Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Evaluating Uncertainty and Quality of Visual Language Action-enabled Robots

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Binary success rates hide robot task quality; motion metrics expose it.

desk verdict Useful empirical study of VLA execution-quality metrics against human labels, but the time-averaging aggregation and multiple testing need work before the metric rankings are trusted. read the letter →

arxiv 2507.17049 v3 pith:JWJNCFNB submitted 2025-07-22 cs.SE cs.RO

Pablo Valle , Chengjie Lu , Shaukat Ali , Aitor Arrieta This is my paper
classification cs.SEcs.RO
keywords VisualLanguageActionmodelsuncertaintyquantificationqualitymetricsroboticmanipulationhumanevaluationtasksuccessrateruntimemonitoringmotioninstability
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

Binary success rates, the standard way to evaluate vision-language-action (VLA) robots, hide large differences in how well a task is performed: a run that succeeds after repeated drops, collisions, and stalls is counted the same as one that is smooth and confident. This paper proposes eight uncertainty metrics and five quality metrics that can be computed from the model's predicted actions and the robot's observed trajectory, then tests them on 908 successful executions by three VLA models across four manipulation tasks. Human experts labeled each execution high, medium, or low quality, and several metrics—notably Action Velocity Instability, Action Acceleration Instability, and Trajectory Velocity Instability—showed moderate to strong correlation with those labels. One metric, Optimal Trajectory Difference, consistently separated successful from failed executions even in the absence of a symbolic oracle. If these correlations hold, the metrics provide an automatic proxy for human-judged execution quality and a runtime signal for monitoring and adapting VLA robots.

What carries the argument

The load-bearing machinery is a family of per-time-step metrics aggregated by averaging over each run. Three action-instability metrics (A-PI, A-VI, A-AI) take first-, second-, and third-order differences of the predicted action vector; four token-based metrics (TB-TP, TB-PCS, TB-D, TB-E) convert the language backbone's token probability distribution into confidence scores; and EV measures the standard deviation of repeated inferences. On the quality side, TCP-PI, TCP-VI, TCP-AI, and TI use first-, second-, and third-order differences (and RMS jerk) of the end-effector trajectory, while OT normalizes the derivative of the distance-to-goal into $[0,1]$. Differences are computed over a sliding window of eight time steps. The mechanism is that instability in actions or velocity and divergence from progress toward the target proxy low quality and failure; correlation against expert labels tests whether the proxy holds.

What would settle it

Re-run the study on a held-out set of executions from a fourth VLA model or a third robot embodiment, with fresh experts who have not seen the original labels, and check whether A-VI, A-AI, and TCP-VI still show Spearman $\rho > 0.49$ and whether OT still separates success from failure with large effect sizes; if the correlations drop to near zero or reverse sign, the metrics are not general proxies for human-judged quality.

Watch

Extended reading notes

Core claim

The paper's central claim is that measurable properties of a VLA robot's behavior—how much its action commands oscillate, how smooth the end-effector's velocity is, how faithfully the trajectory converges on the goal—carry real information about execution quality that the binary success oracle discards. On 908 successful executions labeled by domain experts, the paper reports Spearman correlations up to the strong range ($\rho > 0.49$) for the best uncertainty and quality metrics, with A-VI, A-AI, and TCP-VI the most consistently significant across the 12 task-model combinations. It also reports that OT has statistically significant, often large Vargha-Delaney effect sizes separating successful from failing runs for all three models and all four tasks. In the paper's framing, these findings show that success-rate-only evaluation is inadequate and that quality-aware test oracles and runtime monitors can be built from inexpensive, behavior-derived metrics.

Load-bearing premise

The whole correlation analysis assumes that the human quality labels assigned by two, and occasionally three, experts are a stable, transferable ground truth, and that averaging each metric over the time steps of a run is the right way to summarize it; if either assumption fails, the reported correlations only show how the metrics align with this particular labeling procedure.

Editorial extensions

If this is right

  • Benchmarking VLA models by success count alone can rank models misleadingly; the paper finds one model's successful 'Pick up' runs were mostly low-quality while another's were mostly high-quality despite similar success counts.
  • Quality metrics such as TCP-VI can act as a design-time oracle, grading executions when no symbolic final-state oracle is available or when the oracle mislabels false successes.
  • Uncertainty metrics such as A-VI and A-AI can be computed at negligible overhead and used for runtime monitoring of execution quality.
  • EV is not recommended: its computational cost is seconds per inference and it rarely correlates with expert labels.
  • None of the metrics capture final object orientation, so tasks where orientation is critical require either extended metrics or separate checks.

Reading between the lines

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

  • Because the correlation analysis averages each metric over a run, it may obscure phase-specific failures such as the early blocking states the paper observes; peak or late-run aggregations could be tested as stronger signals.
  • The same metric family could be used to build a human-in-the-loop triage system where cheap metrics pre-screen runs and humans only review ambiguous cases.
  • OT's reference distances require knowledge of object and goal poses; in a real deployment that knowledge comes from perception, so OT's oracle-free promise depends on the accuracy of the perception pipeline.
  • The metric definitions are not tied to any particular VLA architecture, so they are testable on new models, embodiments, and tasks, but the paper's evidence covers only two robots and four tasks, so transfer remains an open empirical question.
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

4 major / 4 minor

Summary. The paper proposes eight uncertainty metrics and five quality metrics for VLA-enabled robotic manipulation, computes them from per-timestep action, trajectory, and token data across 908 successful executions of three VLA models (OpenVLA, pi0, SpatialVLA) on four tasks in the SimplerEnv benchmark, and aggregates each run to a single scalar by averaging. It compares these run-level values with human quality labels (high/medium/low) assigned by two domain experts with 85% Cohen's kappa and a third tie-breaker. The study reports that success rate alone masks large quality differences (RQ1), that several metrics show moderate to strong Spearman correlation with human labels (RQ2), that some metrics discriminate successful from failing executions (RQ3), and that most metrics have low overhead except EV (RQ4). The central claim is that the proposed metrics are useful proxies for human-judged execution quality and model confidence, with A-VI, A-AI, and TCP-VI recommended as the most consistent indicators and OT as the strongest success/failure discriminator.

Significance. If the results hold, the paper would give practitioners a concrete, practical alternative to binary success-rate evaluation for VLA robots, including metrics for runtime monitoring and an oracle-free discrimination tool. The study has several genuine strengths: a large manually labeled dataset of 908 executions, three state-of-the-art models, four tasks, two embodiments, a replication package, and honest reporting of negative results such as the failure of EV and the orientation limitation in the Move Near task. The correlation analysis is not circular in the fitting sense, since the metrics are defined independently of the human labels and no parameters are fitted to those labels. The main risks are that the run-level time-averaging aggregation is arbitrary and may drive the reported correlations, that multiple-comparison corrections are absent, and that some headline consistency claims rest on very small per-cell samples.

major comments (4)
  1. [V-G, Table I] Section V-G states that each metric is aggregated over a run by averaging per-timestep values, but Table I defines quality through localized events such as 'one drop', collision 'maximum once', 'multiple drops', and 'pauses where the robot stops to compute or replan'. A mean over a full trajectory dilutes exactly the spikes that determine these labels, and Finding 9 confirms the failure mode: static unsuccessful runs have near-zero instability and look high-quality under the average. Because RQ2 and RQ3 both use this run-level scalar, the correlations in Table III and the A12 effects in Tables V-VII may be artifacts of the aggregation rule rather than properties of the metrics. Please report a sensitivity analysis over alternative aggregations (for example, max, 95th percentile, count or duration of threshold crossings, and area under the curve) and show whether the ranking of metrics and the main conclusions survive; also address the unequal averaging windows between early-blocking failures and complete successful runs.
  2. [V-G, Tables III-VII] The paper reports significance tests for a large family of comparisons (13 metrics x 4 tasks x 3 models in Table III, and similar families in Tables V-VII) without any correction for multiple comparisons. Under this many tests, several moderate Spearman values will be significant by chance, so the claim that A-VI and A-AI are reliable in '8 out of 12' task-model combinations is not yet supported. Please apply a family-wise or false-discovery-rate correction such as the Benjamini-Hochberg procedure, and either restrict the consistency claims to adjusted results or explicitly frame the analysis as exploratory.
  3. [Table II, Section VI-B] OpenVLA has only 9 successful Put-in and 6 successful Put-on executions, and the per-quality-level cells are as small as n=2 for OpenVLA in Put on. The text acknowledges that these samples are too small for statistically sound claims, yet the corresponding correlations are still displayed in Table III and counted in cross-task consistency statements such as Finding 4 and the '8 out of 12' summary. Please exclude or clearly suppress these cells in the summary statistics and recompute the 'most consistent metric' conclusions without them.
  4. [IV-E, Eqs. (26)-(29); Finding 6] Equation (29) defines q_t = 0.5(1 + Δd_t) and states that it is normalized to [0,1], but Δd_t is unbounded, so the normalization statement is mathematically incorrect. More substantively, OT is computed as the distance to the goal object or target pose, which is the same information used by the success oracle in RQ3; Finding 6, which identifies OT as the strongest and most consistent success/failure discriminator, may therefore be partly true by construction rather than a discovery about execution quality. Please test whether OT's discrimination survives when the final fraction of the trajectory is excluded, or compare it against a baseline that uses only the final distance; if the discrimination disappears, the claim should be weakened accordingly.
minor comments (4)
  1. [V-G] Please add the exact software packages and versions used for the Spearman, Mann-Whitney, Shapiro-Wilk, and Vargha-Delaney computations, since these are part of the reproducibility claims.
  2. [Tables III-VII] The tables indicate statistical significance only by color, which makes exact p-values impossible to recover; please provide a supplementary table with the exact p-values and sample sizes for every cell.
  3. [Figures 7-9] The y-axis is independently scaled for each task and metric, which makes cross-panel comparisons of effect size visually misleading; please state this explicitly in the caption or use a common scale for at least one representative metric per family.
  4. [Throughout] There are several typographical and notation errors that should be fixed in a revision, including 'Wihtney' (Section V-G), 'theshold' (Finding 9), 'corresonding' (Section VI-D), 'T PC-PI' (Section VI-C), 'Pirsmatic-7b' (Appendix A-A), and 'Birdge' (Appendix A-A).

Circularity Check

1 steps flagged · score 4.0 of 10

OT success-discrimination finding is built into the metric's definition, but the core quality-correlation claims rest on independently computed metrics and expert labels.

  1. self definitional [Section IV-E (Eqs. 26-29), Section V-C, Section VI-C (Finding 6)]
    "The Optimal Trajectory Difference (OT) evaluates the quality of robotic task execution by measuring the spatial proximity between the robot's end-effector and task-relevant reference positions... dt = ||ptcp(t) - pobj(t)|| ... Finding 6. OT serves as a consistent, model-agnostic indicator of successful and unsuccessful tasks."

    OT is defined as the distance from the end-effector to the goal object/destination pose (Eqs. 26-27), while task success in VLATest is defined by exactly those goal conditions (object lifted 0.02m, placed within 0.05m, placed inside/on target; Section V-C). RQ3 then reports OT as the strongest discriminator between successful and failing executions. Because OT monotonically tracks distance-to-goal, it is a graded version of the symbolic oracle that checks final goal states; its discrimination is a consequence of the metric's definition, not an independent empirical finding. The metric also requires grasp-status and goal-pose knowledge, the same information the oracle uses, so it cannot serve as an oracle-free test.

full rationale

The paper's metrics are computed from action and trajectory data independently of the human quality labels, and no parameters are fitted to those labels, so the RQ2 correlation results (A-VI, A-AI, TCP-VI moderate-to-strong correlations) are not circular. The self-citations present (replication package, prior uncertainty studies) are not load-bearing. The one genuine circular step is the RQ3 emphasis on OT as the best success/failure discriminator: OT is defined using the same goal-object and destination poses that the symbolic success oracle uses, making its discriminating power definitional rather than discovered. This affects a highlighted conclusion but not the paper's central human-quality proxy claim, which retains independent content. The time-averaging aggregation is a validity threat (it can dilute event-based defects such as drops and collisions, as the paper's own Finding 9 shows), but it is not a circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claims rest on two tuned parameters (window size, EV sample count), the assumption that human labels are ground truth, and the assumption that smoothness and token probability reflect quality and confidence. No new physical entities are postulated.

free parameters (2)
  • Temporal window size = 8 time steps
    Selected in Section V-D based on empirical observation that smaller windows (5-6 steps) gave less stable estimates for A-AI and TCP-AI; affects all difference-based metrics.
  • Number of EV inference samples = 4
    Section V-D: highest number of additional model instances loadable on the available hardware; EV metric depends on this.
assumptions (4)
  • domain assumption Human expert labels are a valid ground truth for execution quality
    All RQ2 correlations compare metrics to these labels (Section V-F, Table I); if labels are unreliable, the central claims weaken.
  • domain assumption Smoothness and jerk correlate with perceived quality
    Justified by prior work (Flash and Hogan minimum-jerk; Hogan and Sternad) cited in Section IV-D; used to define TI and interpret smoothness metrics.
  • standard math Token probability distributions reflect model confidence
    Standard probabilistic interpretation used in Sections III-D to III-G for TB-TP, TB-PCS, TB-D, TB-E.
  • domain assumption SimplerEnv simulation is representative of real VLA deployments
    External validity relies on this; acknowledged as a threat in Section VII.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluating Uncertainty and Quality of Visual Language Action-enabled Robots." pith.science (2026). https://pith.science/paper/JWJNCFNB

@misc{pith2026250717049,
  author       = {Pith},
  title        = {Pith review of: Evaluating Uncertainty and Quality of Visual Language Action-enabled Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JWJNCFNB}},
  note         = {Machine review of arXiv:2507.17049}
}
read the original abstract

Vision-Language-Action (VLA)-enabled robots integrate visual perception, natural language understanding, and action planning to interpret their environment, comprehend instructions, and perform embodied tasks autonomously. Such robots are typically evaluated through task success rates, i.e., whether a robot performs its intended task, which are commonly used as test oracles for evaluating such robots. Such an evaluation fails to capture the quality of task execution and the robot's confidence in its decisions. In this paper, we adapt eight uncertainty metrics and five quality metrics specifically designed for VLA-enabled robotic manipulation tasks. We assess their effectiveness through a large-scale empirical study involving 908 successful task executions from three state-of-the-art VLA models across four representative robotic manipulation tasks and two robot embodiments. Human domain experts manually labeled task quality, enabling us to analyze the correlation between our proposed metrics and expert judgments, serving as a human oracle for testing such robots. The results reveal that several metrics show moderate to strong correlation with human assessments, highlighting their utility for evaluating task quality and model confidence. Furthermore, we found that some metrics can discriminate between high-, medium-, and low-quality executions from unsuccessful tasks, which is useful when test oracles are absent. Our findings challenge the adequacy of current evaluation practices that rely solely on binary success rates and pave the way for improved real-time monitoring and adaptive enhancement of VLA-enabled robots.

Figures

Figures reproduced from arXiv: 2507.17049 by the authors.

Figure 1
Figure 1. Overview of the VLA Model Architecture and obser [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Screenshot of the web-based application for tagging [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Comparison between correct and incorrect orientations [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: Architecture of π0 SpatialVLA [7] is a 3B-parameter Vision-Language-Action (VLA) model pretrained on 1.1 million real-robot demonstra￾tions from the Open X-Embodiment [24] and RH20T [88] datasets. The architecture of SpatialVLA consists of three main components (see […
Figure 4
Figure 4. Figure 4: Architecture of OpenVLA π0 [5] is a 3.3B parameter VLA model trained with condi￾tional flow matching [81], [82], [21] on a diverse mixture of datasets including the Open X-Embodiment dataset [24], the Birdge dataset [36] and DROID dataset [83] high-frequency dexterous …
Figure 6
Figure 6. Figure 6: Architecture of SpatialVLA Llama2 language model (see [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]
Figure 7
Figure 7. Figure 7: Distribution of average values across all metrics for each task type. The columns ”High”, ”Medium”, and ”Low” [PITH_FULL_IMAGE:figures/full_fig_p020_7.png]
Figure 8
Figure 8. Figure 8: Distribution of average values for the π0 model [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Distribution of average values for the SpatialVLA model [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. UAOR: Uncertainty-aware Observation Reinjection for Vision-Language-Action Models

    cs.CV 2026-02 conditional novelty 6.0 of 10

    Uncertainty-aware observation reinjection into FFN layers improves VLA manipulation success rates across LIBERO, SIMPLER, CALVIN, and real-robot tasks with no training.

  2. LIBERO-PRO: Towards Robust and Fair Evaluation of Vision-Language-Action Models Beyond Memorization

    cs.CV 2025-10 conditional novelty 6.0 of 10

    SOTA VLA models like OpenVLA and pi0 collapse when object positions are perturbed, indicating that standard LIBERO scores reward memorization; LIBERO-PRO provides a systematic perturbed evaluation suite.

Reference graph

Works this paper leans on

91 extracted references · 43 canonical work pages · cited by 2 Pith papers

  1. [1]

    Human–robot interaction review and challenges on task planning and programming,

    P. Tsarouchi, S. Makris, and G. C. and, “Human–robot interaction review and challenges on task planning and programming,”International Journal of Computer Integrated Manufacturing, vol. 29, no. 8, pp. 916– 931, 2016

    2016

  2. [2]

    Artificial intelligence, machine learning and deep learning in advanced robotics, a review,

    M. Soori, B. Arezoo, and R. Dastres, “Artificial intelligence, machine learning and deep learning in advanced robotics, a review,” Cognitive Robotics, vol. 3, pp. 54–70, 2023

    2023

  3. [3]

    Large language models for robotics: Opportunities, challenges, and perspectives,

    J. Wang, E. Shi, H. Hu, C. Ma, Y . Liu, X. Wang, Y . Yao, X. Liu, B. Ge, and S. Zhang, “Large language models for robotics: Opportunities, challenges, and perspectives,” Journal of Automation and Intelligence , 2024

    2024

  4. [4]

    Gr00t n1: An open foundation model for generalist humanoid robots,

    NVIDIA, :, J. Bjorck, F. Casta ˜neda, N. Cherniadev, X. Da, R. Ding, L. J. Fan, Y . Fang, D. Fox, F. Hu, S. Huang, J. Jang, Z. Jiang, J. Kautz, K. Kundalia, L. Lao, Z. Li, Z. Lin, K. Lin, G. Liu, E. Llontop, L. Magne, A. Mandlekar, A. Narayan, S. Nasiriany, S. Reed, Y . L. Tan, G. Wang, Z. Wang, J. Wang, Q. Wang, J. Xiang, Y . Xie, Y . Xu, Z. Xu, S. Ye, Z...

    2025

  5. [5]

    π0: A vision-language-action flow model for general robot control,

    K. Black, N. Brown, D. Driess, A. Esmail, M. Equi, C. Finn, N. Fusai, L. Groom, K. Hausman, B. Ichter, S. Jakubczak, T. Jones, L. Ke, S. Levine, A. Li-Bell, M. Mothukuri, S. Nair, K. Pertsch, L. X. Shi, J. Tanner, Q. Vuong, A. Walling, H. Wang, and U. Zhilinsky, “ π0: A vision-language-action flow model for general robot control,” 2024

    2024

  6. [6]

    Openvla: An open- source vision-language-action model,

    M. J. Kim, K. Pertsch, S. Karamcheti, T. Xiao, A. Balakrishna, S. Nair, R. Rafailov, E. Foster, G. Lam, P. Sanketi, et al. , “Openvla: An open- source vision-language-action model,” arXiv preprint arXiv:2406.09246, 2024

    arXiv 2024

  7. [7]

    Spatialvla: Exploring spatial representations for visual-language-action model,

    D. Qu, H. Song, Q. Chen, Y . Yao, X. Ye, Y . Ding, Z. Wang, J. Gu, B. Zhao, D. Wang, et al., “Spatialvla: Exploring spatial representations for visual-language-action model,” arXiv preprint arXiv:2501.15830 , 2025

    arXiv 2025

  8. [8]

    Vlatest: Testing and evaluating vision-language-action models for robotic manip- ulation,

    Z. Wang, Z. Zhou, J. Song, Y . Huang, Z. Shu, and L. Ma, “Vlatest: Testing and evaluating vision-language-action models for robotic manip- ulation,” Proceedings of ACM Software Engineering , vol. 2, July 2025

    2025

Show all 91 references
  1. [9]

    Valle, C

    P. Valle, C. Lu, S. Ali, and A. Arrieta, “Vla-uq.” https://github.com/ pablovalle/VLA UQ, 2025

    2025

  2. [10]

    Results of evaluating uncertainty and quality of visual language action-enabled robots

    P. Valle, C. Lu, and S. A. A. Arrieta, “Results of evaluating uncertainty and quality of visual language action-enabled robots.” Zenodo, July

  3. [11]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE , vol. 86, no. 11, pp. 2278–2324, 2002

    2002

  4. [12]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” Advances in neural informa- tion processing systems , vol. 25, 2012

    2012

  5. [13]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

    2014 arXiv

  6. [14]

    Going deeper with convolutions,

    C. Szegedy, W. Liu, Y . Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V . Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 1–9, 2015

    2015

  7. [15]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

    2017

  8. [16]

    Gpt-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat, et al. , “Gpt-4 technical report,” arXiv preprint arXiv:2303.08774 , 2023

    2023 arXiv

  9. [17]

    Models overview - anthropic

    “Models overview - anthropic.” https://docs.anthropic.com/en/docs/ about-claude/models/overview, 2025

    2025

  10. [18]

    Gemini: a family of highly capable multimodal models,

    G. Team, R. Anil, S. Borgeaud, J.-B. Alayrac, J. Yu, R. Soricut, J. Schalkwyk, A. M. Dai, A. Hauth, K. Millican,et al., “Gemini: a family of highly capable multimodal models,”arXiv preprint arXiv:2312.11805, 2023

    2023 arXiv

  11. [19]

    The llama 3 herd of models,

    A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Vaughan, et al. , “The llama 3 herd of models,” arXiv preprint arXiv:2407.21783 , 2024

    2024 arXiv

  12. [20]

    Mistral 7b,

    A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. de las Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier, L. R. Lavaud, M.-A. Lachaux, P. Stock, T. L. Scao, T. Lavril, T. Wang, T. Lacroix, and W. E. Sayed, “Mistral 7b,” 2023

    2023

  13. [21]

    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,” arXiv preprint arXiv:2304.13705, 2023

    2023 arXiv

  14. [22]

    Octo: An open-source generalist robot policy,

    O. M. Team, D. Ghosh, H. Walke, K. Pertsch, K. Black, O. Mees, S. Dasari, J. Hejna, T. Kreiman, C. Xu, et al. , “Octo: An open-source generalist robot policy,” arXiv preprint arXiv:2405.12213 , 2024

    2024 arXiv

  15. [23]

    Rt-1: Robotics transformer for real-world control at scale,

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, J. Dabis, C. Finn, K. Gopalakrishnan, K. Hausman, A. Herzog, J. Hsu, et al. , “Rt-1: Robotics transformer for real-world control at scale,” arXiv preprint arXiv:2212.06817, 2022

    2022 arXiv

  16. [24]

    Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collab- oration 0,

    A. O’Neill, A. Rehman, A. Maddukuri, A. Gupta, A. Padalkar, A. Lee, A. Pooley, A. Gupta, A. Mandlekar, A. Jain,et al., “Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collab- oration 0,” in 2024 IEEE International Conference on Robotics and Aut...

    2024

  17. [25]

    Automated testing of hybrid simulink/stateflow controllers: industrial case studies,

    R. Matinnejad, S. Nejati, and L. C. Briand, “Automated testing of hybrid simulink/stateflow controllers: industrial case studies,” in Proceedings of the 2017 11th Joint Meeting on Foundations of Software Engineering , pp. 938–943, 2017

    2017

  18. [26]

    Diffusion policy: Visuomotor policy learning via action diffusion,

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song, “Diffusion policy: Visuomotor policy learning via action diffusion,” The International Journal of Robotics Research , 2024

    2024

  19. [27]

    Scaling diffusion policy in transformer to 1 billion parameters for robotic manipulation,

    M. Zhu, Y . Zhu, J. Li, J. Wen, Z. Xu, N. Liu, R. Cheng, C. Shen, Y . Peng, F. Feng, et al. , “Scaling diffusion policy in transformer to 1 billion parameters for robotic manipulation,” arXiv preprint arXiv:2409.14411, 2024

    2024 arXiv

  20. [28]

    Sensitivity of smoothness measures to movement duration, amplitude, and arrests,

    N. Hogan and D. Sternad, “Sensitivity of smoothness measures to movement duration, amplitude, and arrests,” Journal of motor behavior , vol. 41, no. 6, pp. 529–534, 2009

    2009

  21. [29]

    Systems design and implementation with jerk-optimized trajectory generation for uav calligraphy,

    S. K. Phang, S. Lai, F. Wang, M. Lan, and B. M. Chen, “Systems design and implementation with jerk-optimized trajectory generation for uav calligraphy,” Mechatronics, vol. 30, pp. 65–75, 2015

    2015

  22. [30]

    The coordination of arm movements: an ex- perimentally confirmed mathematical model,

    T. Flash and N. Hogan, “The coordination of arm movements: an ex- perimentally confirmed mathematical model,” Journal of neuroscience , vol. 5, no. 7, pp. 1688–1703, 1985

    1985

  23. [31]

    openvla/openvla-7b · hugging face

    openvla, “openvla/openvla-7b · hugging face.” https://huggingface.co/ openvla/openvla-7b

  24. [32]

    Haomingsong/lerobot-pi0-fractal · hugging face

    H. Song, “Haomingsong/lerobot-pi0-fractal · hugging face.” https:// huggingface.co/HaomingSong/lerobot-pi0-fractal, Feb. 2025

    2025

  25. [33]

    Haomingsong/lerobot-pi0-bridge · hugging face

    H. Song, “Haomingsong/lerobot-pi0-bridge · hugging face.” https:// huggingface.co/HaomingSong/lerobot-pi0-bridge, Feb. 2025

    2025

  26. [34]

    Ipec-community/spatialvla-4b-mix-224- pt · hugging face

    IPEC-COMMUNITY, “Ipec-community/spatialvla-4b-mix-224- pt · hugging face.” https://huggingface.co/IPEC-COMMUNITY/ spatialvla-4b-mix-224-pt, July 2025

    2025

  27. [35]

    Evaluating real-world robot manipulation policies in simulation,

    X. Li, K. Hsu, J. Gu, K. Pertsch, O. Mees, H. R. Walke, C. Fu, I. Lunawat, I. Sieh, S. Kirmani, S. Levine, J. Wu, C. Finn, H. Su, Q. Vuong, and T. Xiao, “Evaluating real-world robot manipulation policies in simulation,” arXiv preprint arXiv:2405.05941 , 2024

    2024 arXiv

  28. [36]

    Bridgedata v2: A dataset for robot learning at scale,

    H. R. Walke, K. Black, T. Z. Zhao, Q. Vuong, C. Zheng, P. Hansen- Estruch, A. W. He, V . Myers, M. J. Kim, M. Du, et al., “Bridgedata v2: A dataset for robot learning at scale,” in Conference on Robot Learning, pp. 1723–1736, PMLR, 2023

    2023

  29. [37]

    Cohen, Statistical power analysis for the behavioral sciences

    J. Cohen, Statistical power analysis for the behavioral sciences . Rout- ledge, 2013

    2013

  30. [38]

    Exploring methods for evaluating group differences on the nsse and other surveys: Are the t-test and cohen’sd indices the most appropriate choices,

    J. Romano, J. D. Kromrey, J. Coraggio, J. Skowronek, and L. Devine, “Exploring methods for evaluating group differences on the nsse and other surveys: Are the t-test and cohen’sd indices the most appropriate choices,” in annual meeting of the Southern Association for Instituti...

    2006

  31. [39]

    A review of uncertainty quantification in deep learning: Techniques, applications and challenges,

    M. Abdar, F. Pourpanah, S. Hussain, D. Rezazadegan, L. Liu, M. Ghavamzadeh, P. Fieguth, X. Cao, A. Khosravi, U. R. Acharya,et al., “A review of uncertainty quantification in deep learning: Techniques, applications and challenges,” Information fusion, vol. 76, pp. 243–297, 2021

    2021

  32. [40]

    Bayesian layers: A module for neural network uncertainty,

    D. Tran, M. Dusenberry, M. Van Der Wilk, and D. Hafner, “Bayesian layers: A module for neural network uncertainty,” Advances in neural information processing systems , vol. 32, 2019

    2019

  33. [41]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning,

    Y . Gal and Z. Ghahramani, “Dropout as a bayesian approximation: Representing model uncertainty in deep learning,” in international conference on machine learning , pp. 1050–1059, PMLR, 2016

    2016

  34. [42]

    Monte carlo dropblock for mod- eling uncertainty in object detection,

    S. H. Yelleni, D. Kumari, et al. , “Monte carlo dropblock for mod- eling uncertainty in object detection,” Pattern Recognition , vol. 146, p. 110003, 2024

    2024

  35. [43]

    Simple and scalable predictive uncertainty estimation using deep ensembles,

    B. Lakshminarayanan, A. Pritzel, and C. Blundell, “Simple and scalable predictive uncertainty estimation using deep ensembles,” Advances in neural information processing systems , vol. 30, 2017

    2017

  36. [44]

    Pretrain, prompt, and transfer: Evolving digital twins for time-to-event analysis in cyber-physical systems,

    Q. Xu, T. Yue, S. Ali, and M. Arratibel, “Pretrain, prompt, and transfer: Evolving digital twins for time-to-event analysis in cyber-physical systems,” IEEE Transactions on Software Engineering , 2024. 17

    2024

  37. [45]

    Uncertainty-aware prediction validator in deep learning models for cyber-physical system data,

    F. O. Catak, T. Yue, and S. Ali, “Uncertainty-aware prediction validator in deep learning models for cyber-physical system data,” ACM Trans- actions on Software Engineering and Methodology (TOSEM) , vol. 31, no. 4, pp. 1–31, 2022

    2022

  38. [46]

    Towards safe autonomous driving: Capture uncertainty in the deep neural network for lidar 3d vehicle detection,

    D. Feng, L. Rosenbaum, and K. Dietmayer, “Towards safe autonomous driving: Capture uncertainty in the deep neural network for lidar 3d vehicle detection,” in 2018 21st international conference on intelligent transportation systems (ITSC) , pp. 3266–3273, IEEE, 2018

    2018

  39. [47]

    What uncertainties do we need in bayesian deep learning for computer vision?,

    A. Kendall and Y . Gal, “What uncertainties do we need in bayesian deep learning for computer vision?,” Advances in neural information processing systems, vol. 30, 2017

    2017

  40. [48]

    Assessing the uncertainty and robustness of the laptop refurbishing software,

    C. Lu, J. Wu, S. Ali, and M. L. Olsen, “Assessing the uncertainty and robustness of the laptop refurbishing software,” in 2025 IEEE Conference on Software Testing, Verification and Validation (ICST) , pp. 406–416, 2025

    2025

  41. [49]

    A survey on uncertainty quantification of large language models: Taxonomy, open research challenges, and future directions,

    O. Shorinwa, Z. Mei, J. Lidard, A. Z. Ren, and A. Majumdar, “A survey on uncertainty quantification of large language models: Taxonomy, open research challenges, and future directions,” ACM Computing Surveys , 2025

    2025

  42. [50]

    Look before you leap: An exploratory study of uncertainty mea- surement for large language models,

    Y . Huang, J. Song, Z. Wang, S. Zhao, H. Chen, F. Juefei-Xu, and L. Ma, “Look before you leap: An exploratory study of uncertainty mea- surement for large language models,” arXiv preprint arXiv:2307.10236, 2023

    2023 arXiv

  43. [51]

    A survey on evaluation of large language models,

    Y . Chang, X. Wang, J. Wang, Y . Wu, L. Yang, K. Zhu, H. Chen, X. Yi, C. Wang, Y . Wang, W. Ye, Y . Zhang, Y . Chang, P. S. Yu, Q. Yang, and X. Xie, “A survey on evaluation of large language models,” ACM Trans. Intell. Syst. Technol., vol. 15, Mar. 2024

    2024

  44. [52]

    Larger and more instructable language models become less reliable,

    L. Zhou, W. Schellaert, F. Mart ´ınez-Plumed, Y . Moros-Daval, C. Ferri, and J. Hern´andez-Orallo, “Larger and more instructable language models become less reliable,” Nature, vol. 634, no. 8032, pp. 61–68, 2024

    2024

  45. [53]

    On the impor- tance of uncertainty in decision-making with large language models,

    N. Felicioni, L. Maystre, S. Ghiassian, and K. Ciosek, “On the impor- tance of uncertainty in decision-making with large language models,” arXiv preprint arXiv:2404.02649 , 2024

    2024 arXiv

  46. [54]

    Uncertainty quantification with pre-trained language mod- els: A large-scale empirical analysis,

    Y . Xiao, P. P. Liang, U. Bhatt, W. Neiswanger, R. Salakhutdinov, and L.- P. Morency, “Uncertainty quantification with pre-trained language mod- els: A large-scale empirical analysis,” arXiv preprint arXiv:2210.04714, 2022

    2022 arXiv

  47. [55]

    Uncertainty quantification in large language models through convex hull analysis,

    F. O. Catak and M. Kuzlu, “Uncertainty quantification in large language models through convex hull analysis,” Discover Artificial Intelligence , vol. 4, no. 1, p. 90, 2024

    2024

  48. [56]

    Generating with confidence: Uncertainty quantification for black-box large language models,

    Z. Lin, S. Trivedi, and J. Sun, “Generating with confidence: Uncertainty quantification for black-box large language models,” 2024

    2024

  49. [57]

    Map: Multimodal uncertainty-aware vision-language pre-training model,

    Y . Ji, J. Wang, Y . Gong, L. Zhang, Y . Zhu, H. Wang, J. Zhang, T. Sakai, and Y . Yang, “Map: Multimodal uncertainty-aware vision-language pre-training model,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pp. 23262–23271, 2023

    2023

  50. [58]

    Unveiling uncertainty: A deep dive into calibration and performance of multimodal large language models,

    Z. Chen, W. Hu, G. He, Z. Deng, Z. Zhang, and R. Hong, “Unveiling uncertainty: A deep dive into calibration and performance of multimodal large language models,” arXiv preprint arXiv:2412.14660 , 2024

    2024 arXiv

  51. [59]

    Data-driven grasp synthesis—a survey,

    J. Bohg, A. Morales, T. Asfour, and D. Kragic, “Data-driven grasp synthesis—a survey,” IEEE Transactions on robotics , vol. 30, no. 2, pp. 289–309, 2013

    2013

  52. [60]

    Dex-net 2.0: Deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics,

    J. Mahler, J. Liang, S. Niyaz, M. Laskey, R. Doan, X. Liu, J. A. Ojea, and K. Goldberg, “Dex-net 2.0: Deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics,” arXiv preprint arXiv:1703.09312, 2017

    2017 arXiv

  53. [61]

    S. M. LaValle, Planning algorithms. Cambridge university press, 2006

    2006

  54. [62]

    Prob- abilistic roadmaps for path planning in high-dimensional configuration spaces,

    L. E. Kavraki, P. Svestka, J.-C. Latombe, and M. H. Overmars, “Prob- abilistic roadmaps for path planning in high-dimensional configuration spaces,” IEEE transactions on Robotics and Automation , vol. 12, no. 4, pp. 566–580, 2002

    2002

  55. [63]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in 2009 IEEE conference on computer vision and pattern recognition , pp. 248–255, Ieee, 2009

    2009

  56. [64]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in European conference on computer vision , pp. 740–755, Springer, 2014

    2014

  57. [65]

    Bleu: a method for automatic evaluation of machine translation,

    K. Papineni, S. Roukos, T. Ward, and W.-J. Zhu, “Bleu: a method for automatic evaluation of machine translation,” in Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pp. 311–318, 2002

    2002

  58. [66]

    Rouge: A package for automatic evaluation of summaries,

    C.-Y . Lin, “Rouge: A package for automatic evaluation of summaries,” in Text summarization branches out , pp. 74–81, 2004

    2004

  59. [67]

    Ladev: A language-driven testing and evaluation platform for vision- language-action models in robotic manipulation,

    Z. Wang, Z. Zhou, J. Song, Y . Huang, Z. Shu, and L. Ma, “Ladev: A language-driven testing and evaluation platform for vision- language-action models in robotic manipulation,” arXiv preprint arXiv:2410.05191, 2024

    2024 arXiv

  60. [68]

    Arnold: A benchmark for language- grounded task learning with continuous states in realistic 3d scenes,

    R. Gong, J. Huang, Y . Zhao, H. Geng, X. Gao, Q. Wu, W. Ai, Z. Zhou, D. Terzopoulos, S.-C. Zhu, et al., “Arnold: A benchmark for language- grounded task learning with continuous states in realistic 3d scenes,” in Proceedings of the IEEE/CVF International Conference on Computer...

    2023

  61. [69]

    Maniskill2: A unified benchmark for generalizable manipulation skills,

    J. Gu, F. Xiang, X. Li, Z. Ling, X. Liu, T. Mu, Y . Tang, S. Tao, X. Wei, Y . Yao, et al. , “Maniskill2: A unified benchmark for generalizable manipulation skills,” arXiv preprint arXiv:2302.04659 , 2023

    2023 arXiv

  62. [70]

    Behavior-1k: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation,

    C. Li, R. Zhang, J. Wong, C. Gokmen, S. Srivastava, R. Mart ´ın-Mart´ın, C. Wang, G. Levine, M. Lingelbach, J. Sun, et al. , “Behavior-1k: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation,” in Conference on Robot Learning, pp. 80–93, PMLR, 2023

    2023

  63. [71]

    Isaac gym: High performance gpu-based physics simulation for robot learning,

    V . Makoviychuk, L. Wawrzyniak, Y . Guo, M. Lu, K. Storey, M. Macklin, D. Hoeller, N. Rudin, A. Allshire, A. Handa, et al. , “Isaac gym: High performance gpu-based physics simulation for robot learning,” arXiv preprint arXiv:2108.10470, 2021

    2021 arXiv

  64. [72]

    Calvin: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks,

    O. Mees, L. Hermann, E. Rosete-Beas, and W. Burgard, “Calvin: A benchmark for language-conditioned policy learning for long-horizon robot manipulation tasks,” IEEE Robotics and Automation Letters, vol. 7, no. 3, pp. 7327–7334, 2022

    2022

  65. [73]

    robosuite: A modular simulation framework and benchmark for robot learning,

    Y . Zhu, J. Wong, A. Mandlekar, R. Mart´ın-Mart´ın, A. Joshi, S. Nasiriany, and Y . Zhu, “robosuite: A modular simulation framework and benchmark for robot learning,” arXiv preprint arXiv:2009.12293 , 2020

    2009 arXiv

  66. [74]

    Libero: Benchmarking knowledge transfer for lifelong robot learning,

    B. Liu, Y . Zhu, C. Gao, Y . Feng, Q. Liu, Y . Zhu, and P. Stone, “Libero: Benchmarking knowledge transfer for lifelong robot learning,” Advances in Neural Information Processing Systems , vol. 36, pp. 44776–44791, 2023

    2023

  67. [75]

    Palm-e: An embodied multimodal language model,

    D. Driess, F. Xia, M. S. Sajjadi, C. Lynch, A. Chowdhery, A. Wahid, J. Tompson, Q. Vuong, T. Yu, W. Huang, et al., “Palm-e: An embodied multimodal language model,” 2023

    2023

  68. [76]

    Rt-2: Vision-language-action models transfer web knowledge to robotic control,

    B. Zitkovich, T. Yu, S. Xu, P. Xu, T. Xiao, F. Xia, J. Wu, P. Wohlhart, S. Welker, A. Wahid, et al. , “Rt-2: Vision-language-action models transfer web knowledge to robotic control,” in Conference on Robot Learning, pp. 2165–2183, PMLR, 2023

    2023

  69. [77]

    Prismatic vlms: Investigating the design space of visually- conditioned language models,

    S. Karamcheti, S. Nair, A. Balakrishna, P. Liang, T. Kollar, and D. Sadigh, “Prismatic vlms: Investigating the design space of visually- conditioned language models,” in Forty-first International Conference on Machine Learning , 2024

    2024

  70. [78]

    Llama 2: Open foundation and fine-tuned chat models,

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al. , “Llama 2: Open foundation and fine-tuned chat models,” arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  71. [79]

    Sigmoid loss for language image pre-training,

    X. Zhai, B. Mustafa, A. Kolesnikov, and L. Beyer, “Sigmoid loss for language image pre-training,” in Proceedings of the IEEE/CVF international conference on computer vision , pp. 11975–11986, 2023

    2023

  72. [80]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, et al. , “Dinov2: Learning robust visual features without supervision,” arXiv preprint arXiv:2304.07193, 2023

    2023 arXiv

  73. [81]

    Flow matching for generative modeling,

    Y . Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” arXiv preprint arXiv:2210.02747 , 2022

    2022 arXiv

  74. [82]

    Rectified flow: A marginal preserving approach to optimal transport,

    Q. Liu, “Rectified flow: A marginal preserving approach to optimal transport,” arXiv preprint arXiv:2209.14577 , 2022

    2022 arXiv

  75. [83]

    Droid: A large-scale in-the-wild robot manipulation dataset,

    A. Khazatsky, K. Pertsch, S. Nair, A. Balakrishna, S. Dasari, S. Karam- cheti, S. Nasiriany, M. K. Srirama, L. Y . Chen, K. Ellis, et al., “Droid: A large-scale in-the-wild robot manipulation dataset,” arXiv preprint arXiv:2403.12945, 2024

    2024 arXiv

  76. [84]

    Paligemma: A versatile 3b vlm for transfer,

    L. Beyer, A. Steiner, A. S. Pinto, A. Kolesnikov, X. Wang, D. Salz, M. Neumann, I. Alabdulmohsin, M. Tschannen, E. Bugliarello, et al. , “Paligemma: A versatile 3b vlm for transfer,” arXiv preprint arXiv:2407.07726, 2024

    2024 arXiv

  77. [85]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer,

    N. Shazeer, A. Mirhoseini, K. Maziarz, A. Davis, Q. Le, G. Hinton, and J. Dean, “Outrageously large neural networks: The sparsely-gated mixture-of-experts layer,” arXiv preprint arXiv:1701.06538 , 2017

    2017 arXiv

  78. [86]

    Gshard: Scaling giant models with conditional computation and automatic sharding,

    D. Lepikhin, H. Lee, Y . Xu, D. Chen, O. Firat, Y . Huang, M. Krikun, N. Shazeer, and Z. Chen, “Gshard: Scaling giant models with conditional computation and automatic sharding,” arXiv preprint arXiv:2006.16668, 2020

    2006 arXiv

  79. [87]

    Playground v3: Improving text-to-image alignment with deep-fusion large language models,

    B. Liu, E. Akhgari, A. Visheratin, A. Kamko, L. Xu, S. Shrirao, C. Lambert, J. Souza, S. Doshi, and D. Li, “Playground v3: Improving text-to-image alignment with deep-fusion large language models,” arXiv preprint arXiv:2409.10695, 2024

    2024 arXiv

  80. [88]

    Rh20t: A comprehensive robotic dataset for learning diverse 18 skills in one-shot,

    H.-S. Fang, H. Fang, Z. Tang, J. Liu, C. Wang, J. Wang, H. Zhu, and C. Lu, “Rh20t: A comprehensive robotic dataset for learning diverse 18 skills in one-shot,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , pp. 653–660, IEEE, 2024

    2024

  81. [89]

    Paligemma 2: A family of versatile vlms for transfer,

    A. Steiner, A. S. Pinto, M. Tschannen, D. Keysers, X. Wang, Y . Bit- ton, A. Gritsenko, M. Minderer, A. Sherbondy, S. Long, et al. , “Paligemma 2: A family of versatile vlms for transfer,” arXiv preprint arXiv:2412.03555, 2024

    2024 arXiv

  82. [90]

    Zoedepth: Zero-shot transfer by combining relative and metric depth,

    S. F. Bhat, R. Birkl, D. Wofk, P. Wonka, and M. M ¨uller, “Zoedepth: Zero-shot transfer by combining relative and metric depth,” arXiv preprint arXiv:2302.12288, 2023. APPENDIX A IMPLEMENTATION GUIDELINES In this section, we provide relevant information about the architecture ...

    2023 arXiv

  83. [2025]

    https://doi.org/10.5281/zenodo.16315133

    doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.