and Place use w=10, Insert uses w=9)
    Window length for detecting sustained high variance, tuned alongside n_peaks in the same grid search.
  • Component switch count threshold = more than 2 of 7 action components flagged
    Hand-chosen rule in Section III-C to avoid false triggers on inherently uncertain actions such as grasping; it is not derived from data.
assumptions (4)
  • domain assumption Ensemble action variance correlates monotonically with expected prediction error and unsafe states
    Stated in Section III-B as a hypothesis. If false, the thresholded switch cannot separate safe from unsafe states. The paper's own unrecoverable-state examples in Section IV-E show where it fails.
  • domain assumption The rule-based backup controller can continue the task safely from any state where switching fires
    The switching logic assumes pi_B is a valid fail-safe. Section IV-E reports that the backup controller can tip objects and cannot reset mis-seated vials, so this assumption is only partially true.
  • domain assumption Calibration rollouts are representative of deployment rollouts
    Thresholds are estimated from a collection of ensemble rollouts, and the paper does not describe a train/test split for calibration, so distribution shift between calibration and deployment could change threshold validity.
  • domain assumption The 10 human demonstrations expanded by MimicGen capture expert behavior for each task
    Used in Section IV-A to build D; if the expanded demonstrations are not representative, the ensemble and its variance signal are biased.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAFE-CHEM: Uncertainty-Aware Policy Switching for Robust Robotic Chemistry." pith.science (2026). https://pith.science/paper/YWVSGDR4

@misc{pith2026260809303,
  author       = {Pith},
  title        = {Pith review of: SAFE-CHEM: Uncertainty-Aware Policy Switching for Robust Robotic Chemistry},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YWVSGDR4}},
  note         = {Machine review of arXiv:2608.09303}
}
read the original abstract

The deployment of autonomous robotic systems in chemistry laboratories is accelerating experimental workflows and providing the foundational data for AI-driven scientific discovery. However, despite the success of data-driven methods in acquiring dexterous skills, safety remains a primary barrier to their deployment in high-risk domains, such as early-stage materials chemistry experiments. Specifically, learning-based policies frequently struggle to distinguish between safe and unsafe actions, leading to overconfident extrapolation and potentially catastrophic failures. To mitigate these safety risks, we propose SAFE-CHEM, an uncertainty-aware framework designed for robust, learning-based robotic chemists. Our approach leverages an ensemble of recurrent neural network-based imitation learning policies to quantify epistemic uncertainty online through the variance of action predictions. By characterising the success-conditioned density of this variance using kernel density estimation, we introduce a hybrid control architecture that autonomously switches from the learned policy to a deterministic, rule-based backup controller when uncertainty exceeds a calibrated safety threshold. We evaluate SAFE-CHEM across three fundamental laboratory manipulation tasks, where our empirical results demonstrate that this hybrid strategy improves overall task success rates and reduces critical safety violations compared to traditional single-policy baselines. Finally, we demonstrate the practical viability of the framework through zero-shot sim-to-real transfer onto a physical Franka Production 3 robot manipulator.

Figures

Figures reproduced from arXiv: 2608.09303 by the authors.

Figure 1
Figure 1. An overview of our method SAFE-CHEM, which consists of four stages: (1) training an ensemble-based imi￾tation learning policy (πL) in simulation for robotic chemistry manipulation tasks; (2) estimating the epistemic uncertainty σ through ensemble variance; (3) action selection that monitors σ to gate control between πL and a fail-safe rule-based backup robot controller (πB); and (4) real-world demonstration on a rob… view at source ↗
Figure 2
Figure 2. Overall system block diagram. Our framework demonstrates how a learned ensemble of RNN-based policies provide [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The robotic chemistry manipulation simulated tasks. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Confidence thresholds calculated at the θ0.90, θ0.95 and θ0.99 bounds. Due to certain actions such as grasping being inherently uncertain, we trigger a switch only when the condition in Equation 7 is met by more than two action components. We utilised a sliding window …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 24 canonical work pages

  1. [1]

    Accelerating discovery in natural science laboratories with AI and robotics: Perspectives and chal- lenges,

    A. I. Cooper, P. Courtney, K. Darvishet al., “Accelerating discovery in natural science laboratories with AI and robotics: Perspectives and chal- lenges,”Science Robotics, vol. 10, no. 106, p. eadv7932, 2025. [Online]. Available: https://www.science.org/doi/abs/10.1126/scirobotics.adv7932

    doi:10.1126/scirobotics.adv7932 2025

  2. [2]

    Self-driving laboratories for chemistry and materials science,

    G. Tom, S. P. Schmid, S. G. Bairdet al., “Self-driving laboratories for chemistry and materials science,”Chemical Reviews, 2024

    work page 2024

  3. [3]

    Self-driving laboratories in japan,

    N. Yoshikawa, Y . Asano, D. N. Futabaet al., “Self-driving laboratories in japan,”Digital Discovery, vol. 4, pp. 1384–1403, 2025. [Online]. Available: http://dx.doi.org/10.1039/D4DD00387J

    doi:10.1039/d4dd00387j 2025

  4. [4]

    Autonomous laboratories in china: an embodied intelligence-driven platform to accelerate chemical discovery,

    J. Li, C. Ding, D. Liuet al., “Autonomous laboratories in china: an embodied intelligence-driven platform to accelerate chemical discovery,”Digital Discovery, vol. 4, pp. 1672–1684, 2025. [Online]. Available: http://dx.doi.org/10.1039/D5DD00072F

    work page doi:10.1039/d5dd00072f 2025

  5. [5]

    A mobile robotic chemist,

    B. Burger, P. M. Maffettone, V . V . Gusevet al., “A mobile robotic chemist,”Nature, vol. 583, no. 7815, pp. 237–241, Jul 2020. [Online]. Available: https://doi.org/10.1038/s41586-020-2442-2

    doi:10.1038/s41586-020-2442-2 2020

  6. [6]

    Toward autonomous laborato- ries: Convergence of artificial intelligence and experimental automation,

    Y . Xie, K. Sattari, C. Zhang, and J. Lin, “Toward autonomous laborato- ries: Convergence of artificial intelligence and experimental automation,” Progress in Materials Science, vol. 132, no. 0079-6425, p. 101043, 2023

    work page 2023

  7. [7]

    FLIP: Flowability-Informed Powder Weighing,

    N. Radulov, A. Wright, T. Littleet al., “FLIP: Flowability-Informed Powder Weighing,” in2026 IEEE International Conference on Robotics and Automation (ICRA), 2026

    work page 2026

  8. [8]

    Matterix: toward a digital twin for robotics-assisted chemistry laboratory automation,

    K. Darvish, A. Sohal, A. Mandalet al., “Matterix: toward a digital twin for robotics-assisted chemistry laboratory automation,”Nature Computational Science, vol. 6, pp. 67 – 82, 2025. [Online]. Available: https://api.semanticscholar.org/CorpusID:284363658

    work page 2025

Show all 29 references
  1. [9]

    Accelerating Laboratory Automation Through Robot Skill Learning For Sample Scraping*,

    G. Pizzuto, H. Wang, H. Fakhruldeenet al., “Accelerating Laboratory Automation Through Robot Skill Learning For Sample Scraping*,” in 2024 IEEE 20th International Conference on Automation Science and Engineering (CASE), 2024

    2024

  2. [10]

    Steering towards safe self-driving laboratories,

    S. X. Leong, C. Griesbach, R. Zhang,et al., “Steering towards safe self-driving laboratories,” 12 2024

    2024

  3. [11]

    Prevent: Proactive risk evaluation and vigilant execution of tasks for mobile robotic chemists using multi-modal behavior trees,

    S. Veeramani, Z. Zhou, F. Munguia-Galeano, H. Fakhruldeen, T. Roddelkopf, M. F. R. Al-Okby, K. Thurow, and A. I. Cooper, “Prevent: Proactive risk evaluation and vigilant execution of tasks for mobile robotic chemists using multi-modal behavior trees,” 2025. [Online]. Available...

    2025

  4. [12]

    RobInHood: A Robotic Chemist in a Fume Hood,

    Louis Longley and Francisco Munguia-Galeano and Yushu Han and others , “RobInHood: A Robotic Chemist in a Fume Hood,”ChemRxiv, vol. 2026, no. 0112, 2026. [Online]. Available: https://chemrxiv.org/doi/abs/10.26434/chemrxiv-2026-s2619

    2026 doi

  5. [13]

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

    C. Xu, T. K. Nguyen, E. Dixonet al., “Can we detect failures without failure data? uncertainty-aware runtime failure detection for imitation learning policies,” inProceedings of Robotics: Science and Systems (RSS), 2025

    2025

  6. [14]

    Learning robotic powder weighing from simulation for laboratory automation,

    Y . Kadokawa, M. Hamaya, and K. Tanaka, “Learning robotic powder weighing from simulation for laboratory automation,” inThe IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2023

    2023

  7. [15]

    Revolutionizing clean energy labs: Robotic imitation learning for efficient fabrication ai-powered electrical units assembly platform,

    X. Xu, Y . Gu, T. Zhang, J. Yu, and S. Skinner, “Revolutionizing clean energy labs: Robotic imitation learning for efficient fabrication ai-powered electrical units assembly platform,”Energy and AI, vol. 21, p. 100517, 2025. [Online]. Available: https://www.sciencedirect.com/s...

    2025

  8. [16]

    Compact task-aligned imitation learning for laboratory automation,

    K. Suzuki, H. Nakamurama, K. Miyamoto, and T. Ogata, “Compact task-aligned imitation learning for laboratory automation,” 2026. [Online]. Available: https://arxiv.org/abs/2603.01110

    2026 arXiv

  9. [17]

    Failure predic- tion at runtime for generative robot policies,

    R. R ¨omer, A. Kobras, L. Worbis, and A. P. Schoellig, “Failure predic- tion at runtime for generative robot policies,” inAdvances in Neural Information Processing Systems (NeurIPS), 2025

    2025

  10. [18]

    Uncertainty-aware deep imitation learning and deployment for autonomous navigation through crowded intersections,

    Z. Zhu, S. Wang, and H. Zhao, “Uncertainty-aware deep imitation learning and deployment for autonomous navigation through crowded intersections,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 13 980–13 987

    2024

  11. [19]

    Alvinn: An autonomous land vehicle in a neural net- work,

    D. Pomerleau, “Alvinn: An autonomous land vehicle in a neural net- work,” inProceedings of (NeurIPS) Neural Information Processing Systems, D. Touretzky, Ed. Morgan Kaufmann, December 1989, pp. 305 – 313

    1989

  12. [20]

    B. W. Silverman,Density estimation for statistics and data analysis. Routledge, 2018

    2018

  13. [21]

    Isaac lab: A gpu- accelerated simulation framework for multi-modal robot learning,

    NVIDIA, M. Mittal, P. Roth, J. Tigueet al., “Isaac lab: A gpu- accelerated simulation framework for multi-modal robot learning,”

  14. [22]

    The franka emika robot: A reference platform for robotics research and education,

    S. Haddadin, S. Parusel, L. Johannsmeieret al., “The franka emika robot: A reference platform for robotics research and education,”IEEE Robotics & Automation Magazine, vol. 29, no. 2, pp. 46–64, 2022

    2022

  15. [23]

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

    A. Mandlekar, D. Xu, J. Wonget al., “What matters in learning from offline human demonstrations for robot manipulation,” inConference on Robot Learning (CoRL), 2021

    2021

  16. [24]

    Mimicgen: A data generation system for scalable robot learning using human demonstrations,

    A. Mandlekar, S. Nasiriany, B. Wenet al., “Mimicgen: A data generation system for scalable robot learning using human demonstrations,” in7th Annual Conference on Robot Learning, 2023

    2023

  17. [25]

    Go with the flow: deep learning methods for autonomous viscosity estimations,

    Walker, Michael and Pizzuto, Gabriella and Fakhruldeen, Hatem and others, “Go with the flow: deep learning methods for autonomous viscosity estimations,”Digital Discovery, vol. 2, 2023

    2023

  18. [26]

    Modular, multi-robot integration of laboratories: an autonomous workflow for solid-state chemistry,

    A. M. Lunt, H. Fakhruldeen, G. Pizzutoet al., “Modular, multi-robot integration of laboratories: an autonomous workflow for solid-state chemistry,”Chem. Sci., vol. 15, 2024

    2024

  19. [27]

    Warp: Differentiable spatial computing for python,

    M. Macklin, “Warp: Differentiable spatial computing for python,” in ACM SIGGRAPH 2024 Courses, ser. SIGGRAPH ’24 Courses. New York, NY , USA: Association for Computing Machinery, 2024. [Online]. Available: https://doi.org/10.1145/3664475.3664543

    2024

  20. [28]

    Moveit! task construc- tor for task-level motion planning,

    M. G ¨orner, R. Haschke, H. Ritter, and J. Zhang, “Moveit! task construc- tor for task-level motion planning,” 05 2019, pp. 190–196

    2019

  21. [2025]

    Available: https://arxiv.org/abs/2511.04831

    [Online]. Available: https://arxiv.org/abs/2511.04831

    arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.