Pith. sign in

REVIEW 3 major objections 4 minor 33 references

XS-VLA: Coupling Coarse-grained Spatial Distillation with Latent Flow Matching for Lightweight Robotic Control

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A 0.25-billion-parameter VLA reaches 90% average success on LIBERO, beating 2.2B and 7B models.

desk verdict A serious lightweight VLA with a promising architecture, but the reported evidence doesn't back the headline SOTA claim: 10 episodes per task, no error bars, and internally inconsistent speedup numbers. read the letter →

arxiv 2607.04171 v2 pith:2CAGGPG4 submitted 2026-07-05 cs.RO cs.LG

Iok Tong Lei , Qingchen Xie , Wei Huang , Ying Jie Yap , Yujie Zhang , Qianzhi Li , Xiaolong Liu , Zhidong Deng This is my paper · ORCID
classification cs.ROcs.LG
keywords vision-language-actionmodelslightweightroboticcontrolspatialdistillationflowmatchingconditionalvariationalautoencoderLIBERObenchmarkknowledgemanipulationpolicies
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 very small vision-language-action model can become a strong robotic manipulation policy if its native 'spatial blindness' is fixed by distilling coarse spatial descriptions from a much larger teacher, and its action head is built as a latent flow matching policy rather than a deterministic controller. The authors report 90.0% average success on LIBERO with 0.25B parameters, above the 88.8% of a 2.25B vanilla model and 76.5% of a 7B OpenVLA, with the largest gain on long-horizon tasks. If correct, this means parameter count is not the dominant factor for precise control; targeted spatial supervision and generative action modeling can make tiny models deployable on edge hardware.

What carries the argument

The carrying mechanism is the combination of coarse-grained spatial distillation and latent flow matching. Spatial distillation converts the teacher's keypoint predictions into nine discrete directional labels (top, bottom-left, center-right, etc.) and trains the student to autoregressively predict those labels, turning localization into a compact classification-like task. Latent flow matching then replaces a deterministic policy head: a CVAE encoder extracts a low-dimensional intent variable z from proprioception and action chunks (with KL warmup to avoid posterior collapse), and a flow matching transformer regresses the velocity field u = ε - A_t under Huber loss, with interleaved cross-at

What would settle it

Take the spatially distilled backbone and test it on real-robot scene images or on LIBERO scenes with object arrangements and backgrounds not in the training set; if its predicted directional captions are no better than the untuned backbone, or if training the same pipeline with randomly shuffled captions retains the success-rate gains, the claim that spatial distillation causes the improvement is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a two-stage recipe—(1) fine-tuning a 0.25B SmolVLM2 backbone on coarse-grained spatial captions synthesized by Qwen3-VL-4B for simulated LIBERO images, then (2) coupling that backbone with a Conditional Variational Autoencoder that encodes action 'style' and a flow matching head that regresses an action velocity field—yields state-of-the-art success among sub-0.5B VLAs. XS-VLA averages 90.0% across LIBERO suites, including 86.0% on LIBERO-Long versus 63.0% for the vanilla 0.25B baseline, and runs 3.2x faster than the previous lightweight flow matching policy in task execution.

Load-bearing premise

The claim rests on the assumption that Qwen3-VL-4B's automatically generated keypoint and direction captions on simulated LIBERO images are accurate enough that fine-tuning the 0.25B student on them produces genuine spatial understanding, and because those same simulated images are used for evaluation, the LIBERO gains could reflect benchmark-specific overfitting rather than transferable spatial grounding.

Editorial extensions

If this is right

  • Sub-0.5B VLAs can be serious candidates for real-time, closed-loop manipulation on edge devices, not just compact classifiers.
  • Spatial grounding for small models can be injected synthetically from a large teacher instead of requiring large-scale human spatial annotation.
  • Generative latent flow policies handle multimodal human demonstrations better than deterministic behavior cloning, reducing mean-seeking jitter.
  • The 86% LIBERO-Long result suggests long-horizon stability improves when actions are sampled as a distribution rather than a point estimate.
  • The 3.2x speedup in execution time indicates that flow matching with a small backbone can lower the compute barrier for VLA deployment.

Reading between the lines

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

  • If the strength of spatial distillation comes from the coarse directional tokenization, the same nine-region scheme could be applied to real-robot data and other simulators, potentially transferring beyond LIBERO.
  • Because the teacher's keypoint and caption accuracy is never measured, a direct test would be to evaluate the student's spatial-description accuracy on held-out real images; the paper's claims do not yet establish that genuine spatial understanding was learned.
  • The real-robot carrot transfer result rests on only 10 trials per model; treating 7.5 vs 7.0 vs 6.5 as a decisive edge is plausible but not statistically strong.
  • The truncation to 16 layers and the straight-line nature of flow-matching trajectories may explain part of the speedup independently of the spatial distillation, so the two contributions could be disentangled by runtime profiling.
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

3 major / 4 minor

Summary. The paper proposes XS-VLA, a 0.25B vision-language-action model for robotic manipulation. The method has two stages: (1) coarse-grained spatial distillation, in which Qwen3-VL-4B annotates LIBERO images with two grasp keypoints and nine-region directional descriptions, and the student SmolVLM2-0.25B is fine-tuned on these annotations; (2) a latent flow matching policy, in which a CVAE latent variable conditions a flow-matching action expert built on the spatially fine-tuned backbone. The main empirical claim is that XS-VLA achieves an average success rate of 90.0% on LIBERO, surpassing Vanilla SmolVLA-2.25B (88.8%) and OpenVLA-7B (76.5%), and that it is faster than earlier lightweight flow-matching policies. The paper also reports ablations isolating the spatial-distillation and latent-flow-matching components, plus a 10-trial real-robot carrot-transfer experiment.

Significance. If the headline results were established, the paper would make a useful contribution: it shows a plausible recipe for injecting spatial inductive bias into a sub-0.5B VLA and a compact generative action head. The strengths are the automatically generated spatial-description pipeline, the combination of CVAE intent modeling with flow matching, the explicit KL warmup to avoid posterior collapse, the ablations that separate the two contributions, and the real-robot deployment on three hardware platforms. The paper is not circular in the derivation-level sense: the success rates are empirical measurements, and the teacher and base models are external. However, the central comparative claims rest on very small evaluation budgets and unmatched baseline protocols, so the evidence as currently reported does not support the headline superiority over SmolVLA-2.25B. The efficiency claims also contain internal inconsistencies. The idea is defensible and likely worth further development, but the quantitative support needs to be substantially strengthened or reframed.

major comments (3)
  1. [§V, Table I] The paper reports success rates averaged over only 10 episodes per task, with no variance, error bars, or multiple seeds. With 10 tasks per LIBERO suite, this is n=100 per suite. For the Long suite, XS-VLA 86% vs. SmolVLA-2.25B 77% has a standard error of roughly sqrt(0.86×0.14/100 + 0.77×0.23/100) ≈ 5.5 points, so the +9-point gap is not significant at the 5% level; the aggregate 90.0% vs. 88.8% (+1.2 points) is entirely within noise. In addition, the SmolVLA, OpenVLA, Dita, FPC-VLA, and other baseline numbers are taken from external publications with no demonstration that episode counts, seeds, environment versions, or success criteria are matched. The load-bearing claim that a 0.25B model outperforms SmolVLA-2.25B is therefore not statistically supported. The authors should report confidence intervals, run multiple seeds with matched evaluation, or substantially scale the evaluation b
  2. [§V-B, Table I] The efficiency claims are internally inconsistent and misleading. Table I reports Time/Epoch: XS-VLA (w/o backbone pre-trained) = 14 s, XS-VLA = 58 s, SmolVLA-PD = 186 s. Section V-B first says XS-VLA reduces execution time to just 14 s, compared to 186 s for SmolVLA-PD, i.e., a 13.3× speedup; the abstract and later text claim a 3.2× speedup. The 14 s figure belongs to the variant without spatial pretraining, whereas the final XS-VLA takes 58 s, giving 186/58 ≈ 3.2×. Thus the more striking 13.3× number is obtained from a different model. Since efficiency is a central contribution, the paper must clearly state which variant achieves which time, report inference latency on identical hardware with the same evaluation protocol, and reconcile the 3.2× and 13.3× statements.
  3. [§III-A, §IV-A, §V-B] The spatial distillation data are generated from simulated LIBERO images (D_raw) and the student is then evaluated on the same LIBERO benchmark. The paper never measures the teacher's keypoint/caption accuracy or the student's spatial-description accuracy. Consequently, the large ablation gain credited to spatial distillation (e.g., +18.0 points on LIBERO-Long from SmolVLA-PD relative to Vanilla SmolVLA-0.25B) could reflect adaptation to LIBERO's visual appearance rather than transferable spatial grounding. The 10-trial real-robot carrot experiment is a useful pilot but too small to resolve this concern. The authors should report teacher annotation quality, student spatial-description accuracy on held-out images, and at least one cross-domain or out-of-distribution evaluation (e.g., a second simulator or a different visual setup) to support the claim that the distilled spatial knowledge
minor comments (4)
  1. [§V-C, Table II] Table II has a typo: 'Trails' should be 'Trials'. Also, reporting only the total score over 10 trials makes it hard to judge variance; a per-trial breakdown or success counts would be more informative for a 1-point difference (7.5 vs. 7.0 vs. 6.5).
  2. [Abstract and Contributions] The claim '82.8%↑90%' is written in an abbreviated, unclear way. It should be explicit that 82.8% is Vanilla SmolVLA-0.25B and 90.0% is XS-VLA, and whether these are the same evaluation protocol.
  3. [References] Reference [2] is cited as 'Qwen3-VL' but the title and arXiv ID given are for the original 'Qwen-VL' paper. Please cite the correct Qwen3-VL release. Also 'OpernARM' in §V-C should be 'OpenARM'.
  4. [§V-B] The ablation wording 'By introducing the Latent Flow Matching action expert head, XS-VLA further improves the overall average success rate from 82.8% to 87.4%' is confusing: the 87.4% row is labeled 'XS-VLA (w/o backbone pre-trained)', not full XS-VLA. Please clarify that this is the LFM-only ablation, not the final model, and update the surrounding text accordingly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity identified: the central success-rate results are empirical measurements, not outputs derived by construction from fitted inputs or self-citations.

full rationale

I walked the paper's derivation chain looking for definitional circularity, fitted inputs called predictions, load-bearing self-citations, imported uniqueness theorems, ansatz smuggling via citation, or renaming known results. The central claim is an empirical success-rate comparison on LIBERO (Section V-A). The reported numbers are outcomes of training with explicit losses (Eq. 1 for spatial fine-tuning; L_FM and L_KL for the flow-matching policy) and evaluation over 10 episodes per task. No equation defines the reported success rates in terms of the teacher labels or the spatial-description targets; a policy could in principle train on the same spatial descriptions and still fail on LIBERO. The teacher model Qwen3-VL-4B and the SmolVLA baselines are external works with no author overlap, and no 'uniqueness theorem' or prior self-authored result is invoked to force the architectural choice. The spatial-distillation data being generated from simulated LIBERO images and evaluation also occurring on LIBERO is a distributional and potential-overfitting concern, but it is not circularity: the method's outputs are not identical to its inputs by construction, nor is any prediction algebraically forced by a fitted parameter. The paper contains internal inconsistencies in the ablation text (e.g., stating that latent flow matching improves from 82.8% to 87.4% when comparing against SmolVLA-PD at 88.8%), but these are correctness/consistency issues, not circularity under the criteria. Accordingly, the circularity score is 0.

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

The central claim rests on the accuracy of the teacher-generated spatial labels, the transferability of coarse discretized spatial supervision, and several hand-set training hyperparameters (KL warmup, truncation depth, loss weights, training steps). The flow-matching and interleaved-attention foundations are taken as standard background from the cited literature. No new physical or theoretical entities are introduced; the CVAE latent variable z is standard machinery. The largest concern is the domain assumption that distillation on LIBERO images yields a general spatial prior rather than simulator-specific adaptation.

free parameters (5)
  • Directional region vocabulary (9 discrete regions) = top, top-left, top-right, center, center-left, center-right, bottom, bottom-left, bottom-right
    Choice of discretization for the spatial distillation target; the granularity of spatial supervision is set by hand and determines what the student learns (Section III-A/Fig. 2).
  • KL warmup schedule length = 10,000 training steps
    Linear warmup of λ_KL is chosen to prevent posterior collapse; this training hyperparameter directly affects the CVAE/flow-matching balance (Section III-C.3).
  • Backbone truncation depth = 16 layers
    SmolVLM2-0.25B is truncated to 16 layers for speed; this capacity-versus-latency choice affects both success rate and the reported speedup (Section III-B).
  • Training steps = 160,000 steps
    XS-VLA is trained for 160,000 steps on Lerobot-Libero, following SmolVLA; the stopping point influences final success rates (Section IV-A).
  • Flow matching objective hyperparameters = λ_FM, λ_KL, Huber δ, Beta(α,β) for τ, chunk size k (values not reported)
    Loss weights, Huber threshold, flow-time sampling distribution, and action chunk size are chosen by hand and not numerically specified, yet they define the policy objective (Section III-C).
assumptions (5)
  • domain assumption The teacher Qwen3-VL-4B generates accurate grasp keypoints and spatial descriptions
    Invoked in Section III-A; no accuracy metric for the teacher's keypoints or captions is reported, so the entire distillation signal rests on unverified teacher correctness.
  • domain assumption Fine-tuning on coarse-grained directional regions transfers to continuous action control
    Invoked in Sections III-B and V; the paper assumes 9-region classification supervision induces useful spatial features in the visual encoder, but no probe of the student's spatial accuracy is given.
  • standard math Flow matching objective with Huber loss and Beta-sampled τ is a valid generative model for action chunks
    Background from cited literature [4,22]; the paper applies flow matching without re-deriving its guarantees.
  • domain assumption SmolVLA's interleaved attention architecture is an effective VLA backbone
    Borrowed from SmolVLA [5] without independent justification of the design choices (Section III-C.2).
  • domain assumption LIBERO benchmark domain is representative enough that improvements transfer to physical robots
    Real-world evaluation covers one bimanual task with 10 trials (Section V-C), while the headline SOTA claim is on the same simulator used to generate the distillation data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of XS-VLA: Coupling Coarse-grained Spatial Distillation with Latent Flow Matching for Lightweight Robotic Control." pith.science (2026). https://pith.science/paper/2CAGGPG4

@misc{pith2026260704171,
  author       = {Pith},
  title        = {Pith review of: XS-VLA: Coupling Coarse-grained Spatial Distillation with Latent Flow Matching for Lightweight Robotic Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2CAGGPG4}},
  note         = {Machine review of arXiv:2607.04171}
}
read the original abstract

Large Vision-Language Models (LVLMs) have shown strong multimodal understanding and spatial grounding, but their computational cost limits real-time robotic control. In contrast, lightweight models are suitable for edge deployment but often suffer from "spatial blindness", namely weak native spatial prediction ability. Training Vision-Language-Action (VLA) models on mixed human demonstrations can also degrade policy performance due to highly diverse behaviors. To address these limitations, we propose XS-VLA, a two-stage framework for efficient and spatially grounded robotic manipulation. First, we distill spatial semantic knowledge from Qwen3-VL-4B into the SmolVLM2-0.25B backbone by fine-tuning on curated coarse-grained spatial descriptions, turning the lightweight model into a spatially grounded engine. Second, we use this enhanced backbone to condition a Latent Flow Matching policy. Unlike deterministic controllers, our policy combines a Conditional Variational Autoencoder (CVAE) with Flow Matching dynamics to model complex multimodal action distributions. On the LIBERO benchmark, XS-VLA achieves state-of-the-art performance among models with fewer than 0.5B parameters. It improves average success rates by up to 7.2 percent, including a 23 percent gain on LIBERO-Long, over the SmolVLA 0.25B baseline, and outperforms the larger 2.2B vanilla SmolVLA. Ablations show that spatial tuning and generative latent flow control substantially improve lightweight VLA performance, delivering a 3.2 times speedup in mission execution over the previous lightweight flow matching policy.

Figures

Figures reproduced from arXiv: 2607.04171 by the authors.

Figure 1
Figure 1. The two-stage spatial description generation pipeline. In Stage 1, Qwen3-VL-4B predicts two grasp keypoints (in integer coordinates) from LIBERO [26] scene images using instruction-based prompts. These keypoints are visualized on the original images. In Stage 2, the annotated images, together with additional prompts, are used by Qwen3-VL-4B to generate spatial text descriptions. Middle Middle Middle Middle Left side… view at source ↗
Figure 2
Figure 2. Keypoint annotations and Position description. Each sets shows the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. XS-VLA Pipeline Architecture. The base Vision-Language Model is established by spatially pre-training SmolVLM2 (0.25B) and truncating it to 16 layers. In the post-training phase, a Transformer encoder extracts a latent style variable, z, to condition the Latent Flow Matching policy. During inference, the policy synthesizes continuous actions conditioned on multi-view images, physical state, language instructions, an… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Demonstration of our XS-VLA on a Libero-Long task. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: shows video frames for the XLerobot dual-arms manipulation task [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: OpenARM: Tidy the table. VI. Discussion and Limitations Our evaluation demonstrates that lightweight 0.25B VLMs can achieve competitive performance on spatial manipulation tasks without the massive computational cost of 7B+ parame￾ter models. However, limitations persi…
Figure 7
Figure 7. Figure 7: PiPER: Pick the orange duck. VII. Conclusion In this work, we introduce XS-VLA, a highly efficient 0.25B Vision-Language-Action model developed through a novel two-stage framework. By first distilling coarse-grained spatial grounding capabilities from Qwen3-VL-4B into …

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 8 linked inside Pith

  1. [1]

    GPT-4v(ision) system card,

    OpenAI, “GPT-4v(ision) system card,” OpenAI, Tech. Rep., 2023, accessed: 2024-03-05. [Online]. Available: https://cdn.openai.com/ papers/GPTV System Card.pdf

    2023

  2. [2]

    Qwen-VL: A versatile vision-language model for understanding, localization, text reading, and beyond,

    J. Bai, S. Bai, S. Yang, S. Wang, S. Tan, P. Wang, J. Lin, C. Zhou, and J. Zhou, “Qwen-VL: A versatile vision-language model for understanding, localization, text reading, and beyond,” arXiv preprint arXiv:2308.12966, 2023. [Online]. Available: https: //arxiv.org/abs/2308.12966

    arXiv 2023

  3. [3]

    Learning structured output representa- tion using deep conditional generative models,

    K. Sohn, H. Lee, and X. Yan, “Learning structured output representa- tion using deep conditional generative models,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 28, 2015

    2015

  4. [4]

    π 0: A Vision-Language-Action Flow Model for General Robot Control,

    K. Black and et al., “π 0: A Vision-Language-Action Flow Model for General Robot Control,”arXiv preprint arXiv:2410.24164, 2024

    arXiv 2024

  5. [5]

    SmolVLA: A vision-language-action model for affordable and efficient robotics,

    M. Shukor, D. Aubakirova, F. Capuano, P. Kooijmans, S. Palma, A. Zouitine, M. Aractingi, C. Pascal, M. Russi, A. Marafioti et al., “SmolVLA: A vision-language-action model for affordable and efficient robotics,”arXiv preprint arXiv:2506.01844, 2025. [Online]. Available: https://arxiv.org/abs/2506.01844

    arXiv 2025

  6. [6]

    Flamingo: A visual language model for few- shot learning,

    J.-B. Alayrac and et al., “Flamingo: A visual language model for few- shot learning,” inAdvances in Neural Information Processing Systems (NeurIPS), 2022

    2022

  7. [7]

    Improved baselines with visual instruction tuning,

    H. Liu, C. Li, Y. Li, and Y. J. Lee, “Improved baselines with visual instruction tuning,” inProc. NeurIPS Workshop Instruction Tuning and Instruction Following, 2023. [Online]. Available: https://openreview.net/forum?id=yx3Hkx5ved

    2023

  8. [8]

    Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models,

    J. Li, D. Li, S. Savarese, and S. Hoi, “Blip-2: Bootstrapping language- image pre-training with frozen image encoders and large language models,” inInternational conference on machine learning (ICML). PMLR, 2023, pp. 19 730–19 742

    2023

Show all 33 references
  1. [9]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inProc. Int. Conf. Mach. Learn. (ICML), ser. Proc. Mach. Learn. Res. (PMLR), 2021, pp. 8748–8763

    2021

  2. [10]

    (2024) SmolVLM2: Bringing video understanding to every device

    Hugging Face. (2024) SmolVLM2: Bringing video understanding to every device. Hugging Face Blog. Accessed: Mar. 5, 2026. [Online]. Available: https://huggingface.co/blog/smolvlm2

    2024

  3. [11]

    MobileVLM: A fast, strong and open vision language assistant for mobile devices,

    X. Chu, L. Qiao, X. Lin, S. Xu, Y. Yang, Y. Hu, F. Wei, X. Zhang, B. Zhang, X. Weiet al., “MobileVLM: A fast, strong and open vision language assistant for mobile devices,” arXiv preprint arXiv:2312.16886, 2023. [Online]. Available: https: //arxiv.org/abs/2312.16886

    2023 arXiv

  4. [12]

    TinyLLaV A: A framework of small-scale large multimodal models,

    B. Zhou, Y. Hu, X. Weng, J. Jia, J. Luo, X. Liu, J. Wu, and L. Huang, “TinyLLaV A: A framework of small-scale large multimodal models,”arXiv preprint arXiv:2402.14289, 2024. [Online]. Available: https://arxiv.org/abs/2402.14289

    2024 arXiv

  5. [13]

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

    B. Zitkovich and et al., “RT-2: Vision-language-action models transfer web knowledge to robotic control,” inConference on Robot Learning (CoRL), 2023

    2023

  6. [14]

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

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

    2024 arXiv

  7. [16]

    ThinkAct: Vision-language-action reasoning via reinforced visual latent planning,

    C.-P. Huang, Y.-H. Wu, M.-H. Chen, Y.-C. F. Wang, and F.-E. Yang, “ThinkAct: Vision-language-action reasoning via reinforced visual latent planning,” inProc. 39th Conf. Neural Inf. Process. Syst. (NeurIPS), 2025. [Online]. Available: https://openreview.net/forum? id=72UR53jN7T

    2025

  8. [17]

    TraceVLA: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies,

    R. Zheng, Y. Liang, S. Huang, J. Gao, H. Daum ´e III, A. Kolobov, F. Huang, and J. Yang, “TraceVLA: Visual trace prompting enhances spatial-temporal awareness for generalist robotic policies,” inProc. 13th Int. Conf. Learn. Representations (ICLR), 2025. [Online]. Available: ht...

    2025

  9. [18]

    SpatialVLM: Endowing vision-language models with spatial reasoning capabilities,

    B. Chen and et al., “SpatialVLM: Endowing vision-language models with spatial reasoning capabilities,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024

    2024

  10. [19]

    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. Wanget al., “SpatialVLA: Exploring spatial representations for visual-language-action model,”arXiv preprint arXiv:2501.15830, 2025. [Online]. Available: https://arxiv.org/abs/ 2501.15830

    2025 arXiv

  11. [20]

    FPC-VLA: A vision- language-action framework with a supervisor for failure prediction and correction,

    Y. Yang, Z. Duan, T. Xie, F. Cao, P. Shen, P. Song, C. Zhao, P. Jin, G. Sun, S. Xu, Y. You, and J. Liu, “FPC-VLA: A vision- language-action framework with a supervisor for failure prediction and correction,”Expert Syst. Appl., p. 131742, 2026

    2026

  12. [21]

    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, vol. 44, no. 10-11, pp. 1684–1704, 2025

    2025

  13. [22]

    Flow matching for generative modeling,

    Y. Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” inProc. 11th Int. Conf. Learn. Representations (ICLR), 2023. [Online]. Available: https://openreview.net/forum?id=PqvMRDCJT9t

    2023

  14. [23]

    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,” inProc. Robotics: Science and Systems (RSS), Daegu, Republic of Korea, Jul. 2023

    2023

  15. [24]

    VL2Lite: Task-specific knowledge distil- lation from large vision-language models to lightweight networks,

    J. Jang, C. Ma, and B. Lee, “VL2Lite: Task-specific knowledge distil- lation from large vision-language models to lightweight networks,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), 2025, pp. 30 073–30 083

    2025

  16. [25]

    Knowledge distillation and dataset distillation of large language models: Emerging trends, challenges, and future directions,

    L. Fang, X. Yu, J. Cai, Y. Chen, S. Wu, Z. Liuet al., “Knowledge distillation and dataset distillation of large language models: Emerging trends, challenges, and future directions,”Artif. Intell. Rev., vol. 59, no. 1, p. 17, 2025

    2025

  17. [26]

    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,” inAd- vances in Neural Information Processing Systems (NeurIPS), vol. 36, 2023, pp. 44 776–44 791

    2023

  18. [27]

    Smollm-blazingly fast and remarkably powerful,

    L. B. Allal, A. Lozhkov, E. Bakouch, L. von Werra, and T. Wolf, “Smollm-blazingly fast and remarkably powerful,”Hugging Face Blog, 2024, accessed: 2026-03-05. [Online]. Available: https: //huggingface.co/blog/smollm

    2024

  19. [28]

    BERT: Pre- training of deep bidirectional transformers for language understand- ing,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- training of deep bidirectional transformers for language understand- ing,” inProc. Conf. North Amer. Chapter Assoc. Comput. Linguistics: Hum. Lang. Technol. (NAACL-HLT), 2019, pp. 4171–4186

    2019

  20. [29]

    MuJoCo: A physics engine for model-based control,

    E. Todorov, T. Erez, and Y. Tassa, “MuJoCo: A physics engine for model-based control,” inProc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS). IEEE, 2012, pp. 5026–5033

    2012

  21. [30]

    Dita: Scaling diffusion transformer for generalist vision-language-action policy,

    Z. Hou, T. Zhang, Y. Xiong, H. Duan, H. Pu, R. Tong, C. Zhao, X. Zhu, Y. Qiao, J. Daiet al., “Dita: Scaling diffusion transformer for generalist vision-language-action policy,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 7686–7697

    2025

  22. [31]

    G. Wang, Z. Lu, Y. Huang, and Y. Liu. (2025) XLeRobot: A practical low-cost household dual-arm mobile robot design for general manipulation. GitHub Repository. Accessed: Mar. 5, 2026. [Online]. Available: https://github.com/Vector-Wangel/XLeRobot

    2025

  23. [32]

    Openarm,

    “Openarm,” accessed: 2026-03-05. [Online]. Available: https:// openarm.dev/

    2026

  24. [33]

    Piper arm 6 dof,

    “Piper arm 6 dof,” accessed: 2026-03-05. [Online]. Available: https://global.agilex.ai/products/piper

    2026

  25. [2024]

    Available: https://arxiv.org/abs/2405.12213

    [Online]. Available: https://arxiv.org/abs/2405.12213

    arXiv

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.