Pith. sign in

REVIEW 4 major objections 5 minor 7 cited by

RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Sampling many robot actions and verifying the best one lifts real-world task success by 25 percentage points.

desk verdict A useful generate-and-verify wrapper for VLAs, but the headline gains are not cleanly attributed to the verifier—needs closed-loop baselines, error bars, and a more honest framing of the scaling law. read the letter →

arxiv 2506.17811 v2 pith:MHJA455F submitted 2025-06-21 cs.RO cs.AIcs.SYeess.SY

Jacky Kwok , Christopher Agia , Rohan Sinha , Matt Foutter , Shulu Li , Ion Stoica , Azalia Mirhoseini , Marco Pavone This is my paper · ORCID
classification cs.ROcs.AIcs.SYeess.SY
keywords test-timescalingvision-language-actionmodelsactionverificationpreferencelearningroboticmanipulationinference-timelawssyntheticdatagenerate-then-verify
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 argues that a robot policy that emits one action per observation leaves useful performance on the table: repeatedly generating candidate actions and then picking the best one with a learned verifier makes the same policy substantially more reliable. The authors first show that action error falls as an exponentiated power law as the number of candidate samples grows, across several generalist policies. They then build RoboMonkey, a wrapper that samples a small batch of actions, perturbs them with a Gaussian fit and majority voting to form a cheap proposal distribution, and scores the candidates with a vision-language-model verifier trained on synthetic pairwise comparisons. In their experiments the wrapper raises success on out-of-distribution real-robot tasks by 25 percentage points and on in-distribution simulation by 9 percentage points.

What carries the argument

The load-bearing mechanism is the generate-then-verify loop with a learned proxy verifier. Formally, RoboMonkey samples $\hat N$ actions from a VLA with positive temperature, fits $\mathcal{N}(\mu,\Sigma)$ to the six continuous action deltas, sets the gripper by majority vote, samples $\hat K$ refined candidates from that Gaussian, and executes $\arg\max_{\tilde a} R_\phi(\tilde a,s,I)$. The reward model $R_\phi$ is trained with a modified Bradley-Terry objective on synthetic preference pairs $(\text{winner},\text{loser})$ whose labels come from whether a candidate is closer in RMSE to the recorded expert action, with a margin term $\alpha\,|\Delta^*-\hat\Delta|^2$ that helps the verifier separate clearly different actions. The observed relation $\log(\text{error})\approx\log(a)+b\log(k)$ is the phenomenon the machinery exploits: because error falls predictably with more samples, a cheap proposal distribution plus a verifier can harvest that gain without expensive repeated calls to the base policy.

What would settle it

Take a task with a deliberately imperfect demonstration, such as a teleoperator who slips and corrects, train the verifier with the paper's RMSE-to-expert labels, and compare closed-loop rollouts against a verifier trained on outcome-based labels from successful and failed trials. If the outcome-trained verifier wins, or if the closest-to-expert candidate consistently fails while a more distant candidate succeeds, then the expert-action proxy, not sampling and verification alone, is carrying the reported gains.

Watch

Extended reading notes

Core claim

The paper's central discovery is that test-time compute can be traded for control quality in vision-language-action models: when an oracle always picks the candidate closest to the expert action, action error decreases regularly with the number of samples, with $\log(\text{error})\approx\log(a)+b\log(k)$ across several off-the-shelf policies. RoboMonkey converts this observation into a deployable loop. At each step it samples a handful of actions, fits a Gaussian over their translation and rotation deltas, takes the gripper state by majority vote, draws a larger candidate set from that Gaussian, and lets a 7-billion-parameter VLM-based verifier choose the action. The verifier is learned on automatically generated comparisons whose labels are derived from RMSE to the expert demonstration, and the paper reports that scaling this synthetic preference dataset yields roughly log-linear task improvement. The upshot is that pairing this wrapper with existing policies yields 25 percentage points absolute improvement on out-of-distribution real-robot tasks, 9 points on in-distribution simulation, and 7 points when both policy and verifier are fine-tuned on a new robot benchmark.

Load-bearing premise

The load-bearing premise is that the recorded expert action is the best action at each state, because every synthetic preference label is just closeness to that one demonstration; if the human demonstration is suboptimal or the deployment state was never demonstrated, the verifier is trained to imitate rather than to succeed.

Editorial extensions

If this is right

  • The base policy does not need to be retrained: sampling and verification turn any off-the-shelf generalist VLA into a stronger closed-loop controller.
  • More synthetic comparison data produces better verification, and downstream success grows roughly log-linearly with that data.
  • Gaussian perturbation delivers most of the benefit of full policy resampling at a fraction of the latency, making the gains available near 1.5 Hz on a single GPU.
  • On a new robot setup, fine-tuning the verifier together with the policy adds about 7 percentage points over fine-tuning the policy alone.
  • The learned verifier is markedly less sensitive to visual and semantic distribution shift than the base policy, which is why the largest gains appear on out-of-distribution tasks.

Reading between the lines

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

  • An implicit corollary is that the same wrapper could be tuned to a compute budget: given the fitted power-law exponent and the verifier cost, there is an optimal number of candidates per second of latency, and the paper's latency analysis reports the trade-off without deriving the optimum.
  • Because preference labels are RMSE to the recorded expert action, a natural extension is to label synthetic pairs with actual task success from rollouts or a task reward, which would remove the dependence on demonstration optimality.
  • The verifier currently scores single-step actions; a testable extension is to score multi-step action chunks or whole trajectories, which speaks directly to long-horizon benchmarks where local action error is not the right criterion.
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 / 5 minor

Summary. The paper studies test-time scaling for vision-language-action (VLA) models. It first presents an analysis on Bridge V2 in which action error, measured by RMSE to the expert action, decreases as the number of sampled actions grows, and claims this follows an exponentiated power law across OpenVLA, CogACT, Octo, and SpatialVLA. It then introduces RoboMonkey, a generate-then-verify framework: at deployment it samples \hat{N} actions from a VLA, fits a Gaussian over translation and rotation components, uses majority voting for the gripper, samples \hat{K} candidates from that proposal distribution, and selects the action with the highest score from a fine-tuned LLaVA-7B verifier. The verifier is trained on synthetic pairwise preferences whose labels are derived from RMSE to the expert action. Experiments report a 25% absolute improvement on four out-of-distribution real-world WidowX tasks, a 9% improvement on four in-distribution SIMPLER tasks, a 6.7% improvement on LIBERO-Long after fine-tuning, and a scalable synthetic-data pipeline. The paper also contributes an SGLang-based serving engine for batched OpenVLA sampling.

Significance. If the central attribution claim holds, this is a useful and timely contribution: it provides one of the first characterizations of inference-time scaling for VLA policies, offers a fully synthetic pipeline for training action verifiers without human preference labels, and demonstrates consistent gains across simulation and hardware. The efficiency analysis is also a genuine strength, since test-time scaling is often impractical; the reported batching and Gaussian-perturbation results make the framework deployable at roughly 1.5 Hz. The paper is generally clearly written and the experimental scope, spanning two embodiments and three evaluation settings, is substantially broader than many concurrent works. However, the headline improvements are not yet cleanly attributed to the learned verifier: the closed-loop experiments lack selection-rule ablations over the same candidate pool, the simulation results are reported as best-over-sweep, and the real-world results have small sample sizes without variance estimates. These issues are fixable within the scope of the manuscript, so the work is promising but needs revision before the core claims are fully supported.

major comments (4)
  1. [Sections 5.2-5.3 and Appendix C] The closed-loop results do not isolate the contribution of the learned verifier. In all closed-loop evaluations, RoboMonkey is compared with single-action OpenVLA and V-GPS, but there is no random-selection, majority-voting, or best-of-N baseline over the same candidate pool generated by Algorithm 1. Appendix C does compare these selection rules, but only on RMSE-to-expert action error over Bridge V2 states, which is exactly the proxy used to train the verifier (Section 4.2) and may not track closed-loop success under distribution shift. Since Section 3 shows that sampling alone, even random sampling, reduces RMSE relative to greedy decoding, the reported 9% SIMPLER gain and 25% real-world gain cannot yet be attributed to verification. Please add closed-loop controls for random selection, majority voting, and best-of-N over the identical proposal distribution, at least on the SIMPLER tasks and ideally on the real-world tasks.
  2. [Appendix B.2 and Section 5.3] The reported SIMPLER results are selected as the best over \hat{N} in {5,9} and \hat{K} in {8,16,32} for each task, while the V-GPS baseline is also swept and reported at its best; this turns the headline comparison into a best-case comparison rather than a comparison of fixed deployment configurations. Additionally, Section 5.3 reports 10 trials per real-world task with no error bars or confidence intervals. With 10 binary trials, a one-trial difference is 10 percentage points, so the per-task differences in Figure 3 are not established as robust. Please report either a single fixed configuration or all swept configurations, and provide per-task standard errors or confidence intervals for the real-world results.
  3. [Section 3 and Figure 1] The paper's first contribution, the inference-time scaling law, is not supported by the reported fit statistics. The text says the relationship follows an 'exponentiated power law' and gives the model log(e) ≈ log(a) + b log(k), but no fitted values of a and b, no R² values, no residual diagnostics, and no confidence intervals are reported for OpenVLA, CogACT, Octo, or SpatialVLA. The term 'exponentiated power law' is also inconsistent with the stated log-linear functional form, which is an ordinary power law. Please provide the fitted parameters and goodness-of-fit for each model and sampling method, or soften the claim to a qualitative observation.
  4. [Sections 4.2, 4.3, and Appendix F] The verifier is trained and evaluated against RMSE to the single expert action, which is an imitation proxy. The ablation in Appendix F claims that preference-based learning 'generalizes better in OOD settings,' but this is tested only on the same RMSE metric, so it does not establish robustness of closed-loop success. Since the entire pipeline rests on this labeling rule, the OOD success results do not disambiguate verifier quality from proposal-distribution quality. Please add at least one closed-loop comparison (for example on SIMPLER) between a verifier trained on RMSE-to-expert labels and a verifier trained on an alternative objective, or provide a direct analysis showing that the RMSE-based labels are necessary for the reported success-rate gains.
minor comments (5)
  1. [Section 3] Use consistent terminology: the model e = a k^b should be called a power law, not an 'exponentiated power law'.
  2. [Section 5.5 and Figure 6] The claim that task performance 'grows nearly log-linearly with synthetic dataset size' is based on four points without error bars or a fitted curve; please show the fitted trend or state this as an observation.
  3. [Section 5.2] Please clarify the OpenVLA baseline average in the SIMPLER comparison: the text reports RoboMonkey at 47.5% and a 9% average improvement, but the exact OpenVLA average is not stated in the main text.
  4. [Section 2 and Table of Notation] The notation lists at as [Δx, Δy, Δz, Δu, Δv, Δw, g]' and states both the state and action spaces are 7-dimensional; please clarify whether the state also includes the gripper as a binary state or only as an action component.
  5. [Abstract and Title] There are formatting artifacts in the title and abstract ('T est-Time', 'V erification'); these should be cleaned in the camera-ready version.

Circularity Check

1 steps flagged · score 3.0 of 10

Action-error ablations reuse the verifier's own RMSE training label; closed-loop OOD claims are independent.

  1. fitted input called prediction [Section 4.2 (Synthetic Data Generation Pipeline) and Appendix C (Ablation Over Action Selection Methods and Number of Samples)]
    "Subsequently, we construct K-choose-2 pairwise comparisons and compute the RMSE between each sampled action ... and the ground-truth action ... the "winning" action ... and the "losing" action ... are determined as follows: ... if RMSE(a^i_t, a*_t) < RMSE(a^j_t, a*_t) ... We report the normalized RMSE between the ground-truth action and the selected action for each method. ... RoboMonkey reduces the action error by 21% relative to the greedy decoding baseline, highlighting the effectiveness of our verifier in improving action precision."

    The preference labels that train the verifier are defined by exactly the same RMSE-to-expert metric used in the Appendix C action-error evaluation. Therefore, RoboMonkey's reported 21% action-error reduction measures how well the verifier reproduces its own training signal; it is a fit check, not an independent prediction. The Section 3 oracle scaling analysis uses the same RMSE metric, so the 'scaling law' is defined relative to the same expert-action target rather than to task success. Because the closed-loop SIMPLER, LIBERO, and real-world OOD results are scored by task success rather than by RMSE, they do not inherit this circularity; this step is partial and non-central.

full rationale

The paper's central claims—25% absolute OOD real-world improvement, 9% SIMPLER improvement, and 6.7% LIBERO-Long improvement—are evaluated by closed-loop task success, which is external to the RMSE-to-expert labels used to train the action verifier. The scaling-law analysis in Section 3 is an empirical curve fit to measured action error under an oracle selector, not a first-principles derivation, and it does not assume the conclusion it later tests. The self-citations to prior repeated-sampling work (e.g., [22,30]) are contextual and not load-bearing: the present scaling measurements are made on VLA models in this paper. The one genuinely self-referential element is the action-error ablation in Appendix C, where the verifier is scored on the same RMSE-to-expert metric that generated its synthetic preference labels; that result is a training-objective check rather than an independent validation. The absence of a random-selection or majority-voting closed-loop baseline is an experimental-control concern, not a circularity, because the closed-loop success metric does not reduce to the verifier's training input. Overall, the main empirical contributions remain independently supported, with only a minor partial circularity in the action-error evaluation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the quality of the synthetic preference labels, which are derived from RMSE to expert actions, and on the assumption that Gaussian perturbation around a small number of policy samples covers the space of good actions. The power-law observation is descriptive and does not itself introduce free parameters that feed into the verifier, but the deployment hyperparameters (N_hat, K_hat, temperature) are chosen per task in simulation.

free parameters (5)
  • power law parameters a and b = not reported numerically
    In Section 3, log(e) is modeled as log(a) + b log(k) and a,b are fitted to the measured RMSE curves. These parameters are descriptive, not used predictively, but they underlie the 'scaling law' claim.
  • temperature T for VLA sampling = 1.0 for real-world, 0.5 for SpatialVLA
    Chosen by the authors for deployment; sensitivity not fully explored.
  • number of initial samples N_hat = 5 (real-world), 5 or 9 (simulation)
    Simulation reports best results over N_hat in {5,9} per task, a selection-on-eval choice.
  • number of augmented samples K_hat = 16 (real-world), 8, 16, 32 (simulation)
    Simulation reports best results per task, which can inflate reported success.
  • margin weight alpha = 0.1
    Selected via ablation on 10,000 synthetic pairs (Appendix E).
assumptions (4)
  • domain assumption Expert demonstrations in the training dataset are optimal or near-optimal actions for their states.
    The synthetic preference labels are assigned by RMSE to the expert action (Section 4.2), so the verifier can only learn to imitate the expert, not to discover better actions.
  • domain assumption RMSE in 7D action space is a valid proxy for action quality.
    The verifier is trained and the oracle is defined using RMSE; this assumes Euclidean closeness to the expert action corresponds to task success.
  • domain assumption A Gaussian fit to 5 sampled actions adequately covers the set of high-quality actions.
    The proposal distribution in Section 4.4 is a Gaussian estimated from N_hat=5 samples; with 5 points in 6D the covariance is rank-deficient, yet the method relies on it.
  • standard math The Bradley-Terry preference model with a margin term is an appropriate objective for action ranking.
    Used as the reward modeling loss in Section 4.3, following standard preference learning practice.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models." pith.science (2026). https://pith.science/paper/MHJA455F

@misc{pith2026250617811,
  author       = {Pith},
  title        = {Pith review of: RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MHJA455F}},
  note         = {Machine review of arXiv:2506.17811}
}
read the original abstract

Vision-Language-Action (VLA) models have demonstrated remarkable capabilities in visuomotor control, yet ensuring their robustness in unstructured real-world environments remains a persistent challenge. In this paper, we investigate test-time scaling through the lens of sampling and verification as means to enhance the robustness and generalization of VLAs. We first demonstrate that the relationship between action error and the number of generated samples follows an exponentiated power law across a range of VLAs, indicating the existence of inference-time scaling laws. Building on these insights, we introduce RoboMonkey, a test-time scaling framework for VLAs. At deployment, RoboMonkey samples a small set of actions from a VLA, applies Gaussian perturbation and majority voting to construct an action proposal distribution, and then uses a Vision Language Model (VLM)-based verifier to select the optimal action. We propose a synthetic data generation pipeline for training such VLM-based action verifiers, and demonstrate that scaling the synthetic dataset consistently improves verification and downstream accuracy. Through extensive simulated and hardware experiments, we show that pairing existing VLAs with RoboMonkey yields significant performance gains, achieving a 25% absolute improvement on out-of-distribution tasks and 9% on in-distribution tasks. Additionally, when adapting to new robot setups, we show that fine-tuning both VLAs and action verifiers yields a 7% performance increase compared to fine-tuning VLAs alone.

Figures

Figures reproduced from arXiv: 2506.17811 by the authors.

Figure 1
Figure 1. Inference-Time Scaling Law: We observe that action error consistently decreases as we scale the number of generated actions across multiple sampling approaches, assuming the presence of an oracle verifier. Repeatedly sampling actions from robot policies, applying Gaussian perturbation to a few sampled actions, and even random sampling all outperform single-attempt OpenVLA. We also find that the relationship between … view at source ↗
Figure 2
Figure 2. Stage 1: Training the Action Verifier. Given an imitation learning dataset, we sample N candidate actions per state from a generalist robot policy, and apply clustering to reduce them to K representative actions. We construct K 2  synthetic action comparisons and assign preferences based on the RMSE between each sampled action and the ground-truth action. This synthetic preference dataset is then used to fine-tune … view at source ↗
Figure 3
Figure 3. Scaling test-time compute significantly improves the precision and robustness of generalist robot [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Example tasks across SIMPLER, real-world, and LIBERO environments. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Left: Repeated sampling can exploit KV Cache optimizations and batch processing to achieve higher throughput than greedy decoding. Therefore, we extended SGLang’s capabilities to properly support OpenVLA. Our optimized implementation substantially outperforms the naive…
Figure 6
Figure 6. Figure 6: Average success rates across four tasks on SIMPLER as a function of synthetic dataset size. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Representative task executions in real-world, SIMPLER, and LIBERO environments. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Comparison of action error (average RMSE) across different selection methods as the number of [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Effect of scaling test-time compute with RoboMonkey across different generalist robot policies. [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]
Figure 10
Figure 10. Figure 10: Comparison between preference-based learning and RMSE regression across number of samples. [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Action error (average RMSE) as a function of computational overhead for policy sampling and [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 7 Pith papers

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

  1. Q-VGM: Q-Value-Gradient Matching for Off-Policy Reinforcement Learning of Flow-Matching VLA

    cs.RO 2026-06 unverdicted novelty 7.0 of 10

    Q-VGM offline RL fine-tuning converts critic Q-gradients into residual velocity targets for flow-matching VLAs, raising LIBERO success from 75.0% to 92.5%.

  2. World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry

    cs.LG 2026-04 accept novelty 7.0 of 10

    WAV self-improves action-conditioned world models by cycle-consistent verification of state plausibility and sparse action reachability, doubling sample efficiency and lifting policy reward by over 22% on nine tasks.

  3. Verifier-free Test-Time Sampling for Vision-Language-Action Models

    cs.RO 2025-10 conditional novelty 7.0 of 10

    A verifier-free test-time sampling method for vision-language-action models that selects actions by KL divergence to a condition-masked reference distribution, improving task success rates.

  4. Action Chunk Scheduling for Batched Robot Policy Serving

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A lookahead scheduler that simulates each robot's action-queue state before choosing batches improves throughput in heterogeneous multi-robot policy serving by up to 18% in real-world tests.

  5. Addressing the Orchestration Gap in Generalist Robots via Physical Agency

    cs.RO 2026-07 conditional novelty 6.0 of 10

    A closed-loop inference-time orchestrator using a frontier VLM lifts frozen robot policies from 12.8% to 53.3% on LIBERO-PRO and from near-zero to over 90% on reasoning-limited real-robot tasks, without retraining the...

  6. EVE: A Generator-Verifier System for Generative Policies

    cs.RO 2025-12 conditional novelty 6.0 of 10

    Zero-shot VLM verifiers, ensembled and fused via guided diffusion, improve frozen generative robot policies' success rates by 1-2 percentage points on simulated manipulation tasks.

  7. AtlasVLA: Persistent World-Ego State Modeling for Vision-Language-Action Models

    cs.RO 2026-08 conditional novelty 5.0 of 10

    Wrist-only VLA manipulation with persistent world and ego memory reports state-of-the-art success on LIBERO, RLBench, and real-world long-horizon tasks.

Reference graph

Works this paper leans on

57 extracted references · 6 canonical work pages · cited by 7 Pith papers

  1. [1]

    Black, N

    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. URLhttps://arxiv.org/a...

    arXiv 2024

  2. [2]

    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

  3. [3]

    Driess, F

    D. Driess, F. Xia, M. S. M. Sajjadi, C. Lynch, A. Chowdhery, B. Ichter, A. Wahid, J. Tompson, Q. Vuong, T. Y u, W. Huang, Y . Chebotar, P . Sermanet, D. Duckworth, S. Levine, V . V anhoucke, K. Hausman, M. Toussaint, K. Greff, A. Zeng, I. Mordatch, and P . Florence. Palm-e: An embodied multimodal language model, 2023. URLhttps://arxiv.org/abs/2303.03378

    arXiv 2023

  4. [4]

    Brohan, N

    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

    arXiv 2022

  5. [5]

    Brohan, N

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, X. Chen, K. Choromanski, T. Ding, D. Driess, A. Dubey, C. Finn, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control. arXiv preprint arXiv:2307.15818, 2023

    arXiv 2023

  6. [6]

    Khazatsky, K

    A. Khazatsky, K. Pertsch, S. Nair, A. Balakrishna, S. Dasari, S. Karamcheti, 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

    arXiv 2024

  7. [7]

    O’Neill, A

    A. O’Neill, A. Rehman, A. Gupta, A. Maddukuri, A. Gupta, A. Padalkar, A. Lee, A. Pooley, A. Gupta, A. Mandlekar, et al. Open x-embodiment: Robotic learning datasets and rt-x models. arXiv preprint arXiv:2310.08864, 2023

    arXiv 2023

  8. [8]

    C. Agia, R. Sinha, J. Y ang, Z. ang Cao, R. Antonova, M. Pavone, and J. Bohg. Unpacking failure modes of generative policies: Runtime monitoring of consistency and progress, 2024. URL https://arxiv.org/abs/2410.04640

    arXiv 2024

Show all 57 references
  1. [9]

    Sinha, A

    R. Sinha, A. Elhafsi, C. Agia, M. Foutter, E. Schmerling, and M. Pavone. Real-time anomaly detection and reactive planning with large language models, 2024. URL https://arxiv.org/abs/2407.08735

    2024 arXiv

  2. [10]

    Z. Zhou, P . Atreya, A. Lee, H. Walke, O. Mees, and S. Levine. Autonomous improvement of instruc- tion following skills via foundation models, 2024. URLhttps://arxiv.org/abs/2407.20635

    2024 arXiv

  3. [11]

    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. InConference on Robot Learning, pages 1723–1736. PMLR, 2023

    2023

  4. [12]

    Hejna, C

    J. Hejna, C. Bhateja, Y . Jiang, K. Pertsch, and D. Sadigh. Re-mix: Optimizing data mixtures for large scale imitation learning, 2024. URLhttps://arxiv.org/abs/2408.14037

    2024 arXiv

  5. [13]

    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, page 02783649241273668, 2023. 10

    2023

  6. [14]

    Cheang, G

    C.-L. Cheang, G. Chen, Y . Jing, T. Kong, H. Li, Y . Li, Y . Liu, H. Wu, J. Xu, Y . Y ang, H. Zhang, and M. Zhu. Gr-2: A generative video-language-action model with web-scale knowledge for robot manipulation. arXiv preprint arXiv:2410.06158, 2024

    arXiv

  7. [15]

    T. Z. Zhao, V . Kumar, S. Levine, and C. Finn. Learning fine-grained bimanual manipulation with low-cost hardware, 2023. URLhttps://arxiv.org/abs/2304.13705

    2023 arXiv

  8. [16]

    Zawalski, W

    M. Zawalski, W. Chen, K. Pertsch, O. Mees, C. Finn, and S. Levine. Robotic control via embodied chain-of-thought reasoning, 2025. URLhttps://arxiv.org/abs/2407.08693

    2025 arXiv

  9. [17]

    Clark, S

    J. Clark, S. Mirchandani, D. Sadigh, and S. Belkhale. Action-free reasoning for policy generalization,

  10. [18]

    Q. Zhao, Y . Lu, M. J. Kim, Z. Fu, Z. Zhang, Y . Wu, Z. Li, Q. Ma, S. Han, C. Finn, A. Handa, M.-Y . Liu, D. Xiang, G. Wetzstein, and T.-Y . Lin. Cot-vla: Visual chain-of-thought reasoning for vision-language-action models, 2025. URLhttps://arxiv.org/abs/2503.22020

    2025 arXiv

  11. [19]

    Zhang, K

    Z. Zhang, K. Zheng, Z. Chen, J. Jang, Y . Li, S. Han, C. Wang, M. Ding, D. Fox, and H. Y ao. Grape: Generalizing robot policy via preference alignment, 2025. URL https://arxiv.org/abs/2411.19309

    2025 arXiv

  12. [20]

    Zhang, Y

    B. Zhang, Y . Zhang, J. Ji, Y . Lei, J. Dai, Y . Chen, and Y . Y ang. Safevla: Towards safety alignment of vision-language-action model via safe reinforcement learning, 2025. URL https://arxiv.org/abs/2503.03480

    2025 arXiv

  13. [21]

    D. Li, J. Ren, Y . Wang, X. Wen, P . Li, L. Xu, K. Zhan, Z. Xia, P . Jia, X. Lang, N. Xu, and H. Zhao. Finetuning generative trajectory model with reinforcement learning from human feedback, 2025. URL https://arxiv.org/abs/2503.10434

    2025

  14. [22]

    Brown, J

    B. Brown, J. Juravsky, R. Ehrlich, R. Clark, Q. V . Le, C. R´e, and A. Mirhoseini. Large language monkeys: Scaling inference compute with repeated sampling.arXiv preprint arXiv:2407.21787, 2024

    2024 arXiv

  15. [23]

    Snell, J

    C. Snell, J. Lee, K. Xu, and A. Kumar. Scaling llm test-time compute optimally can be more effective than scaling model parameters.arXiv preprint arXiv:2408.03314, 2024

    2024 arXiv

  16. [24]

    Saad-Falcon, A

    J. Saad-Falcon, A. G. Lafuente, S. Natarajan, N. Maru, H. Todorov, E. Guha, E. K. Buchanan, M. Chen, N. Guha, C. R´e, et al. Archon: An architecture search framework for inference-time techniques. arXiv preprint arXiv:2409.15254, 2024

    2024 arXiv

  17. [25]

    L. Chen, J. Q. Davis, B. Hanin, P . Bailis, I. Stoica, M. Zaharia, and J. Zou. Are more llm calls all you need? towards scaling laws of compound inference systems.arXiv preprint arXiv:2403.02419, 2024

    2024 arXiv

  18. [26]

    Y . Song, G. Wang, S. Li, and B. Y . Lin. The good, the bad, and the greedy: Evaluation of llms should not ignore non-determinism.arXiv preprint arXiv:2407.10457, 2024

    2024 arXiv

  19. [27]

    D. Li, S. Cao, C. Cao, X. Li, S. Tan, K. Keutzer, J. Xing, J. E. Gonzalez, and I. Stoica. S*: Test time scaling for code generation, 2025. URLhttps://arxiv.org/abs/2502.14382

    2025 arXiv

  20. [28]

    G. Chen, M. Liao, C. Li, and K. Fan. Alphamath almost zero: process supervision without process. arXiv preprint arXiv:2405.03553, 2024

    2024 arXiv

  21. [29]

    Cobbe, V

    K. Cobbe, V . Kosaraju, M. Bavarian, M. Chen, H. Jun, L. Kaiser, M. Plappert, J. Tworek, J. Hilton, R. Nakano, et al. Training verifiers to solve math word problems.arXiv preprint arXiv:2110.14168, 2021

    2021 arXiv

  22. [30]

    Schaeffer, J

    R. Schaeffer, J. Kazdan, J. Hughes, J. Juravsky, S. Price, A. Lynch, E. Jones, R. Kirk, A. Mirho- seini, and S. Koyejo. How do large language monkeys get their power (laws)?, 2025. URL https://arxiv.org/abs/2502.17578. 11

    2025 arXiv

  23. [31]

    Sartor and N

    S. Sartor and N. Thompson. Neural scaling laws in robotics, 2025. URL h t t p s : //arxiv.org/abs/2405.14005

    2025 arXiv

  24. [32]

    F. Lin, Y . Hu, P . Sheng, C. Wen, J. Y ou, and Y . Gao. Data scaling laws in imitation learning for robotic manipulation, 2025. URLhttps://arxiv.org/abs/2410.18647

    2025 arXiv

  25. [33]

    Q. Li, Y . Liang, Z. Wang, L. Luo, X. Chen, M. Liao, F. Wei, Y . Deng, S. Xu, Y . Zhang, X. Wang, B. Liu, J. Fu, J. Bao, D. Chen, Y . Shi, J. Y ang, and B. Guo. Cogact: A foundational vision-language-action model for synergizing cognition and action in robotic manipulation, 20...

    2024 arXiv

  26. [34]

    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

  27. [35]

    D. Qu, H. Song, Q. Chen, Y . Y ao, X. Y e, Y . Ding, Z. Wang, J. Gu, B. Zhao, D. Wang, and X. Li. Spatialvla: Exploring spatial representations for visual-language-action model, 2025. URL https://arxiv.org/abs/2501.15830

    2025 arXiv

  28. [36]

    Ouyang, J

    L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. L. Wainwright, P . Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Ray, J. Schulman, J. Hilton, F. Kelton, L. Miller, M. Simens, A. Askell, P . Welinder, P . Christiano, J. Leike, and R. Lowe. Training language models to follow instructio...

    2022 arXiv

  29. [37]

    Touvron, L

    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

  30. [38]

    H. Liu, C. Li, Q. Wu, and Y . J. Lee. Visual instruction tuning, 2023

    2023

  31. [39]

    Z. Sun, S. Shen, S. Cao, H. Liu, C. Li, Y . Shen, C. Gan, L.-Y . Gui, Y .-X. Wang, Y . Y ang, K. Keutzer, and T. Darrell. Aligning large multimodal models with factually augmented rlhf, 2023. URL https://arxiv.org/abs/2309.14525

    2023 arXiv

  32. [40]

    Radford, J

    A. Radford, J. W . Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P . Mishkin, J. Clark, G. Krueger, and I. Sutskever. Learning transferable visual models from natural language supervision, 2021. URLhttps://arxiv.org/abs/2103.00020

    2021 arXiv

  33. [41]

    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

  34. [42]

    Nakamoto, O

    M. Nakamoto, O. Mees, A. Kumar, and S. Levine. Steering your generalists: Improving robotic foundation models via value guidance, 2025. URLhttps://arxiv.org/abs/2410.13816

    2025 arXiv

  35. [43]

    Zheng, L

    L. Zheng, L. Yin, Z. Xie, C. Sun, J. Huang, C. H. Y u, S. Cao, C. Kozyrakis, I. Stoica, J. E. Gonzalez, C. Barrett, and Y . Sheng. Sglang: Efficient execution of structured language model programs, 2024. URL https://arxiv.org/abs/2312.07104

    2024 arXiv

  36. [44]

    Jiang, A

    Y . Jiang, A. Gupta, Z. Zhang, G. Wang, Y . Dou, Y . Chen, L. Fei-Fei, A. Anandkumar, Y . Zhu, and L. Fan. Vima: General robot manipulation with multimodal prompts. arXiv preprint arXiv:2210.03094, 2(3):6, 2022

    2022 arXiv

  37. [45]

    H.-S. Fang, H. Fang, Z. Tang, J. Liu, J. Wang, H. Zhu, and C. Lu. Rh20t: A robotic dataset for learning diverse skills in one-shot. InRSS 2023 W orkshop on Learning for T ask and Motion Planning, 2023

    2023

  38. [46]

    V aswani, N

    A. V aswani, 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.Advances in neural information processing systems, 30(2017), 2017. 12

    2017

  39. [47]

    Sinha, A

    R. Sinha, A. Sharma, S. Banerjee, T. Lew, R. Luo, S. M. Richards, Y . Sun, E. Schmerling, and M. Pavone. A system-level view on out-of-distribution data in robotics, 2023. URL https://arxiv.org/abs/2212.14020

    2023 arXiv

  40. [48]

    E. Zhu, M. Levy, M. Gwilliam, and A. Shrivastava. Nerf-aug: Data augmentation for robotics with neural radiance fields, 2025. URLhttps://arxiv.org/abs/2411.02482

    2025 arXiv

  41. [49]

    Mitrano and D

    P . Mitrano and D. Berenson. Data augmentation for manipulation, 2022. URL https://arxiv.org/abs/2205.02886

    2022 arXiv

  42. [50]

    Liang, W

    J. Liang, W. Huang, F. Xia, P . Xu, K. Hausman, B. Ichter, P . Florence, and A. Zeng. Code as policies: Language model programs for embodied control, 2023. URL https://arxiv.org/abs/2209.07753

    2023 arXiv

  43. [51]

    Y . Mu, Q. Zhang, M. Hu, W. Wang, M. Ding, J. Jin, B. Wang, J. Dai, Y . Qiao, and P . Luo. Embodiedgpt: Vision-language pre-training via embodied chain of thought, 2023. URL https://arxiv.org/abs/2305.15021

    2023 arXiv

  44. [52]

    L. X. Shi, B. Ichter, M. Equi, L. Ke, K. Pertsch, Q. Vuong, J. Tanner, A. Walling, H. Wang, N. Fusai, A. Li-Bell, D. Driess, L. Groom, S. Levine, and C. Finn. Hi robot: Open- ended instruction following with hierarchical vision-language-action models, 2025. URL https://arxiv.o...

    2025 arXiv

  45. [53]

    Y . Wu, R. Tian, G. Swamy, and A. Bajcsy. From foresight to forethought: Vlm-in-the-loop policy steering via latent alignment, 2025. URLhttps://arxiv.org/abs/2502.01828

    2025 arXiv

  46. [54]

    Y . Wang, L. Wang, Y . Du, B. Sundaralingam, X. Y ang, Y .-W. Chao, C. Perez-D’Arpino, D. Fox, and J. Shah. Inference-time policy steering through human interactions, 2025. URL https://arxiv.org/abs/2411.16627

    2025 arXiv

  47. [55]

    Ehrlich, B

    R. Ehrlich, B. Brown, J. Juravsky, R. Clark, C. R´e, and A. Mirhoseini. Codemonkeys: Scaling test-time compute for software engineering, 2025. URLhttps://arxiv.org/abs/2501.14723

    2025 arXiv

  48. [56]

    B. Liu, Y . Zhu, C. Gao, Y . Feng, Q. Liu, Y . Zhu, and P . Stone. Libero: Benchmarking knowledge transfer for lifelong robot learning.arXiv preprint arXiv:2306.03310, 2023. 13 A Evaluation T asks As described in Section 5.3 and illustrated in Figure 4, we evaluate RoboMonkey,...

    2023 arXiv

  49. [2025]

    URL https://arxiv.org/abs/2502.03729

    arXiv

Pith tools

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